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
1912.11450	1	1912	2019-12-24T17:37:07	Room-temperature Single Photon Emitters in Cubic Boron Nitride Nanocrystals	[ "cond-mat.mes-hall" ]	Color centers in wide bandgap semiconductors are attracting broad attention as platforms for quantum technologies relying on room-temperature single-photon emission (SPE), and for nanoscale metrology applications building on the centers' response to electric and magnetic fields. Here, we demonstrate room-temperature SPE from defects in cubic boron nitride (cBN) nanocrystals, which we unambiguously assign to the cubic phase using spectrally resolved Raman imaging. These isolated spots show photoluminescence (PL) spectra with zero-phonon lines (ZPLs) within the visible region (496-700 nm) when subject to sub-bandgap laser excitation. Second-order autocorrelation of the emitted photons reveals antibunching with $g^{2}$ ~ 0.2 and a decay constant of 2.75 ns that is further confirmed through fluorescence lifetime measurements. The results presented herein prove the existence of optically addressable isolated quantum emitters originating from defects in cBN, making this material an interesting platform for opto-electronic devices and quantum applications.	cond-mat.mes-hall	cond-mat	Room-temperature Single Photon Emitters in Cubic Boron Nitride Nanocrystals Gabriel I. López-Morales1,2,3, Aziza Almanakly1,4, Sitakanta Satapathy1, Nicholas V. Proscia1, Harishankar Jayakumar1, Valery N. Khabashesku5,6, Pulickel M. Ajayan6, Carlos A. Meriles1,7, 1Department of Physics, City College of the City University of New York, New York, NY 10031 Vinod M. Menon1,7 USA 2Department of Chemistry, Lehman College of the City University of New York, Bronx, NY 10468, USA 3Ph. D. Program in Chemistry, The Graduate Center of the City University of New York, New York, NY 10016, USA 4Department of Electrical Engineering, The Cooper Union for the Advancement of Science and 5Baker Hughes, Center for Technology Innovation, Houston, TX 77040, USA 6Department of Materials Science and NanoEngineering, Rice University, Houston, TX 77005, Art, New York, NY 10003, USA 7Ph. D. Program in Physics, The Graduate Center of the City University of New York, New York, USA *Email: [email protected], [email protected] NY 10016, USA Color centers in wide bandgap semiconductors are attracting broad attention as platforms for quantum technologies relying on room-temperature single-photon emission (SPE), and for nanoscale metrology applications building on the centers' response to electric and magnetic fields. Here, we demonstrate room-temperature SPE from defects in cubic boron nitride (cBN) nanocrystals, which we unambiguously assign to the cubic phase using spectrally resolved Raman isolated spots show photoluminescence (PL) spectra with zero-phonon lines (ZPLs) within the visible region (496 -- 700 nm) when subject laser excitation. Second-order autocorrelation of the emitted photons reveals antibunching with ɡ2(0) ~ 0.2, and a decay constant of 2.75 ns that is further confirmed through fluorescence lifetime measurements. The results presented herein prove the existence of optically addressable isolated quantum emitters originating from defects in cBN, making this material an interesting platform for opto-electronic devices and quantum applications. to sub-bandgap imaging. These With an atomic structure equivalent to that of diamond, refractive index close to 2.1, an optical bandgap exceeding 10 eV, and the possibility of n- and p-type doping, cubic boron nitride (cBN) is attracting interest as a host for atomic defects with optical features across a wide region of the electromagnetic spectrum.1-3 A combination of photoluminescence (PL) and electron paramagnetic resonance (EPR) studies have shed light on the optical and electronic properties of a range of point defects, from vacancies to heavy-metals and rare-earth elements.4-9 Computational methods have also been used to study native and extrinsic defects in cBN, predicting the existence of an oxygen-related defects having isoelectronic properties to the well- known NV -- center in diamond.10-12 Thus far, however, experimental signatures from single atomic defects in this material have proven elusive. Mainly, this has been due to its hardness along with its synthesis requirements, which pose limitations on defect engineering methods; as well as the energy separation between the defect levels within the bandgap, often large enough to make them inaccessible through optical excitation. Cathodoluminescence (CL), which provides the necessary energies to reach deep intra-band defect levels in cBN has been successfully used in 1 this context. However, electron excitation usually results in synchronized emission from nearby defect centers that can be excited by the same incoming electron, posing difficulties for single- defect studies and antibunching measurements which are key to unambiguously identify individual defects.13, 14 Here, we report on room-temperature single-photon emission (SPE) originating from atomic defects in cBN, some of which show spectral features like those previously reported for color centers in single-crystal cBN.1, 4 We overcome some of the aforementioned limitations by using nanocrystalline cBN flakes, which have a high density of near-surface atomic defects accessible through optical laser excitation. By employing confocal Raman spectroscopy, we show that the optical regions of interest are indeed dominated by the cubic phase of BN, while small hexagonal boron nitride (hBN) quantities arising as a synthesis byproduct are found to phase- segregate in nm-sized regions. cBN nanocrystals (dispersed in ethanol) obtained from PlasmaChem GmbH are drop-cast onto a Si/SiO2 substrate, kept at 60 oC under ambient conditions. The as-prepared samples show a variable distribution of cBN nanocrystal sizes (Fig 1 (a)) that strongly depends upon drop-cast parameters, such as solution concentration and dispersion. Individual nanocrystals range in size (80 -- 450 nm), but are usually found to be around 160 nm. These high-quality cBN crystals can be obtained either by high-pressure/high-temperature (HPHT) synthesis from boron- and nitrogen- containing precursors, or by HPHT treatments of hBN in the presence of catalysts.15 As a result, Fig 1. Confocal Raman spectroscopy of cBN nanocrystals. (a) Laser reflection imaging reveals large area coverage and conglomeration of cBN nanoflakes. (b) Raman spectrum collected by averaging over the squared region in (a) shows cBN LO and TO phonon modes strongly dominate over the E2g mode of hBN. (c-e) Confocal Raman spectroscopy for select phonon energies (shaded areas in (b)) show that the cubic phase of BN is dominant and uniform across the sample, while hBN regions seem to be localized possibly due to phase segregation. Scale bar = 2 μm. 2 1000112512501375 E2g(hBN) TO(cBN) LO(cBN) Intensity (arb. units)Raman shift (cm-1)1.801.00Intensity (Kcounts/s)1050 cm-11254 cm-11360 cm-1Intensity (Mcounts/s)(a)(b)(c)(d)(e) there are usually small fractions of hBN present in cBN samples, especially in polycrystalline, powder-like and cBN nanocrystals.15, 16 Thus, an important part of this study consist in characterizing possible phase-segregation of hBN within the nanocrystalline cBN, while ensuring that the optically-probed domains correspond to cBN-dominated regions. Raman studies on BN show that cBN and hBN have spectrally separated vibrational signatures that can be used to identify and differentiate the two phases.17-19 Specifically, both cBN and hBN have an optical phonon mode centered at Γ point which, due to ionic contributions, splits into longitudinal and transverse optical phonon modes (LO and TO) only in the case of cBN. These cBN modes are usually centered around 1304 and 1055 cm-1, respectively, while the E2g- symmetric phonon mode of hBN is centered around 1365 cm-1.17-19 Performing both optical and Raman confocal spectroscopy thus allows us to probe the phase profile of the spectral regions, while providing proof as to whether the collected fluorescence originates from defects in cBN nanocrystals or in residual hBN nanoflakes. We perform Raman spectroscopy via a confocal Raman microscope (alpha300 R; WITec GmbH), equipped with a 100x (1.0NA) microscope objective and an SHG Nd:YAG laser (532 nm). Collected photons are sent to a spectrometer with a CCD-camera (1024128 pixel, Peltier cooled to 65C) rejecting reflected and elastically scattered photons by using an edge filter. The nominal spectral resolution is found to be ~5 cm−1 per CCD pixel. Due to low scattering cross sections, Raman signatures from both BN phases tend to be close to the noise level, especially when dealing with nanocrystalline samples. A well-resolved Raman spectrum from the different BN phases is thus obtained by averaging over specific scanned areas (dashed square in Fig 1 (a)) while integrating for 0.5 s and illuminating at ~12.0 mW laser power (Fig 1 (b)). The main Raman spectral features extracted from this and several other locations include two dominant peaks centered at ~1050 and ~1254 cm-1, which arise from TO and LO modes in cBN, respectively, and a third, lower intensity peak centered at ~1360 cm-1 arising from the E2g phonon mode in hBN. For confocal Raman imaging, the 50 µm core of a multimode fiber serves as the pinhole, which leads to a focal depth of ~1 µm, while the diffraction-limited spot results in a lateral resolution of Fig 2. Steady-state confocal spectroscopy of cBN nanoflakes. (a) Confocal imaging of a representative cBN region reveals an array of isolated bright spots. White circles denote spots that showed sharp and spectrally separated ZPLs. (b-d) Representative optical spectra for different bright spots identified with (orange, green and blue) arrows in (a). While (b, c) show spectral features comparable to those reported for defects in cBN (black arrows), (d) reveals a phonon replica matching the E2g phonon energy in hBN demonstrating that both cBN and hBN defects are present across the sample. Scale bar denotes 2 μm. 3 (a)0.90Intensity (Mcounts/s)(b)500550600650 Fluorescence (arb. units)Wavelength (nm)(c)500550600650 Fluorescence (arb. units)Wavelength (nm)(d)500550600650 Fluorescence (arb. units)Wavelength (nm) ~1 -- 2 m. Scanning over the same region while only collecting one of the specific phonon energies in Fig 1 (b) reveals an almost uniform distribution for both TO and LO cBN modes (Fig 1 (c, d) respectively), while showing a non-uniform distribution for the E2g hBN mode (Fig 1 (e)). The small amount of residual hBN is thus observed to be localized at specific spots, mainly resulting from phase segregation. Further, the ratio, 𝐼(E2g) 𝐼(TO+LO) = 0.0615 ± 0.0053 averaged over multiple regions, confirms that these are indeed cBN-dominated nanocrystalline domains. We observe variable linewidths and small energy shifts in all the BN phonon modes throughout different sample locations, possibly due to size effects, local strain and relaxation of both phases to nanocrystalline states.17, 19 However, all Raman measurements demonstrate that the cubic phase of BN dominates across the sample, while small hBN quantities are found to segregate within nm- sized regions. To elucidate the optoelectronic properties of defect-related color centers in cBN nanocrystals, room-temperature confocal spectroscopy is performed via a home-built confocal microscope, equipped with a 50x (0.83 NA) objective and a 460 nm continuous-wave (CW) laser (L462P1400MM, ThorLabs). The collected fluorescence is filtered from the excitation source by using a 500 nm long-pass filter (LPF) and sent through a fiber beam splitter allowing for real-time spectral analysis and time-resolved measurements. The excitation source, with spot-size of about ~ 1 μm, is kept at an average power below 200 μW to avoid/suppress blinking and bleaching of single emitters. Despite the inhomogeneity in the spatial distribution of flakes owing to the drop- casting technique, a collection of isolated bright spots across large sample areas is systematically observed upon sub-bandgap excitation (Fig 2 (a)), indicating the presence of defect centers within the cBN nanoflakes independent of the sample preparation process. Most of the isolated bright spots inside the cBN nanoflakes usually give broad fluorescence features. Usually, two main cases are observed: (1) real-time spectral analysis shows substantial blinking of different zero- phonon lines (ZPLs) associated with different centers inside the same diffraction-limited spot (defect ensembles); (2) individual ZPLs arise on top of a broadband background, which is present in the nearby nanoflake regions. The majority of the ZPLs arising from single defects, however, are found to share similar spectral response, as can be observed by comparing Fig 2 (b) and Fig 2 (c), some of which resemble features previously reported from color centers in cBN, e.g., under UV-excitation.4 Other ZPLs tend to resemble those observed in hBN defects (Fig 2 (d)), e.g., with phonon replicas matching the E2g phonon mode energy (~165 meV). Thus, it is noted that both cBN and hBN defects are found across the sample. However, the majority of the observed emitters are unlikely to originate from hBN defects due to: (1) their uniform distribution across the sample and the high extent of hBN localization (Fig 1); (2) mismatch between the obtained spectral features and those reported for hBN defects; (3) the similarities between most of the observed emitters and those previously reported in cBN. To further study the dynamics and photon correlation of defects in cBN nanoflakes, we turn to room-temperature time-resolved and correlation spectroscopy. CW photon correlation measurements are performed via Hanbury-Brown-Twiss (HBT) interferometry (Picoquant, Picoharp 300) using two avalanche photodiode detectors (APD) and 460 nm laser excitation. On the other hand, lifetime measurements are performed using the 488 nm line of an 80 MHz, 500- fs pulsed tunable laser (Toptica FemtoFiber TVIS). Results from these studies are summarized in Fig 3. The obtained spectrum from a single defect center in cBN is shown in Fig 3 (a), with 4 features indicated by arrows matching the shape and distribution of those previously reported.4 In our study, however, these features appeared much more separated (~130 meV), while preserving equidistance from the main peak. Given the similarities between these shifts and the phonon energies for cBN (~130 and ~160 meV), they are likely to be phonon-mediated replicas of the Fig 3. Time-resolved fluorescence spectroscopy of a single defect center in cBN. (a) Optical spectrum of an isolated color center under 460 nm continuous-wave (CW) excitation. Blue arrows indicate features that match those previously observed for the GB-1 color center under UV-excitation. Inset shows a zoomed-in confocal scan of this emitter (dotted circle). (b) Time-trace of the collected fluorescence showing fast transitions to a dark-state, characteristic of most of single-emitters. (c) Second-order autocorrelation function, showing photon antibunching at zero-time delay with ɡ2(0) = 0.258. (d) Fluorescence lifetime under 488 nm pulsed excitation. Solid line represents a biexponential decay fit to the experimental data (orange circles). Scale bar denotes 1 μm. main peak, hinting at a unique electronic defect structure with electron-phonon coupling that may be characteristic of defects in cBN. The fluorescence time-trace of this single defect center under continuous excitation showed substantial blinking (Fig 3 (b)), a signature characteristic of isolated quantum emitters, likely due to different charge-states of the excited defect. Photon correlation of the collected fluorescence through the second-order autocorrelation function, ɡ2(τ), gives rise to a clear dip at zero-time delay (Fig 3 (c)), with ɡ2(0) = 0.229 proving the quantum nature of the emitter. A slight bunching component with time constant of ~30 ns is observed, which points at a possible metastable and/or charge state that may correlate to the blinking behavior observed during fluorescence time traces.5, 13, 20 Finally, lifetime measurements reveal an exponential decay (Fig 3 (d)) from which a radiative time constant of 2.753 ns is obtained. This exponential decay is further confirmed directly from CW ɡ2(τ) measurements and falls well within the range of fluorescence lifetimes previously reported from point defects in cBN.1, 5 Exposure to external magnetic fields of up to 0.1 T and/or circularly polarized excitation yield no discernible change in 5 -60-30030600.00.51.01.52.0g2(0) = 0.229 g2()Time delay (ns)500550600650 Fluorescence (arb. units)Wavelength (nm)~156 meV0246810121 = 2.753 ns Fluorescence (arb. units)Time (ns) Exp. Fit0255075100 Fluorescence (arb. units)Time (s)Dark state(a)(b)(c)(d) the fluorescence, hence suggesting magneto-optically-active defects -- including the ONVB center, predicted to have a response similar to that of the NV- in diamond -- may not be widespread. We warn, however, most emitters tend to fade upon prolonged exposure to the laser beam, hence indicating additional work -- e.g., exploiting time-modulated magnetic fields -- will be required to expose weak magneto-optical contrast. In summary, we have demonstrated room-temperature SPE from defects in cBN nanoflakes. Confocal images show isolated bright spots across the nanoflakes, while Raman spectroscopy confirms that these are indeed cBN-dominated nanocrystalline domains with the ratio, I(E2g)/I(TO+LO) ~ 0.06. The amount of residual hBN is localized at different spots (phase segregation) and further confirms the cBN-origin of the observed color centers. Fluorescence spectroscopy reveals numerous ZPLs in the visible region, mainly covering 496 -- 700 nm. Interestingly, the optical signatures of some of the studied defects resemble those previously reported4 for color centers in cBN, with a clear shift (~130 meV) on the separation between the main peak and its low-intensity features. These features hint at an electronic structure unique to these color centers, while suggesting electron-phonon coupling similar to that of defects in cBN, e.g., the GB-1 centers.1, 4 These results open interesting possibilities in defect engineering based on cBN, combining properties similar to its hexagonal counterpart while providing a wider bandgap, ease of being ion-bombarded and a new group of diverse defect-related color centers. Acknowledgements: V.M.M. and C.A.M. acknowledge support from the National Science Foundation through grants NSF-1906096 and NSF-1726573. C.A.M. also acknowledges support from Research Corporation for Science Advancement through a FRED Award; all authors acknowledge partial funding and access to the facilities and research infrastructure of the NSF CREST IDEALS, grant number NSF-HRD-1547830. VM acknowledges useful discussions with Prof. Igor Aharonovich. References: 1. A. Tararan, S. di Sabatino, M. Gatti, T. Taniguchi, K. Watanabe, L. Reining, L. H. G. Tizei, M. Kociak, A. Zobelli, Phys. Rev. B 98, 094106 (2018). 2. L. Vel, G. Demazeau, J. Etourneau, Materials Science and Engineering B 10 (2),149 (1991). 3. G. Cappellini, G. Satta, Phys. Rev. B 64, 035104 (2001). 4. E. M. Shishonok, J. Appl. Spectroscopy 74 (2), (2007). 5. R. Buividas, I. Aharonovich, G. Seniutinas, X. W. Wang, L. Rapp, A. V. Rode, T. Taniguchi, S. Juodkazis, Optics Letters 40 (24),5711(2015). 6. E. M. Shishonok, S. V. Leonchik, J. W. Steeds, Phys. Stat. Sol. (b) 244 (6), 2172 (2007). 7. E. M. Shishonok, S. V. Leonchik, J. W. Steeds, D. Wolverson, Diamond & Related Materials 16, 1602 (2007). 8. S.V. Nistora, M. Stefana, D. Schoemaker, G. Dinca, Solid State Comms. 115, 39 (2000). 9. S. V. Nistor, M. Stefan, D. Ghica, E. Goovaerts, Appl. Magn. Reson. 39, 87 (2010). 10. W. Orellana, H. Chacham, Phys. Rev. B 63, 125205 (2017). 11. P. Piquini, R. Mota, T. M. Schmidt, A. Fazzio, Phys. Rev. B 56 (7), 3556 (1997). 12. T. A. Abtew, W. Gao, X. Gao, Y. Y. Sun, S. B. Zhang, P. Zhang, Phys. Rev. Lett. 113, 136401 (2014). 13. T. T. Tran, C. Zachreson, A. M. Berhane, K. Bray, R. G. Sandstrom, L. H. Li, T. Taniguchi, K. Watanabe, I. Aharonovich, M. Toth, Phys. Rev. Applied 5, 034005 (2016). 6 14. S. Meuret, L. H. G. Tizei, T. Cazimajou, R. Bourrellier, H. C. Chang, F. Treussart, M. Kociak, Phys. Rev. Lett. 114, 197401 (2015). 15. S. Zuo-Peng, D. Yong-Hui, J. Xiao-Rui, Y. Da-Peng, Y. Xu-Xin, G. Xi-Liang, Z. Tie-Chen, Chinese Phys. Lett. 23, 2285 (2006). 16. Y. Chong, K. Ma, K. Leung, C. Chan, Q. Ye, I. Bello, W. Zhang, S. Lee, Chemical Vapor Deposition 12, 33 (2006). 17. S. Reich, A. C. Ferrari, R. Arenal, A. Loiseau, I. Bello, J. Robertson, Phys. Rev. B 71, 205201 (2005). 18. T. Werninghaus, J. Hahn, F. Richter, D. R. T. Zahn, Appl. Phys. Lett. 70, 958 (1997). 19. O. Kutsay, C. Yan, Y. M. Chong, Q. Ye, I. Bello, W. J. Zhang, J. A. Zapien, Z. F. Zhou, Y. K. Li, V. Garashchenko, A. G. Gontar, N. V. Novikov, S. T. Lee, Diamond & Related Materials 19, 968 (2010). 20. Z. Shotan, H. Jayakumar, C. R. Considine, M. Mackoit, H. Fedder, J. Wrachtrup, A. Alkauskas, M. W. Doherty, V. M. Menon, C. A. Meriles, ACS Photonics 3, 2490 (2016). 7
1810.02662	2	1810	2019-01-21T09:44:08	Spin-Dependent Electron Transmission Model for Chiral Molecules in Mesoscopic Devices	[ "cond-mat.mes-hall" ]	Various device-based experiments have indicated that electron transfer in certain chiral molecules may be spin-dependent, a phenomenon known as the Chiral Induced Spin Selectivity (CISS) effect. However, due to the complexity of these devices and a lack of theoretical understanding, it is not always clear to what extent the chiral character of the molecules actually contributes to the magnetic-field-dependent signals in these experiments. To address this issue, we report here an electron transmission model that evaluates the role of the CISS effect in two-terminal and multi-terminal linear-regime electron transport experiments. Our model reveals that for the CISS effect, the chirality-dependent spin transmission is accompanied by a spin-flip electron reflection process. Furthermore, we show that more than two terminals are required in order to probe the CISS effect in the linear regime. In addition, we propose two types of multi-terminal nonlocal transport measurements that can distinguish the CISS effect from other magnetic-field-dependent signals. Our model provides an effective tool to review and design CISS-related transport experiments, and to enlighten the mechanism of the CISS effect itself.	cond-mat.mes-hall	cond-mat	Spin-Dependent Electron Transmission Model for Chiral Molecules in Mesoscopic Devices Zernike Institute for Advanced Materials, University of Groningen, NL-9747AG Groningen, The Netherlands Xu Yang,∗ Caspar H. van der Wal, and Bart J. van Wees (Dated: January 23, 2019) Various device-based experiments have indicated that electron transfer in certain chiral molecules may be spin-dependent, a phenomenon known as the Chiral Induced Spin Selectivity (CISS) ef- fect. However, due to the complexity of these devices and a lack of theoretical understanding, it is not always clear to what extent the chiral character of the molecules actually contributes to the magnetic-field-dependent signals in these experiments. To address this issue, we report here an electron transmission model that evaluates the role of the CISS effect in two-terminal and multi- terminal linear-regime electron transport experiments. Our model reveals that for the CISS effect, the chirality-dependent spin transmission is accompanied by a spin-flip electron reflection process. Furthermore, we show that more than two terminals are required in order to probe the CISS effect in the linear regime. In addition, we propose two types of multi-terminal nonlocal transport mea- surements that can distinguish the CISS effect from other magnetic-field-dependent signals. Our model provides an effective tool to review and design CISS-related transport experiments, and to enlighten the mechanism of the CISS effect itself. I. INTRODUCTION Developments in the semiconductor industry have al- lowed integrated circuits to rapidly shrink in size, reach- ing the limit of conventional silicon-based electronics. One idea to go beyond this limit is to use the spin degree- of-freedom of electrons to store and process information (spintronics).1 A spintronic device usually contains two important components: a spin injector and a spin de- tector, through which electrical or optical signals and spin signals can be interconverted. Conventionally, this conversion is done with bulky solid-state materials, but the recently discovered Chiral Induced Spin Selectivity (CISS) effect suggests that certain chiral molecules or their assemblies are capable of generating spin signals as well. This effect describes that electrons acquire a spin polarization while being transmitted through certain chi- ral (helical) molecules. Notably, experimental observa- tions of the CISS effect suggest its existence, but com- plete theoretical insight in its origin is still lacking.2,3 The CISS effect is thus not only relevant for spintronic applications, but also fundamentally interesting. systems ranging (helical) The CISS effect has been experimentally re- ported in chiral from such as dsDNA4,5 large biological units to small molecules such as helicenes.6,7 Typically, these ex- periments can be categorized into either electron photoemission experiments4,7 -- 13 or magnetotransport measurements.5,6,14 -- 20 The latter, in particular, are usu- ally based on solid-state devices and are of great im- portance to the goal of realizing chiral-molecule-based spintronics. Important to realize, in such devices, the CISS-related signals may often be overshadowed by other spurious signals that arise from magnetic components of the devices. Therefore, it is essential to understand the exact role of chiral molecules in these devices and to distinguish between the CISS-related signals and other magnetic-field-dependent signals. However, this has not been addressed, and an effective tool to perform such analyses is still missing. We provide here a model that is based on the Landauer-Buttiker-type of analysis of linear-regime elec- tron transmission and reflection. Unlike other theoretical works,21 -- 29 our model is derived from symmetry theo- rems that hold for electrical conduction in general and does not require any assumptions about the CISS ef- fect on a molecular level. With this model, we quan- titatively demonstrate how the CISS effect leads to spin injection and detection in linear-regime devices, and ana- lyze whether typical two-terminal and four-terminal mea- surements are capable of detecting the CISS effect in the linear regime. II. MODEL We consider a solid-state device as a linear-regime cir- cuit segment whose constituents are described by the fol- lowing set of rules: • A contact (pictured as a wavy line segment perpen- dicular to the current flow, see e.g. Fig. 1) is de- scribed as an electron reservoir with a well-defined chemical potential µ, which determines the energy of the electrons that leave the reservoir. A reser- voir absorbs all incoming electrons regardless of its energy or spin; • A node (pictured as a circle, see later figures for four-terminal geometries) is a circuit constituent where chemical potentials for charge and spin are defined. It is described by two chemical potentials µ→ and µ←, one for each spin species with the ar- rows indicating the spin orientations. At a node a spin accumulation µs is defined (µs = µ→ − µ←). Inside a node the momentum of electrons is ran- domized, while the spin is preserved. The func- 9 1 0 2 n a J 1 2 ] l l a h - s e m . t a m - d n o c [ 2 v 2 6 6 2 0 . 0 1 8 1 : v i X r a tion and importance of the node will be further addressed in the discussion section; • A CISS molecule (pictured as a helix, color-coded and labeled for its chirality, see e.g. Fig. 1), a fer- romagnet (a filled square, see e.g. Fig. 1), and a non-magnetic barrier (a shaded rectangle, see e.g. Fig. 2) are viewed as two-terminal circuit con- stituents with energy-conserving electron transmis- sions and reflections. Each of them is described by a set of (possibly spin-dependent) transmission and reflection probabilities; • The above constituents are connected to each other via transport channels (pictured as line segments along the current flow, see e.g. Fig. 1), in which both the momentum and the spin of electrons are preserved. Before proceeding with introducing the model, we would like to highlight the important role of dephasing in the generation of the CISS effect. In a fully phase- coherent two-terminal electron transport system, time- reversal symmetry prohibits the production of spin po- larization by a charge current.30 Consequently, a CISS molecule requires the presence of dephasing in order to exhibit a CISS-type spin-polarizing behavior. The necessity of dephasing has already been addressed by other theoretical works.31 -- 33 Here we emphasize that the Landauer-Buttiker type of analysis, on which our model is based, does not require phase coherence.34,35 More- over, dephasing can be naturally provided by inelastic processes such as electron-phonon interactions under ex- perimental conditions. Therefore, it is reasonable to as- sume that a CISS molecule is able to generate a spin polarization in a linear-regime circuit segment, and our discussions focus on whether this spin polarization can be detected as a charge signal. In the following part of this article, we first derive a key transport property of CISS molecules and then in- troduce a matrix formalism to quantitatively describe linear-regime transport devices. Later, in the discussion section, we provide analyses for a few experimental cir- cuit geometries. A. Reciprocity and spin-flip reflection by chiral molecules In order to characterize the CISS effect without having to understand it on a molecular level, we look at universal rules that apply to any conductor in the linear regime, namely the law of charge conservation and the reciprocity theorem. The reciprocity theorem states that for a multi- terminal circuit segment in the linear regime, the mea- sured conductance remains invariant when an exchange of voltage and current contacts is accompanied by a re- versal of magnetic field H and magnetization M (of all magnetic components).36,37 Mathematically we write Gij,mn(H, M ) = Gmn,ij(−H,−M ), 2 (1) where Gij,mn is the four-terminal conductance measured using current contacts i and j and voltage contacts m and n. In two-terminal measurements, this theorem reduces to Gij(H, M ) = Gij(−H,−M ), (2) meaning that the two-terminal conductance remains con- stant under magnetic field and magnetization reversal. This theorem emphasizes the universal symmetry inde- pendent of the microscopic nature of the transport be- tween electrical contacts. It is valid for any linear-regime circuit segment regardless of the number of contacts, or the presence of inelastic scattering events.37 FIG. 1: A two-terminal circuit segment with a P -type CISS molecule and a ferromagnet between contacts 1 and 2 (with chemical potentials µ1 and µ2). The notion P -type represents the chirality of the molecule and indicates that it allows higher transmission for spins parallel to the electron momentum. (The opposite chirality allows higher transmission for spins anti-parallel to the electron momentum, and is denoted as AP -type.) The ferromagnet (FM ) is assumed to allow higher transmission of spins parallel to its magnetization direction, which can be controlled to be either parallel or anti-parallel to the electron transport direction. By applying the reciprocity theorem to a circuit seg- ment containing CISS molecules, one can derive a spe- cial transport property of these molecules. For exam- ple, in the two-terminal circuit segment shown in Fig. 1, the reciprocity theorem requires that the two-terminal conductance remains unchanged when the magnetiza- tion direction of the ferromagnet is reversed. Since the two-terminal conductance is proportional to the trans- mission probability between the two contacts (Landauer- Buttiker),35 this requirement translates to T21(⇒) = T21(⇐), (3) where T21 describes the transmission probability of elec- trons injected from contact 1 to reach contact 2, and ⇒ and ⇐ indicate the magnetization directions of the ferromagnet. This requirement gives rise to a necessary spin-flip process associated with the CISS molecule, as described below. For ease of illustration, we assume an ideal case where both the ferromagnet and the CISS molecule allow a 100% transmission of the favored spin and a 100% re- flection of the other (the general validity of the conclu- sions is addressed in Appendix A). We consider electron FM 𝜇1 𝜇2 P FM 𝜇1 𝜇2 P transport from contact 1 to contact 2 (see Fig. 1) and compare the two transmission probabilities T21(⇒) and T21(⇐). For T21(⇒), the P -type CISS molecule (favors spin parallel to electron momentum, see figure caption) allows the transmission of spin-right electrons, while it reflects spin-left electrons back to contact 1. At the same time, the ferromagnet is magnetized to also only allow the transmission of spin-right electrons. Therefore, all spin-right (and none of the spin-left) electrons can be transmitted to contact 2, giving T21(⇒) = 0.5. As for T21(⇐), while the P -type CISS molecule still allows the transmission of spin-right electrons, the ferromag- net no longer does. It reflects the spin-right electrons towards the CISS molecule with their momentum anti- parallel to their spin. As a result, these electrons are reflected by the CISS molecule and are confined between the CISS molecule and the ferromagnet. This situation gives T21(⇐) = 0, which is not consistent with Eqn. 3. In order to satisfy Eqn. 3, i.e. to have T21(⇐) = 0.5, a spin-flip process has to take place for the spin-right electrons, so that they can be transmitted to contact 2 through the ferromagnet. Such a process does not exist for the ideal and exactly aligned ferromagnet. Therefore, a spin-flip electron reflection process must exist for the CISS molecule. Further analysis (see Appendix A) shows that such a spin-flip reflection process completely meets the broader restrictions from Eqn. 2. In addition, the conclusion that a spin-flip reflection process must exist is valid for general cases where the ferromagnet and the CISS molecule are not ideal (see Appendix A). In these derivations, the only assumption regarding the CISS molecule is that it allows higher transmission of one spin than the other, which is a conceptual description of the CISS effect itself. Therefore, the spin-flip reflection process has to be regarded as an inherent property of the CISS effect in a linear-regime transport system, and this is guaranteed by the universal symmetry theorems of electrical conduction.36,37 B. Matrix formalism and barrier-CISS center-barrier (BCB) model for CISS molecules We use matrices to quantitatively describe the spin- dependent transmission and reflection probabilities of CISS molecules and other circuit constituents, as shown in Fig. 2. At the top of the figure, the general form of these matrices is introduced. Matrix element tαβ (or rαβ), where α and β is either left (←) or right (→), rep- resents the probability of a spin-α electron being trans- mitted (or reflected) as a spin-β electron, and α (cid:54)= β indicates a spin-flip process. Here 0 ≤ tαβ, rαβ ≤ 1, and the spin orientations are chosen to be either parallel or anti-parallel to the electron momentum in later discus- sions. Next, the transmission and reflection matrices of a non-magnetic barrier are given. These matrices are spin-independent and are fully determined by a trans- mission probability t (0 ≤ t ≤ 1), which depends on 3 FIG. 2: Transmission and reflection matrices (T and R) for a non-magnetic barrier (subscript B, here we use the term barrier, but it refers to any circuit constituent with spin-independent electron transmission and reflection), a ferromagnet (subscript FM ), and ideal P -type (superscript P ) and AP -type (superscript AP ) CISS molecules. For the CISS molecules the subscripts R (right) and L (left) denote the direction of the incoming electron flow, and the indicator 0 in the subscripts means these matrices are for an ideal cases where all the matrix elements are either 1 or 0. The matrices for AP -type molecules are derived from those for P -type molecules under the assumption that opposite chiral enantiomers are exact mirror images of each other, and therefore selects opposite spins with equal probability. Each matrix element represents the probability of a spin-dependent transmission or reflection, with the column/row position indicating the corresponding spin orientations before/after the transmission or reflection (see general form in the top row). the material and dimensions of the barrier. Here we use the term barrier, but it refers to any circuit constituent with spin-independent electron transmission and reflec- tion. In the third row, we show the transmission and reflection matrices of a ferromagnet. These matrices are spin-dependent, and are determined by the polarization PF M (0 < PF M ≤ 1) of the ferromagnet. Finally, for P - type and AP -type CISS molecules, we show here an ideal case where all the matrix elements are either 1 or 0. The non-zero off-diagonal terms in the reflection matrices rep- resent the characteristic spin-flip reflections. These ideal CISS molecules are later referred to as CISS centers, and will be generalized for more realistic situations. In accordance with the matrix formalism, we use col- umn vector µ = to describe chemical potentials, (cid:18)µ→ µ← (cid:19) (cid:18)I→ I← (cid:19) and column vector I = to describe currents, where each vector element describes the contribution from one spin component (indicated by arrow). A non-ideal CISS molecule with C2 symmetry (two- fold rotational symmetry with an axis perpendicular to ॻ=𝑡→→𝑡←→𝑡→←𝑡←← ℝ=𝑟→→𝑟←→𝑟→←𝑟←← Transmission Reflection Nonmagnetic barrier Ferromagnet P-type ideal CISS molecule ॻ𝐵=𝑡00𝑡 ℝ𝐵=1−𝑡001−𝑡 1+𝑃FM2001−𝑃FM2 1−𝑃FM2001+𝑃FM2 ॻ0,𝑅𝑃=1000 ℝ0,𝑅𝑃=0100 ॻ0,𝑅𝐴𝑃=ॻ0,𝐿𝑃 ℝ0,𝑅𝐴𝑃=ℝ0,𝐿𝑃 ॻ0,𝐿𝑃=0001 ℝ0,𝐿𝑃=0010 ॻ0,𝐿𝐴𝑃=ॻ0,𝑅𝑃 ℝ0,𝐿𝐴𝑃=ℝ0,𝑅𝑃 AP-type ideal CISS molecule AP General form ॻFM= ℝFM= P FM FIG. 3: A generalized Barrier-CISS Center-Barrier (BCB) model for P -type CISS molecules. The ideal, 100%-spin-selective CISS Center in the middle introduces the directional spin transmission in a CISS molecule, while the two identical non-magnetic barriers (with transmission probability t) contribute the non-ideal electron transmission and reflection behavior. The overall transmission and reflection matrices of the entire BCB module are fully determined by t and have all elements taking finite values between 0 and 1. the electron transport path) can be modeled as a lin- ear arrangement of two identical barriers sandwiching an ideal CISS center, as shown in Fig. 3 (only the P -type is shown). In this Barrier-CISS Center-Barrier (BCB) model we consider that all spin-dependent linear-regime transport properties of a CISS molecule exclusively orig- inate from an ideal CISS center inside the molecule, and the overall spin-dependency is limited by the multiple spin-independent transmissions and reflections at other parts (non-magnetic barriers) of the molecule. There- fore, the barrier transmission probability t (0 < t ≤ 1) fully determines the transmission and reflection matri- ces of the entire BCB molecule, and consequently deter- mines the spin-related properties of the molecule. The use of an identical barrier on each side of the CISS cen- ter is to address the C2 symmetry. However, we stress that not all CISS molecules have this symmetry, and the BCB model is still a simplified picture. The model can be further generalized by removing the restriction of the CISS center being ideal, and this case is discussed in Ap- pendix C. Despite being a simplified picture, the BCB model captures all qualitative behaviors of a non-ideal CISS molecule, and at the same time keeps quantitative analyses simple. Therefore, we further only discuss the case of BCB molecules, instead of the more generalized CISS molecules. III. DISCUSSIONS In this section, we use different approaches to sepa- rately analyze two-terminal and multi-terminal circuit geometries. For two-terminal geometries, we evaluate the conductance of the circuit segment by calculating the electron transmission probability T21 between the two contacts. In contrast, for multi-terminal geometries, we take a circuit-theory approach to evaluate the spin accu- mulation µs at the nodes. A major difference between the two approaches is the inclusion of nodes in multi-terminal geometries. In our description, a node is the only location where spin ac- cumulation can be defined. It can be experimentally realized with a diffusive electron transport channel seg- 4 ment that is much shorter (along the electron transport direction) than the spin-diffusion length λs of the chan- nel material. Due to its diffusive nature, a node emits electrons to all directions, so it can be considered as a source of electron back-scattering. Notably, adding a node to a near-ideal electron transport channel (with transmission probability close to 1) significantly alters its electron transmission probability. Nonetheless, this does not affect the validity of our approach because we only address non-ideal circuit segments where electron back-scattering (reflection) already exists due to other circuit constituents (CISS molecules, ferromagnets, or non-magnetic barriers). Note that even when we discuss the use of ideal CISS molecules or ideal ferromagnets, the entire circuit segment is non-ideal due to the reflection of the rejected spins. In the following discussion, we consider only the P - type BCB molecule, and we use expressions TP R,L and RP R,L to describe transmission and reflection matrices of the entire BCB module, where the subscripts consider electron flow directions. The derivations of these matri- ces can be found in Appendix B. A. Two-terminal geometries We discuss here two geometries that are relevant for two-terminal magnetoresistance measurements.38 FIG. 4: An FM-BCB geometry where a ferromagnet and a BCB molecule are connected in series in a two-terminal circuit segment. The magnetization reversal of the ferromagnet does not change the two-terminal conductance. The first is an FM-BCB geometry, as shown in Fig. 4. It simulates a common type of experiment where a layer of chiral molecules is sandwiched between a ferromag- netic layer and a normal metal contact. The other side of the ferromagnetic layer is also connected to a nor- mal metal contact (experimentally this may be a wire that connects the sample with the measurement instru- ment). Due to the spin-dependent transmission of the chiral molecules and the ferromagnet, one might expect a change of the two-terminal conductance once the mag- netization of the ferromagnet is reversed. However, this change is not allowed by the reciprocity theorem (Eqn. 2), which can be confirmed with our model, as explained be- low. In order to illustrate this, we calculate the elec- tron transmission probabilities between the two con- tacts for opposite ferromagnet magnetization directions, (⇐), where the arrows in- T F M−BCB dicate the magnetization directions. (⇒) and T F M−BCB 21 21 P FM FM 𝜇1 𝜇2 P FM 𝜇1 𝜇2 P FM 𝜇1 𝜇2 P For the magnetization direction to the right (⇒), we first derive the transmission and reflection matrices with the combined contribution from the ferromagnet and the BCB molecule + R 21 + (⇒) I + RF M (⇒) · RP TF M−BCB R · = TP (cid:32) (cid:17)2 (cid:16)RF M (⇒) · RP (cid:33) R ·(cid:16)I − RF M (⇒) · RP = RF M (⇒) + TF M (⇒) ·(cid:16)I − RP · TF M (⇒) RF M−BCB = TP + ··· (⇒) 11 R (cid:17)3 (cid:16)RF M (⇒) · RP (cid:17)−1 · TF M (⇒), R R (cid:17)−1 (4b) R · RF M (⇒) · RP (cid:18)1 0 (cid:19) R · TF M (⇒), T F M−BCB 21 (⇒) = RF M−BCB 11 (⇒) = 0 1 is the identity matrix. The addi- where the I = tion of the multiple reflection terms is due to the mul- tiple reflections between the ferromagnet and the BCB molecule. Next, we include the contribution from the contacts and derive the transmission and reflection prob- abilities accounting for both spins (cid:18)1/2 (cid:19) (cid:18)1/2 (cid:19) 1/2 1/2 (⇒) (⇒) (cid:16) (cid:16) 1, 1 1, 1 21 (cid:17)TF M−BCB (cid:17)RF M−BCB (cid:18)1/2 (cid:19) 11 1/2 , , (5a) (5b) (cid:16) (cid:17) where the column vector describes the normal- ized input current from contact 1 with equal spin-right and spin-left contributions, and the row vector is an operator that describes the absorption of both spins into contact 2 (calculates the sum of the two spin com- ponents). For the opposite magnetization direction (⇐), we change the magnetization-dependent terms in Eqn. 4 and Eqn. 5 accordingly. Detailed calculations (see Ap- pendix B) prove 1, 1 T F M−BCB 21 (⇒) ≡ T F M−BCB 21 (⇐) (6) for all BCB transmission probabilities t and all ferromag- net polarizations PF M . Therefore, it is not possible to detect any variation of two-terminal conductance in this geometry by switching the magnetization direction of the ferromagnet. In Appendix B we show that it is also not possible to detect any variation of two-terminal conduc- tance by reversing the current, and that the above con- clusions also hold for the more generalized CISS model (Appendix C). These conclusions also agree with earlier reports on general voltage-based detections of current- induced spin signals.39 5 (4a) FIG. 5: A Spin Valve geometry with a BCB molecule placed in between two ferromagnets. Unlike a conventional spin valve, here the magnetization reversal of one ferromagnet does not change the two-terminal conductance due to the presence of the BCB molecule. The second geometry, as shown in Fig. 5, contains two ferromagnets, and is similar to a spin valve. In a con- ventional spin valve (a non-magnetic barrier sandwiched between two ferromagnets), the magnetization reversal of one ferromagnet leads to a change of the two-terminal conductance38 (this does not violate the reciprocity theo- rem since switching one ferromagnet does not reverse all magnetizations of the entire circuit segment), whereas in the geometry shown in Fig. 5, this change does not happen due to the presence of spin-flip electron reflec- tions in the BCB molecule. (See Appendix B for more details.) As a result, this geometry is not able to quan- titatively measure the CISS effect. We emphasize that here the absence of the spin-valve behavior is unique for the BCB model, which contains an ideal CISS Center. In Appendix C we show that a further-generalized CISS model regains the spin-valve behavior. Nevertheless, one cannot experimentally distinguish whether the regained spin-valve behavior originates from the CISS molecule or a normal non-magnetic barrier, and therefore cannot draw any conclusion about the CISS effect. In general, it is not possible to measure the CISS effect in the linear regime using two-terminal experiments. B. Four-terminal geometries and experimental designs Four-terminal measurements allow one to completely separate spin-related signals from charge-related signals, and therefore allow the detection of spin accumulations created by the CISS effect.40 Here we analyze two geome- tries that are relevant for such measurements. In the first geometry, we use a node connected to BCB molecules to illustrate how spin injection and detection can oc- cur without using magnetic materials (Fig. 6). In the second geometry, we use two nodes to decouple a BCB molecule from electrical contacts and illustrate the spin- charge conversion property of the molecule (Fig. 7). In addition, we propose device designs that resemble these two geometries and discuss possible experimental out- comes (Fig. 8). Fig. 6a) shows a geometry where a node is connected to four contacts. Two of the contacts contain BCB P FM FM 𝜇1 𝜇2 P 6 (7) vector µnode = , and each of the four kinj = FIG. 6: a). A four-terminal geometry that includes a node. Two of the contacts contain BCB molecules, and the other two are coupled to the node via tunnel barriers (with transmission probability tB). The node is characterized by a spin-dependent chemical potential (cid:18)µnode→ (cid:19) µnode← contacts is characterized by a spin-independent chemical potential µi, with i = 1, 2, 3, 4. b). Calculated ratio between four-terminal and two-terminal resistances for this geometry, plotted as a function of t (transmission probability of the barriers in BCB molecules) for three tB (transmission probability of the barriers at the contacts) values. molecules, and the other two contain non-magnetic (tun- nel) barriers. We consider an experiment where contacts 1 and 2 are used for current injection and contacts 3 and 4 are used for voltage detection. In terms of spin injec- tion, we first assume that the voltage contacts 3 and 4 are weakly coupled to the node, and do not contribute to the spin accumulation in the node. This means that the chemical potentials of contacts 1 and 2 fully determine the spin-dependent chemical potential (column vector) of the node µnode = . We also assume µ2 = 0 for convenience since only the chemical potential difference between the two contacts is relevant. Under these as- (cid:18)µnode→ (cid:19) µnode← sumptions, the node receives electrons only from contact 1, but emits electrons to both contact 1 and 2. There- fore, the incoming current (column vector) into the node is G e , TP Rµ1 (cid:18)1 (cid:19) L ) + (1 − rB)I(cid:17) 1 Iin = (cid:16) and the outgoing current (column vector) from the node is Iout = G e (I − RP µnode, (8) (cid:16) where G = N e2/h is the N -channel, one-spin conduc- tance of the channels connecting the node to each of the contacts, and rB (0 ≤ rB ≤ 1) is the reflection probabil- ity of the tunnel barrier between the node and contact 2 (different from the barriers in BCB molecules). Due to the spin-preserving nature of the node, at steady state the incoming current is equal to the outgoing current (for both spin components), or Iin = Iout. From this relation we derive (1 + tB)I − RP L , 1 Rµ1 µnode = (9) where tB = 1 − rB is the transmission probability of the tunnel barrier. Next, we derive the spin accumulation in the node µs = µnode→ − µnode←= 1,−1 where a row vector is used as an operator to calculate the difference between the two spin chemical potentials, and µnode = kinjµ1, (10) (cid:16) (cid:17) 1,−1 (cid:17) (cid:16) (cid:17)−1TP (cid:18)1 (cid:19) (cid:16) 1,−1 (cid:17)(cid:16) (1+tB)I − RP with 0 < kinj ≤ 1 4 L , (cid:17)−1TP R (cid:18)1 (cid:19) 1 , (11) is the spin injection coefficient for these current contacts. This expression shows that the spin accumulation in the node depends linearly on the chemical potential differ- ence between the current contacts, and the coefficient kinj is determined by both the BCB molecule (with pa- rameter t) and the tunnel barrier connected to contact 2 (with parameter tB). With regard to spin detection, we discuss whether the established spin accumulation µs in the node can lead to a chemical potential difference (and thus a charge volt- age) between the weakly coupled voltage contacts 3 and 4. A contact cannot distinguish between the two spin components, therefore only the charge current (sum of (cid:16) (cid:17) both spins, calculated by applying an operator 1, 1 to a current column vector) is relevant. At steady state, there is no net charge current at any of the voltage con- tacts, (cid:16) (cid:17)(cid:32) I3 = G e 1, 1 (cid:18)1 (cid:19) 1 (1 − rB)µ3 − tBµnode = 0, (12a) (cid:33) 𝜇1 𝜇2 𝜇3 𝜇4 𝝁𝒏𝒐𝒅𝒆 P P a) b) 𝑡𝐵 𝑡𝐵=1 𝑡𝐵=0.1 𝑡𝐵=0.01 𝑡 (cid:16) I4 = G e 1, 1 (cid:17)(cid:32) (I−RP L )µ4 (cid:18)1 (cid:19) 1 −TP Rµnode (cid:33) = 0, (12b) which gives where µ4 − µ3 = kdetµs, (cid:18) 1−1 (cid:19) (cid:0)1, 1(cid:1) TP (cid:19) , (cid:18)1 (cid:0)1, 1(cid:1) TP R kdet = 1 2 L 1 with 0 < kdet ≤ 1 2 , (13) (14) is the spin detection coefficient for these voltage contacts. This expression shows that the chemical potential differ- ence between the two voltage contacts depends linearly on the spin accumulation in the node, and the coefficient kdet is exclusively determined by the BCB molecule (with parameter t). Combining Eqn. 10 and Eqn. 13 we obtain R4T R2T = µ4 − µ3 µ1 − µ2 = kinjkdet, (15) where R4T is the four-terminal resistance (measured us- ing contacts 3 and 4 as voltage contacts, while using contacts 1 and 2 as current contacts), and R2T is the two-terminal resistance (measured using contacts 1 and 2 as both voltage and current contacts). This ratio is determined by both the BCB molecule (with parameter t) and the tunnel barrier connected to contact 2 (with parameter tB), and can be experimentally measured to quantitatively characterize the CISS effect. As an example, for t = tB = 0.5, we have kinj ≈ 0.11, kdet ≈ 0.17, and R4T /R2T ≈ 0.02. In Fig. 6b) we plot R4T /R2T as a function of t for three different tB values. Similar plots for kinj and kdet are shown in Appendix B. The above results show that it is possible to inject and detect a spin accumulation in a node using only BCB molecules and non-magnetic (tunnel) barriers, and these processes can be quantitatively described by the injec- tion and detection coefficients. We stress that the signs of the injection and detection coefficients depend on the type (chirality) of the BCB molecule and the position of the molecule with respect to the contact. Switching the molecule from P -type to AP -type leads to a sign change of the injection or detection coefficient. The sign change also happens if the contact is connected to the opposite side of the BCB molecule. For example, in Fig. 6a), contacts 1 and 4 are both connected to the node via P - type BCB molecules, but contact 1 is on the left-hand side of a molecule, while contact 4 is on the right-hand side. Electrons emitted from these two contacts travel in opposite directions through the (same type of) BCB molecules before arriving at the node. As a result, us- ing contact 4 instead of contact 1 as a current contact 7 leads to a sign change of kinj. Similarly, using contact 1 instead of contact 4 as a voltage contact leads to a sign change of kdet. Experimentally, one can use three BCB contacts to observe this sign change: A fixed current con- tact (thus a fixed kinj) in combination with two voltage contacts that use the same type of BCB molecule but are placed on opposite sides of a node (thus opposite signs for kdet). The voltages measured by the two voltage con- tacts (with respect to a common reference contact) will differ by sign. This can be experimentally measured as a signature of the CISS effect. FIG. 7: A four-terminal geometry involving two nodes A and B, which are connected to each other via a BCB molecule. A spin accumulation difference between the two nodes results in a (charge) chemical potential difference between them, and vice versa. Fig. 7 shows a geometry where a BCB molecule is be- tween two nodes A and B, and is decoupled from the contacts. The nodes themselves are connected to con- tacts in a similar fashion as in the previous geometry. In node A, we consider a chemical potential vector µA and a spin accumulation µsA, which are fully determined by the current contacts 1 and 2. In node B, we con- sider weakly coupled voltage contacts 3 and 4, so that its chemical potential vector µB and its spin accumulation µsB, are fully determined by µA. At steady state, there is no net charge or spin current in node B, which leads to µB = (I − RP L )−1TP RµA. (16) Note that here the matrices only refer to the molecule be- tween the two nodes. For BCB molecules, this expression always gives µsB = 0, but for a more generalized CISS molecule (as described in Appendix C), this expression can give µsB (cid:54)= 0. This shows that a spin accumula- tion at one side of a CISS molecule can generate a spin accumulation at the other side of the molecule. Most importantly, for both the BCB model and the more gen- eralized model, Eqn. 16 predicts that a spin accumula- tion difference across a CISS molecule creates a charge voltage across the molecule, and vice versa (spin-charge conversion via a CISS molecule). Mathematically writ- ten, the expression always provides µnA (cid:54)= µnB when µsA (cid:54)= µsB, and µsA (cid:54)= µsB when µnA (cid:54)= µnB, where µnA (or µnB) is the average chemical potential of the two spin components in node A (or in node B). A more detailed description of this geometry can be found in Appendix C. Fig. 8 shows two types of nonlocal devices that resem- ble the two geometries introduced above. We realize the 𝜇′←=𝜇→ 𝜇3=𝜇′→ 𝜇1 𝜇2 𝜇3 𝜇4 A B P P P 𝝁𝑨 𝝁𝑩 8 (one on the left and the other on the right). A variation of this device is obtained by replacing con- tact 1 (together with the CISS molecules underneath it) with a ferromagnet, as shown in the inset of Fig. 8a). This variation allows one to control the sign of Rnl by controlling the magnetization direction of the ferromag- net, which should be aligned parallel or anti-parallel to the helical (chiral) axis of the CISS molecules (out-of- plane, as indicated by the arrows). The nonlocal resis- tance is therefore Rnl(⇑) = −Rnl(⇓) = PF M kdetRλe − d λs , (18) where the arrows indicate the magnetization directions, PF M is the polarization of the ferromagnet, and Rλ is the spin resistance of graphene (see Appendix B for more details). In this device, the reversal of the magnetization direction of the ferromagnet leads to a sign change of the nonlocal resistance. Under experimental conditions,42,43 this nonlocal resistance change ∆Rnl = Rnl(⇑) − Rnl(⇓) can reach tens of Ohms (Ω) and is easily detectable. The second type of device is depicted in Panel b). It is a variation of the geometry in Fig. 7, where contact 2 is replaced by a ferromagnet. In this device, instead of traveling through the CISS molecules, the electrons travel through the graphene channel underneath the molecules. It is assumed that due to the proximity of the CISS molecules, the electrons in graphene also experience a (weaker) CISS effect. Whether this assumption is valid remains to be proven. The nonlocal signals produced by this device are derived in Appendix B. FIG. 8: Nonlocal device designs with CISS molecules adsorbed on graphene. a). A device where electrons travel through CISS molecules. All contacts are non-magnetic and are numbered in agreement with Fig. 6a). A variation of this device can be achieved by replacing contact 1 with a ferromagnet, see inset. b). A device where electrons travel in proximity to CISS molecules. A ferromagnetic contact 2 is used for spin injection, but one can also use only non-magnetic contacts, as in Fig. 7. node function with graphene, chosen for its long spin life- time and long spin diffusion length.41 The first type, as shown in Fig. 8a), represents the ge- ometry in Fig. 6a), where spin injection and detection are both achieved using CISS molecules. A current Iinj is injected from contact 1 through CISS molecules into graphene, then driven out to a normal metal contact 2. This current induces a spin accumulation in the graphene layer underneath the current contacts, which then dif- fuses to the voltage contacts. The voltage contacts then pick up a charge voltage Vdet in a similar fashion as ex- plained in Fig. 6a). With this, the nonlocal resistance can be determined Rnl = Vdet/Iinj. Further, we can de- rive (see Appendix B for details) Rnl = −kinjkdetRinje − d λs , (17) where Rinj is the resistance measured between the cur- rent contacts 1 and 2, d is the distance between contacts 1 and 4, and λs is the spin diffusion length of graphene. It is assumed here that the spacing between the current contacts (1 and 2) and the spacing between the voltage contacts (3 and 4) are both much smaller than λs. The minus sign comes from the fact that the injection and the detection contacts are on the same side of the graphene channel (both on top), unlike the example in Fig. 6a) IV. CONCLUSION In summary, we demonstrated that a spin-flip electron reflection process is inherent to the chiral induced spin se- lectivity (CISS) effect in linear-regime electron transport. Furthermore, we developed a set of spin-dependent elec- tron transmission and reflection matrices and a general- ized Barrier-CISS Center-Barrier (BCB) model to quan- titatively describe the CISS effect in mesoscopic devices. Based on this formalism, we demonstrated that more than two terminals are needed in order to probe the CISS effect in linear-regime transport experiments. Moreover, we also showed several ways of injecting and detecting spins using CISS molecules and demonstrated that CISS molecules can give rise to spin-charge conversion. In ad- dition, we proposed two types of graphene-based nonlocal devices which can be used to directly measure the CISS effect in the linear regime. We stress again that the above discussions and pro- posed devices are all based on linear-regime electron transport. Therefore, our conclusions cannot exclude the two-terminal detection of the CISS effect in the non-linear regime. However, the spin signals in non- linear-regime measurements should approach zero as the two-terminal bias approaches zero (entering the linear regime), and the mechanism that may contribute to such 1 + - V Substrate Graphene Tunnel barrier Normal contact CISS molecules Ferromagnet a) 𝐼𝑖𝑛𝑗 𝑉𝑑𝑒𝑡 2 4 3 + - V b) 𝐼𝑖𝑛𝑗 𝑉𝑑𝑒𝑡 1 2 FM 3 4 FM signals has to be different from spin-dependent electron transmission and reflection. A recent work shows that the CISS effect in electron photoemission experiments (three-terminal) can be explained by losses due to spin- dependent electron absorption in chiral molecules,44 but whether a similar process can lead to the detection of the CISS effect in non-linear two-terminal measurements re- mains to be investigated. In general, our model captures the fundamental role of the CISS effect in linear-regime mesoscopic devices without assuming any microscopic electron transport mechanism inside CISS molecules. Re- cently, a new type of spin-orbit coupling was predicted for one-dimensional screw dislocations in semiconductor crystals, which has a one-dimensional helical effective electric field. This type of spin-orbit coupling can lead to an enhanced spin lifetime for electrons traveling along the helical axis.45 Future theoretical work should study whether similar effects exist in chiral or helical molecules. In general, our model helps to analyze and understand device-based CISS experiments without having to under- stand the CISS effect on a molecular level. It provides a guideline for future reviewing and designing of CISS- based mesoscopic spintronic devices. V. ACKNOWLEDGMENTS The authors acknowledge the financial support from the Zernike Institute for Advanced Materials (ZIAM), and the Spinoza prize awarded to Prof. B. J. van Wees by the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO). XY acknowledges helpful discussions with G. E. W. Bauer, V. Mujica, H. Zacharias, and R. Naaman. the sum of these probabilities is therefore unity. a + b + A = 1, c + d + D = 1. Therefore we have RCISS = (cid:18)1 − a − b 0 (cid:19) 0 1 − c − d 9 (A2) (A3) In addition, we adopt transmission and reflection matri- ces of a ferromagnet from Fig. 2. Next, we consider that the CISS effect exists in the linear regime. This means that (according to the de- scription of the CISS effect) with an input of spin non- polarized electrons, the CISS molecule gives a spin- polarized transmission output (a non-zero spin current Is). Here we do not make assumptions about the chi- rality of the molecule or the electron flow direction, so that our conclusions hold for the most general situations. Therefore, we do not assume the sign of Is, and write (cid:54)= 0, (cid:17)TCISS (a + c) − (b + d) (cid:18)1/2 (cid:19) 1,−1 (cid:16) Is = = 1/2 2 (A4) where the column vector indicates the spin non-polarized input current, and the row vector is an operator that cal- culates the difference between the two spin components in the output current. In order to illustrate the discrepancy between the as- sumption of not having any spin-flip reflection and the conceptual description of the CISS effect (Eqn. A4), we apply the reciprocity theorem to the circuit segment shown in Fig. 1. For this circuit segment, we calculate the total transmission matrix accounting for the contribution from both the CISS molecule and the ferromagnet, and obtain (cid:32) (cid:33)−1 Appendix A: General validity of the spin-flip reflection process T21 = TF M · I − RCISS · RF M · TCISS. (A5) In the main text, we stated that a spin-flip electron re- flection process has to exist in order for spin-dependent transmission through CISS molecules to be allowed by the reciprocity theorem. Mathematically, this statements means that at least one off-diagonal term in the reflec- tion matrices of CISS molecules has to be non-zero. Now we prove the general validity of our statement by lim- iting all off-diagonal terms of the reflection matrices to zero, and derive violations against the description of the CISS effect. Under this limit, the general form of the transmission and reflection matrices of a CISS molecule is (cid:18)a c (cid:19) b d (cid:18)A 0 (cid:19) 0 D TCISS = , RCISS = , (A1) where 0 ≤ a, b, c, d, A, D ≤ 1. For electrons traveling to- wards the CISS molecule, each spin component can either be transmitted (with or without spin-flip) or reflected, The transmission probability accounting for both spin species is therefore (cid:16) (cid:17)T21 (cid:18)1/2 (cid:19) 1/2 T21 = 1, 1 , (A6) where the row vector is an operator that calculates the sum of both spin species. By substituting the matrices, we can write T21 as a function of PF M T21(PF M ) = a(1 + PF M ) 1 + a + b + PF M (1 − a − b) d(1 − PF M ) + 1 + c + d − PF M (1 − c − d) (A7) . Here the magnetization reversal of the ferromagnet is equivalent to a sign change of PF M . Therefore, the broader reciprocity theorem requires36,37 T12(PF M ) ≡ T12(−PF M ) (A8) 10 (B2a) (B2b) (B2c) the CISS center is ideal. Now we consider Part 1 as one unit and combine it with the second normal barrier (cid:19) L · RB)−1 · TP 1 σ(1 − t) σ R , (cid:19) L )−1 · RB · TP 1 σ/(1 − t) R , σ(1 − t)2 − t + 1) (cid:19) L )−1 · TB L · (I − RB · RP 1 σ/(1 − t) σ σ , RP R = RP 1 L · (I − RB · RP 1 = R = TB · (I − RP 1 TP σ = R + TP 1 (cid:18)σ/(1 − t) (cid:18)σ(1 − t)2 − t + 1 (cid:18)σ(1 − t) (cid:18)σ(1 − t)2 − t + 1 TP L = TP 1 σ(1 − t) = σ/(1 − t) L = RB + TB · (I − RP 1 RP L · RB)−1 · RP 1 σ(1 − t) (cid:19) L · TB σ(1 − t)2 − t + 1) t2(1 − t) (B2d) , where σ = −t4 + 4t3 − 6t2 + 4t . 1. Derivation of the BCB transmission and reflection matrices = for all 0 < PF M ≤ 1. This requirement gives and therefore, a = d, and b = c, (a + c) − (b + d) = 0, which violates Eqn. A4. Therefore, we proved that the spin-flip electron reflec- tion process has to exist in order for the CISS effect to exist in the linear transport regime, and this is a direct requirement from the Reciprocity Theorem. Appendix B: Further discussions about the BCB model In the BCB model, we consider a CISS molecule as two normal barriers sandwiching an ideal CISS center. The three parts together determine the transmission and re- flection matrices of the molecule. We derive these matri- ces in two steps. First, we calculate the combined contri- bution of the left-most barrier and the ideal CISS center as Part 1 (superscript P 1). Then, we add the second barrier to Part 1 and calculate the total effect. We adopt the transmission and reflection matrices for a normal barrier and a CISS center from Fig. 2. Here we only discuss P -type CISS molecules, as the AP -type can be derived using the relations given in Fig. 2. For Part 1 we have TP 1 R = (TP 0,R + TP (cid:18)t t(1 − t) (cid:19) 0,R · RB · RP = 0 0 , 0,R) · TB (cid:18)1 − t (cid:19) R = RB + TB · RP 0,R · TB RP 1 t2 1 − t = 0 , (cid:18)0 t(1 − t) (cid:19) L = (TB + TB · RP 0,R · RB) · TP TP 1 0,L = 0 t , RP 1 L = RP 0,L + TP (cid:18)0 (1 − t)2 (cid:19) 0,R · RB · RP = 1 0 . 0,R · RB · TP 0,L (B1a) (B1b) (B1c) (B1d) Note that here we have a finite number of reflections between the first barrier and the CISS center because These results show that in the BCB model, one pa- rameter t (0 < t ≤ 1) determines the entire set of trans- mission and reflection probability matrices. Therefore it is possible to plot various transmission or reflection anal- ysis results as functions of t, as shown in the supplemen- tary information. With different t values the BCB model is able to represent a large spectrum of CISS molecules with different "strengths" of the CISS effect. While it is sufficient to illustrate the fundamental role of a CISS molecule in a solid-state device, it is still a simplified pic- ture. We will introduce a more generalized model later in Appendix C. 2. Discussion on the FM-BCB geometry In the main text, we discussed that the two-terminal transmission remains constant under magnetization re- versal in the FM-BCB geometry, here we illustrate the same result under current reversal. (Note that this is also a restriction from the linear regime.) The transmission and reflection matrices from contact 2 to contact 1 are TF M−BCB 12 (⇒) (cid:32) =TF M (⇒) · I − RP R · RF M (⇒) (cid:33)−1 · TP L , (B3a) RF M−BCB 22 (⇒) (cid:32) I − RF M (⇒) · RP R =RP R · L + TP · RF M (⇒) · TP L , (cid:33)−1 (B3b) 11 FIG. 9: Normalized total transmission and reflection of an FM-BCB segment as a function of t (BCB barrier transmission). Red and magenta labels (lower group) are for total transmissions accounting for both spin species for different magnetization orientations (FM=R or L, for magnetization right or left) and different electron flow directions (I=R or L, for electron flow from left to right or from right to left). Blue and cyan labels (upper group) are for total reflections. The polarization of the ferromagnet is chosen as PF M = 0.1, comparable to experimental conditions with Co contacts. FIG. 10: Normalized total transmission and reflection of a spin-valve segment as a function of t (BCB barrier transmission). Red and magenta labels (lower group) are for total transmissions accounting for both spin species for different ferromagnet configurations (similar notation as before, but for two ferromagnets, FM=RR, RL, LR, or LL) and different current flow directions (I=R or L). Blue and cyan labels (upper group) are for total reflections. The polarization of the ferromagnet is chosen as PF M = 0.1, comparable to experimental conditions with Co contacts. and the corresponding transmission and reflection prob- abilities accounting for both spins are T F M−BCB (⇒) = (⇒) (B4a) 1, 1 , 12 (cid:16) (cid:16) (cid:17)TF M−BCB (cid:17)RF M−BCB 12 22 (cid:18)1/2 (cid:19) (cid:18)1/2 (cid:19) 1/2 1/2 RF M−BCB 22 (⇒) = 1, 1 (⇒) . (B4b) With these expressions, we can calculate the transmis- sion and reflection probabilities as a function of t (BCB transmission probability) for four situations: two current directions and two magnetization directions, and the re- sults are plotted in Fig. 9. Note that for all four situa- tions, the transmission curves (or the reflection curves) completely overlap with each other, this means that nei- ther magnetization reversal nor current reversal can lead to a signal change in the two-terminal conductance. Fur- thermore, we are able to quantitatively analyze the con- tribution of each spin component in each of the four sit- uations, and this will be shown in the supplementary information. 3. Discussion on the spin valve geometry Similar to the FM-BCB geometry, in order to calculate the two-terminal transmission and reflection probability of the spin valve geometry, we first calculate the transmis- sion and reflection matrices for this geometry. For this, we treat the spin valve geometry as an FM-BCB module and a ferromagnet connected in series, and derive (cid:32) 21 (⇒,⇒) TSV =TF M (⇒) · · TF M−BCB 21 22 I − RF M−BCB (⇒), (cid:33)−1 (⇒) · RF M (⇒) (B5a) 11 (⇒,⇒) RSV (cid:32) =RF M−BCB 11 (⇒) + TF M−BCB (⇒) 21 · · RF M (⇒) · TF M−BCB I − RF M (⇒) · RF M−BCB (⇒), 22 12 (⇒) (cid:33)−1 (B5b) (cid:32) I − RF M (⇒) · RF M−BCB 22 12 (⇒,⇒) TSV =TF M−BCB · TF M (⇒), 12 (⇒) · (cid:33)−1 (⇒) (B5c) 00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81Totaltransmissionorre.ectiontrans, FM=R, I=Rtrans, FM=R, I=Ltrans, FM=L, I=Rtrans, FM=L, I=Lrefl, FM=R, I=Rrefl, FM=R, I=Lrefl, FM=L, I=Rrefl, FM=L, I=L00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81Totaltransmissionorre.ectiontrans, FM=RR, I=Rtrans, FM=RR, I=Ltrans, FM=RL, I=Rtrans, FM=RL, I=Ltrans, FM=LR, I=Rtrans, FM=LR, I=Ltrans, FM=LL, I=Rtrans, FM=LL, I=Lrefl, FM=RR, I=Rrefl, FM=RR, I=Lrefl, FM=RL, I=Rrefl, FM=RL, I=Lrefl, FM=LR, I=Rrefl, FM=LR, I=Lrefl, FM=LL, I=Rrefl, FM=LL, I=L(cid:33)−1 (⇒) · RF M (⇒) (B5d) 12 22 (⇒,⇒) RSV (cid:32) =RF M (⇒) + TF M (⇒) I − RF M−BCB · · RF M−BCB 22 22 (⇒) · TF M (⇒), where the two arrows in the brackets on the left-hand side of the equations indicate the magnetization direction of the two ferromagnets respectively. For the case where the magnetization of one of the ferromagnets is reversed, we can substitute the corresponding magnetization direction with an opposite arrow. For the two-terminal transmission and reflection prob- abilities accounting for both spins, we have (cid:16) (cid:16) 1, 1 1, 1 ij (cid:17) · TSV (cid:17) · RSV ij (cid:18)1/2 (cid:19) (cid:19) (cid:18)1/2 1/2 1/2 · · T SV ij = RSV ij = FIG. 11: Injection coefficient kinj as a function of t (BCB barrier transmission) for various tB (contact barrier transmissiion), for the geometry described in Fig. 6a) in the main text. , , (B6a) (B6b) and we can calculate these probabilities as a function of t (BCB transmission probability) for eight situations: the two ferromagnets each with two magnetization direc- tions, and two opposite current directions, as shown in Fig. 10. Note that for all eight situations, the transmis- sion curves (or the reflection curves) completely overlap with each other, this means that neither magnetization reversal (for either ferromagnet) nor current reversal can lead to a signal change in the two-terminal conductance. Furthermore, we can quantitatively analyze the contribu- tion of each spin component in this geometry, as shown in the supplementary information. 4. Discussion on injection and detection coefficients In the main text, we derived the injection and detection coefficients kinj and kdet for the four-terminal geometry shown in Fig. 6a), and we showed that the product of these two geometries represents the ratio between four- terminal and two-terminal resistances. The injection co- efficient depends both on the BCB barrier transmission t and the transmission probability tB of the barrier at contact 2, whereas the detection coefficient only depends on the BCB barrier transmission t. This difference is due to our assumption of weakly coupled detection contacts. Here we plot the injection coefficient kinj, the detection coefficient kdet, and the 4T/2T resistance ratio R4T /R2T as a function of t (transmission probability of the barrier in the BCB molecule). Especially, for kinj and R4T /R2T , we set tB (transmission probability of the barrier in con- tact 2) to a few different values and illustrate its influ- ence. FIG. 12: Detection coefficient kdet as a function of t (BCB barrier transmission) for the geometry described in Fig. 6a) in the main text. 5. Discussion on spin-charge conversion In the main text, we illustrated the spin-charge conver- sion property of CISS molecules. The chemical potential vectors in the two nodes in Fig. 7 are related to each other following Eqn. 16. We also introduced the scalers of charge chemical potential µn (the average of two spin chemical potentials) and spin accumulation µs (the dif- ference between two spin chemical potentials). Here we show how these scaler chemical potentials relate to each other. A vector chemical potential of a node can be rewritten 00.20.40.60.81BCBbarriertransmissiont00.050.10.150.20.25Injectioncoe/cientkinjtB=0.01tB=0.1tB=0.5tB=0.900.20.40.60.81BCBbarriertransmissiont00.10.20.30.40.5Detectioncoe/cientkdet13 where kconv = t/(4 − 2t) (0 < kconv ≤ 0.5) is the spin- charge conversion coefficient. Eqn. B11a shows that µsB = µsA − ∆µs = 0, meaning that a spin accumulation in node A cannot generate a spin accumulation in node B. This result is special for the BCB model and does not hold for a more generalized CISS model (as will be introduced later in Appendix C). Eqn. B11b shows that the charge voltage across a BCB molecule linearly depends on the spin accumulation dif- ference across the molecule, and vice versa (spin-charge conversion). The conversion coefficient kconv depends on t (BCB barrier transmission). Notably, for BCB molecules this coefficient has the same value as the de- tection coefficient (kconv = kdet). This is because the result µsB = 0 makes node B equivalent to a contact: it is not able to distinguish between two spin components. For a more generalized model (as will be introduced later in Appendix C) the two coefficient take different values. These conclusions were mentioned in the main text, here we showed the proof. 6. Discussion on nonlocal signals We first discuss the non-local resistance measured with device geometries shown in Fig. 8a). In this picture the axes of the CISS molecules are vertical rather than hor- izontal, therefore the spin orientations are described as spin-up or spin-down. As a result, the spin accumulation is defined as µs = (µ↑ − µ↓). First, we discuss the case where the spin injection is done through a BCB molecule. The injected spin accu- mulation underneath contact 1 is µs,inj = −kinjµ1 = −ekinjIinjRinj, (B12) FIG. 13: The ratio between four-terminal and two-terminal resistances as a function of t (BCB barrier transmission) for various tB (contact barrier transmissiion), for the geometry described in Fig. 6a) in the main text. as a sum of two column vectors µ = = (cid:18)µ→ (cid:19) (cid:19) (cid:18)1 µ← = µn + 1 (cid:18)µn + 1 (cid:18) 1−1 µn − 1 1 2 µs 2 µs 2 µs (cid:19) (cid:19) . (B7) By rewriting both µA and µB in this fashion, and sub- stituting them into Eqn. 16, we obtain (cid:18)1 (cid:19) 1 ∆µn + 1 2 ∆µs (cid:32) (cid:19) (cid:18) 1−1 = I − (I − RP L )−1TP R µA, (cid:33) (B8) where ∆µn = µnA− µnB is the charge chemical potential difference between the two nodes, and ∆µs = µsA−µsB is the spin accumulation difference between the two nodes. where the minus sign is due to the fact that the elec- trons are traveling downwards through CISS molecules into graphene, as a result, the injected spin accumula- tion is negative (mostly spin-down). The resistance For the BCB model, I − (I − RP L )−1TP R = (cid:18)1 + 1 t−2 −1 − 1 t−2 − 1 1 t−2 t−2 (cid:19) (B9) where t is the BCB transmission probability. Substitut- ing this matrix into Eqn. B8 gives (cid:18)1 (cid:19) 1 ∆µn + 1 2 ∆µs (cid:18) 1−1 (cid:19) (cid:19) (cid:18)1 + 1 t−2 1 t−2 = µsA . (B10) Note that here the charge chemical potentials in node A (µnA) drops out of the equation. We emphasize that this is a unique result for the BCB model. Solving the above vector equation gives ∆µs = µsA, ∆µn = −kconv∆µs, (B11a) (B11b) Rinj = R12 = (B13) µ1 − µ2 eIinj is the resistance measured between the two injection con- tacts 1 and 2. Inside graphene, the spin accumulation diffuses to all directions, and therefore the spin accumulation at the detection contact is µs,det = µs,inje − d λs , (B14) where λs is the spin diffusion length in graphene, and d is the distance between the inner injection and detection contacts (1 and 4) (we have assumed that this distance is much larger than the separation of the two injection contacts or the separation of the two detection contacts). Further, the voltage detected by the detection contacts is (following Eqn. 13) Vdet = 1 e kdetµs,det. (B15) 00.20.40.60.81BCBbarriertransmissiont00.050.1R4T=R2TtB=0.01tB=0.1tB=0.5tB=0.9With this, we have Appendix C: Generalized CISS Model 14 Rnl = Vdet Iinj = −kinjkdetRinje − d λs , (B16) as in Eqn. 17. For the case where the spin injection is obtained through a ferromagnet, the spin injection becomes µs,inj = ±ePF M IinjRλ, (B17) where PF M is the polarization of the ferromagnet (with magnetization direction out-of-plane), and Rλ is the spin resistance of graphene. This spin resistance is deter- mined by the spin relaxation length in graphene and the shape of the graphene channel, and is defined as Rλ = Rsqλs/W , where Rsq is the square resistance of graphene and W is the width of the graphene channel (assuming the channel width remains the same across the spin diffusion length). The sign of the injected spin accu- mulation is determined by the magnetization direction of the ferromagnet, with magnetization-up for positive spin accumulation and magnetization-down for negative spin accumulation. The diffusion and detection mechanisms are the same as in the previous case. Therefore, the non-local resis- tance for this situation is Rnl = ±kdetPF M Rλe − d λs , (B18) as in Eqn. 18. With the help of the ferromagnet, it is possible to switch the sign of the non-local resistance. Next, for the device shown in Fig. 8b), the CISS molecules are aligned in-plane of the device, therefore we assume the ferromagnet also has in-plane magne- tization, and we describe spin accumulation again as µs = µ→ − µ←. We simplify the discussion by assum- ing the spin injection is mainly contributed by the ferro- magnet, and the spin detection is achieved through the spin-charge conversion mechanism of the CISS molecules. The spin accumulation underneath the detection contacts can be generated by two mechanisms: the spin diffusion in graphene (as in the previous case), and the spin-charge conversion between the injection node and the detection node. However, in the BCB model, the spin-charge con- version does not contribute to a spin accumulation under- neath the detector contacts. Therefore, we only consider the spin diffusion mechanism. Similar to the previous case, we can derive the nonlocal signal Rnl = ±kconvPF M Rλe (B19) − d λs , where kconv is the spin-charge conversion coefficient described before, but here it concerns the proximity- induced CISS effect in the graphene channel, rather than the CISS molecules themselves. Important to realize, due to the proximity effect, the diffusion length λs and the spin resistance Rλ of graphene may differ from the pre- vious case. The BCB model is a simplified model where the spin- dependent characteristics of a CISS molecule are exclu- sively originated from an ideal CISS center. However, the assumption of having an ideal spin-flip core in a molecule may not be accurate. Therefore, we assume a general form of transmission and reflection matrices, (cid:18)a, c (cid:19) b, d (cid:18)A, C (cid:19) B, D , RP R = , (C1) TP R = where 0 ≤ a, b, c, d, A, B, C, D ≤ 1, (C2a) a + b + A + B = c + d + C + D = 1, (C2b) a + c > b + d. (C2c) Here the first restriction Eqn. C2a comes from the fact that all matrix elements are probabilities. The second restriction Eqn. C2b addresses that for each spin compo- nent, the sum of its probabilities of being transmitted and being reflected equals 1. The third restriction Eqn. C2c is the conceptual description of the CISS effect, which shows that a spin polarization arises after transmission through a CISS molecule. This restriction is similar to Eqn. A4, but here we determine a sign for the polariza- tion because we have assumed the chirality (P -type) of the molecule and the electron flow direction (R for right- wards). We still assume that the molecule has C2 symmetry, so that the transmission and reflection matrices for reversed current can be written as (cid:18)0 1 (cid:19) (cid:18)0 1 (cid:19) 1 0 1 0 (cid:18)0 1 (cid:19) (cid:18)0 1 (cid:19) 1 0 1 0 R · · TP · RP R · TP L = RP L = , . (C3a) (C3b) The matrices for AP -type molecules can be derived ac- cording to Fig. 2. Before we proceed with the mathe- matical proof, we discuss the validity of these symmetry assumptions. Since we only consider the electron transport in CISS molecules, the C2 symmetry that we require only refers to the symmetry in electron transport. Furthermore, this symmetry is also a requirement of the linear regime. A highly asymmetric molecule can still be treated with our model, but only under the assumption that it is composed of a symmetric (C2) part which contributes to CISS, and asymmetrically distributed normal barriers. In terms of the symmetry relations between the two chiralities, it is by definition that the two chiral enantiomers are exact mirror images of each other, and thereby select opposite spins with equal probability (when placed in the same circuit environment). We apply the reciprocity theorem to a geometry simi- lar to Fig. 4 in the main text, but now the BCB molecule is replaced by a generalized CISS molecule. The general reciprocity theorem requires that the two-terminal resis- tance remains unchanged under magnetization reversal regardless of the polarization of the ferromagnet. There- fore, we assume the ferromagnet is 100% polarized for convenience. Following the same steps as in the FM- BCB geometry, the two terminal transmission probabil- ities considering two magnetization directions and two current directions become T21(⇒) = a + b + (c + d) T21(⇐) = c + d + (a + b) T12(⇒) = b + d + (a + c) T12(⇐) = a + c + (b + d) B 1 − D , C 1 − A , C 1 − D , B 1 − A , (C4a) (C4b) (C4c) (C4d) where the arrows indicate the magnetization direction of the ferromagnet, and the subscripts indicate the electron flow direction. The reciprocity theorem requires these four expressions to have the same value T12(⇒) = T12(⇐) = T21(⇒) = T21(⇐) = T, (C5) which can be written as a vector equation  1 C 1−A C 1−D 1  · a  = T 1  . (C6) b c d 1 1 1 1 C 1−A 1 B 1−A B 1−D 1 C 1−D 1 B 1−D 1 1 B 1−A In order for this equation to be self-consistent, relation A = D is required. Consequently, we can derive b = c, a − d = C − B. (C7a) (C7b) Therefore, the generalized form of the transmission and reflection matrices of a CISS molecule is (cid:18)a (cid:19) TP R = b b a(1 − s) , RP R = (cid:18)A B + as (cid:19) B A , (C8) with a + b + A + B = 1, and s (0 ≤ s ≤ 1) being a quantitative description of the strength of the CISS effect in a molecule. 15 FIG. 14: Normalized total transmission and reflection of an FM-CISS segment as a function of s (generalized CISS strength). Red and magenta labels (lower group) are for total transmissions accounting for both spin species for different magnetization orientations (FM=R or L) and different electron flow directions (I=R or L). Blue and cyan labels (upper group) are for total reflections. The polarization of the ferromagnet is chosen as PF M = 0.7 to be consistent with Fig. 15. Next, we discuss the results given by this generalized model for the geometries discussed in the main text, and compare it with the BCB model. We arbitrarily choose the following transmission and reflection matrices for the generalized model (cid:19) (cid:18) 0.2 0.05 + 0.6s (cid:19) 0.05 0.2 , TP R = 0.15 0.15 0.6(1 − s) , RP R = (cid:18) 0.6 (C9) and use parameter s as the variable to tune the CISS strength. For the FM-BCB geometry (which now becomes FM- CISS geometry), we plot the two-terminal transmission and reflection probabilities as a function of s for four situations: two magnetization directions and two cur- rent directions. The results show that all four situations give identical results for all CISS strength s, as shown in Fig. 14. This is consistent with the BCB model, just as required by the reciprocity theorem. For the spin valve geometry, we plot the two-terminal transmission and reflection probabilities as a function of s for eight situations: two ferromagnets each with two mag- netization directions, and two current directions. The results are shown in Fig. 15. Unlike the BCB model, here we find different transmission and reflection proba- bilities for cases where the two magnetization directions are parallel vs. anti-parallel. The transmission probabil- ity is in general higher when the two ferromagnets are magnetized parallel compared to anti-parallel. Notably, the difference between the two configurations decreases as the CISS strength s increases. This is because with increasing s, electrons have a higher probability of en- 00.20.40.60.81CISSstrengths0.350.40.450.50.550.60.65Totaltransmissionorre.ectiontrans, FM=R, I=Rtrans, FM=R, I=Ltrans, FM=L, I=Rtrans, FM=L, I=Lrefl, FM=R, I=Rrefl, FM=R, I=Lrefl, FM=L, I=Rrefl, FM=L, I=L16 FIG. 15: Normalized total transmission and reflection of a spin-valve segment as a function of s (generalized CISS strength). Red and magenta labels (lower group) are for total transmissions accounting for both spin species for different ferromagnet configurations (two ferromagnets, FM=RR, RL, LR, or LL) and different current flow directions (I=R or L). Blue and cyan labels (upper group) are for total reflections. The polarization of the ferromagnet is chosen as PF M = 0.7 in order to amplify the deviation between parallel and anti-parallel ferromagnet configurations. This deviation is larger when the ferromagnet polarization is higher. countering spin-flip reflections. This also explains why in a BCB model, where an ideal spin-flip process is present, switching the magnetization directions does not lead to any conductance variation. The direction of the current has no effect on the transmission and reflection probabil- ities, as is required by the linear regime. FIG. 16: Injection coefficient kinj as a function of s (generalized CISS strength) for various tB (contact barrier transmissiion), for the geometry described in Fig. 6a) in the main text. FIG. 17: Detection coefficient kdet as a function of s (generalized CISS strength) for the geometry described in Fig. 6a) in the main text. FIG. 18: The ratio between four-terminal and two-terminal resistances as a function of s (generalized CISS strength) for various tB (contact barrier transmission), for the geometry described by Fig. 6a) in the main text. For four terminal geometries, we calculate here the spin injection and detection coefficients using the formulas de- rived in the main text, and we plot these coefficients as a function of the CISS strength s. Fig. 16 shows the in- jection coefficient as a function of s for a few tB values. Fig. 17 shows the detection coefficient as a function of s, and Fig. 18 shows the ratio between four-terminal and two-terminal resistances as a function of s. Last but not least, we discuss the spin-charge conver- sion property of the generalized CISS molecule. Sub- stituting the corresponding transmission and reflection matrices into Eqn. B8, and solving the vector equation, we can obtain relations between the conversion coefficient kconv and the chemical potential vector µA. Unlike the BCB model, here it is not possible to drop out either µnA 00.20.40.60.81CISSstrengths0.20.30.40.50.60.70.8Totaltransmissionorre.ectiontrans, FM=RR, I=Rtrans, FM=RR, I=Ltrans, FM=RL, I=Rtrans, FM=RL, I=Ltrans, FM=LR, I=Rtrans, FM=LR, I=Ltrans, FM=LL, I=Rtrans, FM=LL, I=Lrefl, FM=RR, I=Rrefl, FM=RR, I=Lrefl, FM=RL, I=Rrefl, FM=RL, I=Lrefl, FM=LR, I=Rrefl, FM=LR, I=Lrefl, FM=LL, I=Rrefl, FM=LL, I=L00.20.40.60.81CISSstrengths00.050.10.150.2Injectioncoe/cientkinjtB=0.01tB=0.1tB=0.5tB=0.900.20.40.60.81CISSstrengths00.050.10.150.20.250.3Detectioncoe/cientkdet00.20.40.60.81CISSstrengths00.020.040.06R4T=R2TtB=0.01tB=0.1tB=0.5tB=0.9or µsA from the equation. The final result gives 17 kconv = − ∆µn ∆µs = (A − B − 1)sµnA + 1 2 (b + B)sµsA 2(A + B − 1)sµnA + (b + B)(−2 + 2A + 2B + s)µsA , (C10) where a, b, A, B, s are the parameters in the transmission and reflection matrices of the generalized CISS model. This equation shows that a non-zero charge voltage dif- ference (∆µn) can give rise to a spin accumulation differ- ence (∆µs) across a generalized CISS molecule, and vice versa (spin-charge conversion). Interestingly, the con- version coefficient kconv depends not only on the trans- mission and reflection matrices of the molecule, but also on the spin accumulation in the nodes connected to the molecule. If s = 0, i.e. the molecule does not exhibit any CISS effect, the spin-charge conversion property also diminishes (kconv = 0). With these discussions, we demonstrated that a gen- eralized CISS molecule shows differences from the sim- plified BCB model. Nonetheless, these differences are rather quantitative than qualitative, and are not easily measurable in experiments. Furthermore, the general- ized model also introduces extra degrees-of-freedom to calculations as it uses four variables to describe a CISS molecule, compared to one in the BCB model. Having taken the above into consideration, in the main text we only showed the results of the BCB model. ∗ [email protected] 1 SA Wolf, DD Awschalom, RA Buhrman, JM Daughton, S Von Molnar, ML Roukes, A Yu Chtchelkanova, and DM Treger, "Spintronics: a spin-based electronics vision for the future," Science 294, 1488 -- 1495 (2001). 2 R Naaman and David H Waldeck, "Chiral-induced spin se- lectivity effect," The Journal of Physical Chemistry Letters 3, 2178 -- 2187 (2012). 3 Ron Naaman and David H Waldeck, "Spintronics and chi- rality: Spin selectivity in electron transport through chi- ral molecules," Annual Review of Physical Chemistry 66, 263 -- 281 (2015). 4 B Gohler, V Hamelbeck, T Z Markus, M Kettner, G F Hanne, Z Vager, R Naaman, and H Zacharias, "Spin se- lectivity in electron transmission through self-assembled monolayers of double-stranded DNA," Science 331, 894 -- 897 (2011). 5 Zouti Xie, Tal Z Markus, Sidney R Cohen, Zeev Vager, Rafael Gutierrez, and Ron Naaman, "Spin specific elec- tron conduction through DNA oligomers," Nano Letters 11, 4652 -- 4655 (2011). 6 Vankayala Kiran, Shinto P Mathew, Sidney R Cohen, Irene Hern´andez Delgado, J´erome Lacour, and Ron Naaman, "Helicenesa new class of organic spin filter," Advanced Ma- terials 28, 1957 -- 1962 (2016). 7 Matthias Kettner, Volodymyr V Maslyuk, Daniel Nurenberg, Johannes Seibel, Rafael Gutierrez, Gianaure- lio Cuniberti, Karl-Heinz Ernst, and Helmut Zacharias, "Chirality-dependent electron spin filtering by molecular monolayers of helicenes," The Journal of Physical Chem- istry Letters 9, 2025 -- 2030 (2018). 8 K Ray, S P Ananthavel, D H Waldeck, and R Naaman, "Asymmetric scattering of polarized electrons by organized organic films of chiral molecules," Science 283, 814 -- 816 (1999). 9 Itai Carmeli, Viera Skakalova, Ron Naaman, and Zeev Vager, "Magnetization of chiral monolayers of polypep- tide: a possible source of magnetism in some biological membranes," Angewandte Chemie International Edition 41, 761 -- 764 (2002). 10 S G Ray, S S Daube, G Leitus, Z Vager, and R Naa- man, "Chirality-induced spin-selective properties of self- assembled monolayers of DNA on gold," Physical Rreview Letters 96, 036101 (2006). 11 Debabrata Mishra, Tal Z Markus, Ron Naaman, Matthias Kettner, Benjamin Gohler, Helmut Zacharias, Noga Friedman, Mordechai Sheves, and Claudio Fontanesi, "Spin-dependent electron transmission through bacteri- orhodopsin embedded in purple membrane," Proceedings of the National Academy of Sciences 110, 14872 -- 14876 (2013). 12 Miguel ´Angel Nino, Iwona Agnieszka Kowalik, Fran- cisco Jes´us Luque, Dimitri Arvanitis, Rodolfo Miranda, and Juan Jos´e de Miguel, "Enantiospecific spin polariza- tion of electrons photoemitted through layers of homochi- ral organic molecules," Advanced Materials 26, 7474 -- 7479 (2014). 13 M Kettner, B Gohler, H Zacharias, D Mishra, V Kiran, R Naaman, C Fontanesi, David H Waldeck, S(cid:32)lawomir Sek, Jan Paw(cid:32)lowski, et al., "Spin filtering in electron trans- port through chiral oligopeptides," The Journal of Physical Chemistry C 119, 14542 -- 14547 (2015). 14 Karen Michaeli, Vaibhav Varade, Ron Naaman, and David H Waldeck, "A new approach towards spintronics -- spintronics with no magnets," Journal of Physics: Con- densed Matter 29, 103002 (2017). 15 Oren Ben Dor, Shira Yochelis, Shinto P Mathew, Ron Naa- man, and Yossi Paltiel, "A chiral-based magnetic memory device without a permanent magnet," Nature Communi- cations 4, 2256 (2013). 16 Prakash Chandra Mondal, Nirit Kantor-Uriel, Shinto P Mathew, Francesco Tassinari, Claudio Fontanesi, and Ron Naaman, "Chiral conductive polymers as spin filters," Ad- vanced Materials 27, 1924 -- 1927 (2015). 17 Vaibhav Varade, Tal Markus, Kiran Vankayala, Noga Friedman, Mordechai Sheves, David H Waldeck, and Ron 18 Adame, G Cuniberti, and Qing-feng Sun, "Contact effects in spin transport along double-helical molecules," Physical Review B 89, 205434 (2014). 34 M Buttiker, "Role of quantum coherence in series resis- tors," Physical Review B 33, 3020 (1986). 35 M Buttiker, Y Imry, R Landauer, and S Pinhas, "Gener- alized many-channel conductance formula with application to small rings," Physical Review B 31, 6207 (1985). 36 M Buttiker, "Four-terminal phase-coherent conductance," Physical Review Letters 57, 1761 (1986). 37 M Buttiker, "Symmetry of electrical conduction," IBM Journal of Research and Development 32, 317 -- 334 (1988). 38 Gary A Prinz, "Magnetoelectronics," Science 282, 1660 -- 1663 (1998). 39 I Adagideli, Gerrit E W Bauer, and Bertrand I Halperin, "Detection of current-induced spins by ferromagnetic con- tacts," Physical Review Letters 97, 256601 (2006). 40 Friso Jacobus Jedema, A T Filip, and B J van Wees, "Electrical spin injection and accumulation at room tem- perature in an all-metal mesoscopic spin valve," Nature 410, 345 (2001). 41 Marc Drogeler, Christopher Franzen, Frank Volmer, To- bias Pohlmann, Luca Banszerus, Maik Wolter, Kenji Watanabe, Takashi Taniguchi, Christoph Stampfer, and Bernd Beschoten, "Spin lifetimes exceeding 12 ns in graphene nonlocal spin valve devices," Nano Letters 16, 3533 -- 3539 (2016). 42 Josep Ingla-Ayn´es, Marcos H D Guimaraes, Rick J Mei- jerink, Paul J Zomer, and Bart J van Wees, "24-µm spin relaxation length in boron nitride encapsulated bi- layer graphene," Physical Review B 92, 201410 (2015). 43 Mallikarjuna Gurram, Siddhartha Omar, Simon Zihlmann, P´eter Makk, QC Li, YF Zhang, Christian Schonenberger, and Bart J van Wees, "Spin transport in two-layer-CVD- hBN/graphene/hBN heterostructures," Physical Review B 97, 045411 (2018). 44 Daniel Nurenberg and Helmut Zacharias, "Spin-flip scat- tering in chiral induced spin selectivity using the Riccati equation," arXiv 1808.00269 (2018). 45 Lin Hu, Huaqing Huang, Zhengfei Wang, W Jiang, Xi- aojuan Ni, Yinong Zhou, V Zielasek, MG Lagally, Bing Huang, and Feng Liu, "Ubiquitous spin-orbit coupling in a screw dislocation with high spin coherency," Physical Review Letters 121, 066401 (2018). Naaman, "Bacteriorhodopsin based non-magnetic spin fil- ters for biomolecular spintronics," Physical Chemistry Chemical Physics 20, 1091 -- 1097 (2018). 18 Brian P Bloom, Vankayala Kiran, Vaibhav Varade, Ron Naaman, and David H Waldeck, "Spin selective charge transport through cysteine capped CdSe quantum dots," Nano Letters 16, 4583 -- 4589 (2016). 19 Shinto P Mathew, Prakash Chandra Mondal, Hagay Moshe, Yitzhak Mastai, and Ron Naaman, "Non-magnetic organic/inorganic spin injector at room temperature," Ap- plied Physics Letters 105, 242408 (2014). 20 Kazi M Alam and Sandipan Pramanik, "Spin filtering through single-wall carbon nanotubes functionalized with single-stranded DNA," Advanced Functional Materials 25, 3210 -- 3218 (2015). 21 R Gutierrez, E D´ıaz, R Naaman, and G Cuniberti, "Spin- selective transport through helical molecular systems," Physical Review B 85, 081404 (2012). 22 Sina Yeganeh, Mark A Ratner, Ernesto Medina, and Vladimiro Mujica, "Chiral electron transport: scatter- ing through helical potentials," The Journal of Chemical Physics 131, 014707 (2009). 23 Joel Gersten, Kristen Kaasbjerg, and Abraham Nitzan, "Induced spin filtering in electron transmission through chiral molecular layers adsorbed on metals with strong spin-orbit coupling," The Journal of Chemical Physics 139, 114111 (2013). 24 Ernesto Medina, Luis A Gonz´alez-Arraga, Daniel Finkelstein-Shapiro, Bertrand Berche, and Vladimiro Mu- jica, "Continuum model for chiral induced spin selectiv- ity in helical molecules," The Journal of Chemical Physics 142, 194308 (2015). 25 Karen Michaeli and Ron Naaman, "Origin of spin dependent tunneling through chiral molecules," arXiv 1512.03435 (2015). 26 Volodymyr V Maslyuk, Rafael Gutierrez, Arezoo Di- anat, Vladimiro Mujica, and Gianaurelio Cuniberti, "En- hanced magnetoresistance in chiral molecular junctions," The Journal of Physical Chemistry Letters 9, 5453 -- 5459 (2018). 27 Ai-Min Guo and Qing-feng Sun, "Spin-selective transport of electrons in DNA double helix," Physical Review Letters 108, 218102 (2012). 28 Ai-Min Guo and Qing-Feng Sun, "Spin-dependent electron transport in protein-like single-helical molecules," Pro- ceedings of the National Academy of Sciences 111, 11658 -- 11662 (2014). 29 Ting-Rui Pan, Ai-Min Guo, and Qing-Feng Sun, "Spin- polarized electron transport through helicene molecular junctions," Physical Review B 94, 235448 (2016). 30 Jens H Bardarson, "A proof of the Kramers degeneracy of transmission eigenvalues from antisymmetry of the scat- tering matrix," Journal of Physics A: Mathematical and Theoretical 41, 405203 (2008). 31 Shlomi Matityahu, Yasuhiro Utsumi, Amnon Aharony, Ora Entin-Wohlman, and Carlos A Balseiro, "Spin- dependent transport through a chiral molecule in the presence of spin-orbit interaction and nonunitary effects," Physical Review B 93, 075407 (2016). 32 Shlomi Matityahu, Amnon Aharony, Ora Entin-Wohlman, and Carlos A Balseiro, "Spin filtering in all-electrical three- terminal interferometers," Physical Review B 95, 085411 (2017). 33 Ai-Min Guo, E D´ıaz, C Gaul, R Gutierrez, F Dom´ınguez- 1 Supplementary information: additional figures Fig. S1 to Fig. S4 show the transmission and reflection probability of each spin component for each of the four situations (two magnetization directions and two current directions) of the FM-BCB geometry. In these four situations, even though the total transmissions (or reflections) concerning both spin components are identical, the transmissions and reflections for each spin component, and hence the spin polarizations of the electrons entering the contact are different. This difference cannot be measured as a charge signal because the contacts cannot distinguish between spin components. Fig. S5 to Fig. S12 show the same analysis for the eight situations (two ferromagnets each with two magnetization directions, and two current directions) of the spin valve geometry. Similarly, spin polarization can be generated at the contacts, and are different for all situations. However, due to the limitation of the two-terminal geometry, the total transmissions (or reflections) concerning both spin components remain the same for all eight situations. FIG. S1: Normalized one-spin transmission or reflection for the FM-BCB case, where the magnetization of the ferromagnet is "R" (pointing to the right) and electron flow "R" (from left to right). 00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81One-spintransmissionorre.ectionspin R trans, FM=R, I=Rspin L trans, FM=R, I=Rspin R refl, FM=R, I=Rspin L refl, FM=R, I=R2 FIG. S2: Normalized one-spin transmission or reflection for the FM-BCB case, where the magnetization of the ferromagnet is "R" (pointing to the right) and electron flow "L" (from right to left). FIG. S3: Normalized one-spin transmission or reflection for the FM-BCB case, where the magnetization of the ferromagnet is "L" (pointing to the left) and electron flow "R" (from left to right). 00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81One-spintransmissionorre.ectionspin R trans, FM=R, I=Lspin L trans, FM=R, I=Lspin R refl, FM=R, I=Lspin L refl, FM=R, I=L00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81One-spintransmissionorre.ectionspin R trans, FM=L, I=Rspin L trans, FM=L, I=Rspin R refl, FM=L, I=Rspin L refl, FM=L, I=R3 FIG. S4: Normalized one-spin transmission or reflection for the FM-BCB case, where the magnetization of the ferromagnet is "L" (pointing to the left) and electron flow "L" (from right to left). FIG. S5: Normalized one-spin transmission or reflection for the spin valve case, where the magnetization of the ferromagnet is "RR" (both pointing to the right) and electron flow "R" (from left to right). 00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81One-spintransmissionorre.ectionspin R trans, FM=L, I=Lspin L trans, FM=L, I=Lspin R refl, FM=L, I=Lspin L refl, FM=L, I=L00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81Transmissionorre.ectionspin R trans, FM=RR, I=Rspin L trans, FM=RR, I=Rspin R refl, FM=RR, I=Rspin L refl, FM=RR, I=R4 FIG. S6: Normalized one-spin transmission or reflection for the spin valve case, where the magnetization of the ferromagnet is "RR" (both pointing to the right) and electron flow "L" (from right to left). FIG. S7: Normalized one-spin transmission or reflection for the spin valve case, where the magnetization of the ferromagnet is "RL" (the first one pointing to the right, and the second one to the left) and electron flow "R" (from left to right). 00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81Transmissionorre.ectionspin R trans, FM=RR, I=Lspin L trans, FM=RR, I=Lspin R refl, FM=RR, I=Lspin L refl, FM=RR, I=L00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81Transmissionorre.ectionspin R trans, FM=RL, I=Rspin L trans, FM=RL, I=Rspin R refl, FM=RL, I=Rspin L refl, FM=RL, I=R5 FIG. S8: Normalized one-spin transmission or reflection for the spin valve case, where the magnetization of the ferromagnet is "RL" (the first one pointing to the right, and the second one to the left) and electron flow "L" (from right to left). FIG. S9: Normalized one-spin transmission or reflection for the spin valve case, where the magnetization of the ferromagnet is "LR" (the first one pointing to the left, and the second one to the right) and electron flow "R" (from left to right). 00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81Transmissionorre.ectionspin R trans, FM=RL, I=Lspin L trans, FM=RL, I=Lspin R refl, FM=RL, I=Lspin L refl, FM=RL, I=L00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81Transmissionorre.ectionspin R trans, FM=LR, I=Rspin L trans, FM=LR, I=Rspin R refl, FM=LR, I=Rspin L refl, FM=LR, I=R6 FIG. S10: Normalized one-spin transmission or reflection for the spin valve case, where the magnetization of the ferromagnet is "LR" (the first one pointing to the left, and the second one to the right) and electron flow "L" (from right to left). FIG. S11: Normalized one-spin transmission or reflection for the spin valve case, where the magnetization of the ferromagnet is "LL" (both pointing to the left) and electron flow "R" (from left to right). 00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81Transmissionorre.ectionspin R trans, FM=LR, I=Lspin L trans, FM=LR, I=Lspin R refl, FM=LR, I=Lspin L refl, FM=LR, I=L00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81Transmissionorre.ectionspin R trans, FM=LL, I=Rspin L trans, FM=LL, I=Rspin R refl, FM=LL, I=Rspin L refl, FM=LL, I=R7 FIG. S12: Normalized one-spin transmission or reflection for the spin valve case, where the magnetization of the ferromagnet is "LL" (both pointing to the left) and electron flow "L" (from right to left). 00.20.40.60.81BCBbarriertransmissiont00.20.40.60.81Transmissionorre.ectionspin R trans, FM=LL, I=Lspin L trans, FM=LL, I=Lspin R refl, FM=LL, I=Lspin L refl, FM=LL, I=L
1806.10289	1	1806	2018-06-27T04:19:17	Josephson effect in silicene-based SNS Josephson junction: Andreev reflection and free energy	[ "cond-mat.mes-hall", "cond-mat.supr-con" ]	We investigate the Josephson effcet in superconductor-normal-superconductor junction (SNS) base on the doped unbiased silicene under the perpendicular electric field and off-resonance circularly polarized light. The Andreev reflection (including the retroreflection and specular one) during the subgap transport, the free energy, and the reversal of the Josephson effect as well as the emergence of $\phi_{0}$-junction are exploited. The Andreev reflection is complete in the NS interface even for the clean interface and without the Fermi wave vector mismatch, which is opposite to the case of ferromagnet-superconductor interface. The important role played by the dynamical polarization of the degrees of freedom to the $0-\pi$ transition and the generation of $\phi_{0}$-junction are mentioned in this paper. The scattering by the charged impurity in the substrate affects the transport properties in the bulk as well as the valley relaxation, which can be taken into consider by the macroscopic wave function. In short junction limit, the approximated results about the Andreev level and free energy are also discussed. Beside the low-energy limit of the tight-binding model, the finite-size effect need to be taken into account as long as the spacing model is much larger than the superconducting gap.	cond-mat.mes-hall	cond-mat	a Josephson effect in silicene-based SNS Josephson junction: Andreev reflection and free energy Key Laboratory of Atomic & Molecular Physics and Functional Materials of Gansu Province, College of Physics and Electronic Engineering, Northwest Normal University, Lanzhou 730070, China Chen-Huan Wu ∗ June 28, 2018 We investigate the Josephson effcet in superconductor-normal-superconductor junc- tion (SNS) base on the doped unbiased silicene under the perpendicular electric field and off-resonance circularly polarized light. The Andreev reflection (including the retroreflection and specular one) during the subgap transport, the free energy, and the reversal of the Josephson effect as well as the emergence of φ0-junction are ex- ploited. The Andreev reflection is complete in the NS interface even for the clean interface and without the Fermi wave vector mismatch, which is opposite to the case of ferromagnet-superconductor interface. The important role played by the dynami- cal polarization of the degrees of freedom to the 0− π transition and the generation of φ0-junction are mentioned in this paper. The scattering by the charged impurity in the substrate affects the transport properties in the bulk as well as the valley relax- ation, which can be taken into consider by the macroscopic wave function. In short junction limit, the approximated results about the Andreev level and free energy are also discussed. Beside the low-energy limit of the tight-binding model, the finite-size effect need to be taken into account as long as the spacing model is much larger than the superconducting gap. 1 Introduction In this paper, we consider a silicene-based superconductor-normal-superconductor Josephson junction with the silicene nanoribbons lies along the kx (zigzag) direction in a two-terminal geometry as shown in Fig.1. The silicene is divided into three parts, the middle part is deposited on the SiO2 substrate, and thus with the dielectric constant ǫ = 2.45 (ǫSiO2 = 3.9), while the left- and right-side parts are deposited on the conventional superconductor electrodes with the s-wave superconductors realized by the superconducting proximity effect. The perpendicular electric field and off-resonance circularly polarized light are applied on the middle normal region. We set a finite chemical potential µn in the middle region by slightly doping, while at the superconductive regions, the chemical potential µs required by the high carriers density is much larger than the Fermi wave-vector kF and the Dirac-mass mησz τz , which has kF = D − U)2/vF with U the electrostatic potential induced by the doping or gate voltage in the superconducting region, which breaks the electron-hole symmetry, and lifts the zero-model (between the lowest conduction band and the highest valence band) above the Fermi level (imaging the zero Dirac-mass here)[1]. The U has been experimentally proved to be valid in controlling the phase shift of the φ0-junction as well as the 0 − π transition[2] like the bias voltage. Note that the Dirac-mass here is related to the band gap by ∆ = 2mησz τz . Thus we can know that µs & p(mησz τz D − U)2 and the incident angle is larger than the transmission s − (mησz τz pµ2 D D ∗[email protected] 1 angle due to the relation kF sinθn = µssinθs[3] where θn is the incident angle from the normal region and θs is the transmission angle in the superconducting region. Furthermore, we can estimate the transmission angle as θs ≈ 0 to obtain the zero scattering angle and the smooth propagation. The tight-binding Hamiltonian of the monolayer silicene reads H =vF (ητxkx + τyky) + ηλSOCτzσz + aλR2ητz(kyσx − kxσy) ∆ 2 − E⊥τz + λR1 2 (ησyτx − σxτy) + Mssz + Mc, (1) where E⊥ is the perpendicularly applied electric field, a = 3.86 is the lattice constant, ∆ is the buckled distance in z-diraction between the upper sublattice and lower sublattice, σz and τz are the spin and sublattice (pseudospin) degrees of freedom, respectively. η = ±1 for K and K' valley, respectively. Ms is the spin-dependent exchange field and Mc is the charge- dependent exchange field. λSOC = 3.9 meV is the strength of intrinsic spin-orbit coupling (SOC) and λR2 = 0.7 meV is the intrinsic Rashba coupling which is a next-nearest-neightbor (NNN) hopping term and breaks the lattice inversion symmetry. λR1 is the electric field- induced nearest-neighbor (NN) Rashba coupling which has been found that linear with the applied electric field in our previous works[6, 4, 5, 7, 8], which as λR1 = 0.012E⊥. For circularly polarized light with the electromagnetic vector potential has A(t) = A(±sin Ωt, cos Ωt), where ± denotes the right and left polarization, respectively. Due to the perpendicular electric field E⊥ and the off-resonance circularly polarized light which with frequency Ω > 1000 THz, the Dirac-mass and the corresponding quasienergy spectrum of the normal region are mησz τz Dn =ηλSOCszτz − εn =sr(q2v2 ∆ 2 E⊥τz + Mssz + Mc − ηv2 F A Ω, (2) F k2 + (mησz τz Dn )2 + sµn)2, respectively, where the dimensionless intensity A = eAa/ is in a form similar to the Bloch frequency, and s = ±1 is the electron/hole index, and the subscript e and h denotes the electron and hole, respectively. The off-resonance circularly polarized light results in the asymmetry band gap in two valleys (see Ref.[8]) and breaks the time-reversal symmetry in the mean time, and thus provides two pairs of the different incident electrons that may leads to the josephson current reversal due to the valley-polarization. The index "n" in above equations is to distinct them from the Dirac-mass and quasienergy spectrum in superconducting regions, which are mησz τz D =ηλSOCszτz + Mssz + Mc, ε =sr(q2v2 F k2 + (mησz τz D )2 + sµs)2 + ∆2 s, (3) where ∆s is the superconducting gap (complex pair potential) which obeys the BCS relation and can be estimated as ∆s = ∆0tanh(1.74pTc/T − 1)eiφ/2 (here we only consider the right superconducting lead)[9, 10] with φ the macroscopic phase-difference between the left and right superconducting leads, ∆0 the zero-temperature energy gap which estimated as 0.001 eV[1] here and Tc the superconducting critical temperature which estimated as 5.66 × 10−4 eV. The superconducting gap is often used to compared with the excitation gap in normal region, thus we simply use ∆s to replace the ∆s +mηszτz in below. Obviously, the quasienergy spectrum here is distinct from the one obtained by the Floquet technique in low-momentum limit as presented in the Refs.[11, 8]. Note that since the exchange field considered here (Ms = Mc = 0.0039 eV), the critical electric field (for the zero Dirac-mass) is no more 0.017 eV/A, but . 0.051 eV/Afor small light-intensity. D 2 2 Andreev bound state The Andreev reflection (AR) happen in the middle normal region (or insulating barrier) for the bias voltage smaller than ∆s and U (gate voltage) which been applied across the junction[1]. We consider the quasi-normal incidence of the electrons from normal region, i.e., with con- stant ky and non-constant kx. The Andreev bound state, which is common in the d-wave superconductor[12], exist in the middle normal region unless there is a band insulator, that can be realized by control the electric field and the light field which takes effect in the middle normal region. For SNS Josephson junction, in contrast to the normal reflection (NR; like the elas- tic cotunneling (ECT) process in ferromagnet (normal)/superconductor/ferromagnet (normal) (NSN) junction) whose reflection and transmission coefficients obeys the relation re2+te2 = 1, while the AR process obeys rh2 +te2 = 1, where rh is the reflection coefficient of the reflected hole from conduction band (retro-like) or valence band (specular), and te is the transmission coefficient of the electron-like quasiparticle. The ECT relys on the coherent superposition of states through the, e.g., quantum dot orbit in the parallel configuration between two electrodes. For the AR process, here the reflection coefficient reads rh = ei(ηθn−φ)kF cosθn ikF sinβcosθn + εncosβ , (4) where θn = asin µssinθs can be obtained by Eq.(2) kF and s = ±1 is the electron(+1)/hole(-1) index and the Fermi wave vector kF = p(εn − sµn)2 + (mηsz τz vF D )2 . (5) We can see that the reflection probability is rh2 and it equas to one for vertical incidence θn = 0 and β = 0 (i.e., εn = ∆s). In Fig.1, we present the Andreev reflection probability rh2 versus the phase difference with different superconducting gap ∆s and electric field. We can obtained that, in the case of εn > ∆s, the Andreev reflection probability decrease with the increase of Dirac-mass mηsz τz or ∆s. The factor β here is associate with the relation between the supra-gap or subgap excitation energy and the superconducting gap Dn β =(cid:26) −iacosh εn ∆s acos εn ∆s , , εn > ∆s, εn < ∆s, (6) for a propagating scattering waves and a evanescent scattering waves, respectively, and we have eiβ = εn ∓pε2 ∆s n − ∆2 s , (7) where the sign " ∓ " takes " + " for εn < ∆s, and takes " − " for εn > ∆s. Then the transmission coefficient can be obtained as te =s1 − kF cosθ2e2(Imφ−Im(ηφ)) εncosβ + ikF sinβcosθ2 , (8) where Im denotes the imaginary part. Both the reflection coefficient and transmission coefficient implies that the AR is complete in the NS interface (as depicted in Fig.1) even when the interface is clean (without impurity) and without the Fermi wave vector mismatch, that's opposite to the case of ferromagnet-superconductor interface[13]. In N-S junction, to ensure the AR to occur, the excitation energy in the normal region must be smaller than the band gap in the superconducting regions, which including the proximity-induced superconducting 3 gap ∆s[3], i.e., only the subgap excitation energy is allowed, which would leads to the coherent superposition of the electron and hole excitations in small excitation energy. In this case of εn < ∆s, electrons can enter into the superconductor lead only by forming a Cooper pair which consisted of a electron with up-spin in K valley and a electron with up-spin in K' valley, or it can't penetrate into the superconductor lead due to the small excitation energy[13]. Further, when εn > µn + mD the AR is specular (interband), while it's retro-like (intraband) for εn < µn + mD. For the former one, the large excitation energy also results in the thermal transport between two superconducting leads by the propagating model[14] and it's related to the Fermi distribution term tanh(εn/2kBT ), while for the latter one, it contributes only to the localized model. Then we focus on the dispersion of the Andreev bound level, which is εA = s ∆s√2s1 − A(C − cosφ) + szpB2[A2 + B2 − (C − cosφ)2] A2 + B2 , (9) where we have use the definitions A =C1C2 + , f1 F k2 (S1S2( f2 f1 + 1)( f4 + 1 2 ) S1C2( f3 2f1 f3 − 1)) yf2/f4 + 1q−f4 4q2v2 f3 q−2v2 F k2 C1S2( f4 − q(2v2 yf1)/f3 + 1pf2/f1 yf2/f4 + 1p−f4/f3q−(2v2 yf1/f3 + 1q f2 yf2)/f4 + 1p−f4/f3 yf1)/f3 + 1pf2/f1 ) + (−1) · Θ(−εn + µn + mηsz τz 2f2 − 1 2) yS1S2 2v2 F k2 F k2 ] D F k2 (S1S2(f2/f1 − 1)(f4/f3 + 1)) yf2/f4 + 1p−f4/f3q−(2v2 F k2 yf1)/f3 + 1pf2/f1 , B = C = F k2 F k2 q−(2v2 q2v2 − [1 · Θ(εn − µn − mηsz τz × 4q2v2 F k2 D with the Heaviside step function Θ which distinguish the two kinds of AR: retroreflection and specular AR, and thus makes this expression valid for both of these two case. The wave vectors f1 ∼ f4 and parameters C1, C2, S1, S2 are defined as , (10) (11) f1 = mησz τz f3 = εn − mησz τz Dn + εn + µs, f2 = mησz τz Dn + µs, f4 = mησz τz Dn + εn − µs, Dn − εn + µs, F − k2 y), F − k2 y), F − k2 y), F − k2 y). C1 = cos(Lqf1f3/2v2 C2 = cos(Lq−f2f4/2v2 S1 = sin(Lqf1f3/2v2 S2 = sin(Lq−f2f4/2v2 The x-component of the wave vector for the electron channel and hole channel, kxe and kxh, are incorporated in the above wave vectors, specially, the electron and hole wave vectors here are both complex, which implies the inclusion of the subgap solutions with the evanescent scattering waves[1], and it has ∆s 2cosβ = 1 for εn < ∆s. Note that we only consider the 2εn dc Josephson effcet in the thermodynamic equilibrium state here rather than the ac Josephson effcet which with time-dependent phase difference (e.g., by a time-dependent bias voltage). 4 There exist a critical angles for AR process. When the incident angle of the scattering wave excess this critical angle, it becomes exponentially-decayed. The critical angle for AR process is θAR = asin , (12) f2(−f4) f1f3 when the quasienergy and chemical potential µn are much smaller than the Dirac-mass mηsz τz Dn , the above critical angle can be simplified as θAR = asin (−f4) , which is consistent with the result f3 of Ref.[15]. While for the critical angle of ECT in the NSN junction, it is related to the electron/hole index: s = 1 for the charge channel which corresponds to the parallel configuration[15], and s = −1 for the hole channel which corresponds to the antiparallel configuration. The critical angle of ECT can be written as sinθECT = f3δs,1 + f1δs,−1 f1 . (13) While for the case of heavily doping in the middle normal region, that's µn ≫ εn, then the AR is dominated by the retro one. In this case, when the chemical potential is comparable with the Dirac-mass, the AR process is suppressed by the destructive interband interference[16] among the Cooper pairs due to the large Dirac-mass (and thus the large band gap) which brings a large dissipative effect and reduce the AR. The minimum free energy in this SNS planar junction is φ- and temperature-dependent, and local st the 0 or π-junction. With the defined incident angle θs for the quasiparticle injected from the one of the superconducting electrodes to the middle normal region, the minimum free energy with AR process can be written as E(φ, T ) = −kBT XηszτzZ π/2 −π/2 Dln[2cosh( εA 2kBT )]cosθndθn. (14) The reversal of the Josephson effect will be rised by the valley, spin, and pseudospin po- larization, which with dramatic dipole oscillation between the two components induced by the off-resonance circularly polarized light as we have discussed[8], and results in a nonzero center- of-mass (COM) wave vector, just like the Josephson current realized by a ferromagnetic middle silicene (in superconductor/ferromagnet/superconductor ballistic Josephson junction[10]). It's found that the oscillation of valley polarization is related to the carrier-phonon scattering due to the photoexcitation which has a relaxation time in picosecond range[17] (since the frequency of light is setted in the terahertz range) and larger than that of the electron-eletron scattering. On the other hand, since the silicene in normal region is been deposited on a SiO2 substrate, the relaxation of the valley-polarization is also associated with the screened scattering by the charged impurities within the substrate. It's found that for a certain concentration of the impurity and the vertical distance to the middle silicene layer d, the relaxation time can be obtained, which is in a dimensionless form nimpεn 3v2 F 1 + Nf τ =  2πe2 ǫ0ǫq 2πe2 ǫ0ǫq Π(q, ω)!2  , (15) where ǫ0 and ǫ are the vacuum dielectric constant and background dielectric constant, respec- tively. Nf is the number of the degenerate which can be treated as Nf = gsgv = 4 here. q is the scattering wave vector and Π(q, ω) is the dynamical polarization within the random-phase- approximation (RPA) which contains the screening effect by the high energy state with large charge density of state (DOS) D = Wf1/(πvF ) where W is the width of the silicene ribbron. For the case that the screening is ignored, the relaxation time has a simply relation wih the 5 distance to the impurity, τ ∝ ed/ℓ[18], where ℓ = 0.47 Afor silicene. For ballistic Josephson junction, the diffusive effect is not considered, however, due to the existence of the screened (or unscreened) charged impurities in the substrate, the diffusive effect (e.g., to the conductivity or the valley-polarization) need to be taken into accout. Note that we don't consider the edge states in the wave vector space, however, if the middle normal region is replaced by a ferromagnetic one, e.g., by applying the out-of-plane ferro- magnetic exchange field on both the upper edge and the lower edge of middle region, the junction will always be the 0(π)-junction (unless the length of the upper edge and lower edge is unequal) when without applying the electric field or off-resonance light. That's because the out-of-plane ferromagnetic exchange field won't gap out the edge models (when without the Rashba-coupling) while the in-plane ferromagnetic exchange field[19, 20] which can easily gap out the gapless edge models even in the absence of Rashba-coupling. In such case, the Curie temperature is required to much larger than the superconducting critical temperature to pre- vent the dramastic variety of the dipole polarization. If we apply an in-plane ferromagnetic exchange field, the gapless edge model will be gapped out and the time revesal invariance will be broken, and thus the edge states-supported Josephson effect disappear. While for the out-of- plane antiferromagnetic exchange field, which may leads to the single valley characteristic, both the edge and bulk state support the Josephson effect in the normal region[21]. In this case, the Andreev level of the upper edge and lower edge can be easily obtained just by inversing the re- lated degrees of freedom, due to the chiral character-induced opposite phase between the upper edge and lower edge, e.g., the helical edge model in quantum spin Hall phase which with up- and down-spin carriers flow toward opposite directions in each edge; while for the chiral edge model in the quantum anomalous Hall phase, the edge models with up- and down-spin carriers flow toward the same direction in each edge. Further, for the a ferromagnetic middle region, the spin-rotation symmetry no more exist, and thus leads to the anomalous conductance[13]. In short junction case (L ≤ 50 A), the Andreev level can be approximatively obtained as ε∗AR = ∆scos(cid:20) 1 2(cid:18)φ − L( f3 vF + f2 vF )(cid:19)(cid:21) , (16) which can be further simplified as ε∗AR = ∆scos(φ/2) in the case that the εn is small and that's consistent with the result of Refs.[22, 23], where the transmission coefficient equals 1 here due to the partial Klein tunneling of Cooper pairs which happen at zero Dirac-mass[24] (while the AR doesn't requires the zero Dirac-mass). 3 Results and discussion Due to the presence of the degrees of freedom η, sz, τz and the electron/hole index s in the above expression of the Andreev bound level, it should has 24 = 16 levels under a certain electric field and light field, however, there may be less levels due to the degenerative case. The Josephson current is a nonlocal supercurrent carried by the cooper pairs in the superconducting leads (while it's by quasiparticles in the middle normal region), it can be found in such SNS junction, and its slope is affected by the dynamical polarizations of the degrees of freedom as mentioned above[8], for the case of, e.g., pεn − m+++ 6=pεn − m−−−Dn . For simplicity, we only present the result of one of the levels, where all the degrees of freedom are takes 1. For a Cooper pair penetrated into the bulk state, the macroscopic wave function is affected by the scattering process by the charged impurity, and thus has Ψ(x) ∝ e−x/ξed/ℓeiφ/2, where ξ is the superconducting coherance length and x < L is the mean free path in the bulk region. The Josephson effect here only consider the low-temperature condition (∼ Tc), which is dominated by the elastic scattering, while at higher temperature, the rised inelastic scattering may leads Dn 6 to the switching of the Fermi parity[23], and the frequency-dependent noise which is induced by the current-current correlation in nonequilibrium Josephson setup leads to the 4π period of the Josephson current due to the existence of Majorana bound states. That also implies that the dissipation (related to the Dirac-mass) plays a important role during the quasiparticle transportation. For simulation, we set the parameters as follows: The frequency of the off-resonance light is setted larger than 3000 THz which is much higher than the critical value. The critical value here is 3t = 4.8 eV= 1200 THz for silicene. In the case that frequency is 3000 THz, the dimensionless intensity is A = 0.3 which is much larger than the SOC parameter λSOC. The length of the normal region L is much shorter than the superconducting coherence length L ≪ vF , i.e., √3 L ≪ 5350 A, where we estimate vF = 2 at = 5.35 here. The y-component of the momentum as well as the off-resonance circularly polarized light (which we only use the right-polarization from now on) is conserved due to the translational invariance, thus it's a good quantum number in the computation, and we set ky = 2 meV. k is in an order of 0.01 eV here. While the detail setting of the parameters are labeled in the each following plot. ∆0 Fig.3 depicts the dispersion of the Andreev bound level for the AR process in 0 state (we don't consider the breaking of inversion symmetry by the Rashba-coupling here since it's very small). We can easily see than the period of Andreev level is 4π, which obeys the generate relation εAR ∝ cos(φ/2)[21], while the period of εAR/∆s is 2π as shown in Fig.4, which is consistent with the results of Ref.[10]. We detect the effects of the electric field, off-resonance circularly polarized light, chemical potential, and length of the normal region to the Andreev bound level numerically. We found that, in condition: L = 2000 A, the slope of the Andreev level is always negative in the range [0, π) and positive in the range [π, 2π], i.e., in the 0(π)- junction (which depends on the product between the sign and slope of the Andreev level). From Fig.3(b), we can see that the amplitude of Andreev level exhibits non-monotonous change with the increase of length L. In this case, when we increase the light-intensity parameter A, the slope of the Andreev level changes, and we found the A should be < 0.7 in the condition: E⊥ = 0.034 eV, µn = 2 eV, L = 2000 A, for a available Andreev level (or the slope is nearly vanish for A ≥ 0.7). The anomalous Josephson effect as well as the φ0-junction can be found in Fig.3(b). From the blue dash-dot line (L = 1500 A) and green line (L = 1200 A) in Fig.3(b), we can see that the sign reversal happen even in the range [0, π), this phenomenon emerges also in yellow line in Fig.3(a) which with E⊥ = 2 eV. While for other levels, the sign change can happen only at φ = π. In Fig.4(a), we present the Andreev level εAR/∆s whose period is 2π[9]. The slope of the blue line, which corresponds to the condition: E⊥ = 0.01 eV, µn = 0.03 eV, A = 0.03, L = 2000 A, is opposite to the green line which corresponds to the condition: E⊥ = 0.26 eV, µn = 0.03 eV, A = 0.02, L = 2000 A. That implies the occurence of 0 − π transition which with the reversed current compared to the usual 0-junction. Thus we can obtain that the change of electric field and the intensity of light is valid for the generation of 0 − π transition because of the sublattice degree of freedom and the valley degree of freedom consider here, respectively. Thus if the subalttice degree of freedom of electric field is not considered, the variety of electric field won't support the 0− π transition or the transition to the φ0-junction, which is consistent with the result in Ref.[21, 9]. In the absence of the electric field, off-resonance light and the antiferromagnetic exchange field, it is always be the 0-junction due to the presence of time-reversal invariance (and chiral symmetry), and the Josephson supercurrent vanishes at φ = nπ (n is integer) in this case. However, the broken symmetry or chiral leads to the nonzero supercurrent at φ = nπ and the φ0-junction which can also be implemented by a external magnetic field in nanowire quantum dots-based junction (SQUID)[2]. Fig.5 shows the free energy versus phase difference, we can see that change of electric field 7 or the length of normal region are also valid for the generation of φ0 transition, which with a phase shift φ0 ∈ (0, π) (or ∈ (π, 2π)) for the minimum free energy. The period of free energy is 2π, which is consistent with the theory[25] and experimental[26] results, and it's similar to the Josephson current whose period is 2π in thermodynamic equilibrium state[27], and becomes 4π in nonequilibirum state due to the majorana bounding[28]. Note that the free energy here is related to the width W of the nanoribbon due to the finite-size effect. We find that the 0-π transition and the φ0-junction can be realized by changing the length of the normal region or the electric field. The approximated Andreev level for short length L is shown in Fig.6 according to Eq.(16), where the 0 − π transition (see the black line and gray line) and the φ0-junction can be implemented by changing the electric field, A, and the length L. Phenomenologically, through Fig.6, the Eq.(16) can be rewritten as ε∗AR = ∆scos(cid:2) 1 2(φ + φ0)(cid:3) with φ0 ∈ [0, 2π), the φ0 here is controlled by the variables presented in the plot, and this expression is similar to that of anomalous switching current[2]. The usual current-phase relation can be deduced from the derivative of ε∗AR as J = − 2e 2(φ + φ0)(cid:3) in low-temperature limit. Note that these results (including the anomalous Josephson effects) are all base on the first-order perturbation theory with the perturbation term V which couples the left and right superconducting leads. sin(cid:2) 1 The free energy of the approximated Andreev level in Fig.6 is presented in Fig.7, where we show the free energy in one period, 2π. It's obvious that the free energy exhibits the characteristic of the φ0-junction with the minimum free energy (i.e., the maximum value of the -E(φ, T ) in the plot) change their positions under different parameters. By comparing the black, purple, yellow lines in Fig.7(a), we can obviously see the phase shift induced by the variety of length of normal region. While for the temperature near (but larger than) the critical one Tc, as shown in Fig.7(b), it doesn't affects the existence of 0(π)-junction and the time reversal invariance, and thus the minimum free energy is always localed at φ = 3.5. For temperature lower than the critical one Tc as shown in Fig.6(c), the minimum free energy localed at φ = 5.24 for temperature lower than 0.6Tc, while it localed at φ = 5.48 for temperature larger or equals 0.6Tc as shown in Fig.6(b). In contrast to the Rashba-coupling-induced helical p-wave spin- triplet superconductor, the Josephson current of conventional s-wave superconductor saturates at low-temperature (e.g., at T < 0.6Tc)[29], and the free energy doesn't change gradually with the variational temperature. 4 Conclusion In the interface between the normal nanoribbon (strip) region and the superconducting region, the AR is related to the width W (the number of edge states along the armchair direction) of the ribbon: For W is even, the AR supressed by the opposite pseudospin degree of freedom between first sublattice and last subalttice in the ky direction, while for W is odd, the AR is allowed. In this strip scenario, the model spacing between bands above the Dirac level depends on W and the NN hopping when consider the finite-size effect, the model spacing is δ∆ = 3πt/(3W − 2) for W is even and δ∆ = πt/(W − 1) for W is odd. The band structure of the zigzag silicene nanoribbon in strip geometry is shown in Fig.8, where the model spacing δ∆ is indicated in the inset. The finite-size effect here is important in large energy range, but it can be ignored for low-energy limit as the tight-binding model depicted in Sec.1. In this paper we investigate the Josephson effcet in superconductor-normal-superconductor junction base on the doped silicene with a dc Josephson current in zero voltage. We found that the dynamical polarizations of the degrees of freedom mentioned above can induce the 0 − π transition and the emergence of the φ0-junction. For example, the change of the electric field or off-resonance circularly polarized light may induce the 0−π transition by the pseudospin degree of freedom or valley degree of freedom, respectively; the interaction between antiferromagnetic exchange field and the SOC may induce the 0 − π transition by the valley polarization[21]; the 8 interaction between the internal exchange field and Josephson superconducting current may induce the φ0-junction when the normal region is noncentrosymmetric[30]. References [1] Rainis D, Taddei F, Dolcini F, et al. Andreev reflection in graphene nanoribbons[J]. Physical Review B, 2009, 79(11): 115131. [2] Szombati D B, Nadj-Perge S, Car D, et al. Josephson ? 0-junction in nanowire quantum dots[J]. Nature Physics, 2016, 12(6): 568. [3] Linder J, Yokoyama T. Superconducting proximity effect in silicene: Spin-valley-polarized Andreev reflec- tion, nonlocal transport, and supercurrent[J]. Physical Review B, 2014, 89(2): 020504. [4] Wu C H. Tight-binding model and ab initio calculation of silicene with strong spin-orbit coupling in low- energy limit[J]. arXiv preprint arXiv:1804.01695, 2018. [5] Wu C H. Interband and intraband transition, dynamical polarization and screening of the monolayer and bilayer silicene in low-energy tight-binding model[J]. arXiv preprint arXiv:1805.07736, 2018. [6] Wu C H. Geometrical structure and the electron transport properties of monolayer and bilayer silicene near the semimetal-insulator transition point in tight-binding model[J]. arXiv preprint arXiv:1805.00350, 2018. [7] Wu C H. Integer quantum Hall conductivity and longitudinal conductivity in silicene under the electric field and magnetic field[J]. arXiv preprint arXiv:1805.10656, 2018. [8] Wu C H. Anomalous Rabi oscillation and related dynamical polarizations under the off-resonance circularly polarized light [J]. arXiv preprint arXiv:1806.03592, 2018. [9] Zhou X, Jin G. Light-modulated 0-π transition in a silicene-based Josephson junction[J]. Physical Review B, 2016, 94(16): 165436. [10] Annunziata G, Enoksen H, Linder J, et al. Josephson effect in S/F/S junctions: Spin bandwidth asymmetry versus Stoner exchange[J]. Physical Review B, 2011, 83(14): 144520. [11] Lpez A, Scholz A, Santos B, et al. Photoinduced pseudospin effects in silicene beyond the off-resonant condition[J]. Physical Review B, 2015, 91(12): 125105. [12] Kashiwaya S, Tanaka Y. Tunnelling effects on surface bound states in unconventional superconductors[J]. Reports on Progress in Physics, 2000, 63(10): 1641. [13] De Jong M J M, Beenakker C W J. Andreev reflection in ferromagnet-superconductor junctions[J]. Physical review letters, 1995, 74(9): 1657. [14] Beenakker C W J. Colloquium: Andreev reflection and Klein tunneling in graphene[J]. Reviews of Modern Physics, 2008, 80(4): 1337. [15] Linder J, Zareyan M, Sudb? A. Spin-switch effect from crossed Andreev reflection in superconducting graphene spin valves[J]. Physical Review B, 2009, 80(1): 014513. [16] Golubov A A, Brinkman A, Tanaka Y, et al. Andreev spectra and subgap bound states in multiband superconductors[J]. Physical review letters, 2009, 103(7): 077003. [17] Kumar S, Anija M, Kamaraju N, et al. Femtosecond carrier dynamics and saturable absorption in graphene suspensions[J]. Applied physics letters, 2009, 95(19): 191911. [18] Boross P, Plyi A. Valley relaxation in graphene due to charged impurities[J]. Physical Review B, 2015, 92(3): 035420. [19] Rachel S, Ezawa M. Giant magnetoresistance and perfect spin filter in silicene, germanene, and stanene[J]. Physical Review B, 2014, 89(19): 195303. [20] An X T, Zhang Y Y, Liu J J, et al. Spin-polarized current induced by a local exchange field in a silicene nanoribbon[J]. New Journal of Physics, 2012, 14(8): 083039. 9 [21] Zhou X, Jin G. Silicene-based and 0 Josephson junctions[J]. Physical Review B, 2017, 95(19): 195419. [22] Kulik I O. Zh. ksp. Teor. Fiz. 57, 1745 1969 Sov. Phys[J]. JETP, 1970, 30: 944. [23] Fu L, Kane C L. Josephson current and noise at a superconductor/quantum-spin-Hall- insulator/superconductor junction[J]. Physical Review B, 2009, 79(16): 161408. [24] Missault N, Vasilopoulos P, Vargiamidis V, et al. Spin-and valley-dependent transport through arrays of ferromagnetic silicene junctions[J]. Physical Review B, 2015, 92(19): 195423. [25] Radovi´c Z, Dobrosavljevi?-Gruji? L, Vuji?i? B. Coexistence of stable and metastable 0 and states in Josephson junctions[J]. Physical Review B, 2001, 63(21): 214512. [26] Baselmans J J A, Heikkil? T T, Van Wees B J, et al. Direct observation of the transition from the conventional superconducting state to the state in a controllable Josephson junction[J]. Physical review letters, 2002, 89(20): 207002. [27] Kwon H J, Sengupta K, Yakovenko V M. Fractional ac Josephson effect in p-and d-wave superconductors[J]. The European Physical Journal B-Condensed Matter and Complex Systems, 2004, 37(3): 349-361. [28] Wu C H. Time Evolution and Thermodynamics for the Nonequilibrium System in Phase-Space[J]. arXiv preprint arXiv:1711.00547, 2017. [29] Asano Y, Yamano S. Josephson effect in noncentrosymmetric superconductor junctions[J]. Physical Review B, 2011, 84(6): 064526. [30] Buzdin A. Direct coupling between magnetism and superconducting current in the Josephson 0 junction[J]. Physical review letters, 2008, 101(10): 107005. 10 Fig.1 (cid:54)(cid:38) (cid:54)(cid:76)(cid:50)(cid:21) (cid:54)(cid:38) (cid:36) (cid:40) (cid:78)(cid:91) (cid:54) (cid:49) (cid:54) (cid:78)(cid:92) Figure 1: (Color online) Schematic of the silicene-based superconductor-normal-superconductor Josephson junc- tion (SNS) with the silicene nanoribbons lies along the kx (zigzag) direction. The middle normal region is deposited on the SiO2 substrate, while the left- and right-side parts are deposited on the conventional supercon- ducting electrodes (SC) with the s-wave superconductors realized by the superconducting proximity effect. The electric field (straight arrow) and off-resonance circularly polarized light (broken arrow) are applied perpendic- ularly on the middle normal region. The process of the Andreev reflection is shown, where the solid black circle stands electron and the hollow circle stands hole and with the short arrows stands the direction od spin. We can see that an electron with down-spin from the conduction band of the normal region incident into the NS interface and reflected as a hole with up-spin from the valence band (specular Andreev reflection), and forms a Cooper pair (within dash-line circle) together with an electron with up-spin generated by the reflected hole. 11 Fig.2 2 h r 1.0 0.8 0.6 0.4 0.2 0.0 E (eV), s/m sz zD 0.034 1.02 1.540 1.02 2.000 1.02 0.034 1.01 0.034 1 0.034 0.9 0.034 0.8 0.034 0.7 0 1 2 3 4 5 6 Figure 2: (Color online) Andreev reflection probability rh2 versus the phase difference with different super- conducting gap ∆s and electric field. The setting of the parameters are: A = 0.3, µn = 0.2 eV, L = 50 A. The superconducting gap ∆s here is setted as always smaller than εn thus β = −iacoshεn/∆s. The Dirac-mass in the normal region mηsz τz are 0.2108 eV, 0.5572 eV, 0.663 eV for E⊥ = 0.034 eV, E⊥ = 1.54 eV, E⊥ = 2 eV, respectively. Dn 12 Fig.3 (a) 10-3 2.0 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 -2.0 R A (b) E (eV),A, n(eV),L(Ang) 0.034,0.3,2,2000 1.54,0.3,2,2000 2,0.3,2,2000 2,0.5,2,2000 0.034,0.3,0.2,2000 0 2 4 6 8 10 12 L=50 Figure 3: (Color online) Andreev bound level versus phase difference φ. The dash-line indicate the half-period at φ = 2π. The spin-dependent exchange field Ms and the charge-dependent exchange field Mc are all setted as 0.0039 eV. For each curve, the corresponding setting of the parameters are labeled in the plot. 13 Fig.4 (a) s / R A 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 (b) 1.0 0.8 s 0.6 / R A E (eV),A, n(eV),L(Ang) 0.01,0.03,0.02,2000 0.02,0.03,0.02,2000 0.26,0.03,0.02,2000 0.01,0.03,0.03,2000 0 2 4 6 8 10 12 L(Ang) L=2000 L=1500 L=1000 L=990 L=500 L=100 L=50 0.4 0.2 0.0 0 2 4 6 8 10 12 Figure 4: (Color online) εAR/∆s versus versus the phase difference φ. For each curve, the corresponding setting of the parameters are labeled in the plot. 14 Fig.5 (a) ) T , ( E 0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 -0.05 -0.10 -0.15 (b) 0.4 0.3 ) 0.2 ( E 0.1 0.0 -0.1 E (eV),A, n(eV),L(Ang) 0.034,0.3,2,2000 1.54,0.3,2,2000 2,0.3,2,2000 0.034,0.5,2,2000 0.034,0.3,0.2,2000 0 2 4 6 8 10 12 L(Ang) L=2000 L=1800 L=1500 L=1000 L=500 L=50 0 2 4 6 8 10 12 Figure 5: (Color online) Free energy versus the phase difference. The corresponding setting of the parameters are labeled in the plot. Here we set width of ribbon as W = 126. 15 Fig.6 s / * R A 1.0 0.5 0.0 -0.5 -1.0 E (eV),A,L(Ang) 0.034,0.3,50 0.5,0.3,50 1.54,0.3,50 2, 0.3,50 0.034,0.5,50 0.034,0.3,40 0.034,0.3,30 0 2 4 6 8 10 12 Figure 6: (Color online) Andreev bound level versus phase difference according to the approximation result of Eq.(16) with short junction length. The temperature setted here is 0.5Tc. The corresponding setting of the parameters are labeled in the plot. 16 Fig.7 (a) ) T , ( E 0.35 0.30 0.25 0.20 0.15 0.10 0.05 (b) ) ( E 0.070 0.065 0.060 0.055 0.050 0.045 0.040 0.035 0.030 0.025 (c) ) T , ( E 0.030 0.025 0.020 0.015 0.010 0.005 0.000 -0.005 -0.010 E (eV),A, n(eV),L(Ang) 0.034,0.3,2,50 1.54,0.3,2,50 2,0.3,2,50 0.034,0.3,0.5,50 0.034,0.5,2,50 0.034,0.3,2,40 0.034,0.3,2,30 0 2 4 6 T/Tc 1.5 1.4 1.3 1.2 1.1 1 0 2 3.5 4 6 T/Tc 0.8 0.6 0.4 0.3 0.2 0.01 0 2 4 5.24 5.48 6 Figure 7: (Color online) Free energy obtained by the approximated Andreev level in Eq.(16). where we show the free energy in one period. The temperature is setted as 1.5 Tc. The corresponding setting of the parameters are labeled in the plot. (b)-(c) show the free energy with different temperatures, while the other parameters as setted as: E⊥ = 0.034 eV, A = 0.3, µn = 0.2 eV, L = 50 A. The dash-line in (b)-(c) indicate the position of minimum free energy. 17 Fig.8 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 -2 0 2 0.75 4 6 8 10 1.25 12 14 Figure 8: Band structure of the zigzag silicene nanoribbon in a strip geometry. The energy E is in unit of t. The middle band at E = 0 is the zero-model edge (Dirac level) which localed in the range [0.75π, 1.25π]. The zero-model level is at E = 0 for zero chemical potential and zero electrostatic potential which guarantees the electron-hole symmetry. Inset: the enlarged view near the valley K at kxa = 0.75π. 18
1803.01656	2	1803	2018-04-26T14:25:19	Electric-field modification of interfacial spin-orbit field-vector	[ "cond-mat.mes-hall" ]	Current induced spin-orbit magnetic fields (iSOFs), arising either in single-crystalline ferromagnets with broken inversion symmetry1,2 or in non-magnetic metal/ferromagnetic metal bilayers3,4, can produce spin-orbit torques which act on a ferromagnet's magnetization,thus offering an efficient way for its manipulation.To further reduce power consumption in spin-orbit torque devices, it is highly desirable to control iSOFs by the field-effect, where power consumption is determined by charging/discharging a capacitor5,6. In particular, efficient electric-field control of iSOFs acting on ferromagnetic metals is of vital importance for practical applications. It is known that in single crystalline Fe/GaAs (001) heterostructures with C2v symmetry, interfacial SOFs emerge at the Fe/GaAs (001) interface due to the lack of inversion symmetry7,8. Here, we show that by applying a gate-voltage across the Fe/GaAs interface, interfacial SOFs acting on Fe can be robustly modulated via the change of the magnitude of the interfacial spin-orbit interaction. Our results show that, for the first time, the electric-field in a Schottky barrier is capable of modifying SOFs, which can be exploited for the development of low-power-consumption spin-orbit torque devices.	cond-mat.mes-hall	cond-mat	Electric-field modification of interfacial spin- orbit field-vector L. Chen1, M. Gmitra2, M. Vogel1, R. Islinger1, M. Kronseder1, D. Schuh1, D. Bougeard1, J. Fabian2, D. Weiss1 and C. H. Back1 1Institute of Experimental and Applied Physics, University of Regensburg, 93040 Regensburg, Germany. 2Institute of Theoretical Physics, University of Regensburg, 93040 Regensburg, Germany. Correspondence and requests for materials should be addressed to C.H.B. (e-mail: [email protected]). Current induced spin-orbit magnetic fields (iSOFs), arising either in single- crystalline ferromagnets with broken inversion symmetry1,2 or in non-magnetic metal/ferromagnetic metal bilayers3,4, can produce spin-orbit torques which act on a ferromagnet's magnetization, thus offering an efficient way for its manipulation. To further reduce power consumption in spin-orbit torque devices, it is highly desirable to control iSOFs by the field-effect, where power consumption is determined by charging/discharging a capacitor5,6. In particular, efficient electric-field control of iSOFs acting on ferromagnetic metals is of vital 1 importance for practical applications. It is known that in single crystalline Fe/GaAs (001) heterostructures with C2v symmetry, interfacial SOFs emerge at the Fe/GaAs (001) interface due to the lack of inversion symmetry7,8. Here, we show that by applying a gate-voltage across the Fe/GaAs interface, interfacial SOFs acting on Fe can be robustly modulated via the change of the magnitude of the interfacial spin-orbit interaction. Our results show that, for the first time, the electric-field in a Schottky barrier is capable of modifying SOFs, which can be exploited for the development of low-power-consumption spin-orbit torque devices. Experiments in non-magnetic metal (NM)/ferromagnetic metal (FM) hetero- structures, have shown that current induced interfacial non-equilibrium spin polarization due to spin-orbit interaction, originating either from the Rashba effect at the NM/FM interface3 or the spin Hall effect in the bulk of NM4, can be used to reverse the magnetization direction of ferromagnets, thus paving the way for the developement of low-power consumption magnetic memory devices. It is convenient to parameterize the current-induced non-equilibrium spin polarization as effective spin-orbit magnetic field (iSOFs)1,9, where the magnitude of iSOFs is proportional to the applied current density in the NM10. To efficiently reorientate the magnetization using low currents, materials with large charge-to-spin conversion efficiency have been searched for and remarkable progress has been made in the past few years11-16. An alternative route towards efficient manipulation of magnetization is electric-field 2 control, e.g., using a capacitor structure where the energy consumption is, in principle, determined by charging/discharging the capacitor6. Consequently, to further reduce power consumption in spin-torque devices, field- effect control of the iSOFs is highly desirable. Recently, it has been shown that electric-field control of current-induced effective magnetic fields can be realised in a magnetic topological insulator via a top gate by which the Fermi level within the Dirac bands gets shifted17. However, research on electric-field control of iSOFs acting on ferromagnetic metals, i.e., in NM/FM bi-layers, is just at its beginning18. Here, the experimental difficulty is that spin-orbit interaction at NM/FM interfaces can not be tuned effectively as the short screening length in metals shields external electric- fields18,19. Besides NM/FM bilayer systems, also single crystalline ferromagnets lacking space-inversion symmetry, feature iSOFs which can induce spin-orbit torques acting on the ferromagnet itself1,2. Recent examples are bulk (Ga,Mn)As1,2,20,21, NiMnSb22 and Fe/GaAs (001)8. For (Ga,Mn)As and NiMnSb, the iSOFs occur mainly in the bulk resulting from the ferromagnets' bulk inversion asymmetry; while for Fe/GaAs (001), the iSOFs have an interfacial origin7, thus enabling electric-field control. Here, we report electric-field control of iSOFs at the Fe/GaAs (001) interface. By applying a gate-voltage across the interface, the interfacial spin-orbit field-vector acting on Fe can be varied. 3 a VG = 0.88 V b VG = 0.1 V eVG Fe 3.2 2.8 nGaAs Fe nGaAs c ) / m V 7 0 1 ( y k ] 0 1 0 [ kx [100] [100] d e n Ei 2.4 -1.0 -0.5 VG (V) 0.0 Figure 1 Electric-field modulation of spin-orbit fields at the Fe/GaAs (001) interface. Schematic band bending of a Fe/n-GaAs Schottky diode under reverse (a) and forward (b) bias. (c) Calculated interfacial electric-field Ein obtained by solving the one-dimensional Poisson equation for the non-constant doping profile of n-GaAs (Methods). The magnitude of interfacial spin-orbit interaction is proportional to Ein. Wave-vector k dependence of Rashba- (d) and Dresselhaus (e) spin-orbit fields which dominate at the epitaxial Fe/GaAs interface with C2v symmetry. The mechanism of the electric-field modulation is shown in Fig. 1. Due to the C2v symmetry of the interface, Bychkov-Rashba- and Dresselhaus-like spin-orbit interaction emerges at the Fe/GaAs (001) interface, and the corresponding k-space dependence of the equilibrium spin-orbit magnetic fields, SOFs, heff can be written as23 4 0h  eff    1  B k k    , x k y k x y , (1) where 0 is the magnetic constant, B is Bohr's magneton, kx (ky) is the wave-vector k along the [100] ([010]) direction, and ( characterizes the strength of the Dresselhaus (Bychkov-Rashba) spin-orbit coupling. The magnitude of  and depends linearly on the interfacial electric-field Ein,  = CEin and  = CEin, (2) where C (C) is the coefficient of Dresselhaus (Bychkov-Rashba) spin-orbit interaction24. The sample investigated here is a Fe/n-GaAs heterostructure, in which Ein consists of the internal built-in electric-field (pointing from GaAs to Fe due to the larger work function of n-GaAs25) and the external electric-field with its direction depending on the polarity of gate-voltage VG. The advantage of the Fe/n-GaAs Schottky tunnel junction is that the depleted n-GaAs below the Fe layer serves as a dielectric insulator across which an external electric-field can be applied (Figs. 1a and b), e.g., biasing in reverse (forward) direction increases (decreases) Ein. Thus, by tuning the gate-voltage, both strength and direction of the SOFs can be tuned. The calculated VG dependence of Ein of the device layout is presented in Fig. 1c (Methods), which shows an essentially linear behavior within the investigated VG range. Below we demonstrate electric-field tuning of iSOFs in epitaxially grown Fe/n- GaAs samples with different Fe thickness and n-GaAs doping concentrations. All samples show consistent results and we focus here on data taken from the sample with 5 a n-GaAs IG V Fe VG M M [001] [100] [010] b ) V  ( t e s f f Vo V  12 M = 225o VG = 6 0 -6 72 8 7 6  0.88 V  0.34 V + 0.07 V f = 12.5 GHz 76 80 0H (mT) 84 vsym va-sym 0.0 -0.5 VG (V)  0.88 V + 0.07 V Fit c 12 ) V  ( t e s f f Vo V  e ) V  ( m y s - Va Va-sym 76 80 0H (mT) 6 0 -6 72 8 4 0 -4 -8 Exp. Fit Vsym d ) V  ( V 84 -1.0 f ) V  ( m y Vs 8 4 0 -4 -8  0.88 V  0.34 V + 0.07 V Fit 0 90 180 270 360 0 90 180 270 360 M (deg) M (deg) Figure 2 Electric-field modulation of induced spin-orbit fields measured by spin-orbit ferromagnetic resonance. (a) Schematic of the circuit and device structure for spin-orbit ferromagnetic resonance measurements. A microwave current j(t), generated by a signal generator, passes through a bias tee into the Fe/GaAs (001) layer, where time dependent interfacial iSOFs h(t) are generated due to the spin-orbit interaction at the Fe/GaAs interface. h(t) drives the magnetization dynamics M(t), and leads to a resistance variation R(t) due to the anisotropic magneto-resistance of Fe. The coupling between R(t) and j(t) produces a rectified d.c. voltage V. By measuring V, the magnitude of the interfacial iSOFs can be quantified. To modulate the iSOFs by an electric-field, Fe/n-GaAs samples have been used, where a Schottky barrier is formed at the interface. Thus, by adding a gate-voltage VG between Fe and n-GaAs, the magnitude of the iSOFs can be controlled. Here, M is the magnetization angle 6 with respect to the [100] orientation of GaAs. (b) d.c. voltage spectra for several gate- voltages for a [100]-oriented device measured at a microwave frequency of 12.5 GHz and M = 225°. The input microwave power is 10 mW. An offset voltage, Voffset, has been subtracted. (c) Typical Lorentzian fit of the voltage spectrum, from which the magnitudes of Vsym and Va-sym are obtained. (d) VG dependence of Va-sym and Vsym for M = 225°. Magnetization angle M dependence of Va-sym (e) and Vsym (f) under different VG for a [100]-oriented device. The solid lines in (e) and (f) are fits to Eqs. 3 and 4, respectively. 4-nm Fe thickness. Fe stripes with dimension of 10 m × 200m oriented along four main crystal directions, i.e., [100], [110], [010], and [110], which determine the prevailing k-vector directions, are fabricated by electron-beam lithography (for example, see Fig. 2a for a [100]-orientated device and Methods). To quantify the magnitude of the interfacial iSOFs, we use the spin-orbit ferromagnetic resonance technique8,20. Microwaves with a power of 10 mW and a frequency of 12.5 GHz are applied to the Fe stripe through a bias tee. The rectified d.c. voltage spectrum induced by the anisotropic magneto-resistance effect of Fe is measured by sweeping an external magnetic field H across the ferromagnetic resonance (FMR) of Fe. For measuring the gate-voltage dependence of the d.c. voltage, a gate-voltage VG is applied across the Fe/n-GaAs interface (Methods). Since the detection direction of the d.c. voltage is parallel to the microwave current, the dynamic tunneling anisotropic 7 magneto-resistance effect26 and the dynamic Hanle effect27, only detectable transverse to the microwave current, cannot contribute to the signal. Figure 2b shows the d.c. voltage spectra as a function of VG at a magnetization angle M of 225° for the [100]-oriented device measured at room temperature. The modulation of the spectra with VG is evident. As shown in Fig. 2c, each spectrum can be well fitted by combining a symmetric (Lsym = H2 / [4(HHR)2 + H2]) and an anti-symmetric Lorentzian (La-sym = 4H(HHR) / [4(HHR)2 + H2]), VVoffset = VsymLsym + Va-symLa-sym, where HR is H at FMR, H is the full width at half maximum, Voffset is the offset voltage, and Vsym (Va-sym) is the magnitude of the symmetric (anti- symmetric) component of the d.c. voltage. The fitting procedure gives values for HR, H, Vsym and Va-sym. The results show that both HR and H do not depend on VG, indicating that the bulk magnetic properties of Fe are not influenced by the external electric-field (see Supplementary Information). However, a clear modification of Va- sym and Vsym is observed in Fig. 2d. The dependence of Va-sym and Vsym on VG shows distinctly different slopes, indicating different origins for Va-sym and Vsym (see Supplementary Information). Figure 2e shows the dependence of Va-sym on magnetization direction M at VG of 0.88 V, 0.34 V and +0.07 V for the [100]-oriented device. A clear modulation of Va- sym by VG can be seen. By analyzing the angular dependence, the magnitude of the in- plane iSOFs at each VG can be extracted using V [100] a sym    l j   Fe M 2 Re  I   I h  D sin  M  I h R cos  M  sin2  M , (3) 8 a ) T  ( I hR 0 25 20 15 b -15 ) T  ( I -18 hD 0 -21 -1.0 mR I I mD c 10 ) / V T  ( I mD , I mR mR I = -10.7(+0.5) uT/V mD I = 2.0(+0.2) uT/V 0.0 -0.5 VG (V) 5 0 [100] [010] Device orientation Figure 3 Gate-voltage dependence of in-plane induced spin-orbit fields. VG dependence of in-plane Bychkov-Rashba- (a) and Dresselhaus- (b) iSOFs obtained from the [100]-oriented device. The solid lines are linear fits, and the modification rate of Rashba-iSOF mRI and Dresselhaus-iSOF mDI is obtained from the slopes. (c) Absolute value of mRI and mDI for devices along [100] and [010] directions. The magnitude of mR is larger than mD for both orientations. The error bars in (c) are standard deviation from the linear fits. where hDI (hRI) is the current induced in-plane iSOF due to Dresselhaus (Bychkov- Rashba) spin-orbit interaction,  is the magnitude of the anisotropic magneto- resistance of Fe, jFe is the microwave current density in Fe, l is the length of the device, M is the magnetization, and Re(I) is the real part of the diagonal component of the magnetic susceptibility due to in-plane excitation8. The magnitude of , jFe, M 9 and Re(I) is confirmed to be VG independent (see Supplementary Information). Thus, by using constant values of 0M = 2.1 T, = 1.1×109 m, l = 200 m and jFe = 1.0×1010 Am2, the angular dependence of Va-sym at each VG can be well reproduced by Eq. (3) using hDI and hRI as adjustable parameters. The extracted values of hRI and hDI are shown as a function of VG in Figs. 3a and b, respectively. One can see that the absolute value of hRI and hDI increases (decreases) at negative (positive) VG. Besides the polarity, the dependence of hRI (hDI) on VG is basically linear, which is expected from the calculated linear dependence of the interfacial electric-field on VG (see Fig. 1c and Ref. 28). The modification rate can be quantified by the slope m extracted from the linear fits (solid lines). As shown in Fig. 3c, for both [100] and [010] orientations, mRI is ~ 5 times larger than mDI, indicating that the modulation rate of the in-plane iSOFs originating from the Bychkov-Rashba spin-orbit interaction is larger than that from the Dresselhaus one. The different amplitudes of mRI and mDI also exclude the possibility that the observed modulation is an artifact due to changes of jFewith VG. If the modulation were dominated by a VG dependent jFe, mRI would be equal to mDI according to Eq. 3 (see Supplementary Information). We also show in the Supplementary Information that other possible excitations, e.g., the Oersted field as well as the bulk Dresselhaus field induced by the inductive microwave current in un- depleted n-GaAs, are negligibly small. 10 20 uT  0.88 V 0.07 V [010] jy ) V e 9 - 0 1 ( O S  4 2 0 0 2 4 [100] jx Figure 4 Tuning of in-plane induced spin-orbit fields. Polar plot of in-plane iSOFs. The arrows represent direction and relative strength of heff, heff = hRI + hDI, along the four main directions for a current density of 1×1010 Am2. Note that the magnitude of the iSOFs for [110] has been amplified by a factor of 10, and the orange arrows along [110] and [110] orientations have been shifted for clarity. The solid lines show the polar plot of the spin-orbit energy splitting, SO = 2B0heff for VG = 0.88 V and +0.07 V. The full angular dependence of heff is extrapolated from the magnitude of field-vectors along the four main orientations. The complete angular dependence of the in-plane iSOFs, which is obtained from devices oriented along the four main crystal directions, is shown as a polar plot in Fig. 4 for VG = 0.88 V and +0.07 V. In the plot, the lengths of the arrows represent the magnitude (heff = hRI + hDI), and the angles between the arrows and current directions represent the direction of the iSOFs. For the [100] ([010]) direction, both the length and angle of the arrows change with VG; in contrast to the [110] and [110] 11 directions, where only the length of the arrows is modified by VG as expected from the k-space alignment of Bychkov-Rashba and Dresselhaus SOFs (see Figs. 1d and e). These results clearly demonstrate an electric-field modulation of the spin-orbit field- vectors. Note that the length of the arrow along [110] is much larger than that along [110] (the amplitude for [110] has been amplified by a factor of 10), which is due to subtracted along [110] (see Figs. 1d and e). The dramatic amplitude difference for [110] and [110] directions also excludes the Oersted field in Fe as a possible driving the fact that Dresselhaus and Bychkov-Rashba fields add along [110] but get force for the magnetization dynamics. Moreover, as shown by the solid lines in Fig. 4, the modulation effect becomes much clearer from the polar plot of the spin-orbit energy splitting, SO = 2B0heff23. Here, the full angular dependence of heff is extrapolated from the magnitude of the magnetic field-vectors obtained from devices for four main orientations. Besides the in-plane iSOFs, the presence of Vsym also implies the existence of a sizeable out-of-plane iSOF, hz, which is due to the combined effects of in-plane spin polarization and exchange coupling8,21,29. The out-of-plane iSOFs act on the magnetization and produce an anti-damping like torque. Similar out-of-plane spin polarization has also been observed in nonmagnetic semiconductor structures caused by the joint effect of in-plane spin polarization and Zeeman coupling through an external magnetic field30-32. However, the coupling (Zeeman field ~ 105 eV) in semiconductors is much smaller than that in ferromagnets with reduced symmetry (exchange field ~ 30 meV for GaMnAs and ~ 1 eV for Fe). The magnitude of hz is 12 a 80 ) T  ( z hR 0 75 70 b mR z = -8.4(+0.9) uT/V 35 30 25 ) T  ( z hD 0 c ) / V T  ( z mD , z mR 9 6 3 0 mR z z mD [010] [100] Stripe orientation mD z = -4.3(+1.0) uT/V -1.0 0.0 -0.5 VG (V) Figure 5 Gate-voltage dependence of the out-of-plane induced spin-orbit fields. (a) Gate-voltage VG dependence of the out-of-plane induced spin-orbit fields originating from Bychkov-Rashba (a) and Dresselhaus (b) spin-orbit interaction for the [100]-oriented device. The solid lines in (a) and (b) are linear fits to the experimental data. From the slope, the modification rate of hRz and hDz is obtained. (c) Absolute values of mRz and mDz obtained from the linear fits of the out-of-plane iSOFs for devices along [100] and [010] directions. The error bars in (c) are standard deviation from the linear fits. magnetization direction dependent8,21. Taking the [100]-oriented device as an example, the dependence can be expressed as, hz = hRzcosM hDzsinM, where hRz (hDz) originates from Bychkov-Rashba (Dresselhaus) spin-orbit interaction8. Thus, the M dependence of Vsym shown in Fig. 2f can be described by V [100] sym  l j   Fe M 2 Im  O  a  z h R cos  M  z h D sin  M  sin 2  M , (4) 13 where Im(aO) is the imaginary part of the off-diagonal component of the magnetic susceptibility due to the out-of-plane excitation8. The obtained gate-voltage dependence of hRz and hDz is summarized in Figs. 5a and b, respectively. Similar to the in-plane case, a clear modulation can be seen and the modulation amplitude of hRz is larger than hDz, indicating again a stronger modulation of the Bychkov-Rashba spin- orbit interaction. Now, we discuss the mechanism for the electric-field modulation of iSOFs. According to Eqs. 1 and 2, the modulation is directly related to the modulation of the strength of the spin-orbit interaction. This modulation can be in general different for hR and hD, as the two fields originate from two inequivalent spatial crystal orientations [110] and [110] at the Fe/GaAs interface7. We indeed see that the electric-field modification of the Bychkov-Rashba and Dresselhaus field is different, see Figs. 3c and 5c. The smaller mD is consistent with previous measurements in semiconductor quantum wells33-35, where the magnitude of Dresselhaus spin-orbit interaction is less sensitive to the external electric-field compared to Bychkov-Rashba spin-orbit interaction. The different gate dependences for in-plane and out-of-plane components of both Bychkov-Rashba- and Dresselhaus-SOFs have another origin. The in-plane iSOFs are formed by spin accumulation  at the Fermi level36, while, the out-of- plane iSOFs originate from the whole bands, known as intrinsic effect8,21,29. Another possible origin of hz could be the spin-transfer torque effect due to the absorption of in-plane spin accumulation37,38, i.e., hz ~ M×as in case of NM/FM bi-layers. If hz originates dominantly from spin-transfer torque, applying negative gate-voltage can 14 only decrease the absorption and thus hz. Since we observe a clear increase of hz at negative bias, spin-transfer torque effects can also be excluded. The magnitudes of mR and mD are sizeable. Density function theory (DFT) results have predicted an efficient control of interfacial spin-orbit field for transverse electric- fields of the order of 1 V/nm (see supplementary material of Ref. 7). Our electric- fields are much smaller, about 0.01 V/nm. In DFT calculations thin slabs of Fe/GaAs without doping were studied, while our structure comprises a wide (~100 nm) Schottky barrier. In thin slabs large electric-fields are needed to modify the interfacial orbital hybridization between Fe and GaAs atoms to change interfacial spin-orbit field. In our samples we believe that we have interfacial states that penetrate deep enough into GaAs to get influenced by the Schottky barrier. Applying the gate-voltage then changes the barrier which, in turn, changes the overlap of the interfacial electron states with Fe. The modification of SOFs follows, since they are determined by the interfacial overlap. The assumption, that we have an additional conducting interfacial channel, that leads effectively to spin accumulation, is supported by the sizeable interfacial resistance at reverse gate-voltage in the junctions (~ 0.19 k/m2 for VG = 0.5 V, see Fig. S1b). Finally, we want to mention that in conventional approaches of electrically controlling iSOFs acting on a ferromagnetic metal18, a top-gate field-effect transistor has been utilized to change the carrier population of the magnetic material. As a consequence, effective control of the iSOFs acting on the ferromagnetic metal requires that the thickness of the ferromagnet should be ultra-thin (usually below 1 15 nm) due to the short screening length of a metal. However, the method presented here, i.e., voltage control by changing the magnitude of the interfacial spin-orbit interaction in a Schottky diode, is distinctly different from the conventional approach, and not limited by the thickness of the ferromagnetic metal (4 nm in this study). This unique approach thus provides an alternative way of electric-field control. The efficient electric-field control suggests that an electric-field in a Schottky diode can be used for the development of low-power consumption spin-orbit torque devices for magnetization manipulation. Methods: Calculation of interfacial electric-field. To calculate the interfacial electric-field Ein for Fe/n-GaAs, we solve the one-dimensional Poisson equation, d2/dx2 = /r0, where  is electric potential,  is charge density, r is the relative permittivity, and 0 is the permittivity in vacuum. In the calculation, the work function of Fe is set to 4.7 eV, and the electron affinity of GaAs is 4.15 eV. Since the experiment is carried out at room temperature, we consider all the donors of n-GaAs to be ionized, thus the donor concentration ND is equal to the carrier concentration n. The 15 nm graded n→n+ junction is discretized into 5 steps, and the Fermi-Dirac statistics has been employed. Having obtained the distribution of on x, the interfacial electric-field at each gate- voltage can be determined from Ein = d/dxinterface. Sample preparation. The Fe/n-GaAs heterostructure utilized in this experiment is grown on semi-insulating GaAs (001) substrates in a molecular-beam epitaxy cluster. 16 The semiconductor part is grown in a III-V MBE, consisting of the following layers: a 100-nm undoped GaAs buffer layer, 250 nm Si-doped n-GaAs layer with carrier concentration of 4×1016 cm3, a 15 nm n → n+ transition layer, and a 10 nm n+-GaAs layer with carrier concentration of 1×1018 cm3. The thickness and carrier density of the n-GaAs layers have been designed to minimize the inductive microwave current density in non-depleted GaAs, which could drive the magnetization dynamics through Oersted field and bulk-Dresselhaus field (see Supplementary Information). After that, the wafer is transferred to a metal MBE without breaking the vacuum, and 4-nm-thick Fe is deposited epitaxially. To avoid Fe oxidation, 2.5-nm thick Al capping layer is also deposited in the same chamber, which is estimated to be fully oxidized in air. The growth temperature for Fe is 75 oC. Device. First, a Fe/n-GaAs mesa with dimension of 200 m × 900 m is defined by combining Ar ion-beam milling and chemical wet etching (CH3COOH:H2O:H2O2 = 5:5:1) down to the insulating substrate. Then, SO-FMR Fe-stripes with dimensions of 200 m × 10m are defined on the mesa and the residual Fe in the n-GaAs channel region is etched away. We also use Fe/n-GaAs with a large area of 400 m × 300m as remote contacts (Fig. 2a). Finally, Ti/Au electrodes and bonding pads are fabricated from 15 nm Ti and 300 nm Au by evaporation. Devices with different orientations have been fabricated to measure the symmetry of iSOFs under electric-field modulation. Measurements. As shown in Fig. 2a, microwave currents generated by the signal generator pass into the Fe stripe and drive the magnetization dynamics due to current 17 induced SOFs at the Fe/GaAs interface. The d.c. voltage is measured across the stripe due to the coupling of the microwave current and the magnetization dynamics through anisotropic magnetoresistance effect of Fe. A bias tee is used to separate the d.c. voltage from the microwave background. The microwave frequency is 12.5 GHz, and the microwave power is set as low as possible (10 mW in this experiment). Under this condition the IG-VG curve and thus the Schottky barrier remains unaffected by microwaves (see Supplementary Information). To modulate the SOF by external electric-fields, a bias current IG with different polarity is applied between the Fe stripe and the remote contact, which can be converted to a gate-voltage by the interfacial IG- VG characteristic (see Supplementary Information). The measurements are performed between 0.88 V and +0.07 V. Since the d.c. bias-current (~ 2 A) produces a background voltage in the millivolt range, the d.c. voltage is detected by a lock-in amplifier by modulating the microwaves on and off at 888 Hz using a function generator. All measurements are performed at room temperature. References 1. Chernyshov, A. et al. Evidence for reversible control of magnetization in a ferromagnetic material by means of spin-orbit magnetic field. Nature Phys. 5, 656- 659 (2009). 2. Endo, M. et al. Current induced effective magnetic field and magnetization reversal in uniaxial anisotropy (Ga,Mn)As. Appl. Phys. Lett. 97, 222501 (2010). 3. Miron, I. M. et al. Perpendicular switching of a single ferromagnetic layer induced by in-plane current injection. Nature 476, 189-193 (2011). 18 4. Liu, L. Q. et al. Spin-torque switching with the giant spin Hall effect of Tantalum. Science 336, 555-558 (2012). 5. Ohno, H. Electric-field control of ferromagnetism. Nature 408, 944-946 (2000). 6. Matsukura, F. et al. Control of magnetism by electric fields. Nature Nanotech. 10, 209-220 (2015). 7. Gmitra, M. et al. Magnetic control of spin-orbit fields: A first-principle study of Fe/GaAs junctions. Phys. Rev. Lett. 111, 036603 (2013). 8. Chen, L. et al. Robust spin-orbit torque and spin-galvanic effect at the Fe/GaAs (001) interface at room temperature. Nature Commun. 7, 13802 (2016). 9. Kim, J. et al. Layer thickness dependence of the current-induced effective field vector in Ta CoFeB MgO. Nature Mater. 12, 240-245 (2013). 10. Garello, K. et al. Symmetry and magnitude of spin-orbit torques in ferromagnetic heterostructures. Nature Nanotech. 8, 587-593 (2013). 11. Soumyanarayanan, A. et al. Emergent phenomena induced by spin-orbit coupling at surfaces and interfaces. Nature 539, 509-517 (2016). 12. Mellnik, A. R. et al. Spin-transfer torque generated by a topological insulator. Nature 511, 449-451 (2014). 13. Fan, Y. B. et al. Magnetization switching through giant spin-orbit torque in a magnetically doped topological insulator heterostructure. Nature Mater. 13, 699- 704 (2014). 14. Kondou, K. et al. Fermi-level-dependent charge-to-spin current conversion by Dirac surface states of topological insulators. Nature Phys. 12, 1027-1031 (2016). 15. Lesne, E. et al. Highly efficient and tunable spin-to-charge conversion through Rashba coupling at oxide interfaces. Nature Mater. 15, 1261-1266 (2016). 16. Rojas-Sánchez, J. -C. et al. Spin to charge conversion at room temperature by 19 spin pumping into a new type of topological insulator: -Sn films. Phys. Rev. Lett. 116, 096602 (2016). 17. Fan, Y. B. et al. Electric-field control of spin-orbit torque in a magnetically doped topological insulator. Nature Nanotech. 11, 352-359 (2016). 18. Liu, R. H. et al. Control of current-induced spin-orbit effects in a ferromagnetic heterostructures by electric field. Phys. Rev. B. 89, 220409(R) (2014). 19. Weisheit, M. et al. Electric field-induced modification of magnetism in thin- film ferromagnets. Science 315, 349-351 (2007). 20. Fang, D. et al. Spin-orbit-driven ferromagnetic resonance. Nature Nanotech. 6, 413-417 (2011). 21. Kurebayashi, H. et al. An antidamping spin-orbit torque originating from the Berry curvature. Nature Nanotech. 9, 211-217 (2014). 22. Ciccarelli, C. et al. Room-temperature spin-orbit torque in NiMnSb. Nature Phys. 12, 855-860 (2016). 23. Fabian, J. et al. Semiconductor spintronics. Acta Physica Slovaca. 57, 565-907 (2007). 24. Nitta, J. et al. Gate control of spin-orbit interaction in an inverted In0.53Ga0.47As/In0.52Al0.48As heterostructure. Phys. Rev. Lett. 78, 1335-1338 (1997). 25. 26. Sze, S. M. Semiconductor devices 2nd edn (John Wiely & Sons, Inc, 2002). Liu, C. et al. Electrical detection of ferromagnetic resonance in ferromagnet/n- GaAs heterostructures by tunneling anisotropic magnetoresistance. Appl. Phys. Lett. 105, 212401 (2014). 27. Liu, C. et al. Dynamic detection of electron spin accumulation in ferromagnet- semiconductor devices by ferromagnetic resonance. Nature Commun. 7, 10296 20 (2016). 28. Jonker, B. T. et al. Enhanced carrier lifetimes and suppression of midgap states in GaAs at a magnetic metal interface. Phys. Rev. Lett. 79, 4886-4889 (1997). 29. Qaiumzadeh, A., Duine, R. A., & Titov, M. Spin-orbit torques in two- dimensional Rashba ferromagnets. Phys. Rev. B 92, 014402 (2015). 30. Kato, Y. K. et al. Current-induced spin polarization in strained semiconductors. Phys. Rev. Lett. 93, 176601 (2004). 31. Stern, N. P. et al. Current-induced polarization and the spin Hall effect at room temperature. Phys. Rev. Lett. 97, 126603 (2006). 32. Engel, H-A. et al. Out-of-plane spin polarization from in-plane and magnetic fields. Phys. Rev. Lett. 98, 036602 (2007). 33. Ishihara, J. et al. Direct imaging of gate-controlled persistent spin helix state in a modulation-doped GaAs/AlGaAs quantum well. Appl. Phys. Expre. 7, 013001 (2014). 34. Luengo-Kovac, M. et al. Gate control of the spin mobility through the modification of the spin-orbit interaction in two-dimensional system. Phys. Rev. B. 95, 245315 (2017). 35. Dettwiler, F. et al. Stretchable persistent spin helices in GaAs quantum wells. Phys. Rev. X 7, 031010 (2017). 36. Gambardella, P. et al. Current-induced spin-orbit torques. Phil. Trans. R. Soc. A 369, 3175-3197 (2011). 37. Slonczewski, J. C. Current-driven excitation of magnetic multilayers. J. Mag. Mag. Mater. 159, L1-L7 (1996). 38. Berger, L. Emission of spin waves by a magnetic multilayer traversed by a current. Phys. Rev. B 54, 9353-9358 (1996). 21 Acknowledgements L. Chen thanks M. Kammermeier, M. Buchner and C. Gorini for fruitful discussions. L. Chen is grateful for support from Alexander von Humboldt Foundation. This work is support by the German Science Foundation (DFG) via SFB 689 and SFB 1277. Author contribution L.C. planned the study. L.C. and R. I. fabricated the devices. L.C. collected and analysed the data. M.K., D.S. and D.B. grew the samples. M.V did the COMSOL simulations. M.G. and J.F. did the first principle calculations and theoretical input. L.C. wrote the manuscript with input from J. F., C.H.B. and D.W.. All authors discussed the results. Competing financial interests The authors declare no competing financial interests. Data availability The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request. Figure captions Figure 1 Electric-field modulation of spin-orbit fields at the Fe/GaAs (001) interface. Schematic band bending of a Fe/n-GaAs Schottky diode under reverse (a) and forward (b) bias. (c) Calculated interfacial electric-field Ein obtained by solving 22 the one-dimensional Poisson equation for the non-constant doping profile of n-GaAs (Methods). The magnitude of interfacial spin-orbit interaction is proportional to Ein. Wave-vector k dependence of Rashba- (d) and Dresselhaus (e) spin-orbit fields which dominate at the epitaxial Fe/GaAs interface with C2v symmetry. Figure 2 Electric-field modulation of induced spin-orbit fields measured by spin-orbit ferromagnetic resonance. (a) Schematic of the circuit and device structure for spin-orbit ferromagnetic resonance measurements. A microwave current j(t), generated by a signal generator, passes through a bias tee into the Fe/GaAs (001) layer, where time dependent interfacial iSOFs h(t) are generated due to the spin-orbit interaction at the Fe/GaAs interface. h(t) drives the magnetization dynamics M(t), and leads to a resistance variation R(t) due to the anisotropic magneto-resistance of Fe. The coupling between R(t) and j(t) produces a rectified d.c. voltage V. By measuring V, the magnitude of the interfacial iSOFs can be quantified. To modulate the iSOFs by an electric-field, Fe/n-GaAs samples have been used, where a Schottky barrier is formed at the interface. Thus, by adding a gate-voltage VG between Fe and n-GaAs, the magnitude of the iSOFs can be controlled. Here, M is the magnetization angle with respect to the [100] orientation of GaAs. (b) d.c. voltage spectra for several gate- voltages for a [100]-oriented device measured at a microwave frequency of 12.5 GHz and M = 225°. The input microwave power is 10 mW. An offset voltage, Voffset, has been subtracted. (c) Typical Lorentzian fit of the voltage spectrum, from which the magnitudes of Vsym and Va-sym are obtained. (d) VG dependence of Va-sym and Vsym for 23 M = 225°. Magnetization angle M dependence of Va-sym (e) and Vsym (f) under different VG for a [100]-oriented device. The solid lines in (e) and (f) are fits to Eqs. 3 and 4, respectively. Figure 3 Gate-voltage dependence of in-plane induced spin-orbit fields. VG dependence of in-plane Bychkov-Rashba- (a) and Dresselhaus- (b) iSOFs obtained from the [100]-oriented device. The solid lines are linear fits, and the modification rate of Rashba-iSOF mRI and Dresselhaus-iSOF mDI is obtained from the slopes. (c) Absolute value of mRI and mDI for devices along [100] and [010] directions. The magnitude of mR is larger than mD for both orientations. The error bars in (c) are standard deviation from the linear fits. Figure 4 Tuning of in-plane induced spin-orbit fields. Polar plot of in-plane iSOFs. The arrows represent direction and relative strength of heff, heff = hRI + hDI, along the four main directions for a current density of 1×1010 Am2. Note that the magnitude of the iSOFs for [110] has been amplified by a factor of 10, and the orange arrows along [110] and [110] orientations have been shifted for clarity. The solid lines show the polar plot of the spin-orbit energy splitting, SO = 2B0heff for VG = 0.88 V and +0.07 V. The full angular dependence of heff is extrapolated from the magnitude of field-vectors along the four main orientations. Figure 5 Gate-voltage dependence of the out-of-plane induced spin-orbit fields. 24 (a) Gate-voltage VG dependence of the out-of-plane induced spin-orbit fields originating from Bychkov-Rashba (a) and Dresselhaus (b) spin-orbit interaction for the [100]-oriented device. The solid lines in (a) and (b) are linear fits to the experimental data. From the slope, the modification rate of hRz and hDz is obtained. (c) Absolute values of mRz and mDz obtained from the linear fits of the out-of-plane iSOFs for devices along [100] and [010] directions. The error bars in (c) are standard deviation from the linear fits. 25 Supplementary Information Electric-field modification of field-vector interfacial spin-orbit L. Chen1, M. Gmitra2, M. Vogel1, R. Islinger1, M. Kronseder1, D. Schuh1, D. Bougeard1, J. Fabian2, D. Weiss1 and C. H. Back1 1Institute of Experimental and Applied Physics, University of Regensburg, 93040 Regensburg, Germany. 2Institute of Theoretical Physics, University of Regensburg, 93040 Regensburg, Germany. Corresponding author, e-mail: [email protected] Table of Contents: 1. Microwave current distribution in Fe/n-GaAs 2. Bias-voltage dependence of resonance field, linewidth and anisotropic magneto-resistance 3. Symmetry of iSOFs with respect to positive and negative magnetic field 4. Electric-field control of iSOFs for another sample a Fe VG IG jFe X Depleted n-GaAs Non-depleted n-GaAs IG X jn-GaAs b ) A  ( IG 2 0 -2 microwave off P = 10 mW -1.0 0.0 -0.5 VG (V) c ) 3 - m c 6 1 0 1 ( n 4 2 0 VG = 0.1 V 0 V 0.88 V 50 150 100 dGaAs (nm) 200 250 Figure S1 VG - IG characterization of the Schottky barrier and simulated carrier distribution in n-GaAs. (a) Schematic cross section of the device used for Fig. 2a. A microwave current jFe passes through the Fe stripe and an inductive microwave current is induced in the non-depleted n-GaAs. To tune the interfacial spin-orbit interaction, a bias current IG is applied from the middle Fe stripe to the remote side contact, and the corresponding interfacial gate-voltage VG is measured between the middle Fe stripe and another remote side contact. (b) VG - IG characteristics of the Schottky junction with microwave on and off. (c) Simulated carrier distribution profile of n-GaAs at different VG. 1 a Fe G a A s 2 7 5 n m VG = 0.1 V 0 V 0.88 V b ) e jF + s A a G jn- ( / e jF 1.0005 1.0000 0.9995 4 6 8 10 12 14 16 18 20 22 f (GHz) Figure S2 Determination of microwave current distribution in Fe/n-GaAs using COMSOL 5.3 (a) The current density distribution in Fe/n-GaAs stack is calculated by using the COMSOL 5.3 finite elements method software (RF Module). The stack is modeled in 3D using a triangular mesh with a maximum feature size of 3 nm for discretization. The sample size for simulation is scaled down to 8 × 1 m2, and, for clarity, the schematic shown in (a) is further reduced down to 0.8 × 0.1 m2. In the simulation, the applied microwave power is 10 W and the excitation frequency is varied from 5 GHz to 20 GHz. The conductivity of Fe and non-depleted GaAs is set to 3.5×106 S/m and 5.2×103 S/m, respectively. To simplify the calculations, the carrier distribution of GaAs under different VG is assumed to be a step function distribution as indicated by the dashed lines in Fig. S1c. (b) Calculated ratio of the current density in the Fe layer jFe in relation to the total current density jFe+jn-GaAs, jFe/(jFe+jn-GaAs), as a function of microwave frequency. Here, jn-GaAs is the inductive microwave current density in un-depleted n-GaAs. One can see that most of the current flows in the Fe layer, and the inductive current in non-depleted n-GaAs is negligibly small. 2 1. Microwave current distribution in Fe/n-GaAs There could be three kinds of driving forces for magnetization dynamics, which are determined by the microwave current distribution in the Fe layer and the un-depleted n-GaAs: one is the interfacial spin-orbit field due to the current flow in Fe, and the other two are the Oersted field hOe and the Dresselhaus field hDbulk due to the current flow in the un-depleted n-GaAs. Note that the Dresselhaus field in the bulk of n-GaAs hDbulk is essentially different from the interfacial Dresselhaus field at the Fe/GaAs interface discussed in the main text. To minimize hOe and hDbulk, the sample layout is designed by reducing the thickness (275 nm) and carrier concentration (4×1016 cm3) of n-GaAs. Figure S2b shows the COMSOL simulation results1 of the ratio of the current density in Fe jFe with respect to the total current density jFe+jn-GaAs, jFe/(jFe+ jn-GaAs), as a function of microwave frequency. Here jn-GaAs is the inductive current density in un-depleted n-GaAs. The magnitude of jFe/(jFe+ jn-GaAs) is ~ 1, with a variation far less than 1% in the whole frequency range. This indicates that most of the microwave current flows in the Fe layer, and the inductive current flow in the un-depleted n-GaAs is negligibly small. a 2494 b ) (  R 2493 2492 2491 2490 Idc = 10 A 5 10 15 Pin (dBm) 20 ) (  R 2494 2493 2492 2491 2490 0.0 0.4 0.8 Idc (mA) 1.2 Figure S3 Calibration of the microwave current using Joule heating induced resistance change. (a) Resistance of the Fe layer as a function of microwave input power measured at a small d.c. current of 10 A. (b) Resistance of the Fe layer as a function of d.c. current. 3 The actual microwave current density jFe is calibrated by the Joule heating induced resistance increase2. Figure S3a shows the resistance of the Fe stripe as a function of microwave power with the application of a d.c. current of 10 A, and the same measurement is done by changing the magnitude of d.c. current as shown in Fig. S3b. By comparing the resistance change in Figs. S3a and b, the microwave current is determined to be 0.4 mA, and the corresponding microwave current density jFe is 1×1010 Am2. By using jFe/jn-GaAs of 2×104 (Fig. S2b) and jFe of 1×1010 Am2, jn-GaAs is determined to be 5×105 Am2. The Oersted field hOe can be calculated by 0hOe = 0jn-GaAsdn-GaAs/2, which is 0.05 T by using dn-GaAs of 170 nm (Fig. S1c). In comparison with the previous measurement in (Ga,Mn)As2, the Dresselhaus field hDbulk in bulk un-depleted GaAs can be estimated to be 103 T. The magnitude of hOe and hDbulk is three orders smaller than hRI and hDI obtained in the main text (note that, due to the finite spacing between the Fe layer and the un-depleted n-GaAs, the effective current-induced fields in n-GaAs acting on Fe can be even smaller.). These results indicate that a possible excitation induced by inductive microwave current in the un-depleted n-GaAs is negligible, and the main driving force is the spin-orbit field at the Fe/GaAs interface. The simulation shown in Fig. S2b suggests that the value of jFe/(jFe+ jn-GaAs) ~ 1 with variations of the amplitude far less than 1% for different VG. This indicates that the microwave current density in Fe is almost independent of VG. Here, we further exclude this possibility that modification of hR and hD is due to modulation of microwave current density as follows: If the modification of hR and hD would be really due to a modulation of the microwave current density jFe, the following form of the current density would be expected: jFe = j0Fe(1+ mVG) with m defining the modification rate of jFe and j0Fe is jFe at VG = 0 V. In this case one can re-write Eq. 3 of the 4 main text as  V M 2 [100] a sym   I l j Re 0    Fe   I h  1   mV G  I h  D sin  M  I h R cos  M  sin2  M . (S1) Here hRI and hDI are the in-plane Rashba and Dresselhaus iSOF at 0 V with hRI ~ hDI (see Figs. 3a and b in the main text). Therefore, the modification rate of Rashba-iSOF mRI and Dresselhaus-iSOF mDI due to the modulation of jFe would be mRI=mhRI, mDI=mhDI, (S2) Since hRI ~ hDI, we should have mRI ~ mDI. However, our experimental results show that mRI is ~ 5 times larger than mDI, being at odds with the above assumption. Therefore, we exclude the possibility that the modification of hR and hD is due to a modulation of the microwave current density jFe. 2. Bias-voltage dependence of resonance field, linewidth and anisotropic magneto-resistance Figure S4 shows the gate-voltage VG dependence of the resonance field HR and the linewidth H. One can see that both HR and H are independent of VG, indicating that the magnetic properties of Fe, e.g, the magnitude of M, Re(I) and Im(a O), are not modulated by the electric-field. This can be reasonably understood as the follows: the capacitance C of a Schottky contact3 is estimated to be ~ 107 Fcm2, and the modulation of the interfacial electron density can be expressed by n = (C/e)VG, where e is the electronic charge. Under a gate-voltage of 1 V, n is determined to be ~ 1012 cm2, which is at least three orders of magnitude smaller than the sheet carrier concentration of bulk Fe. We 5 also confirm in Fig. S5 that the magnitude of the anisotropic resistance is independent of gate-voltage. [100] a 200 Exp. [010] ( ) T m HR 0 100 6b ( ) T m H  0 4 0.88 V 0.34 V 0.07 V Fit Exp. 0.88 V 0.34 V 0.07 V Fit [010] [100] 360 180 H (deg) [100] 360 ) g e d ( M  180 0 0  = 0.0033 0 90 180 H (deg) 270 360 Figure S4 Gate-voltage dependence of resonance field and linewidth. Magnetic-field angle H dependence of resonance field HR (a) and linewidth H (b) under different gate-voltages for the [100]-orientated device. By fitting the H dependence of HR, one can obtain the magnitude of the effective perpendicular magnetic anisotropy field 0HK of 1506.5 mT, the biaxial magnetic anisotropy field 0HB of 34.0 mT, the uniaxial magnetic anisotropy field 0HU of 72.0 mT and the Landé g factor of 2.06. The inset of (a) shows H as a function of magnetization angle M. By fitting the H dependence of H, the damping constant is obtained to be 0.0033. 6 4.544 4.540 4.536 4.532 ) m  7 - 0 1 (  // VG = 0.8 V 0.4 V 0.1 V Fit  180 -90 0 90 M (deg) Figure S5 Gate-voltage dependence of the anisotropic magneto-resistance. M dependence of the resistivity of the [110]-orientated device, which is gate-voltage VG independent. In the measurement, an external magnetic field of 1 T has been used, which leads to M = H. The solid line is a fit, from which the magnitude of anisotropic magneto-resistance,  = // , is determined to be 1.1×109 m. a ) V  ( m y s - Va 500 0 -500 b 500 0 -500 ) V  ( m y Vs Va-sym (-H) Va-sym (+H) Fit Vsym (-H) Vsym (+H) 0 90 180 M (deg) Fit 270 360 Figure S6 Symmetry of d.c. voltage with respect to positive and negative magnetic-field. (a) Magnetization angle M dependence of Va-sym(H) and Va-sym(+H) for a [100]-orientated device, which shows Va-sym(H) = Va-sym(+H). (b) M dependence of Vsym(H) and Vsym(+H), which shows Vsym(H) = Vsym(+H). The solid lines in (a) and (b) are fits according to Eqs. (3) and (4), respectively. The thickness of the sample is 5 nm, which is measured at a frequency of 11 GHz with a microwave power of 168 mW. 7 3. Symmetry of iSOFs with respect to positive and negative magnetic field The magnitude of Va-sym(H) and Va-sym(+H) as a function of magnetization angle M is shown in Figure S6a, one can see the symmetry of Va-sym(H) = Va-sym(+H); while for Vsym, shown in Figure S6b, one can see that Vsym(H) = Vsym(+H) holds. This is because Va-sym is induced by in-plane iSOFs hI (see Eq. 3 in the main text), which only depends on the current direction with symmetry hI(H) = hI(+H); on the other hand Vsym is induced by out-of-plane SOFs hO, being anti-dampling like and having the symmetry hO(H) = hO(+H) (see Fig. 4 of Ref. 4). 4. Electric-field control of iSOFs for another sample We have measured another sample consisting the following layers: Fe (1.3 nm)/n+-GaAs (10 nm, 5×1017 cm-3)/n+→n-GaAs (15 nm)/n-GaAs (250 nm, 4×1016 cm-3). Here the doping level of n+-GaAs has been reduced by a factor of 2 to increase the width of the Schottky barrier. As shown in Fig. S7a, the interfacial resistance Rin of this sample is ~ 8 times larger than the sample presented in the main text (Fig. S1b). The gate-voltage dependences of iSOFs are shown from Fig. S7b to Fig. S7e and summarized in Table S1. One can see that the modification rate of each iSOF becomes smaller in comparison with the higher n+-doped sample. The smaller modification rate could result from the smaller change of Ein with VG, being due to the increased width of the Schottky barrier. 8 a ) A  ( IG 2 1 0 -1 -2 b ) T  ( I hR 0 12 10 8 mR I = -5.2(+0.2) uT/V c -11 ) T  ( I -12 hD 0 d 14 12 ) T  ( z hR 0 10 9 e mR Z = -1.9(+0.3) uT/V ) T  ( z 6 hD 0 -13 mD I = 1.4(+0.2) uT/V mD z = -3.1(+0.3) uT/V -0.6 0.0 -0.3 VG (V) 0.3 -0.6 0.0 -0.3 VG (V) 0.3 -0.6 0.0 0.3 -0.3 VG (V) Figure S7 Electric-field control of iSOFs for another sample consisting of the following layers: Fe (1.3 nm)/n+-GaAs (10 nm, 5×1017 cm-3)/n+→n-GaAs (15 nm)/n-GaAs (250 nm, 4×1016 cm-3) (a) VG - IG characterization of the Schottky barrier for the device orientated along [100]. The device dimension is 20 m × 50 m. VG dependence of in-plane Bychkov-Rashba (b), in-plane Dresselhaus (c) out-of-plane Bychkov-Rashba (d) and out-of-plane Dresselhaus (e) iSOF. tFe (nm) 4.0 1.3 n+(cm-3) 1×1018 5×1017 Rin (k/m2) 0.19@0.5 V 0.86@0.5 V I (T/V) mR 10.7±0.5 5.2±0.2 mD I (T/V) 2.0±0.2 1.4±0.2 mR Z (T/V) 8.4±0.9 1.9±0.3 mD Z (T/V) 4.3±1.0 3.1±0.3 Table S1. Comparison of electric-field effect on samples with different Fe thickness and n+ doping levels. References 1. COMSOL rf package, www.comsol.com 9 2. Fang, D. et al. Spin-orbit-driven ferromagnetic resonance. Nature Nanotech. 6, 413-417 (2011). 3. Sze, S. M. Semiconductor Devices 2nd edn (John Wiley & Sons, Inc, 2002). 4. Chen, L. et al. Robust spin-orbit torque and spin-galvanic effect at the Fe/GaAs (001) interface at room temperature. Nature Commun. 7, 13802 (2016). 10
1110.4135	2	1110	2012-04-12T22:52:38	Numerical analysis of carrier multiplication mechanisms in nanocrystal and bulk forms of PbSe and PbS	[ "cond-mat.mes-hall" ]	We report on systematic numerical study of carrier multiplication (CM) processes in spherically symmetric nanocrystal (NC) and bulk forms of PbSe and PbS representing the test bed for understanding basic aspects of CM dynamics. The adopted numerical method integrates our previously developed Interband Exciton Scattering Model (IESM) and the effective mass based electronic structure model for the lead chalocogenide semiconductors. The analysis of the IESM predicted CM pathways shows complete lack of the pathway interference during the biexciton photogeneration. This allows us to interpret the biexciton photogeneration as a single impact ionization (II) event and to explain major contribution of the multiple II events during the phonon-induced population relaxation into the total quantum efficiency (QE). We investigate the role of quantum confinement on QE, and find that the reduction in the biexciton density of states (DOS) overruns weak enhancement of the Coulomb interactions leading to lower QE values in NCs as compared to the bulk on the absolute photon energy scale. However, represented on the photon energy scale normalized by corresponding band gap energies the trend in QE is opposite demonstrating the advantage of NCs for photovoltaic applications. Comparison to experiment allows us to interpret the observed features and to validate the applicability range of our model. Modeling of QE as a function of pulse duration shows weak dependence for the gaussian pulses. Finally, comparison of the key quantities determining QE in PbSe and PbS demonstrates the enhancement of II rate in the latter materials. However, the fast phonon-induced carrier relaxation processes in PbS can lead to the experimentally observed reduction in QE in NCs as compared to PbSe NCs.	cond-mat.mes-hall	cond-mat	Numerical analysis of carrier multiplication mechanisms in nanocrystal and bulk forms of PbSe and PbS Kirill A. Velizhanin∗ and Andrei Piryatinski† Center for Nonlinear Studies (CNLS), Theoretical Division, Los Alamos National Laboratory, Los Alamos, NM 87545 (Dated: November 1, 2018) We report on systematic numerical study of carrier multiplication (CM) processes in spheri- cally symmetric nanocrystal (NC) and bulk forms of PbSe and PbS representing the test bed for understanding basic aspects of CM dynamics. The adopted numerical method integrates our previ- ously developed Interband Exciton Scattering Model (IESM) and the effective mass based electronic structure model for the lead chalocogenide semiconductors. The analysis of the IESM predicted CM pathways shows complete lack of the pathway interference during the biexciton photogeneration. This allows us to interpret the biexciton photogeneration as a single impact ionization (II) event and to explain major contribution of the multiple II events during the phonon-induced population relaxation into the total quantum efficiency (QE). We investigate the role of quantum confinement on QE, and find that the reduction in the biexciton density of states (DOS) overruns weak en- hancement of the Coulomb interactions leading to lower QE values in NCs as compared to the bulk on the absolute photon energy scale. However, represented on the photon energy scale normalized by corresponding band gap energies the trend in QE is opposite demonstrating the advantage of NCs for photovoltaic applications. Comparison to experiment allows us to interpret the observed features and to validate the applicability range of our model. Modeling of QE as a function of pulse duration shows weak dependence for the gaussian pulses. Finally, comparison of the key quantities determining QE in PbSe and PbS demonstrates the enhancement of II rate in the latter materials. However, the fast phonon-induced carrier relaxation processes in PbS can lead to the experimentally observed reduction in QE in NCs as compared to PbSe NCs. PACS numbers: I. INTRODUCTION An ability to produce more then one electron-hole pairs (excitons) per single absorbed photon in semiconduc- tor materials can potentially enhance the solar energy conversion efficiency in photovoltaic devices beyond the fundamental Shockley-Queisser limit.1 -- 8 In the literature this process is referred to as Carrier Multiplication (CM) or multiple exciton generation, and has been extensively studied in the bulk semiconductors for decades.9 -- 13 Re- cent ultrafast optical studies of CM in semiconductor nanocrystals (NC) reported significantly higher QE and much lower activation energy threshold (AET) as com- pared to the bulk materials on the photon energy scale normalized by corresponding band gap values.3,14 -- 23 These reports stimulated extensive experimental and theoretical studies of the CM in the nanostructured ma- terials in which the exciton dynamics is affected by the quantum confinement. It was initially expected that the quantum confinement enhancement of the Coulomb in- teractions between the carriers could result in more ef- ficient multi-exciton production. Furthermore, a relax- ation of the quasi-momentum conservation constrains by breaking the translational symmetry should open addi- tional pathways for CM and reduce the AET. Finally, the presence of the phonon bottleneck in NCs should slow down the intraband phonon-assisted cooling and fur- ther increase the QE. However, the subsequent reports claimed significantly lower QE and even the absence of the CM in NCs.24 -- 27 This controversy possibly rises from experimental inaccuracy,26,28 sample-to-sample variation in surface preparation,29 -- 33 and contribution from ex- traneous effects such as photocharging.34,35 The op- tical measurements of QE in bulk PbS and PbSe semiconductors36 show that, in fact, the bulk QE ex- ceeds validated values in NCs25,34,35,37 -- 39 if compared on the absolute photon energy scale.36 However, due to con- finement induced increase in the band gap energy, Eg, the benefits to photovoltaics is higher in NCs than in bulk. The controversy calls for development of new sen- sitive spectroscopic tools,40,41 alternative photo-current measurements,42,43 and the reassessment of the quantum confinement role in the CM dynamics based on a unified theoretical model.25,44 Accepted theoretical models of CM are based on the many-body Coulomb interactions which correspond to the valence-conduction band transitions conserving the total charge but not the number of electrons and holes.45 -- 47 In the exciton picture, this represents the transitions between the exciton bands of different multi- plicity. Hence, we will refer to these transitions below as the interband exciton transitions, and to the correspond- ing Coulomb interactions as the interband Coulomb in- teractions. Within adopted nomenclature, the intraband Coulomb interactions (i.e., restricted to an exciton band of a certain multiplicity) conserve the total number of electrons and holes and determine the multi-exciton in- teraction (e.g., binding) energy. The simplest theoretical model initially developed to explain CM in bulk materials is the impact ionization (II) model.48 It treats CM dynamics following high energy photoexcitation as the lowest order interband transitions whose rate is given by the Fermi's Golden Rule. This rate depends on the interband Coulomb matrix elements and the final multiple-carriers (e.g., biexciton) density of states (DOS).49,50 The model has been further applied to interpret the CM dynamics in NCs along with various methods for the electronic structure calculations. Franceschetti, An and Zunger performed atomistic pseudopotential calculations of the II and Auger recom- bination rates in a PbSe NC. Assuming the absence of the quasi-momentum constraint and rapid growth of the biexciton DOS, they estimated low value (∼ 2.2Eg) of AET defining this quantity as the energy above which the II rate exceeds the Auger recombination one.51 Rabani and Baer performed screened semiempirical pseudopo- tential calculations of the II rates in CdSe and InAs and found significant reduction of the QE with the NCs size increase. Observed trend makes CM already inefficient in NCs of diameter ∼ 3 nm. The observation has been rationalized by strong size dependence of the interband Coulomb interactions and the trion DOS behavior.52 Using the semiempirical tight binding model Allan and Delerue performed extensive calculations for PbSe and PbSe, InAs and Si NCs.53,54 They found that although the Coulomb interactions are enhanced by the quantum confinement, the quantum-confinement-induced reduc- tion in the biexciton DOS facilitates the reduction of the II rate and subsequently of QE. The comparison of the II rates calculated in PbSe bulk and PbSe NCs shows that the II rates in the NCs does not exceed the bulk one.53 Based on these calculations, the authors of Ref. [36] argue that the experimentally observed drop of QE in NCs can be attributed to the dominant effect of the reduced car- rier's DOS. Lack of the phonon bottleneck, leads to the rapid intraband phonon-assisted relaxation which further reduces the QE.53,54 Schaller, Agranovich, and Klimov, first pointed out that in addition to II, the direct photogeneration of biex- citons can take place in NCs through "virtual-exciton channel".15 Subsequently, Rupasov and Klimov sug- gested that additional contribution to photogeneration QE can rise from the "vitual biexciton channel".55 Us- ing the effective mass model for the electronic structure of PbS and PbSe NCs due to Kang and Wise,56 the au- thors estimated the photogeneration QE and argued that this process dominates the CM dynamics in NCs. Re- cently, Silvestri and Agranovich using the same model, performed detailed calculations for relatively small radii (specifically for 1.95 nm and 3 nm) PbSe NCs.57 They concluded that the contribution of the photogeneration processes in NCs is much weaker than it was claimed be- fore, and clarified that the overestimated QE results from disregarding the effect of selection rules for the interband Coulomb matrix elements and the oscillator strength fac- tors weighting optically allowed transitions. Here, we demonstrate that the absence of the interference between the photogeneration pathways and the presence of small 2 size dispersion further reduces the photogeneration QE. Another model describing the coherent photogenera- tion of biexcitons from the resonant exciton states ini- tially proposed by Shabaev, Efros, and Nozik considered idealized five level system,58 and was subsequently re- fined to account for the biexciton DOS effect.59 Using the effective mass k · p Hamiltonian, the authors of Ref. [59] performed calculations for small radii (i.e., 2 nm and 3 nm) PbSe NCs, and found that dense biexciton DOS leads to the vanishing coherent oscillations. They also pointed out that efficient CM can be expected in consid- ered small NCs. The drawback of the coherent superpo- sition model is the lack of the pure-dephasing effects and the inhomogeneous broadening which as we demonstrate here play important role in the CM dynamics reducing the QE. Finally, the ab initio calculations on small clusters ((cid:46) 1 nm) demonstrated the role of strong Coulomb corre- lations and fast exciton-biexciton dephasing in multiexci- ton photogeneration,60,61 and suggested contributions of the phonon-assisted Auger processes to CM.62 The atom- istic calculations mostly focused on II dynamics occurring during the population relaxation are limited to the small diameter ((cid:46) 3 nm) NCs. However, reported experimen- tal studies consider larger NCs with the diameter varying up to 10 nm. To close this gap and to perform compari- son with bulk, an extrapolation procedure combined with atomistic calculations has been used.63 The photogenera- tion dynamics has been investigated using effective mass models for larger but still small diameter (∼ 4 − 6 nm) NCs. Recently, we have published a letter summarizing the results of our numerical investigation on the direct pho- togeneration and population relaxation processes con- tributing to CM in a broad diameter range PbSe NCs and in PbSe bulk.64 The calculations employ our earlier proposed Interband Exciton Scattering Model (IESM)47 which recovers the models discussed above as limiting cases and which we further parameterized using the ef- fective mass, k · p, electronic structure model proposed by Kang and Wise.56,90 In the letter, we argue that in both cases of the photogeneration and population relax- ation the II is the main mechanisms of CM. This explains weak contribution of the direct photogeneration to the total quantum QE. Analyzing the scaling of the total QE with the NC size, we found that QE in NCs plot- ted on absolute energy scale does not exceed that in the bulk. This is in agreement with reported experimental data and some theoretical studies, and confirms that the quantum-confinement-induced reduction in the biexciton DOS makes a dominant contribution to QE. Present paper discusses in great details the theoretical method and provides extended analysis of the numerical results summarized in Ref. [64]. In addition, we con- sider the effect of the optical pump pulse duration on QE, and provide a comparative analysis of CM in PbS and PbSe materials. The paper is organized as follows. In Sec. II, we review the weak Coulomb interaction limit of the IESM used for numerical calculations, introduce convenient quasicontinuous representation, and further discuss the details of the model parameterization. In Sec. III an extensive analysis of our numerical results on PbSe NCs is given. Comparison of the key CM features in PbSe and PbS is performed in Sec. IV. Sec. V con- cludes the paper by a discussion on the limitations of the adopted model, connection with experiment, and possi- ble improvements of QE in NCs. II. THEORETICAL MODEL AND NUMERICAL PARAMETERIZATION The central quantity describing CM efficiency is QE which can be calculated as QE = 2Nxx + Nx Nxx + Nx . (1) Here Nx and Nxx are the total exciton and biexciton populations, respectively, produced by single absorbed photon in the limit of vanishing pump fluencies. They depend on the delay time measured from the center of the pump pulse and allow one to determine both the QE due to the photogeneration event and the total QE after the population relaxation. A complimentary quantity, often used to describe CM yield, is the biexciton quantum yield, ηxx = QE − 1, defined as a number of biexcitons produced per single absorbed photon. In this section, we describe the weak Coulomb limit of the IESM47 which we use to calculate the time evo- lution of the exciton and biexciton populations entering Eq. (1). We also introduce the quasicontinuous frequency representation which is convenient to analyze the quan- tum confinement signatures and the size scaling of the quantities determining the QE in transition from NC to the bulk limit. Finally, we discuss the model parameter- ization for the numerical calculations of QE in PbSe and PbS materials. A. Biexciton Photogeneration and Population Relaxation Model In the weak Coulomb interaction limit, the leading con- tribution to the population of the a-th exciton state gen- erated by the pump pulse is given by (0) = µx a02I(ωx a − ωpm), nx a (2) where µx a0 is the transition dipole moment between the ground state, x0(cid:105), (i.e., filled valence band) and an exci- ton state xa(cid:105), and (cid:90) ∞ (cid:48)(cid:90) ∞ I(ωx a − ωpm) = × cos [(ωx dt 2 2 a − ωpm) t1]Epm(t −∞ 0 dt1e−γx a t1 )Epm(t (cid:48) (cid:48) − t1), (3) 3 is the pulse self-convolution function depending on the ground state to exciton state transition frequency, ωx a , and dephasing rate, γx a . The pump pulse is characterized by the envelope amplitude, Epm(t), giving the temporal profile, and the central frequency, ωpm. In our calcula- tions, we use the Gaussian form of the pulse envelope function, Epm(t) = E (0) pm , with the mean am- plitude, E (0) pm, and pulse duration, τpm. For the pulses longer than typical dephasing time, τpm (cid:29) 1/γ, the continuous wave (cw) limit is recovered and the pulse self-convolution function becomes proportional to the Lorentzian line shape function. pme−t2/2τ 2 To be consistent with the calculations of the photo- generated biexciton population appearing in the second order interband Coulomb coupling, we account for the first and second order corrections to the exciton popu- lation and disregard the effect of the exciton coherences both introduced in Ref. [47]. The negligible contribution of the latter quantities has been validated by our direct numerical calculations. The photogenerated population of the k-th biexciton state is47 nxx k (2) = k,a µx a0 a − ωpm) (4) (cid:88) a≥1 (cid:12)(cid:12)(cid:12)Λxx,x (cid:88) (cid:12)(cid:12)(cid:12)2 I(ωx (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)2 l,0 (cid:88) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) + Λxx,x k,a µx a0 + kl Λxx,x µxx a≥1 l≥1 I(ωxx k − ωpm), where µxx biexciton states xxk(cid:105) and xxl(cid:105), and kl is the intraband transition dipole between the (cid:104) V xx,x k,a k − ωx ωxx a + iγxx,x k,a (cid:105) , (5) Λxx,x k,a = describes the transition amplitude associated with the single (Born) scattering event between exciton sate, xa(cid:105), and biexciton state, xxk(cid:105). The quantity Λxx,x l,0 , also en- tering Eq. (4), is the transition amplitude between the 0 = 0), and biexciton state, xxl(cid:105). ground state, x0(cid:105) (ωx V xx,x is the interband Coulomb matrix element in the k,a exciton representation giving rise to the interband scat- tering, and γxx,x is the interband dephasing rate. The k − ωpm), entering pulse self-convolution function, I(ωxx k,a Eq. (4) contains ωxx k , re- spectively. According to our calculations the coherence contributions to the biexciton populations47 are also neg- ligible. instead of ωx k and γxx k k and γx Our numerical calculations show that the rapid sign- variation of the interfering terms in Eq. (4) leads to their 4 FIG. 1: Non-interfering biexciton photogeneration pathways associated with the weak Coulomb limit. (a) Indirect biexciton photogeneration via exciton states. (c) Direct biexciton photogeneration via biexciton states. (See details in the text.) (b) Direct biexciton photogeneration via virtual exciton states. cancelation allowing us to write Eq. (4) as (cid:88) (cid:88) (cid:88) a≥1 a≥1 (cid:12)(cid:12)(cid:12)Λxx,x (cid:12)(cid:12)(cid:12)Λxx,x (cid:12)(cid:12)(cid:12)µxx k,a µx a0 k,a µx a0 (cid:12)(cid:12)(cid:12)2 I(ωx (cid:12)(cid:12)(cid:12)2 I(ωxx (cid:12)(cid:12)(cid:12)2 I(ωxx a − ωpm) k − ωpm) nxx k (2) = + + (6) kl Λxx,x l,0 k − ωpm). l≥1 The resulting three non-interfering photogeneration pathways are illustrated in Fig. 1. The first and the second pathways, given by the first and the second terms in Eq. (6), are shown in panels (a) and (b), respectively. These terms describe redistribu- tion of the exciton oscillator strengths (∼ µx 0a2) between exciton and biexciton bands mediated by the interband Born scattering (Λx,xx). First pathway (panel (a)), in- volves the ground-state-to-exciton resonant optical tran- sition and further scattering to final biexciton states dis- tributed around ωpm according to the non-zero compo- nents of Λx,xx. We will refer to this process as the indi- rect biexciton photogeneration throughout this paper.47 Panel (b) illustrates the second pathway, where the exci- ton is virtual and final biexciton state is in resonance with the optical pulse. Thus, this pathway will be refereed to as the direct biexciton photogeneration via virtual exciton states.15 The third pathway (panel (c)) consists of Born scattering between the ground state and biexciton states (Λ0,xx) stabilized by the intraband dipole transition, µxx. Accordingly, the final biexciton state is in resonance with the optical pulse. This process, refereed to as the direct biexciton photogeneration via biexciton states,55 becomes prohibited in the bulk limit by the momentum conserva- tion constraint for optical valence-conduction band tran- sitions. Equation (2) with the higher order corrections and Eqs. (4) -- (6) fully describe the exciton and biexciton pop- ulations prepared by the pump pulse. We use them for a nx k nxx Nx = (cid:80) a, and biexciton, Nxx = (cid:80) the numerical evaluation of the photogenerated exciton, k , popula- tions and QE (Eq. (1)). They also provide the initial conditions to model the population relaxation dynamics using a set of rate equations.47 Using this computational approach, our goal is to clarify the effect of quantum con- finement on QE in transition from NCs to the bulk limit. Since the CM dynamics in NCs takes place in the energy region characterized by high electron and hole DOS, we expect that the CM dynamics in NCs and in the bulk should have common features. Thus, we intend to see how strongly these features are affected by the presence of the confinement potential. Quantitatively, we are going to look at the interplay between the size scalings of the interband Coulomb inter- action and exciton/biexciton DOS determining the QE variation in transition from NC to the bulk limit. First, we define the bulk limit as the thermodynamic limit: V → ∞, V /v → ∞, and v = const., where V is a crystal volume, v is the unit cell volume, and the ratio V /v gives the number of unit cells in the crystal. Then Eqs. (2) and (6), and the quantities determining the population relaxation should be represented in such a form that the effect of the interband Coulomb interactions and exci- ton/biexciton DOS are clearly distinguished. This can be achieved by using the quasicontinuous energy repre- sentation. Since some quantities of interest in the bulk limit have volume scaling, it is convenient to introduce as- sociated intensive (i.e., volume-independent in the bulk limit) variables. Their deviation from the well defined bulk values will provide us with the convenient measure of the quantum confinement effects. We start with the exciton and biexciton populations which in the quasicontinuous energy representation read nx(ω) = nxx(ω) = aδ(ω − ωx nx a ), k δ(ω − ωxx nxx k ), (7) (8) (cid:88) (cid:88) a≥1 k≥1 Λx,xx (c) Λ0,xx µxx Exciton band Bi-exciton band € x0ωpm(b) Exciton band Bi-exciton band µx € x0ωpm(a) Exciton band Bi-exciton band µx Λx,xx € x0ωpmrespectively. Both of them have linear scaling with the volume in the bulk limit. Therefore, we eliminate the latter scaling by multiplying the total populations with the dimensionless prefactors (v/V ) and end up with the following intensive quantities (cid:90) ∞ (cid:90) ∞ 0 0 Nx = Nxx = v V v V nx(ω)dω, nxx(ω)dω. (9) (10) (cid:16) v (cid:16) v V (cid:17)2(cid:88) (cid:17)4(cid:88) a V k (cid:17)(cid:88) (cid:17)4(cid:88) a (cid:16) v (cid:16) v V V Although the prefactor v/V cancels out in the expression for QE (Eq. (1)), indicating that the latter quantity is indeed intensive, it is convenient to keep it in Eqs. (9) and (10) for consistency. According to Appendix A, we define the intensive ex- citon and biexciton DOS as ρx(ω) = ρxx(ω) = δ(ω − ωx a ), δ(ω − ωxx k ), (11) (12) respectively. The associated optically allowed exciton and joint biexciton DOS read ρx(ω) = ρxx(ω1, ω2) = µx a02δ(ω − ωx a ), µxx kl 2 k )δ(ω2 − ωxx l ), kl × δ(ω1 − ωxx (13) (14) respectively. They carry information on the optical se- lection rules reducing the number of states participating in the photogeneration process. Our central quantity is the effective Coulomb term de- fined as the r.m.s. of the interband Coulomb matrix ele- ments which couple states within the frequency intervals [ω1, ω1 + dω1] and [ω2, ω2 + dω2], (cid:19)2(cid:34)(cid:88) (cid:12)(cid:12)V x,xx (cid:12)(cid:12)2 (cid:21)1/2 a,m . V x,xx ef f (ω1, ω2) = (cid:80) δ(ω1 − ωx b δ(ω1 − ωx × a,m a )δ(ω2 − ωxx m ) n δ(ω2 − ωxx n ) v (cid:18) V b )(cid:80) (cid:19)3/2(cid:34)(cid:88) (cid:12)(cid:12)V x,xx 0,m m (cid:12)(cid:12)2 (cid:80) δ(ω − ωxx m ) n δ(ω − ωxx n ) (cid:35)1/2 .(16) Related effective Coulomb term, coupling the ground state and biexciton states, can also be defined as (cid:18) V v V xx ef f (ω) = As we discuss in Appendix A, the volume prefactor (V /v)2 [(V /v)3/2] in Eq. (15) [Eq. (16)] corresponds to a finite effective Coulomb value in the bulk limit. There- fore, the size scaling of such defined interaction with the NC diameter, d, provides quantitative measure of the (15) nxx(ω) = kII (ω)nx(ω) − kAR(ω)nxx(ω) 5 quantum confinement effects. In general, a deviation from the bulk value for the effective Coulomb reflects the net result of the scaling of the Coulomb matrix ele- ments with d, relaxation of the momentum conservation constraints, and the appearance of new selection rules associated with symmetry of the confinement potential. Assuming the cw excitation, the exciton population produced by the pump pulse is simply proportional to the corresponding optically allowed DOS (Eq. (13)), Nx(ωpm) = Aρx(ωpm), (17) where A = 2π3/2τpmE (0) /2. The biexciton population as a function of the pump frequency in the adopted rep- resentation reads pm 2 (18) (cid:48) (cid:90) (cid:90) (cid:90) (cid:48) dω Nxx(ωpm) = A 2 A 2 A 2 dω dω (cid:48) (cid:48) + + [V x,xx ef f (ωpm, ω (cid:48) (ω )]2 ρx(ωpm)ρxx(ω ) (cid:48) − ωpm)2 + γ2 )ρxx(ωpm) (cid:48) − ωpm)2 + γ2 (ω (cid:48) (cid:48) , ωpm)]2 ρx(ω [V x,xx ef f (ω (cid:48) [V xx ef f (ω (cid:48) ωpm) )]2 ρxx(ω (cid:48) 2 ω . As desired, Eq. (18) is volume independent and clearly distinguish the contributions from the effective Coulomb interactions and variously defined DOS. Accordingly, Eqs. (17) and (18) provide central expressions for the interpretation of the numerical simulation results on the photogeneration QE as discussed in the subsequent sec- tion. The population relaxation dynamics is described by the following set of kinetic equations nx(ω) = −kII (ω)nx(ω) + kAR(ω)nxx(ω) 0 (cid:90) ∞ (cid:90) ∞ (cid:90) ∞ (cid:90) ∞ 0 0 − + − + dω(cid:48) Γx(ω(cid:48), ω)nx(ω) dω(cid:48) Γx(ω, ω(cid:48))nx(ω(cid:48)), dω(cid:48) Γxx(ω(cid:48), ω)nxx(ω) dω(cid:48) Γxx(ω, ω(cid:48))nxx(ω(cid:48)). (19) (20) 0 These equations contain both the II and Auger recombi- nation rates kII (ω) = kAR(ω) = 2π (cid:16) v (cid:17)2 2 [V x,xx 2π 2 V ef f (ω)]2ρxx(ω), [V x,xx ef f (ω)]2ρx(ω), (21) (22) respectively. Here the shorthand notation V x,xx ef f (ω) stands for the diagonal component of the effective Coulomb term, i.e., V x,xx ef f (ω, ω). According ef f (ω) ≡ V x,xx to Eq. (21), the II rate is an intensive variable. In con- trast, the Auger recombination rate (Eq. (22)) vanishes in the the bulk limit as d−6. In the case where the II processes dominate the CM dy- namics, the ratio of the II to Auger recombination rates (cid:18) V (cid:19)2 ρxx(ω) v ρx(ω) kII (ω) kAR(ω) = , (23) should determine QE. Since this ratio is proportional to the ratio of the corresponding DOS, it has been proposed as a selection criterion for the materials showing efficient CM.65 However, Eq. (23) clearly shows that besides the material-specific signatures given by the ratio of the in- tensive DOS, it also contains the volume scaling factor. Therefore, we argue that Eq. (23) can only be used for a material selection criterion, after the volume prefactor is eliminated. The intraband relaxation rates, Γx, Γxx, in Eqs. (19) and (20) describe the phonon-assisted cooling. In the ab- sence of the phonon bottleneck and in the region of high exciton DOS, it is expected that single-phonon processes dominate the exciton and biexciton intraband relaxation. Accordingly, we calculate these quantities by using the following expression66,67 Γi(ω(cid:48), ω) = sign(ω(cid:48) − ω) 2Ji(ω(cid:48) − ω) exp[(ω(cid:48) − ω)/kBT ] − 1 , (24) where i = x, xx. kBT is thermal energy, and Ji(∆ω) is the phonon spectral density approximated by the ohmic form with exponential cutoff Ji(∆ω) = λi ∆ω ωc e−∆ω/ωc. (25) Here, the adjustable parameters are electron-phonon cou- pling, λi, with the constraint λxx = 2λx and the phonon frequency cutoff, ωc. Eqs. (19)-(25) represent the quasicontinuous represen- tation of the discrete kinetic equations introduced in Ref. [47]. We found that the former representation is more suitable for numerical integration. The solution of these equations with the initial conditions determined by the photogenerated populations (Eqs. (2) -- (6)) provide closed computational scheme that we use to determine the total QE. Next, we discuss the parameterization of the introduced model and some details on the adopted numerical techniques. B. Model Parameterization An accurate knowledge of single electron and hole wave functions is required to construct the exciton and biex- citon states and to evaluate transition dipoles and the interband Coulomb matrix elements. For this purpose, we adopt the effective mass k · p formalism, originally de- veloped by Mitchell and Wallis68 and Dimmock69 for the 6 bulk lead chalocogenide semiconductors, and further ad- vanced by Kang and Wise56 for the spherically symmet- ric NCs. The formalism is based on a four-band envelope function model explicitly taking into account spin-orbit interaction between valence and conduction bands. In considered PbSe and PbS materials, the band structure anisotropy is small, and therefore, neglected in our cal- culations. An electron and hole wave function obtained within this formalism reads Ψi(r) = 4(cid:88) F i m(r)um(r). (26) m=1 Here um(r) is m-th component of the bulk Bloch wave function associated with the band-edge states in L-valley whose index, m = 1, . . . , 4, denotes the bands.68,69 The envelope eigenfunctions F i m(r) and the corresponding eigenenergies, ωi, are found by solving the k · p Hamil- tonian eigenvalue problem with the infinite wall bound- ary condition, Fi(r = R) = 0, at the surface of a spheri- cal NC of radius R.56 If ωi > 0 (ωi < 0) we identify the state as a conduction band electron (valance band hole) state. The eigenstate index represents a set of quantum numbers, i = {n, π, j, m}, such as primary quantum num- ber describing number of the wave function nods, parity, total angular momentum, and its projection, respectively. In what follows, we will refer to the single particle states, described by Eq. (26), as the Kang-Wise (KW) states. In the bulk limit, we set the envelope functions, F i m(r), to the plane waves. A natural way to introduce the exciton and biexci- ton states is to use the second quantization representa- tion within the basis of the KW-states (Eq. (B1)). Tak- ing into account that the electron-hole Coulomb inter- actions are weak compared to their kinetic energies (i.e., strong confinement regime), we introduce exciton and bi- exciton states as the following configurations of uncorre- lated electron-hole pairs xa(cid:105) = c†q xxk(cid:105) = c†pc†q d†rx0(cid:105), d†r d†sx0(cid:105). (27) (28) As introduced above, the ground state, x0(cid:105), is attributed to the fully filled valence band. An exciton (biexciton) energy in terms of the electron, ωe p, and the hole ωh q q − ωh energies is simply: ωx p + s ). Note that the exciton index a = {p, q} ωe (biexciton index k = {pq, rs}) is a collection of two (four) sets of KW quantum numbers. r −ωh q −ωh a = ωe r (ωxx k = ωe The expressions used to calculate the dipole moments a0 and µxx µx kl , and the interband Coulomb matrix ele- ments, V x,xx, in Eqs. (2), (4), (5), (15), and (16) using associated matrix elements in the KW basis are given in Appendix B. Numerical calculations of the interband Coulomb KW matrix elements are performed by using the multipole expansion in terms of the Clebsch-Gordan coefficients. The dephasing rates for the high excited states are not available from experiment, and in our cal- culations we use some average value of γx = γxx/2 = γx,xx/2 = 50 meV. This value is consistent with the calculations performed on PbSe clusters.60,70,71 The numerical calculations of the photogenerated pop- ulations have been performed according to Eqs. (2) -- (5) including the first and second order corrections to Eq. (2). To account for the degeneracy of the exciton and biexci- ton states associated with the four equivalent L-valleys, we multiply the r.h.s. of Eq. (2) and Eq. (4) by factors 4 and 16, respectively. This assumes that the inter-valley Coulomb scattering processes are negligible. In most cal- culations, the pump pulse duration is set to τpm = 50 fs, which is typical value used in the experimental studies.91 The number of exciton and biexciton states to handle computationally is extremely high. Even for PbSe NCs of moderate size (d ∼ 5 nm), the number of the biex- citon states with energies below 4Eg is ∼ 105 . In the bulk limit, the number of the exciton and and biexciton states becomes infinite rendering the direct summation over these states in Eq. (4) impossible. Therefore, in both cases of finite-size NCs and the bulk, we use the Monte Carlo sampling which allows us to consider sig- nificantly larger NCs than it was possible in the work of Silvestri and Agranovich.57 To perform numerical calculations of the popula- tion relaxation processes, we first evaluated the in- terband Auger recombination and II rates given by Eqs. (21) -- (22). For this purpose, we smeared the delta- functions in the effective Coulomb term (Eq. (15)) and exciton/biexciton DOS (Eqs. (11) and (12)) with the Lorentzian profiles containing the corresponding dephas- ing rates and used the Monte Carlo procedure to sum over the exciton and biexciton states. To evaluate the phonon- induced intraband relaxation rates (Eqs. (24) and (25)), we set the room temperature value for kBT = 25 meV, the cutoff energy to ωc = 50 meV. The electron-phonon coupling λx = λxx/2 was fit to reproduce the experimen- tally observed intraband relaxation time in the range of 0.5 ≤ τph ≤ 5 ps.36,53 The kinetic equations (Eqs. (19) and (20)) have been further numerically integrated on the energy grid. All calculated observables are averaged over ensemble of NCs with the Gaussian size distribution with the standard deviation σ = 5%. III. INTERPLAY OF CM PATHWAYS IN PbSe In this section, we discuss the results of our numer- ical calculations of the photogeneration and population relaxation processes in NC and bulk PbSe to clarify the effect of the CM pathways interplay and quantum con- finement on QE. As we already mentioned, the QE is determined by the interplay between the DOS and the effective Coulomb scalings. Therefore, to set the stage for our analysis we first consider these quantities sepa- rately. Calculated energy dependence of the exciton and biex- 7 FIG. 2: Energy dependence of (a) exciton and (b) biexciton DOS calculated for various diameter, d, PbSe NCs and the bulk. (c) The diagonal component of the effective Coulomb term in PbSe NCs and the bulk. The inset shows actual interband Coulomb interaction in NCs defined as the effective Coulomb term with the volume prefactor added, i.e., V x,xx = (V /v)2V x,xx ef f . C citon DOS (Eqs. (11) and (12)) for different diameter PbSe NCs and the bulk limit are shown in panels (a) and (b) of Fig. 2, respectively. Obviously, the quantum confinement leads to the DOS reduction in NCs com- pared to the bulk. The energy dependence of the DOS follows power law. Specifically for the bulk limit, we find that ρx(ω) ∼ (ω − ωg)2.2 and ρxx(ω) ∼ (ω − 2ωg)6 (a) \x{FFFF}ω(eV)\x{FFFF}ω(eV)(b) (c) ρx(eV−1)ρxx(eV−1)\x{FFFF}ω(eV)Vx,xxeff(meV)Vx,xxC(meV)\x{FFFF}ω(eV)8 FIG. 4: Contributions of the photogeneration pathways to the biexciton quantum yield calculated for d = 15 nm PbSe NC. Black and red lines show the contributions of the two com- ponents associated with the first two pathways schematically shown in panels (a) and (b) of Fig. 1, respectively. The green line shows the contribution of the third pathway illustrated in panel (c) of Fig. 1 The inset presents the relative contribution of the third pathway to the total photogeneration quantum yield in various size NCs. prefactor, i.e., ∼ d−6. However, we remind that the vol- ume prefactor responsible for such dramatic scaling does not enter the quantities determining the QE except the Auger recombination rate (Eq. (17) -- (22)). A. Photogeneration QE The photogeneration QE as a function of the abso- lute photon energy in NC and bulk PbSe is shown in Fig. 3(a). According to the plot, the QE at fixed pho- ton energy monotonically increases with the NC diameter but does not exceed the bulk value. In contrast, the QE plotted on the unitless photon energy scale (Fig. 3(b)) shows the opposite trend useful for photovoltaic applica- tions. However, the absolute energy scale representation is more suitable to understand the physical mechanisms of photogeneration and, therefor, we continue our anal- ysis using this scale. First, we compare our calculations for small, d = 5 nm, NC with the calculations reported by Silvestri and Agranovich57 for d = 6 nm PbSe NC. In Ref. [57], the size dispersion of the NCs is set to 2% resulting in the sharp peaks reaching QE ≈ 1.2. The in- crease in the dispersion to realistic value of 5% (red solid line in Fig. 3(a)) washes out the peaks and reduces the QE by one order of magnitude, i.e. to QE ∼ 1.01. The contributions of different pathways (Fig. 1) to the biexciton photogeneration quantum yield, ηxx = QE − 1, are depicted in Fig. 4. First of all, we point out that for the energy region ωpm > 1.5 eV in which the biexciton FIG. 3: (a) The pump energy dependence of the photogener- ation QE for various diameter, d, NC and bulk PbSe plotted on the absolute pump energy scale. Solid lines represent the results obtained from numerically exact (Eqs. (2) -- (5)) cal- culations, and the dash indicates approximate calculations (Eq. (30)) for d=5,8,15 nm NCs and the bulk using the effec- tive dephasing rates γef f = 50, 40, 55, and 45 meV, respec- tively. (b) Solid lines from (a) plotted on the unitless pump energy scale (i.e., normalized by corresponding NCs and bulk band gap energies, Eg). with ωg = 0.28 eV reflecting contributions of the non- parabolic regions of the electron/hole band structures. Even in the high DOS regions where the bulk behavior is expected in NCs, the DOS converges rather slowly to the limiting bulk values as the NC diameter increases. The effective Coulomb as a function of energy is plotted ef f (ω) ∼ in Fig. 2 (c) and follows power law behavior, V x,xx ω−5. According to the plot, the effective Coulomb in- teraction is only weakly enhanced by quantum confine- ment, except for small, d = 5 nm, NCs. Specifically, the confinement results in a factor two enhancement in NCs with d = 8 nm as compared to the bulk. Therefore, the scaling of the actual interband Coulomb interaction, i.e. VC = (v/V )2V x,xx ef f , in NCs (inset to Fig. 2 (c)) in the region of high DOS should be dominated by the volume \x{FFFF}ωpm(eV)QEQE\x{FFFF}ωpm/Eg(a)(b)\x{FFFF}ωpm(eV)ηxx\x{FFFF}ωpm(eV)Rel.contr.9 tion via biexciton states (green line) has negligibly small contribution compared to the other terms. The inset shows that the relative contribution (i.e., normalized per total photogeneration QE) of the latter pathway for all considered NC sizes is small and has tendency to de- crease with the NC size growth and to vanish in the bulk limit. This contribution becomes small due to the strong off-resonant nature of the ground state to biexciton state effective Coulomb term (Eq. (16)) entering the third term in Eq. (18). The small contribution of the last pathway leads us to an important conclusion that the photogeneration QE is fully determined by a single II event occurring on the dephasing timescale. According to Eqs. (1) and (29) this QE can be approximated by the following simple expres- sion QE(ωpm) = 2kII (ωpm) + γef f kII (ωpm) + γef f ≈ 1 + kII γef f , (30) where kII is the II rate given by Eq. (21), and γef f is effective interband dephasing rate. Note that the optical selection rules do not enter Eq. (30). The approximate form of QE given in Eq. (30) is quite general, since the approximation is based on a general fact of weak inter- band Coulomb interaction. To verify this relation, we calculated the QE using Eqs. (21) and (30), and com- pare the results (dashed lines) in Fig. 3(a). In general, the dashed lines well reproduce the trends in the behav- ior of the associated solid lines confirming the validity of the approximation. Observed small discrepancies are due to the numerical noise, phenomenological origin of the effective dephasing rate neglecting its frequency de- pendence, and neglect of the correlations between the effective Coulomb and DOS fluctuations during the NC ensemble averaging. Both direct biexciton photogeneration pathways via exciton and biexciton states have been studied before.15,55,57 The indirect biexciton generation as we demonstrated above is also important and should not be omitted. Initial studies of the photogeneration pathways15,55 also based on the KW parameterization reported significantly larger QE compared to our results and the results report in Ref. [57]. The cause of the over- estimate is the disregard of the Coulomb coupling se- lection rules and the oscillator strengths factors weight- ing optically allowed transitions.57 Next, we show that the interference related power-scaling of the DOS not ac- counted for in the previous studies further leads to sig- nificant reduction in the photogenerated QE. Provided, the terms in Eq. (4) associated with the direct biexciton photogeneration processes via exciton and biexciton states are interfering constructively, one can show that associated biexciton populations scale as (cid:48) (cid:48) xx]2ρxx, respectively. This is exactly the x]2ρxx and [ρ [ρ (cid:48) xx) is DOS for case considered in Refs. [15,55] Here, ρ the intermediate exciton (biexciton) states, and ρxx the biexciton DOS at the pump frequency. Since, we found that there is no interference between the photogenera- (cid:48) x (ρ FIG. 5: Calculated photogeneration QE as a function of pump energy for PbSe NC of d = 15 nm and PbSe bulk. For compar- ison, we also show experimentally measured total QE in PbSe NCs34 with the diameter varied in the range of 5 ≤ d ≤ 8 nm and in the bulk36. The inset shows calculated QE for PbSe NC of d = 5 nm on extended pump energy scale compared to the experimentally measured total QE for the same size NC. DOS shows steep growth, the indirect biexciton photo- generation (black line) and direct biexciton photogener- ation via exciton state (red line), provide identical con- tributions. This behavior can be interpreted by looking at the related first and second terms in Eq. (18) in which (cid:48)−ωpm)2 +γ2, the resonant nature of the denominator, (ω corresponds to the leading contribution of the diagonal ef f (ωpm) ≡ component of the effective Coulomb term V x,xx V x,xx ef f (ωpm, ωpm). Hence, the integral convolutions can be evaluated resulting in the identical two terms whose net contribution to the biexciton population is (cid:48) xx(ωpm) = kII (ωpm) Nx(ωpm)/γef f . N (29) Here, kII is the II rate given by Eq. (21), and γef f is ef- fective interband dephasing rate. Derived expression has a very clear physical interpretation: The photogenerated biexciton population associated with the first and second pathways is a result of a single II event taking place on the dephasing timescale, γ−1 ef f , and following optical prepara- tion of the exciton states.92 The direct biexciton photogeneration via exciton states is manifested by the direct dependence of the biexciton DOS on ωpm in the second term of Eq. (18). As a result, for ωpm < 1.5 eV, this contribution (red line in Fig. (4)) drops quickly. Accordingly, ωpm = 1.5 eV can be iden- tified as the AET for the photogeneration processes in d = 15 nm NCs. In contrast, the indirect biexciton pho- togeneration (black line in Fig. (4)) is resonant at exciton states, and decreases relatively slow for ωpm (cid:46) 1.5 eV. According to the first term in Eq. (18), this behavior re- flects the behavior of the off-diagonal effective Coulomb component V x,xx (cid:48) (cid:46) 1.5 eV. (cid:48) ef f (ω , ωpm) at ω According to Fig. 4, the direct biexciton photogenera- QE123451234Theory, BulkTheory, NC (d=15 nm)Exp, BulkExp, NC23411.051.1Theory, NC (d=5 nm)Exp, NC (d=5 nm)\x{FFFF}ωpm(eV)10 FIG. 6: Energy dependence of the calculated II rates in PbSe NCs of different diameter and in PbSe bulk. The inset shows associated Auger recombination rates calculated for NCs us- ing (Eq. (23)). (cid:48) (cid:48) tion pathways, the first two terms in Eq. (18) scale as xρxx, and the last one as ρxx ∼ ρxx ρ ρxx. The latter linear dependence on the intermediate exciton and biex- citon DOS versus former quadratic one15,55 significantly decreases the biexciton photogeneration quantum yield and, in fact, allows us to interpret the photogeneration dynamics as a single II event (Eq. (29))! In Fig. 5, we compare the experimentally measured total QE34,36 with the calculated photogeneration QE. The comparison with experimental data for NCs is pos- sible only at the photon energy fixed at ∼ 3.1 eV, where the vertical aligned diamonds represent the QE measured for the NCs with diameters varying in the range between 5 nm and 8 nm. The comparison shows that for the bulk and for the NCs of comparable diameter the experimen- tally measured total QE significantly exceeds the calcu- lated photogeneration QE. Accordingly, we conclude that the biexciton photoexcitation along does not explain the experimentally observed QE values. Next, we investigate the contribution of II events to the total QE taking place during the phonon-assisted cooling. B. Effect of population relaxation and pump pulse duration on total QE Central quantity defining QE during the population relaxation (and as concluded above during the photogen- eration event as well) is the II rate (Eq. (21)). Fig. 6 shows calculated energy dependence of the II rate for NC and bulk PbSe. According to the plot, the rate scales linearly away from AET. The scaling directly fol- lows from Eq. (21) in which one can insert the effec- ef f (ω) ∼ ω−5, and biexciton DOS, tive Coulomb, V x,xx ρxx(ω) ∼ (ω − 2ωg)6, whose scalings were determined FIG. 7: Pump energy dependence of the total QE in PbSe. (a) Calculations performed for bulk PbSe using various intra- band relaxation times, τph. Experimental data from Ref. [36] are shown for comparison. (b) Calculations for both NCs and bulk using τph = 1 ps (solid lines) and τph = 2 ps (dash). For comparison the experimental data from Refs. [34] and [36] are plotted in diamonds and dots, respectively. at the beginning of this section (Fig. 2). In the lit- erature, quadratic and even higher power behaviors of the II rate are obtained near the AET. As proposed in Refs. [50,72,73] such scalings solely follow from the DOS behavior. However, these theories do not account for the long-range interband Coulomb corrections at the energies higher than AET. As we show in Appendix B, the first non-vanishing k · p contribution to the matrix elements of the interband Coulomb term yields additional factor ω−1/2. As a result, one gets kII (ω) ∼ (ω − ω0)2/ω ∼ ω at ω (cid:29) ω0, which is in agreement with our numerical calculations. According to Fig. 6, the II rate increases as the NC diameter increases but does not exceed the bulk values. This observation is a result of the weak effective Coulomb enhancement in NCs observed in Fig. 2 (c). In agreement with previously reported studies,36 we find that this en- hancement is fully suppressed by the reduction in the biexciton DOS (Fig. 2 (b)). As we already pointed out, strong size scaling of the actual interband Coulomb in- kII(ps−1)\x{FFFF}ω(eV)kAR(ps−1)\x{FFFF}ω(eV)12340.511.522.5Theory (!ph=2 ps)Theory (!ph=1 ps)Theory (!ph=0.5 ps)ExpPhotogenerationQE\x{FFFF}ωpm(eV)\x{FFFF}ωpm(eV)QE(b) (a) 11 value. For comparison, the contribution of the photogener- ation to the total QE is shown in Fig. 7 (a). Its small contribution is explained by the fact that the photogener- ation pathways reduce to a single II event taking place on the short (subpicosecond) dephasing timescale (Eq. (30)). In contrast, the population relaxation dynamics occurs on a much longer (picosecond) timescale allowing for the multiple II events that make major input into the total QE. Comparison of the calculated total QE in both NC and bulk PbSe with experimental results34,36 is shown in Fig. 7 (b). The total QE dependence on the NCs diameter both in theory and experiment follows the same trend as the II rate. Solid lines in Fig. 7 (b), show QE calculated for τph = 1.0 ps which gives the best fit for the bulk. However, taking into account that the upper values of the measured QE in the NCs are associated with the NCs diameter of d = 8 nm, we had to increase the phonon relaxation time up to τph = 2 ps for the NC with d = 8 nm (blue dash) in order to get better agreement with experiment. The observed increase of τph can be rationalized by the quantum-confinement-induced increase of the level spacing.74 Determination of AET is more convenient to perform using the unitless energy scale, i.e., the pump energy nor- malized per corresponding Eg of NC or bulk. Fig. 8 presents the same curves as in Fig. 7 (b) for selected τph plotted on this scale. Dashed lines extrapolate the linear portion of the curves at the energies below 3.5 eV. Their intercept points with the horizontal, QE = 1, line provide the AET values. The deviation from linear be- havior at low energy ends and apparent CM below the en- ergy conservation threshold, 2Eg (red and green curves) are merely due to the ensemble averaging entering our calculations.93 The calculated values of AET are summarized in Ta- ble I. For the NCs with d = 5 nm and d = 8 nm, the calculated AET≈ 2.2Eg is in excellent agreement with experiment.34 Furthermore, the photogeneration AET=1.5 eV for d = 15 nm PbSe NCs (determined in Sec. III A, Fig. 4) is close to that, 1.4 eV, listed in Ta- ble I. The latter is an additional indication of the "pure" II nature of CM. For the bulk however, there is a disagree- ment between the calculated, 3.7Eg, and experimentally observed, 6Eg, AET.36 The discrepancy is most likely PbSe d (nm) Eg (eV) AET/Eg AET (eV) NC 5 NC 8 NC 15 Bulk ∞ 1.10 0.66 0.42 0.28 2.3 2.2 3.3 3.7 2.5 1.5 1.4 1.0 TABLE I: AET in PbSe NCs and bulk calculated as the in- tercept point between dashed lines and the solid horizontal line at QE=1 in Fig. 8. FIG. 8: Comparison of the calculated total QE for selected τph and experimental data from Fig. 7 (b) plotted on the unitless pump energy scale (i.e., normalized per corresponding band gap energy, Eg). Dashed lines are extrapolation of the solid lines linear regions to determine the AET values (see Table I). teraction is determined by the volume prefactor (v/V )2 which does not enter the II rate, and therefore, has no effect on the rate. In contrast, the volume prefactor appears in the ex- pression for the Auger recombination rate (Eq. (22)), and as we show in the inset to Fig. 6, makes this effect negli- gible. On the frequency scale, the Auger recombination rate drops as kAR ∼ ω−3 further decreasing its contribu- tion to QE at high energies. As a result, the only region where the Auger recombination processes are significant is near the band edge. However, this region has no sig- nificant contribution to QE, and we conclude that the Auger recombination processes can be fully neglected in the present analysis. Since the Auger recombination processes are weak, ef- ficient CM during the population relaxation is expected if the II rates are comparable to or higher than the phonon- assisted population relaxation rates. To evaluate the up- per boundary for QE in NCs, we first calculate total QE in bulk PbSe for typical population relaxation times τph = 0.5, 1.0, and 2 ps. The results are plotted on the pump energy scale in Fig. 7 (a) and also compared with published experimental data. In general, the calculations reproduce the experimental trends and provide the best fit at τph = 1.0 ps. The discrepancy between theory and experiment in the interval 1.0 < ωpm < 3.5 eV is due to the phenomenological nature of the phonon-assisted re- laxation model that lacks exact knowledge on the spectral dependence of the electron-phonon coupling and phonon DOS. Above 3.5 eV the effect of triexciton generation and possibly the contribution from the higher energy bands take place. Hence, our theory is valid below this energy \x{FFFF}ωpm/EgQE12 FIG. 9: Calculated dependence of the total QE for PbSe NC of d = 8 nm on the pump pulse duration τpm. The relaxation time is set to τph = 1 ps. related to the phenomenological model of the phonon- assisted relaxation. Getting back to Fig. 7 (a), one can clearly see that the best fit to the initial rise of QE can be obtained for τph = 0.5 ps. However, for adopted τph = 1 ps the model fails to accurately reproduce the ini- tial slope resulting in the underestimated values of AET. All calculations discussed above are performed for the cw pulses typically used in ultrafast spectroscopic stud- ies. However, estimated solar light correlation time is about 5 fs. Hence in light of photovoltaic applications, it is natural to calculate QE as a function of the pump pulse duration, τpm. Such a dependence evaluated for various photon energies, ωpm, is shown in Fig. 9. First of all, we notice that the cw regime is reached at τpm > 10 fs. For τpm < 10 fs, the QE associated with ωpm = 3.15 eV shows no variation whereas QE associated with the lower excitation energies show insignificantly small increase. Observed weak dependence of QE on the pulse duration suggests that, for the gaussian pulses, the expected in- crease in the QE due to the increase in the pulse duration is equally compensated by the reduction in the number of states populated by the pulse whose spectral widths is narrowed. IV. COMPARISON OF QE IN PbS AND PbSe Both PbS and PbSe have electronic structures de- scribed by the same KW effective mass model with some different parameters.56 This implies that our conclusions on the interplay between different CM pathways in PbSe fully apply to PbS NCs and bulk materials. In this sec- tion, we perform comparison of the key quantities such as exciton/biexciton DOS, effective Coulomb terms, the II rates, and the resulting QEs calculated for these semicon- ductors. Our numerical tests show that these quantities FIG. 10: Energy dependence of (a) exciton DOS, (b) biex- citon DOS, and (c) effective Coulomb term in the bulk PbS (solid black) and PbSe (dashed black). In all panels, red lines show corresponding quantities calculated using PbSe parame- ters except for the band gap energy replaced with that of PbS. In panel (c), green line shows the effective Coulomb term cal- culated for PbSe with both band gap energy and the effective masses substituted with their corresponding PbS values. are weakly affected by the differences in the Kane mo- mentum and the unit cell volume. Hence, the dominant contributions come from the interplay between the band gap energies, carriers effective masses, and dielectric con- stants. Panels (a) and (b) in Fig. 10 clearly show that both the exciton (Eqs. (11)) and biexciton (Eq. (12)) DOS 0.0010.010.1111.11.21.3QE\x{FFFF}ωpm=3.15eV\x{FFFF}ωpm=2.5eV\x{FFFF}ωpm=2.0eVτpm(ps)(a) \x{FFFF}ω(eV)\x{FFFF}ω(eV)(b) (c) ρx(eV−1)ρxx(eV−1)\x{FFFF}ω(eV)Vx,xxeff(meV)13 FIG. 11: Energy dependence of the photogeneration QE in the bulk PbS (solid black) and PbSe (dashed black). Red (green) line shows the QE calculated for PbSe with the band gap energy (band gap energy and the effective masses) replaced with that (those) of PbS. in bulk PbS exceed the corresponding DOS in PbSe across the whole spectral range of interest. Compared to PbS (Eg = 0.28 eV), PbSe has larger band gap energy (Eg = 0.41 eV) and heavier carriers effective masses.56 In order to isolate the band gap effect, we plot the PbSe DOS calculated with the band gap energy replaced by that of PbS (red dash). This replacement results in a reduction of the DOS values (red dash). Hence, the ma- jor contribution to the steeper growth of the exciton and biexciton DOS in PbS comes from the heavier effective masses. Fig. 10 (c) compares the effective Coulomb terms, V x,xx ef f , (Eq. (15)) calculated for bulk PbSe and PbS. In contrast to the DOS, the effective Coulomb interaction in PbS is weaker than that in PbSe. To understand the trends, we alter the PbSe band gap value in the same way as for the DOS, and observe small reduction in V x,xx ef f (red dash). Altering the effective masses of PbSe with those of PbS, significantly reduces the V x,xx lowering ef f the curve (green dash) below V x,xx in PbS (solid black). ef f The gap between these curves is purely due to the dif- ference in the dielectric constants which in PbS ( = 17) is lower than in PbSe ( = 23). Therefore according to Fig. 10 (c), the narrower band gap value and lighter ef- fective masses make dominant contribution V x,xx in PbS ef f making it smaller than in PbSe. Next, we compare the photogeneration QE in bulk PbS and PbSe plotted in Fig. 11. The plot also shows photo- generation QE values for PbSe with the band gap energy and effective masses altered in the same way as in the case of the effective Coulomb interaction. (The color code is the same as in Fig. 10 (c).) By taking into account the linear dependence of the photogeneration QE on the II rate (Eq. (30)), one can conclude that the trends in the FIG. 12: Energy dependence of (a) II rate and (b) total QE calculated for PbS (solid lines) and PbSe (dash) NCs, and the bulk. For all curves in panel (b), the phonon relaxation time is set to τph = 1.0 ps. behavior of the QE is a result of the interplay between the trends in the biexciton DOS and effective Coulomb term shown in panels (b) and (c) of Fig. 10, respec- tively. Specifically, the lower dielectric constant of PbS contributes to the increase in the photogeneration QE and the competing effect of the lighter effective masses on the DOS and V x,xx results in additional increase of ef f the QE in bulk PbS. This net increase fully overruns the reductions associated with the band gap decrease making the photogeneration QE and the II rate (Fig. 12 (a)) in bulk PbS to be higher than in bulk PbSe. To demonstrate that the size quantization does not change the trends, we plot the II rate calculated for PbS and PbSe NCs and the bulk in Fig. 12 (a). It is clear from the plot that the PbS II rate always exceeds the PbSe II for identical in diameter NCs and for the bulk. Similar to PbSe, the PbS II rate monotonically increases with the NC diameter but does not exceed the bulk values. Finally in panel (b) of Fig. 12, we plot the calculated \x{FFFF}ω(eV)QEkII(ps−1)\x{FFFF}ω(eV)\x{FFFF}ωpm(eV)QE(a) (b) total QE of PbS and PbSe NCs and the bulk using the same phonon relaxation time set to τph = 1.0 ps. The trends are the same as in panel (a) demonstrating that the overall calculated QE in PbS is higher than in PbSe. The latter is a result of the interplay between the band gap energies, effective masses and the dielectric constant values as discussed above. Ultrafast measurements of QE in bulk PbS and PbSe, show that QE for both materials are approximately the same.36 However, the atomistic calculations reported along with the experimental data demonstrate that for identical phonon relaxation times (specifically, τph = 0.5 ps) the calculated QE in PbS exceeds that in PbSe. This observation is in direct agreement with our calcu- lations. Taking into account that both models show the same trend in QE for identical phonon relaxation times, we conclude that the II rate in bulk PbS should be higher than in PbSe, and consequently, the photogeneration QE. Furthermore, we extend this conclusion to PbS NCs ac- cording to the results shown in Fig. 12. Recently reported ultrafast measurements demon- strated significant reduction in QE of large PbS NCs com- pared to PbSe NCs.75 The authors rationalize this obser- vation by estimating the phonon-induced energy loss rate which in PbS turns out to be factor two faster compared to PbSe. Rigorous calculation of the phonon-induced re- laxation is beyond our model capability forcing us to introduce phenomenological τph. Since remaining elec- tronic structure parameters in our and reported atom- istic models are well validated, the calculations directly support the idea of the QE reduction in PbS merely due to the fast phonon-induced relaxation processes charac- terized by τph. V. DISCUSSION AND CONCLUSIONS The effective mass model adopted in this paper has significant limitations accounting only for the L-valley optical transitions.56 It fails to catch contributions from the higher energy transitions originating at other Bril- louin zone points. Specifically in PbSe, the Σ-point tran- sitions show large contribution to the absorption spec- tra above 1.6 eV.76 Experimentally, the CM dynamics is probed using ultrafast transient absorption and fluo- rescence techniques.34 These techniques explicitly deter- mine the number of carriers accumulated near the band edge of the lowest in energy L-valley. Following the ex- citation with a high energy optical pump, the excitons and biexcitons can be generated not only in L-valley (de- noted L-excitons and biexcitons) but also at Σ-point (de- noted Σ-excitons and biexcitons). The contribution of the Σ-excitons to the transient ultrafast signal depends on the mixing between the L and Σ-points. Rigorous calculations of the mixing can be done through atom- istic calculations. However, such calculations become tremendously expensive for the considered large diameter NCs. Below, we provide quantitative discussion of possi- 14 ble mixing mechanisms and rough estimates of the their contribution to QE for the pump energy below 3.5 eV. First, we consider mixing between valleys due to the size quantization which could be efficient in very small (d (cid:46) 3 nm) NCs at low excitation energies. How- ever, for the large NCs, considered here, the excitations with the energies higher than 2.5 eV are effectively of the bulk-type76,77 resulting in the negligible quantum- confinement-induced mixing. Another option is the II event initiated by the optically prepared Σ-exciton and resulting in the direct production of an extra L-exciton. Specifically, the excess energy of the Σ-exciton is trans- ferred to create an extra electron-hole pair within L- valley. The AET for this process can be roughly esti- mated as AETΣL ≈ AETL + (EΣ g ) = 3.3 eV point- ing to an efficient CM at photon energies ωpm (cid:38) 4 eV. Note that efficient CM events fully constrained to Σ-point should occur at even higher energies which are above the photon energy range considered in this paper. g − EL Finally, the optically prepared Σ-exciton can partic- ipate in the II event resulting in the creation of the electron-hole pair through the inter-valley transition. For instance, a higher energy conduction band electron from Σ-point can be transferred to L-valley releasing in the excess energy to create an additional electron-hole pair through the L-valence band to Σ-conduction band tran- sition. As we mentioned above, most of the states in- volved in efficient CM are of the bulk-type with the quasi- momentum being a "good" quantum number. Therefore, the Coulomb matrix elements describing the II processes become ∼ 1/∆q2 where ∆q is the difference between the electron/hole initial and final states. Accordingly, the II events involving the inter-valley transitions require large variation in the quasi-momentum, ∆q, and, therefore, becomes less favorable. Experimental verification of the discussed interplay be- tween the CM pathways contributing to the photogen- eration QE requires direct measurements of the inter- band Coulomb interactions. Distinguishing the inter- and intraband Coulomb interactions is a challenging task, since in the transient absorption and time-resolved fluo- rescence experiments these two components contribute to a measured total energy shift between the exciton and biexciton bands. Recently, we have theoretically shown that the two-dimensional double-quantum coher- ence spectroscopy is capable to probe directly the inter- band Coulomb interactions.78 Alternatively, the photo- generation QE can be determined based on its linear re- lationship to the II rate which has been established above (Eq. (30)). The later rate can be obtained by fitting the total QE as a function of the excitation energy.79 How- ever, the determination of the effective dephasing rate, also entering Eq. (30), might require use of coherent ex- perimental technique (e.g., photon echo) and/or addi- tional theoretical calculations. The absolute photon energy scale, ωpm, is used throughout this paper (except Fig. 3(a) and 8). This scale is more useful to discuss fundamental physical mechanisms behind the CM dynamics, e.g., the role of the quantum confinement.25,63 However, performance of pho- tovoltaic devices can be characterized by various power conversion efficiencies.34,63 In particular, by the QE cal- culated on a dimensionless energy scale which is nor- malized per NC or bulk band gap, ωpm/Eg. Despite lower values of QE in NCs compared to the bulk which show up on the absolute photon energy scale, the perfor- mance of prospective photovoltaic devices based on NCs can overrun their bulk counterpart.35,63,79 By compar- ing Fig. 8 with Fig. 7 (b), one can notice that smaller size NCs have better rise in QE than large ones and the bulk. This results from the quantum-confinement- induced blue-shift of the band gap energy.44 In light of this effect and the requirement to match solar radiation peak energy, semimetal NCs can become more efficient solar energy converters.80 Further enhancement of the interband Coulomb inter- actions in NCs can possibly be reached in NC-metal het- erostructures in which broad surface plasmon response is tuned in resonance with the exciton states participat- ing in CM. This issue requires additional study, since besides the enhancement of the Coulomb interactions, the surface-plasmons-induced ohmic and radiative energy losses can become significant.81 On the other hand, ad- ditional enhancement of QE can in principle be achieved by adding the surface states and ligands. This question is difficult to analyze using adopted effective mass approx- imation. However, the atomistic calculations30,31,62,82 -- 84 combined with the IESM should be extremely helpful in addressing this problem. In conclusion, we have performed systematic numeri- cal investigation of CM mechanisms in NC and bulk PbSe and PbS using our IESM parametrized by the effective mass KW model. We focused on the role of quantum confinement on the photogeneration and total QE. The analysis of the photogeneration pathways resulted in un- expected conclusion that the photogeneration processes reduce to a single II event due to the complete destruction of the associated pathways interference. This allowed us to explains the minor role of the photogeneration in total QE dominated by the multiple II events occurring during the phonon-induced population cooling. Comparison of the size scaling of the effective Coulomb interaction and the biexciton DOS in transition from NC to the bulk limit showed that weak enhancement of the former quantity is overrun by a significant reduction in the latter one. This explains the higher values of the QE in bulk compared to NCs as plotted on the absolute energy scale, and well agrees with the previously reported experimental studies and some theoretical predictions. However, the quantum confinement induced increase in Eg makes NCs more ef- ficient than bulk for practical photovoltaic applications. We have also found that the variation in the pump pulse duration does not significantly change the QE. Compar- ison of QE in PbSe and PbS suggests that the II pro- cesses are more efficient in PbS. However, variation of the material-dependent cooling time can strongly affect the total QE. Finally, we have identified the limitations of our model and defined its applicability range. 15 Acknowledgments K.A.V. acknowledge support of the Center for Ad- vanced Solar Photophysics (CASP), an Energy Frontier Research Center funded by the U.S. Department of En- ergy (DOE), Office of Science, Office of Basic Energy Sciences (BES). A.P. is supported by Los Alamos LDRD program. Both authors acknowledge CNLS and CINT for providing computational facilities, and wish to thank Victor Klimov, Darryl Smith, and Sergei Tretiak for stim- ulating discussions and comments on the manuscript. Appendix A: Volume scalings of DOS, transition dipoles, and the effective Coulomb interaction In this Appendix, we derive the volume normalization prefactors entering Eqs. (11) -- (16). The prefactors can- cel out the volume dependence of the latter quantities making them intensive variables in the bulk (i.e., ther- modynamic) limit. To preserve the dimensionality of the intensive variables, we use the V /v ratio instead of V where v is the unit cell volume. This ratio defines number of unit cells and goes to infinity in the bulk limit. Here, we also use the relationships derived in Appendix B that connect the transition dipole movements and Coulomb matrix elements represented in the single-particle (KW) carrier basis set and in the exciton/biexciton basis. We start with the simple fact that in the bulk limit, the single-particle DOS is proportional to the system volume V.85 Defining the exciton (biexciton) DOS as a joint DOS of an electron and a hole (two electrons and two holes), one immediately finds that they have V 2 (V 4) scalings. Therefore, we introduced v2/V 2 (v4/V 4) prefactor into Eq. (11) (Eq. (12)). To obtain the volume prefactors in the optically al- lowed DOS given by Eqs. (13) and (14), we first notice that the transition dipole matrix element for a single car- rier has no volume scaling and accounts for the total mo- mentum conservation, i.e., Mij ∼ V 0Mki,kj δki,kj .86 Fur- ther using Eq. (B2), we find that the optically allowed exciton DOS scales as(cid:88) (A1) (cid:88) a µx a02δ(ω − ωx M eh a ) ∼ 2δ(ω − ωe kp,kp − ωh kp ) ∼ V. kp kp analysis that for a fixed energy interval,(cid:80) Here and below, we use the same argument as in the DOS ∼ V . Using the first term in the expansion of the intraband biexciton dipole moment given by Eq. (B3), one finds that the joint kp (cid:88) (cid:88) kl optically allowed biexciton DOS scales as k )δ(ω2 − ωxx l ) ∼ − ωe 2δ(ω1 − ωe kl 2δ(ω1 − ωxx µxx (cid:88) M ee kp,kp kp kq (A2) − ωh ks) − ωh kr kpkq krks ×δ(ω2 − ωe (cid:48) kp − ωe kq − ωh kr − ωh ks ) ∼ V 4. According to Eqs. (A1) and (A2), we introduced the pref- actors v/V and (v/V )4 into Eqs. (13) and (14), respec- tively. To determine the volume scaling of the effective Coulomb term (Eq. (15)), we first evaluate the scaling of the following auxiliary quantity V x,xx a,m 2δ(ω1 − ωx a )δ(ω2 − ωxx m ). K(ω1, ω2) = (cid:88) (A3) a,m According to Eq. (B10), the Coulomb matrix ele- ments in the free carriers basis scales as Vij,kl ∼ V −1Vkikj ,kkklδki−kl,kj−kk . Then using Eq. (B6), one finds that K(ω1, ω2) ∼ V −2 (cid:88) V eehe ki,kj ,kk,ki+kj +kk 2 (A4) ×δ(ω2 − ωe ×δ(ω1 − ωe ki − ωe kj ki+kj−kk kikj kkkl − ωh − ωh − ωh kl ) kl ) ∼ V 2. kk By taking into account that K should be normalized by the exciton and biexciton DOS and take square root, we introduce the prefactor (V /v)2 into Eq. (15). The volume prefactor in Eq. (16) can be obtained in the same way with the help of Eqs. (B5), (B10), and (B11). Appendix B: Representation of transition dipoles and interband Coulomb matrix elements in KW basis set In this Appendix, we provide closed expressions for the exciton and biexciton transition dipoles and the in- terband Coulomb matrix elements used in the numerical calculations. To derive these expressions we follow the procedure outlined in Appendices A-C of Ref. [47]. The second quantization is performed using the basis of KW states, {Ψi(r)} defined in Eq. (26), by introducing the following field operators (cid:88) (cid:88) i (cid:104) (cid:104) (cid:105) Θ(Ei)Ψi(r)ci + Θ(−Ei)Ψi(r)d†i (cid:105) Θ(Ei)Ψ∗i (r)c†i + Θ(−Ei)Ψ∗i (r)di , Ψ(r) = Ψ†(r) = , (B1) i where Θ(E) is the step function, ci and di (c†i and d†i ) are conduction band electron and valence band hole an- nihilation (creation) operators, respectively. 16 Using this representation and the definition of the ex- citon and biexciton states given by Eqs. (27) and (28), it is straight forward to show that the transition matrix elements entering Eqs. (2) and (4) are a0 = (cid:104)xa Mx0(cid:105) = Meh µx qr , kl = (cid:104)xxk Mxxl(cid:105) = [δrr(cid:48)δss(cid:48) − δrs(cid:48)δr(cid:48)s] µxx (B2) (B3) × [Mee p(cid:48)pδqq(cid:48) − Mee + [δpp(cid:48)δqq(cid:48) − δpq(cid:48)δp(cid:48)q] × [Mhh s(cid:48)sδrr(cid:48) − Mhh q(cid:48)pδqp(cid:48) − Mee p(cid:48)qδpq(cid:48) + Mee q(cid:48)qδpp(cid:48)], s(cid:48)rδsr(cid:48) − Mhh r(cid:48)sδrs(cid:48) + Mhh r(cid:48)rδss(cid:48)], respectively. In Eq. (B2), the exciton index a = {q, r} and Meh qr is the interband electron-hole transition dipole matrix element calculated in the KW basis set. In Eq. (B3), the biexciton indices are k = {p(cid:48)q(cid:48), r(cid:48)s(cid:48)} and l = {pq, rs} as well as Mee ij are the matrix elements of the intraband transition dipole operator cal- culated using the electron and hole KW wave functions, respectively. ij and Mhh In the envelope function approximation, the KW tran- sition matrix elements entering Eqs. (B2) and (B3) have the following generic form56 Mij = M(1) ij + M(2) ij , 4(cid:88) m=1 (cid:90) (cid:90) M(1) ij = M(2) ij = Plz −[F i m(r), m(r)]∗ pF j dr [F i (cid:16) 1(r)]∗F j [F i dr 4 (r) − [F i (cid:17) 3 (r) + [F i 3(r)]∗F j 1 (r) 2(r)]∗F j 4(r)]∗F j 2 (r) , (B4) where the indices i, j can be associated with the elec- tron and hole states, and p = −i∇. Pl stands for the longitudinal dipole moment component of the band-edge Bloch function, um(r), and z is the unit vector in (cid:104)111(cid:105) direction of the PbSe (PbS) lattice. Both M(1) and M(2) can be evaluated analytically in the bulk limit, where M(1) vanishes identically.94 Using the same approach as above, one can show that the Coulomb matrix elements entering Eqs. (5), (15), (16), and (A4) are (B5) (B6) (cid:1) δs(cid:48),r pqrs − V eehh pqsr , q(cid:48)p(cid:48)r(cid:48)q − V eehe p(cid:48)q(cid:48)r(cid:48)q q(cid:48)p(cid:48)qs(cid:48) − V eeeh q(cid:48)rs(cid:48)r(cid:48) − V ehhh p(cid:48)rr(cid:48)s(cid:48) − V ehhh l,0 = (cid:104)xxl V x0(cid:105) = V eehh V xx,0 V xx,x k,a = (cid:104)xxk V xa(cid:105) =(cid:0)V eehe p(cid:48)q(cid:48)qs(cid:48)(cid:1) δr(cid:48),r q(cid:48)rr(cid:48)s(cid:48)(cid:1) δp(cid:48),q p(cid:48)rr(cid:48)s(cid:48)(cid:1) δq(cid:48),q. 4(cid:88) (cid:2)F i m(r1)(cid:3)∗ F l + (cid:0)V eeeh + (cid:0)V ehhh + (cid:0)V ehhh (cid:90) (cid:90) × 4(cid:88) (cid:2)F j n(r2)(cid:3)∗ F k r1 − r2 dr1dr2 n (r2). e2 m=1 (B7) These quantities depend on the long-range contributions whose matrix elements in the KW basis read Vijkl = m(r1) n=1 Here, denotes the screened dielectric function values evaluated at the optical frequencies. For bulk PbSe and PbS, we set bulk=23 and bulk=17, respectively.56 The dielectric constant in the NCs has been evaluated using the following expression87 where Vki,kj ,kk,kl = (cid:2)F i mF j n (cid:88) m,n e2 (cid:3)∗ F l mF k n 4π ki − kl2 . (B11) 17 N C(d) = 1 + (bulk − 1) (Ebulk (EN C g g + ∆E)2 (d) + ∆E)2 , (B8) where Ebulk g + ∆E = 2.73 eV and 3.14 eV are the energy of the first pronounced absorption peak in the bulk PbSe and PbS, respectively.88,89 To derive the volume scaling of the Coulomb matrix elements given by Eq. (B7), we assume the bulk limit in which the envelope functions become plane waves, i.e. F i m(r1) = F i V . In this basis, the Coulomb ma- trix elements can be written as meikir1/ √ (cid:88) (cid:2)F i (cid:3)∗ F l e2 V 2 (cid:90) (cid:90) Vijkl = × mF j n mF k n (B9) m,n dr1dr2 1 r1 − r2 e−ikir1−ikj k2+ikkr2+iklr1. The integral evaluation leads to the final expression Vijkl = 1 V 2 Vki,kj ,kk,kl V δki−kl,kj−kk , (B10) Eqs. (B10) and (B11) clearly demonstrate a general prop- erty that the Coulomb matrix elements used in the nu- merical calculations of the bulk limit scale inversely pro- portional to the volume.86 The nonparabolicity in the employed k·p-Hamiltonian is crucial for the accurate evaluation of the above Coulomb matrix elements. Specifically, summations over the spinor components in Eq. (B7) implies that the in- terband Coulomb scattering amplitudes vanish exactly if the non-diagonal terms of the k · p-Hamiltonian are set to zero, i.e., in strictly parabolic case.73 In the bulk, where quasimomentum k is a "good" quantum number, diagonal and off-diagonal matrix elements of the k · p- Hamiltonian scale as k2 and k, respectively. At high energies, where the diagonal elements dominate over the off-diagonal ones, the latter can be treated perturbatively giving k−1 ∼ ω−1/2 as the contribution of the hole (elec- tron) states to a high energy electron (hole) wave func- tion. ∗ Electronic address: [email protected] † Electronic address: [email protected] 1 S. Kolodinski, J. Werner, T. Wittchen, and H. Queisser, Appl. Phys. Lett. 63, 2405 (1993). 2 A. J. Nozik, Physica E 14, 115 (2002). 3 R. D. Schaller and V. I. Klimov, Phys. Rev. Lett. 92, 20 J. E. Murphy et al., J. Am. Chem. Soc. 128, 3241 (2006). 21 R. D. Schaller, J. M. Pietryga, and V. I. Klimov, Nano Lett. 7, 3469 (2007). 22 M. C. Beard et al., Nano Lett. 7, 2506 (2007). 23 J. J. H. Pijpers et al., J. Phys. Chem. C 111, 4146 (2007). 24 G. Nair and M. G. Bawendi, Phys. Rev. B 76, 081304(R) 186601 (2004). (2007). 4 M. Hanna and A. Nozik, J. Appl. Phys. 100, 074510 25 G. Nair, S. M. Geyer, L.-Y. Chang, and M. G. Bawendi, (2006). 5 V. I. Klimov, Appl. Phys. Lett. 89, 123118 (2006). 6 M. C. Beard and R. J. Ellingson, Laser & Photon. Rev. 2, 377 (2008). 7 A. J. Nozik, Chem. Phys. Lett. 457, 3 (2008). 8 J. M. Luther et al., Nano Lett. 8, 3488 (2008). 9 S. Koc, Czech. J. Phys. 7, 91 (1957). 10 A. Smith and D. Dutton, J. Opt. Soc. Am. 48, 1007 (1958). 11 V. Vavilov, J. Phys. Chem. Solids 8, 223 (1959). 12 J. Tauc, J. Phys. Chem. Solids 8, 219 (1959). 13 O. Christensen, J. Appl. Phys. 47, 689 (1976). 14 R. J. Ellingson et al., Nano Lett. 5, 865 (2005). 15 R. D. Schaller, V. M. Agranovich, and V. I. Klimov, Na- ture Phys. 1, 189 (2005). 16 R. D. Schaller, M. A. Petruska, and V. I. Klimov, Appl. Phys. Lett. 87, 253102 (2005). 17 R. D. Schaller, M. Sykora, J. M. Pietryga, and V. I. Klimov, Nano Lett. 6, 424 (2006). 18 R. D. Schaller and V. I. Klimov, Phys. Rev. Lett. 96, Phys. Rev. B 78, 125325 (2008). 26 M. Ben-Lulu et al., Nano Lett. 8, 1207 (2008). 27 J. J. H. Pijpers et al., J. Phys. Chem. C 112, 4783 (2008). 28 A. Franceschetti and Y. Zhang, Phys. Rev. Lett. 100, 136805 (2008). 29 M. C. Beard et al., Nano Lett. 9, 836 (2009). 30 S. Kilina, S. Ivanov, and S. Tretiak, J. Am. Chem. Soc. 131, 7717 (2009). 31 G. Allan and C. Delerue, Phys. Rev. B 79, 195324 (2009). 32 P. Tyagi and P. Kambhampati, J. Chem. Phys. 134, 094706 (2011). 33 G. Nootz et al., Phys. Rev. B 83, 155302 (2011). 34 J. A. McGuire et al., Acc. Chem. Res. 41, 1810 (2008). 35 J. A. McGuire et al., Nano Lett. 10, 2049 (2010). 36 J. J. H. Pijpers et al., Nature Phys. 5, 811 (2009). 37 M. T. Trinh et al., Nano Lett. 8, 1713 (2008). 38 M. Ji et al., Nano Lett. 9, 1217 (2009). 39 M. T. Trinh et al., Nano Lett. 11, 1623 (2011). 40 O. Roslyak, G. Gumbs, and S. Mukamel, J. Mod. Opt. 57, 097402 (2006). 2009 (2010). 19 R. D. Schaller, M. Sykora, S. Jeong, and V. I. Klimov, 41 O. Roslyak, G. Gumbs, and S. Mukamel, Nano Lett. 10, J. Phys. Chem. B 110, 25332 (2006). 4253 (2010). 18 42 J. B. Sambur, T. Novet, and B. A. Parkinson, Science 330, 71 H. Kamisaka, S. V. Kilina, K. Yamashita, and O. V. 63 (2010). 43 O. E. Semonin et al., Science 334, 1530 (2011). 44 G. Nair, J. Zhao, and M. G. Bawendi, Nano Lett. 11, 1136 (2011). 45 J. Dai, M. E. Raikh, and T. V. Shahbazyan, Phys. Rev. Lett. 96, 066803 (2006). 46 E. Rabani and R. Baer, Chem. Phys. Lett. 496, 227 (2010). 47 A. Piryatinski and K. A. Velizhanin, J. Chem. Phys. 133, 084508 (2010). 48 E. Antoncik and N. Gaur, J. Phys. D: Solid State Phys. 11, 735 (1978). 49 E. Kane, Phys. Rev. 159, 624 (1967). 50 P. T. Landsberg, Recombination in Semiconductors (Cam- bridge University Press, New York, 2003). Prezhdo, J. Phys. Chem. C 112, 7800 (2008). 72 L. V. Keldysh, Sov. Phys. JETP 20, 1018 (1965). 73 B. K. Ridley, Semicond. Sci. Technol. 2, 116 (1987). 74 C. Bonati et al., Phys. Rev. B 76, 033304 (2007). 75 J. T. Stewart et al., Nano Lett. 12, 622 (2012). 76 R. Koole et al., Small 4, 127 (2008). 77 I. Moreels and Z. Hens, Small 4, 1866 (2008). 78 K. A. Velizhanin and A. Piryatinski, J. Phys. Chem. B 115, 5372 (2011). 79 M. C. Beard et al., Nano Lett. 10, 3019 (2010). 80 G. Allan and C. Delerue, ACS Nano 5, 7318 (2011). 81 L. Novotny and B. Hecht, Principles of Nano-Optics (Cam- bridge University Press, New York, 2008). 82 V. V. Albert, S. A. Ivanov, S. Tretiak, and S. V. Kilina, 51 A. Franceschetti, J. M. An, and A. Zunger, Nano Lett. 6, J. Phys. Chem. C 115, 15793 (2011). 2191 (2006). 52 E. Rabani and R. Baer, Nano Lett. 8, 4488 (2008). 53 G. Allan and C. Delerue, Phys. Rev. B 73, 205423 (2006). 54 G. Allan and C. Delerue, Phys. Rev. B 77, 125340 (2008). 55 V. I. Rupasov and V. I. Klimov, Phys. Rev. B 76, 125321 (2007). 56 I. Kang and F. W. Wise, J. Opt. Soc. Am. B 14, 1632 (1997). 83 S. A. Fischer et al., Nanoscale 4, 904 (2012). 84 S. Kilina et al., (submitted). 85 M. Toda, R. Kubo, and N. Saito, Statistical Physics I, Solid-state sciences, 2nd ed. (Springer, Berlin, 1995). 86 H. Haug and S. W. Koch, Quantum Theory of the Op- tical and Electronic Properties of Semiconductors (World Scientific, London, 2009). 87 A. Franceschetti, H. Fu, L. W. Wang, and A. Zunger, 57 L. Silvestri and V. M. Agranovich, Phys. Rev. B 81, 205302 Phys. Rev. B 60, 1819 (1999). (2010). 88 N. Suzuki, K. Sawai, and S. Adachi, J. Appl. Phys. 77, 58 A. Shabaev, A. L. Efros, and A. J. Nozik, Nano Lett. 6, 1249 (1995). 2856 (2006). 59 W. M. Witzel et al., Phys. Rev. Lett. 105, 137401 (2010). 60 O. V. Prezhdo, Acc. Chem. Res. 42, 2005 (2009). 61 S. A. Fischer, A. B. Madrid, C. M. Isborn, and O. V. Prezhdo, J. Phys. Chem. Lett. 1, 232 (2010). 62 K. Hyeon-Deuk and O. V. Prezhdo, Nano Lett. 11, 1845 (2011). 63 C. Delerue, G. Allan, J. J. H. Pijpers, and M. Bonn, Phys. Rev. B 81, 125306 (2010). 64 K. A. Velizhanin and A. Piryatinski, Phys. Rev. Lett. 106, 207401 (2011). 65 J.-W. Luo, A. Franceschetti, and A. Zunger, Nano Lett. 8, 3174 (2008). 66 R. Zwanzig, Nonequilibrium Statistical Mechanics (Oxford University Press, New York, 2001). 67 M. Grifoni, E. Paladino, and U. Weiss, Eur. Phys. J. B 10, 719 (1999). 68 D. L. Mitchell and R. F. Wallis, Phys. Rev. 151, 581 (1966). 69 J. O. Dimmock, in The Physics of Semimetals and Narrow Gap Semiconductors, edited by D. L. Carter and R. T. Bates (Pergamon, Oxford, 1971). 70 H. Kamisaka, S. V. Kilina, K. Yamashita, and O. V. Prezhdo, Nano Lett. 6, 2295 (2006). 89 H. Kanazawa and S. Adachi, J. Appl. Phys. 83, 5997 (1998). 90 The IESM we use is a complimentary approach to the Hilbert space Green's function method independently pro- posed by Rabani and Baer.46 In contrast to our calcu- lations, the latter method has been used along with the atomistic electronic structure model and subsequently ap- plied to the smaller size NCs. 91 The exception is the calculation of the QE dependence on τpm, where we specifically indicate τpm variation range. −1 ef f , the γ −1 ef f characterizes the timescale of the 92 For τpm > γ pump pulse interaction with NCs. 93 Specifically, the ATE has some energy distribution asso- ciated with the 5% diameter variations. However, the Eg used to normalize the photon energy is an averaged value of the band gap. Therefore, the values of QE below aver- aged 2Eg are due to the contribution of the sub-ensemble with actual Eg values below its mean value. (cid:82) dr(cid:2)F i m(r)(cid:3)∗ 94 In the bulk limit the operator p commutes with the bulk Hamiltonian, and therefore M(1) becomes proportional to m(r), which is zero identically due to the orthogonality of the initial and the final wave func- tions. F j (cid:80)4 m=1
1612.02143	1	1612	2016-12-07T08:29:57	Topological frequency conversion in strongly driven quantum systems	[ "cond-mat.mes-hall", "quant-ph" ]	When a physical system is subjected to a strong external multi-frequency drive, its dynamics can be conveniently represented in the multi-dimensional Floquet lattice. The number of the Floquet lattice dimensions equals the number of {\em irrationally}-related drive frequencies, and the evolution occurs in response to a built-in effective "electric" field, whose components are proportional to the corresponding drive frequencies. The mapping allows to engineer and study temporal analogs of many real-space phenomena. Here we focus on the specific example of a two-level system under two-frequency drive that induces topologically nontrivial band structure in the 2D Floquet space. The observable consequence of such construction is quantized pumping of energy between the sources with frequencies $\omega_1$ and $\omega_2$. When the system is initialized into a Floquet band with the Chern number $C$, the pumping occurs at the rate $P_{12} = -P_{21}= (C/2\pi)\hbar \omega_1\omega_2$, an exact counterpart of the transverse current in a conventional topological insulator.	cond-mat.mes-hall	cond-mat	Topological frequency conversion in strongly driven quantum systems 1 Materials Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA Ivar Martin1,2, Gil Refael3,4, and Bertrand Halperin5 2 Kavli Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106, USA 3 Institute of Quantum Information and Matter and Department of Physics 4 Walter Burke Institute of Theoretical Physics, California Institute of Technology, Pasadena, CA 91125 USA 5 Department of Physics, Harvard University, Cambridge MA 02138, USA (Dated: December 8, 2016) When a physical system is subjected to a strong external multi-frequency drive, its dynamics can be conveniently represented in the multi-dimensional Floquet lattice. The number of the Floquet lattice dimensions equals the number of irrationally-related drive frequencies, and the evolution occurs in response to a built-in effective "electric" field, whose components are proportional to the corresponding drive frequencies. The mapping allows to engineer and study temporal analogs of many real-space phenomena. Here we focus on the specific example of a two-level system under two-frequency drive that induces topologically nontrivial band structure in the 2D Floquet space. The observable consequence of such construction is quantized pumping of energy between the sources with frequencies ω1 and ω2. When the system is initialized into a Floquet band with the Chern number C, the pumping occurs at the rate P12 = −P21 = (C/2π)ω1ω2, an exact counterpart of the transverse current in a conventional topological insulator. I. INTRODUCTION A major goal of quantum condensed matter physics is the control of many-body electronic and atomic states. A periodic drive is emerging as one of the most excit- ing means for achieving such control. Many proposals for phases that could be induced by periodic drive, so- called Floquet phases, have been made recently. Refs. [1–5] demostrate driving a band insulator into a topo- logical phase using circularly polarized radiation, or an alternating Zeeman field. Refs. [6–9] explored topolog- ical invariants unique to periodically driven phases, and predicted the Anderson-Floquet Anomalous Insulator, an unusual system with fully localized bulk but protected edge states. A general invariant for interacting systems was proposed in [10–13]. Refs. [14–17] even showed that a driven disordered system in 1D could spontaneously break the discrete time translation symmetry, forming the long sought-after time crystal, while Refs. [18–20] showed that a periodic drive can delocalize a many body localized system. This long list is a clear indication for the richness of Floquet engineering, with some of the pro- posals already having attracted nascent experimental ef- forts [21–23]. A periodic drive alters the form of a quantum wave function. Each state becomes dressed by all possible har- monics of the drive frequency. The extra degrees of free- dom associated with the amplitudes of the various drive harmonics effectively raise the dimensionality of the sys- tem, and allow it to exhibit new phenomena. While this observation is at the basis of some of the work mentioned above, the role of this extra emergent dimension remains little utilized, and, even worse, little understood. tals onto lower dimensions [24–26]. Most simply, a 1- dimensional (1D) quasicrystal can be constructed by su- perimposing two periodic but mutually incommensurate potentials. Here we show a surprising consequence of combining these two schemes for increasing the dimensionality of a system. The number of extra time dimensions is given by the number of the applied drives with incommensurate frequencies. This can give rise to topological phenomena rooted solely in the time dimension. In particular, we demonstrate that subjecting a single spin-1/2 particle to two elliptically polarized periodic waves can realize the chiral Bernevig-Hughes-Zhang (BHZ) model [27], which usually resides in two spatial dimensions, see Fig. 1. What is the consequence of such a construction? Just as edge states are the earmarks of spatial topological phenomena, the signature of temporal topological phe- nomena arising from incommensurate drives is pumping. We will show that by combining drives into a topological temporal texture, the system will pump energy between the driving fields, drawing energy from one, and feed- ing it into the other. This general principle could be used, for instance, to convert photons between incom- mensurate photonic modes in optical cavities (in con- trast to the perturbative nonlinearities that lead to the more conventional phenomena, such as frequency halving or doubling, e.g. [28]). In addition to establishing the pumping properties of the topological drive textures, we will demonstrate the existence of Floquet eigenstates for quasiperiodic drives, despite the lack of temporal period- icity. These states can be obtained through a convergent limiting procedure of approximating irrational frequency ratio by rational numbers. Interestingly, there is another example where an ex- tra dimensions "magically" emerge: quasicrystals. Qua- sicrystals are aperiodic structures that could be under- stood as projections of higher dimensional periodic crys- Our paper is organized as follows. After providing some background on Floquet theory, and the physics of the Wannier-Stark Ladder, we move on to consider the general properties of a doubly-driven system in Sec. III. 2 number of photons absorbed or emitted by the system. The standard – single-frequency – Floquet theorem can be extended to the case of multiple incommensurate fre- quencies. It has been utilized in the study of intense laser fields acting on atomic and molecular systems [31], as well as to simulate the Anderson localization in dimensions higher than one [32]. In both cases, the physical system subjected to drive was zero-dimensional. The number of independent frequencies translates directly into the num- ber of dimensions of the Floquet lattice. B. Wannier-Stark lattice and Berry curvature A peculiar feature of Floquet lattices generated by in- commensurate frequency drives is that they always ex- perience an effective uniform "electric field" applied in a non-crystallographic direction. The projection of the field onto a particular lattice direction is proportional to the corresponding drive frequency H(cid:126)ω = (cid:126)n · (cid:126)ω, where (cid:126)n is the integer vector of the drive harmonics, and (cid:126)ω is a vector containing the angular frequencies of the drives. By analogy with the one-dimensional Wannier-Stark lad- ders, if the potential energy drop over a lattice constant exceeds the hopping, then the band-structure effects are lost (this corresponds to the weak drive regime). On the other hand, for strong drive, the effective electric field (i.e., frequency) plays a role of a perturbation and the band description provides a convenient framework to study deviations from adiabaticity. What are the physical implications of the topolog- ically nontrivial Floquet band structure in the strong drive case? In conventional materials, bulk topological invariants, and Chern numbers in particular, lead to the appearance of gapless edge modes. In the case of Flo- quet lattice, for a classical coherent drive, the lattice does not have a boundary. Nevertheless, the pseudo-electric field can induce chiral propagation, mimicking the edge physics in a finite crystal. Analogously, we can think of the effective force as inducing an anomalous velocity (cid:126)ω × (cid:126)Ω(cid:126)q, where (cid:126)Ω(cid:126)q is the Berry curvature in the Floquet Brillouin zone, defined below in Eq. (18). The chiral propagation on the Floquet lattice corresponds to the energy transfer between individual drives (different fre- quencies). In the case of two-frequency drive of a two- level system, as we will show, the pumping power is pro- portional to the product of frequencies and the Chern number of the band into which system is initialized. Un- like the standard (perturbative) n-wave mixing, in the strong drive regime, any frequency can be converted into any other, no matter whether they are rationally or irra- tionally related. a. b. c. d. FIG. 1. In this manuscript we present a model that uses 2D topological insulator band structure as a guide to engineering the quantized energy pumping between different frequency modes. (a) A spin 1/2 driven by two sources with incommen- surate frequencies. In the topological phase, the spin trajec- tory (green line) fully covers the Bloch sphere (blue). The in- termode coupling induced by the spin dynamics pumps energy from one drive (red) to the other (blue) at a nearly quantized rate. Lower panel shows effective magnetic field trajectories for (b) m = 0 (gapless phase between two topological),(c) m = 1.8 (just inside topological phase), and (d) m = 2.2 (just outside topological phase). The mass m is depicted as a purple arrow (static Bz field, see Eq. (8) for the definition). Next, in Sec. IV, we introduce the BHZ path to tem- poral topological systems, and explore the model's prop- erties. In Sec. V we explore the temporal BHZ model numerically, and show that the topological regime is char- acterized by energy pumping. In Sec. VI we discuss the connection of our results with other systems and re- cent discoveries, and in Sec. VII we briefly discuss some other directions that can be pursued by means of multi- frequency Floquet engineering. II. BACKGROUND A. Floquet theory and the Floquet lattice The case of a simple periodic drive of arbitrary strength can be conveniently treated by the Floquet the- ory [29], which is a time-analog of the Bloch theory for particles in spatially periodic potentials. In Bloch the- ory [30], particles carry quasi-momentum; analogously, in Floquet theory, energy of eigenstates is replaced by quasi-energy E, and the Floquet quasi-eigenstates have the form Ψ(t) = e−iEtΦ(t). When a system is driven by single frequency ω (and its harmonics), the time-periodic part of the wavefunction can be expanded in the Fourier n e−inωtΦn, and the index of the ex- pansion harmonic n can be interpreted as a position in the Floquet lattice. It has the physical meaning of the series, Φ(t) = (cid:80) 3 single drive. Following Floquet's construction, we write the wave function as ψ(t)(cid:105) = e−iEt (cid:88) Above we introduced the vector notation (cid:126)n· (cid:126)ω =(cid:80) α,n1,n2,... (cid:126)ne−i(cid:126)n·(cid:126)ωt α(cid:105) . φα (4) niωi. i The Hamiltonian can also be expanded in terms of its Fourier components, Hαβ((cid:126)ϕ) = (cid:126)p e−i(cid:126)p·(cid:126)ϕ. hαβ (5) (cid:88) (cid:126)p (cid:88) Next we combine Eqs. tight-binding eigen-problem [33] (3), (4) and (5) and obtain a (E + (cid:126)n · (cid:126)ω)φα (cid:126)n = hαβ (cid:126)p φβ (cid:126)n−(cid:126)p. (6) (cid:126)p Once more, we made use of the vector notation for the set of integers pi. Eq. (6) describes a hopping problem on what we will refer to as a Floquet lattice (see Fig. 2). It has as many dimensions as there are independent drive terms. In ad- dition, it has a tilt, with the potential U ((cid:126)n) = −(cid:126)n·(cid:126)ω. This makes intuitive the interpretation of the Floquet lattice: ni is the number of photons absorbed by the system from drive i. Accordingly, if the energy spectrum of the system in the absence of a drive is bounded, the wavefunction φα (cid:126)n will be exponentially confined to a strip normal to the direction (cid:126)ω, that is, in the direction (−ω2, ω1). Along the strips, the wavefunctions should be non-divergent, which fully specifies the eigenvalue-eigenvector problem (6). Naively, the number of eigenstates is equal to the lat- tice size times the dimension of the local Hilbert space. However, it is easy to see that the number of indepen- dent Floquet eigenstates is the same as for the undriven system. Indeed, equations (6) have a symmetry: if φ(1)α (cid:126)n is a solution with quasienergy E(1), then φ(2)α (cid:126)n = φ(1)α (cid:126)n− (cid:126)m is also a solution with E(2) = E(1) − (cid:126)m· (cid:126)ω for any integer vector (cid:126)m. Indeed, both correspond to the identical solu- tion of the time-dependent Shrodinger equation Eq. (3). This can be verified explicitly, (cid:88) (cid:88) (cid:88) (cid:126)n (cid:126)n ψ(2)α(t) = = = e−iE(2)t−i(cid:126)n·(cid:126)ωtφ(2)α (cid:126)n e−iE(2)t−i(cid:126)n·(cid:126)ωtφ(1)α (cid:126)n− (cid:126)m e−i(E(2)+ (cid:126)m·(cid:126)ω)t−i(cid:126)n·(cid:126)ωtφ(1)α (cid:126)n (cid:126)n = ψ(1)α(t) Any initial wavefunction at t = 0 can be expanded in terms of the unique Floquet eigenstates. If it weren't for the linear onsite potential on the lhs of Eq. (6), the problem would be a translationally invariant tight-binding model in the Floquet space and could be FIG. 2. The Floquet harmonics in a 2-drive system give rise to a 2-dimensional space. Each spot represents the n1ω1 + n2ω2 = (cid:126)n · (cid:126)ω harmonic. If the frequencies are in- commensurate, then the space is infinite. If the ratio ω1/ω2 is rational, however, the infinite lattice contracts to a cylin- der (strip marked in red, with periodic boundary conditions across the strip). The periodic drives appear to exert a force in the (cid:126)n space along the (cid:126)ω direction. The motion that arises due to Berry curvature is normal to the force, and indicated by the d(cid:126)n/dt arrow. III. FLOQUET REPRESENTATION FOR MULTIPLE DRIVE FREQUENCIES The basis for our work is a mapping between a sys- tem of d spatial dimensions subjected to n mutually irra- tional drives, and a d + n-dimensional system. In rough terms, the extra n dimensions in the multidrive Floquet problem emerge when we consider the number of energy quanta absorbed from each drive. As we show below, the numbers of photons absorbed from each drive make up the coordinates in an n-dimensional lattice. The energy associated with each photon, however, gives rise to an effective force in this lattice description, since the Hamil- i niωi with ni the number of absorbed photons from the drive with angular frequency ωi. Below we make this analogy more precise. Consider a system with basis states α(cid:105), with α = 1, 2, . . . (for instance, spin states, though could also be position states in real space) subject to a Hamiltonian tonian will contain diagonal terms (cid:80) H = H αβ(ϕ1, ϕ2, . . .)α(cid:105)(cid:104)β , (1) (cid:88) α,β such that each element H αβ(ϕ1, ϕ2, . . .) is periodic for ϕi → ϕi + 2π. We assume linear time dependence for the ϕi's ϕi(t) = ωit, (2) with the ωi's being mutually incommensurate. The Schrodinger equation, written in terms of the components ψα of the wavefunction ψ(cid:105) =(cid:80) α ψαα(cid:105), is i∂tψα(t) = H αβ[(cid:126)ϕ(t)]ψβ(t). (3) Eq. (3) represents a system under multi-tone drive. It can be analyzed in the spirit of Floquet theorem for a trivially solved by a Fourier transform. The Flqouet wave (cid:126)n((cid:126)ϕ) = φα((cid:126)ϕ)ei(cid:126)ϕ·(cid:126)n. Remarkably, the functions would be φα energy eigenvalues in the absence of the drive (i)((cid:126)ϕ) are the eigenvalues of H((cid:126)ϕ) from Eq. (1). Thus, the angles (cid:126)ϕ play the role of momentum for the driven system. By analogy to the Bloch and Brillouin case, we refer to the region 0 < ϕi < 2π as the Floquet zone. With tilted potential on, the Floquet lattice eigenvalue problem becomes equivalent to Stark ladder in 2D [34]. The tilt produces a change in the momentum-angles as (cid:126)ϕ = (cid:126)ϕ0 + (cid:126)ωt. (7) To find the effects of the tilt on the wave functions, first consider the situation where the driving is strong, which is equivalent to the level spacings of (i)((cid:126)ϕ) being much greater than the largest angular frequency. In this case, we can describe the system using a semiclassical approach following the motion of a particle with momentum (cid:126)ϕ in the (cid:126)n lattice space. In this case, if the band structure is topologically nontrivial, we can obtain chiral "edge" modes that drift along the equi-potential lines in (cid:126)n. This will be central to our work. In the weak driving case, we obtain an effective 1D hopping model that describes quasicrystal with variable onsite potential (and variable hopping, if desired). This regime is not central to the main subject of the paper but interested readers may refer to Appendix A. Let us next consider how the construction above changes when the frequencies have a rational ratio. Con- sider a 2-drive system, with ω1p = ω2q, with p and q mu- tually prime. First, the mapping above of the Floquet problem to a 2D lattice representing the different fre- quency components of the generalized Floquet wave func- tion becomes redundant. The rational ratio can be taken into account by identifying the (n1, n2) lattice point with (n1 + mp, n2 − mq) for any integer m. This compacti- fies the 2D lattice into a strip with periodic boundary conditions, and the vector (p, −q) connects equivalent points across the strip's boundary. So, essentially, the problem is thereby reduced to that of a cylinder of cir- cumference (cid:112)p2 + q2, made of a square lattice (Fig. 2; in the following, for large (cid:112)p2 + q2 there is no signifi- for detailed treatment see Appendix B). As we will show cant difference between the rational and irrational cases as far the energy pumping efficiency is concerned. This is indeed to be expected since for a local Hamiltonian (not many higher harmonics) the system is only sensitive to the local geometry (same for plane and cylinder), and not the global topology. A related observation, interesting from mathematical perspective, is that it is possible to define Floquet eigen- states for quasiperiodic drive, by means of a limiting pro- cedure. Specifically, we can approximate any irrational frequency ratio as a limit of a ratio of two integers both tending to infinity, ω1/ω2 = limi→∞ qi . Solving a se- pi quence of Floquet eigenstate problems for progressively smaller ωi = ω1 we find that the eigenstates indeed qi = ω2 pi 4 FIG. 3. The Floquet zone for two drives. In analogy with a 2D band structure, we draw the Berry curvature (color plot with side legend) vs. the two offset phases ϕ1 and ϕ2. The periodic drives are akin to a motion along this Floquet zone with (cid:126)ϕ = (cid:126)ϕ0 +(cid:126)ωt. In the case of a rational frequency ratio, the system will explore a closed periodic path through the Floquet zone. Drawn here is an example of 3ω1 = 2ω2. The pumping effect will be roughly the integral of the Berry curvature along the path times ω1ω2 (Eq. 15). converge to a limiting state that one can define as the quasiperiodic Floquet eigenstate. The demonstration of this result is given in Appendix C. IV. TEMPORAL TOPOLOGICAL SYSTEMS The analogy between multiple incommensurate drives and multi-dimensional lattice quantum dynamics allows us to constructs 0-dimensional topological systems. The topological effects will arise from the temporal structure of the wave functions of the driven system. Below we construct such a system which consists of a single spin- 1/2, driven by two incommensurate periodic drives. We first define the model, and then explore its topological properties. Particularly, we will consider the semiclassi- cal motion of the system on the Floquet lattice. We will concentrate on the associated (cid:126)ϕ momentum space, espe- cially when there is Berry-curvature associated with the Floquet-lattice momentum states ψα((cid:126)ϕ). Fig. 3 depicts all elements relevant to the discussion, and maps out the motion of the system in the momentum (cid:126)ϕ space, as well as the Berry curvature of the model defined below. A. The BHZ path to temporal topological physics Let us choose hαβ pq in Eq. (6) so that the translation- ally invariant tight binding band structure is topologi- cally non-trivial. One of the simplest band structures of this type is a half of the BHZ model [27], with the Hamiltonian H = vx sin(k1)σx + vy sin(k2)σy + [m − bx cos(k1) − by cos(k2)] σz. (8) The model yields a quantum Hall insulator (band Chern numbers ±1) for −b1 − b2 < m < −b1 − b2 and b1 − b2 < m < b1+b2. The corresponding Flo- quet Hamiltonian is obtained by replacing k1 → ω1t + ϕ1 and k2 → ω2t + ϕ2, H = v1 sin(ω1t + ϕ1)σx + v2 sin(ω2t + ϕ2)σy + [m − b1 cos(ω1t + ϕ1) − b2 cos(ω2t + ϕ2)] σz. (9) 5 The interpretation of this Hamiltonian harks back to the discussion of the emerging Floquet lattice of Sec. III. The operator e−iωit−iϕi absorbs a photon from drive i, thus realizing a hop on the Floquet lattice in the direction i. In order to be able to follow the time evolution on the Floquet lattice, let us explicitly extend the Hilbert space to the direct product of the spin and lattice spaces, with (cid:126)n (cid:126)n(cid:105)α(cid:105) [compare with Eq. (4)]. In this representation, e±iωit → ni ∓ 1(cid:105)(cid:104)ni. We obtain a tight binding Schrodinger equation for the wave function amplitude on the Floquet lattice the wavefunction ψ(cid:105) = (cid:80) α(cid:126)n ψα i∂tψn1,n2 = Hψn1,n2 = 1 2 (iv1σx − b1σz) eiϕ1ψn1−1,n2 + 1 2 (−iv1σx − b1σz) e−iϕ1ψn1+1,n2 2 (iv2σy − b2σz) eiϕ2ψn1,n2−1 + 1 + 1 2 (−iv2σy − b2σz) e−iϕ2 ψn1,n2+1 + (mσz − n1ω1 − n2ω2) ψn1,n2 (10) We could even go further and Fourier transform the hopping part of the Hamiltonian, to obtain its 'momentum representation' on the tight-binding Floquet lattice: H(cid:126)q = v1σx sin(q1 + ϕ1) + v2σy sin(q2 + ϕ2) + [m − b1 cos(q1 + ϕ1) − b2 cos(q2 + ϕ2)] σz H =(cid:80) H(cid:126)n = −(cid:126)n · (cid:126)ω H(cid:126)q n(cid:126)q +(cid:80) H(cid:126)n nn (11) (cid:126)q (cid:126)n where n(cid:126)n is the occupation of site (cid:126)n = (n1, n2) and similarly n(cid:126)q is the occupation of the momentum state (q1, q2) = (cid:126)q. Hamiltonian written in the form Eq. (11) mixes the real (frequency) and momentum (phase/time) space representa- tions. The diagonal part H(cid:126)n = (cid:126)n · (cid:126)ω has the role of a force, which pushes the momenta (cid:126)q in the (cid:126)ω direction. This naturally leads to the momentum evolution (cid:126)q = (cid:126)ωt. (12) B. Photon absorption rates offset phases ϕi act as a lattice vector potential for the photon flow. Therefore, the derivative What would be the manifestation of a topological band structure in the Floquet lattice? When the Berry cur- vature is present, the system will respond to the force implied by H(cid:126)n = −(cid:126)n· (cid:126)ω by moving normal to the force in a preferred direction determined by the band into which the system was initiated. From the practical stand point, the "movement" of the particle within a chiral band cor- responds to a process where photons of one frequency are absorbed and of the other are emitted, in the propor- tions that approximately (up to uncertainty of the scale h(t)) conserve energy. In other words, the rates of work performed by "sources" with frequencies ωi should add up to approximately zero on average, but each one can be substantial. Let us derive expressions for the photon emission and absorption. From Eq. (11) we readily notice that the ji = ∂H ∂ϕi = ∂ni ∂t (13) is the current, or velocity operator, in the Floquet lat- tice. It is the rate of absorption/emission of photons with frequency ωi. The rate of energy absorption (power) is then ∂(cid:104)Ei(cid:105) ∂t = ωi(cid:104)ji(cid:105) = ωi ∂(cid:104)ni(cid:105) ∂t . (14) This intuitive result can be also obtained in another way. Consider a generic double drive Hamiltonian as H = (cid:126)h1(ω1t + ϕ1) · (cid:126)σ + (cid:126)h2(ω2t + ϕ2) · (cid:126)σ. The time- derivative of the total energy is dE/dt = d(cid:104)H(cid:105) /dt = · (cid:104)(cid:126)σ(cid:105). i(cid:104)[H, H](cid:105) +(cid:10) ∂H (cid:11) = ∂(cid:126)h1(ω1t+ϕ1) · (cid:104)(cid:126)σ(cid:105) + ∂(cid:126)h2(ω2t+ϕ2) ∂t ∂t ∂t Therefore, change of energy due to a given source is (cid:28) ∂H (cid:29) ∂φi ∂(cid:104)Ei(cid:105) ∂t = ωi = ∂(cid:126)hi ∂t · (cid:104)(cid:126)σ(cid:105). (15) Note that the same formalism can be applied to deter- mine not only the work performed by a given frequency drive, but even to separate the energy flows between dif- ferent polarizations of the same frequency drive. C. Semiclassical equations of motion and pumping power The evolution of position on the Floquet lattice in the limit of small applied "electric" field (that is, small fre- quency) can be determined from the semiclassical equa- tions of motion. Within a particular band of H(cid:126)q, they are [35], (cid:126)n = ∇(cid:126)q(cid:126)q − ∇nH(cid:126)n × Ω(cid:126)q (cid:126)q = −∇nH(cid:126)n = (cid:126)ω. (16) In the Floquet problem, there is a constant "force", (cid:126)ω, and we take (cid:126)q(t) = (cid:126)ωt, (17) which after substitution into Eq. (11) – as expected – yields the original problem, Eq. (9). The key to our problem is the anomalous velocity, related to the Berry curvature (cid:18) Ω(cid:126)q = zi 1 2 tr P(cid:126)q (cid:21)(cid:19) (cid:20) ∂P(cid:126)q ∂q1 , ∂P(cid:126)q ∂q2 (18) with P(cid:126)q a projector onto a particular band of H(cid:126)q [e.g. for the lower band, P(cid:126)q = (1 − H(cid:126)q/(cid:126)q) /2]. On average, only the Berry curvature pumps. The energy that goes between the two drives is then ∂(E1 − E2) ∂t 1 2 = 1 2 (ω1,−ω2)·((cid:126)ω × Ω(cid:126)q) = Ω(cid:126)qω1ω2. (19) D. Quantum Hall analogy The above result can also be obtained by directly ex- ploiting the analogy with the Hall response in the topo- logical insulators. Namely, we will use the known expres- sion for the Hall current in order to determine the average drift velocity of an individual particle. We take unit cell size 1 and electric field (cid:126)E = (ω1, ω2). The field is pointing at angle α = arctan ω2/ω1 to axis 1. For one fully occu- pied band with Hall conductivity σxy, the application of the electric field leads to the transverse current density j⊥ = σxyE 6 that flows in the direction α + π/2. This current is a product of particle density and velocity, and since the size of a unit cell is one, and the band is fully occupied, the 1 + ω2 2. density is 1. Thus, the drift velocity is v = σxy Its projection onto axis 1 is v1 = v sin α = σxyω2, and, therefore, the rate of energy change along the axis 1, which is the same as the power exerted by mode 1 is (cid:112)ω2 ∂E1 ∂t = σxyω1ω2 = − ∂E2 ∂t , equivalent to Eq. (19). There is also an appealing con- nection to quantum pumping, described in Appendix D. E. Role of incommensurability If ω1/ω2 is a rational number, then, over time, the path in the Floquet zone (cid:126)ωt mod 2π will repeat itself, and only a part of the Berry curvature will be sampled. On the other hand, an irrational ratio would sample uni- formly the entire double Floquet zone. Therefore, for rational ω1/ω2 the pumping power is not generally quan- tized; however, the average over the phases ϕ1 and ϕ2 (which can be interpreted as initial conditions) is. V. RESULTS OF NUMERICAL SIMULATIONS Let us now explore the pumping effects described above numerically. We find that, indeed, in the topological pa- rameter range, as long as the gap in the corresponding band Hamiltonian exceeds the drive frequencies, i.e., for sufficiently strong drive, energy flows between the two drives at a nearly quantized rate. A necessary condition for strong pumping is the high 'fidelity' – large projec- tion of the spin onto the direction of the instantaneous 'magnetic field' (equivalent to the ability of the spin to stay in one topological band). We consider these for both incommensurate and commensurate drive frequen- cies. Interestingly, depending on the initial conditions, the pumping effect for rational frequency ratio can ex- ceed its counterpart for incommensurate systems, and persist even outside the topological regime. The qualita- tive reason lies in the incomplete sampling of the Floquet zone in the case of rational drive, and thus a possibility of preferential sampling of high Berry curvature regions. A. Technical interlude Before diving into the numerical results, several techni- cal aspects of the simulation must be discussed. First, in our simulation we integrated the Schrodinger equation to produce the unitary evolution operator U (t). The Hamil- tonian we use is a special case of Eq. (9): H η = sin(ω1t + φ1)σx + sin(ω2t + φ2)σy + [m − cos(ω1t + φ1) − cos(ω2t + φ2)] σz, (20) with the parameter η defining the overall energy scale. Roughly, the ratio of η to the fastest frequency ωi de- termines whether the strong or weak driving regime is realized. For the gap parameter m we consider the topo- logical (m < 2) and non-topological (m > 2) regimes. In our simulation, we constructed U (t) as a product of the 'infinitesimal' evolution matrices exp [−iH(t)dt]. Our time discretization step was dt = 0.001, such that ω1dt = 10−4. Given the unitary evolution operator, it is possible to calculate the integrated work done by each of the two drives. As discussed in Sec. IV B, the instantaneous power spent or absorbed by drive i is dWi dt = (cid:104)ψ(t) dhi(t) dt ψ(t)(cid:105) = (cid:104)ψ0 U (t)† dhi(t) dt U (t)ψ0(cid:105) (21) [Here, hi(t) = (cid:126)hi(t)·(cid:126)σ]. Therefore we can define the work operator: t(cid:90) Wi = dtU (t)† dhi(t) dt U (t), (22) 0 such that Wi = (cid:104)ψ0 Wi ψ0(cid:105). Alongside U (t), we also calculated the operators Wi for the two drives. Finally, we must discuss the initial conditions for the simulation. We would like to have an initial state ψ0(cid:105) that would maximize the energy transfer between the two drives in the topological case. For this purpose we hark back to the analogy between the time-dependent Hamil- tonian [e.g., as written in Eq. (20)], and the Hamiltonian of a real-space topological insulator. The analogy, and the semiclassical logic of our topological pumping argu- ments suggest that a good choice for the initial state is an eigenstate of the instantaneous Hamiltonian at the be- ginning of the drive. When the drive is strong, η (cid:29) ω1,2, the motion of the system in the photon-number space will be dictated by the semiclassical equations (16), with the Berry curvature determined by the band of the eigenstate ψ0(cid:105). In this strong-drive limit, the eigenstates of the instantaneous initial Hamiltonian coincide with the Flo- quet eigenstates of a periodic system. For weaker drives, it might be advantageous to initialize the system into a Floquet state of a periodic system which approximates the incommensurate two-drive system. We explore this briefly in Appendix E. The fidelity of the initial state evolution, ψ(t) = U (t)ψ0 with respect to the same-band instantaneous eigenstate of H(t), ψi(t), is crucial for effective driving in the incommensurate case. It is possible to write the fidelity as (cid:20)(cid:18) 1 F = (cid:104)ψ(t) Pt ψ(t)(cid:105) − H(0) = tr (cid:19) 2 2(0) (cid:18) 1 2 U†(t) − H(t) 2(t) (cid:19) (cid:21) U (t) .(23) From the definition of the projection operator at time t, Pt, it follows that (cid:34) 1 + (cid:104)(cid:126)σ(t)(cid:105) · (cid:126)h(t) h(t) F = 1 2 7 (cid:35) ≡ cos2 θ(t) 2 , (24) where θ(t) is the angle between the spin expectation and the instantaneous magnetic field at time t. If the spin is not in an instantaneous eigenstate, then the semiclassical equations of motion do not apply, and the pumping effect is suppressed. The fidelity is robust, however, as long as the minimum gap in the band Hamiltonian significantly exceeds the drive frequencies, ∆ = η min(m − 2,m) (cid:29) ω1, ω2. (25) When studying the rational frequency ratio case, it is natural to consider the Floquet eigenstates as initial state for the spin in addition to the instantaneous eigenstates. A rational frequency ratio implies a strict periodicity of the Hamiltonian. For ω1 = 2π/T p and ω2 = 2π/T q, the Hamiltonian is periodic with period τ = T LCM(p, q), with LCM(p, q) the lowest common multiple of p and q. Typically, initialization into Floquet eigenstates results in faster pumping at short and intermediate times, but more importantly, since they are the eigenstates of U (τ ), the pumping power itself is a periodic function of time. B. Incommensurate drives Let us demonstrate the pumping effects in the regime that reflects best the semiclassical limit, in which the Berry curvature effects are dominant. Simulating the sys- tem at η = 2, yields near perfect pumping effects all the way essentially to the topological transition at m = 2. In Fig. 4 we see the energy transfer for several values of the gap parameter m, alongside the fidelity associated with the overlap of the initial state with the instantaneous eigenstate [as defined in Eq. (23)]. 5 + 1(cid:1) ≈ 1.618, the golden ratio. The offset Throughout our analysis here, we use the incommen- surate frequency pair ω1 = 0.1 and ω2 = γω1, with γ = 1 2 phases are φ1 = π/10, φ2 = 0. We will explore the evo- lution up to times t = 104, where the pumping effect is clearly visible. We will keep the parameters v = 1 and b = 1 fixed, and vary m and η in our exploration. (cid:0)√ 2π = 0.00257518, as in Eq. The energy pumping rate indeed saturates to the one expected from the semi-classical analysis. For a sys- tem with Chern number 1 we expect an average Berry curvature of Ω = 1 2π , and an energy pumping rate of dE/dt = ω1ω2 (19). From linear-regression of the plots in Fig. 4 we obtain ex- cellent agreement with the semiclassical theory. Simi- larly, the fidelity also remains near perfect throughout the evolution, except for the case of m = 1.8. There, ∆ = ηm−2b = 0.4, which is comparable to ω2 ≈ 0.1618, and the quasi-adiabaticity condition is expected to break down. 8 FIG. 5. Pumping rates for different η and m. The parameters of the system for this plot are b = v = 1, and frequencies ω1 = 0.1, and ω2 = γω1. The pumping rate was averaged over the time range t < 10000. As η is reduced, the regime of ineffective pumping, where the gap ∆ becomes comparable to or smaller than ω1,2, expands. This is visible in Fig. 5, which shows the pumping rate as a function of m and η for a broad region of parameter space. We explore smaller values of η, as well as the demise of this effect at weaker drives in Appendix F. C. Commensurate drive frequencies The energy pumping effect, and Eq. (15) in particu- lar, can also be explored for a rational frequency ratio. In this case, however, it is not the Chern number that deter- mines the effect, but rather the integral of the Berry cur- vature along a closed path in the two-dimensional 'Flo- quet zone' , as shown in Fig. 3. We study this regime here for the frequency ratio ω1/ω2 = 2/3 (with ω1 = 0.1) and the strong drive, η = 2. Additional results for weaker drives are given in the Appendix F. As discussed in Sec. V A, when the frequency ratio is rational, the Hamiltonian is strictly periodic, and we can initiate the spin into a Floquet eigenstate of H(t). This results in a periodic pumping profile and typically larger pumping power, compared to the initialization into instantaneous eigenstates (detailed comparison between the two initiation procedures is given in Appendix E). The pumping profile for 1.2 < m < 2.2, with offset phases φ1 = π/10, φ2 = 0, and Floquet eigenstate ini- tialization is shown in Fig. 6. As can be seen, the en- ergy pumping effect is present in the entire topological region. Indeed, the pumping rate may exceed the quan- tized theoretical value, since the energy pumping effect is now determined by the Berry curvature along a particular periodic path through the Floquet zone, rather than its average over the whole Floquet zone. Different paths can be selected by varying the offset phases. This dependence a. b. c. FIG. 4. The energy flow and fidelity in the strong-drive regime, η = 2. (a) The total work done by drives 1 and 2 as a function of time for different m's. Each pair of lines is displaced on the time axis for clarity. (b) Fidelity vs. time of a state initialized into the instantaneous eigenstate of H(t = 0), and measured against the eigenstate of H(t). (c) The power pumped as a function of the parameter m averaged up to time t = 2000. The quantized pumping is marked by the gay line at dE/dt = ω1ω2/2π. The strong-drive regime realizes the topological pumping prediction up to m = 1.8, with the energy pumping rate saturating near the expected value of dE/dt = ω1ω2 for a model with Chern-number 1. The initial 2π phases for these plots are φ1 = π/10, φ2 = 0. We see that the fidelity of the evolved state relative to the single-band state deteriorates as we get close to the phase transition. This is due to the closing of the band gap in the corresponding BHZ model. 9 a. b. a. b. FIG. 6. The energy flow in the strong-drive regime, η = 2, for rational frequency ratio, ω2 = 3 2 ω1, with Floquet state initial- ization. (a) The total work done by drives 1 and 2 as a func- tion of time for different m's. Each pair of lines is displaced on the time axis for clarity. (b) The power pumped as a func- tion of the parameter m averaged upto time t = 2000. The pumping in the commensurate case could be stronger than its incommensurate counterpart. Furthermore, pumping may persist in the non-topological regime, since the system does not average the Berry curvature of the entire Floquet zone. In the case of the offset angles chosen, even when m = 2.2, the system explore regions with net positive Berry curvature. The gray line is at dE/dt = ω1ω2/2π, signifying the pumping due to Berry curvature of Chern number C = 1 band. The initial phases for these plots are φ1 = π/10, φ2 = 0. is illustrated in Fig. 7. It is also interesting to note that the offset-phase dependence is stronger for initialization into Floquet eigenstates, rather than into instantaneous eigenstates of the Hamiltonian (not shown). VI. SUMMARY AND DISCUSSION In this work we have demonstrated that it is possible to implement multidimensional topological band struc- tures purely in the frequency space. The implementation relies on the ability to apply strong drives of incommen- FIG. 7. Pumping profile for different offset phases: (a) Ac- cumulated work for Floquet initial states, (b) average power for t < 2000. We take m = b = v = 1, η = 1, and φ1 = nπ/5 with 5 π φ1 = 0, 1, 2, 3, 4. The Floquet zone path of this system for one offset phase combination is shown is Fig. 3). Floquet eigenstates initial conditions were used. For these parame- ters there were no significant differences with instantaneous eigenstate initialization. surate frequencies. The approach was illustrated using the 2-dimensional BHZ [27] model constructed by pump- ing a two-level system ("spin-1/2") by two elliptically polarized waves of incommensurate frequencies. The ob- servable in this case is the quantized energy pumping power between the two drives, which is the direct analog of the quantized transverse Hall conductivity of the orig- inal real-space model. As such, the driving occurs in the topological regime of the model, and remains effective as long as the drives' frequencies are lower than the minimal band gap in the model (quasi-adiabatic regime). There are two ways in which the model can be natu- rally generalized: by considering larger spins, and by al- lowing the driving fields to have their own dynamics. The generalization of our results to larger spins is straightfor- ward. Indeed, the dynamics of the expectation value of the spin is given by the Bloch equations, which are inde- pendent of the spin size [36], (cid:126)S(t) = i[H, (cid:126)S(t)] = (cid:126)S(t) × (cid:126)h(t). (26) The Bloch equations are linear in spin, and therefore the precession frequency is independent of the spin size. The "strong drive" criterion needed for the energy pumping for general spin size, in the language of the Bloch equa- tions, corresponds to having the instantaneous Larmor frequency (cid:126)h(t) higher than the pump frequencies ωi – same condition as for spin-1/2. On the other hand, from Eq. (15), the pumping power depends linearly on the spin size, and thus by taking a large magnetic particle with low magnetic anisotropy (e.g. YIG sphere [37]), or NMR, or ESR systems [38] one can dramatically increase the pumping power, to macroscopic levels. One practical way to include the dynamics of the drive fields is by considering a closed system of spin and two electromagnetic cavities resonating at the drive frequen- cies. The cavities can be initialized into semiclassical coherent states to represent the periodic drives. The en- ergy pumping will act to change both amplitudes (and phases) of the drives, eventually taking the system out of the topological regime. Our preliminary results indeed show that system reaches the topological state bound- ary, after which the pumping direction reverses, albeit at a reduce rate. Treating photons quantum mechanically raises several new questions. Perhaps most appealing is the possibility of using the two-frequency topological effects to pump energy from a laser of one frequency into a cavity with another frequency. Additionally: – What kinds of entangled spin-photon states are achiev- able? – Is there is a spontaneous transition into the pumping regime if one of the cavities is initialized into a vacuum state? – Can a superradiance transition occur in a system of multiple spins in the cavities? We leave these theoretical problems for future study. The quantum limit may be particularly amenable to experimental study using superconducting qubit systems [39]. Their appeal of this scheme is both in the exquisite level of control over the qubit, as well as in the extreme measurement sensitivity to the photon occupation num- bers in the superconducting resonant circuitry. Yet an- other attractive feature of the superconducting devices, and Josephson junctions in particular, is the access to the AC Josephson effect in order to drive the supercon- ducting phase by applying DC voltage. Manipulation of the superconducting phases has indeed been considered as a means to explore and control topological phases in multi-terminal Josephson junctions [40, 41], and is math- ematically analogous to the phase control provided by the AC drive that we consider in the present work. In this work we considered only perfectly stable (fixed- frequency) drive sources. We have seen, however, that there is no significant difference between commensurate 10 and incommensurate drives of approximately equal fre- quencies, as far as the pumping properties are concerned. It would be interesting to consider more general phase dy- namics, deviating from the one given by Eq. (7), either due to the deliberate tuning of drive frequencies, or due to noise. VII. OUTLOOK The multiple-drive paradigm we propose here gives rise to the possibility of engineering emergent band models. These emergent bad structures could produce interesting and potentially useful dynamics of photonic systems as well as access new entangled states of photons and mat- ter. Furthermore, they are not limited to the two-drive example presented in our manuscript. The implementation of BHZ model in the Floquet space that was our focus here is just one example of a model that can be implemented by means of strong AC driving. Many more exotic models can be built and ana- lyzed by the same approach, by applying more drive fre- quencies (equivalent to going to higher-dimensional fre- quency spaces), or by pumping systems with more levels (including spatially extended ones). In particular, by ap- plying three pump frequencies to a four level system, one may be able to construct the 3D extension for BHZ model [42]. The topological invariant in this case is the second Chern number that represents the magneto-electric re- sponse. Again, the "electric field" is built into the Flo- quet model by construction. It will be interesting to de- termine what observable the orbital "magnetic" response would corresponds to in the Floquet implementation. Pumping spatially-extended systems by space-time de- pendent drives is another direction worth exploring. Mix- ing real-space with Floquet-space gives a simple interpre- tation to such classic effects as Thouless pump in terms of 2D quantum Hall effect (Appendix G), and can also help discover new phenomena. E.g., by pumping 1D spatial lattice with two incommensurate frequencies one may be able to construct 3D topological insulators with SU (2) Landau levels [43, 44]. By applying even more drives one may be able to access and study more exotic yet states, such as eight-dimensional quantum Hall effect [45]. VIII. ACKNOWLEDGEMENTS Authors would like to thank J. Sau, M. Sanchez, A. Ya- cobi, V. Manucharyan, M. Gullans, M. Lukin, F. Nathan, and Y. Oreg for useful discussions. IM acknowledges sup- port from Department of Energy, Office of Basic Energy Science, Materials Science and Engineering Division. GR acknowledges support from NSF through DMR-1410435, as well as the Institute of Quantum Information and Mat- ter, a NSF Frontier center funded in part by the Gordon and Betty Moore Foundation, and the Packard Founda- tion. BH acknowledges support from STC Center for In- 11 with H(t) = H(ω1t, ω2t), with H(θ1, θ2) 2π-periodic in θ1,2. Hence H(ω1t, ω2t) = h (cid:126)me−i(cid:126)ω· (cid:126)mt h (cid:126)me−i(cid:126)ω· (cid:126)mt (B2) (cid:88) (cid:88) (cid:126)m "strip" (cid:88) (cid:88) n,m tegrated Quantum Materials, NSF grant DMR-1231319. This research was supported in part by the National Sci- ence Foundation under Grant No. NSF PHY-1125915. We are also grateful to the Aspen Center for Physics where part of the work was done. Appendix A: Weak drive = (cid:113) When the drive is weak then the motion of the system in the Floquet space is highly restricted. This can be seen from (E + ω1n + ω2m)φα nm = hαβ pq φβ n−p,m−q. (A1) Due to large energy mismatches between neighboring sites, the "particle" can hop on the lattice only along a 1D "Manhattan" path that skirts the constant-energy line, e.g., E + ω1n + ω2m = 0. The sites of the corre- sponding 1D model have onsite energies given by unm = ω1n + ω2m = ω2 1 + ω2 2(n sin α + m cos α). This onsite energy is proportional to the distance be- tween a lattice point (n, m) and the line cutting through the lattice at angle α. For tan α irrational, this distance is a quasiperiodic quantity, and hence, the system rep- resents a 1D lattice with quasiperiodic onsite potential. This can be compared to the well studied case of the Fi- bonacci quasicrystal [46]. There, tan α = γ, the onsite energies are all the same, but the hoppings are different if they derive from the "vertical" (m-direction) or the "horizontal" (n-direction) hopping. Clearly, we can also introduce that by choosing h01 (cid:54)= h01, however, the map- ping to 1D only applies if ωi >> h, and thus it is not possible to exactly replicate the limit of pure bond "disor- der". The spectrum of this problem is dense everywhere due to the symmetry E → E + ω1n + ω2m accompanied by wave function translation. Appendix B: Commensurate drive frequencies – detailed treatment. When drives are incommensurate, the wavefunction φ(cid:126)n in Eq. (4) is defined on the whole 2D Floquet plane, since all frequencies nω1 + mω2 are unique. Suppose now that ω1/ω2 = p/q, with p < q mutually prime integers. It is easy to see that if p = 1, then nω1 exhausts all the pos- sible frequencies. For p (cid:54)= 1, the number of independent frequencies is p times that. Thus, in the commensurate case the Floquet space is "compactified" into a strip of width p along ω1 axis. The boundary conditions are pe- riodic along ω2 direction, but with an "offest" of q, in- troducing a shear (twist). Let's see how this comes about in detail. As before, the Schrodinger equation is i∂tψα(t) = H αβ(t)ψβ(t), (B1) where we combined all terms with the same frequencies, thus reducing summation to a strip of frequencies de- scribed above. Similarly, the Floquet representation for the wave func- tion is ψ(t) = e−iEtφ(t) = e−iEt−i(cid:126)ω·(cid:126)ntφ(cid:126)n = e−iEt−i(cid:126)ω·(cid:126)nt φ(cid:126)n. (B3) "strip" Substitution into the Schrodinger equation leads to the expected result: when action of Hamiltonian seems to take the system outside the strip, a shift by ±(−q, p) brings it back into the strip. This corresponds to the "sheared" periodic boundary conditions in the strip. circumference (perimeter(cid:112)p2 + q2). An alternative selection of unique frequencies is a tilted ribbon with the regular (non-sheared) periodic boundary conditions. The construction is the same as in the case of carbon nanotubes with helicity (p, q). The Hamiltonian induces hopping on the "nanotube"; the energy conserv- ing dynamics corresponds to hopping along the nanotube Note that if p, q (cid:29) 1 then the compactification should not be noticeable, and one does not expect any difference between irrationally and such rationally related frequen- cies (e.g. if one constructs topological insulators in the Floquet space). Appendix C: Floquet eigenstates in quasiperiodic potentials When drives are quasiperiodic (irrational frequency ra- tio), Floquet theorem does not directly apply, since one cannot define the evolution operator over the full pe- riod. Nonetheless, a quasiperiodic drive with irrational ω2/ω1 = γ can be approximated by a periodic one, ω2/ω1 = p/q (p and q are positive relatively prime in- tegers). For instance, for γ the Golden mean, q and p can be chosen as consecutive Fibonacci numbers. Then, the drive is periodic, with the period T = pT1 = qT2, and the Floquet theorem can be applied. Hence one can ask, whether in the limit of p, q → ∞, the Floquet eigen- states (FE) converge to unique states. These could be naturally defined as the Floquet eigenstates for the case of incommensurate drives. For a periodic drive, FE have the form Ψ(t) = e−iEtΦ(t), with Φ(t) periodic. The FE are the eigen- states of the evolution operator over the period, lead to topologically distinct trajectories in the Floquet zone. 12 Appendix D: Quantum pump analogy It is tempting to rewrite the Berry curvature of Eq. (18) as Ω(cid:126)q = zi (cid:18) tr P(cid:126)q 1 22 (cid:126)q (cid:16) (cid:21)(cid:19) ∂H(cid:126)q ∂q2 , ∂q1 (cid:20) ∂H(cid:126)q (cid:17) (cid:20) ∂H(cid:126)q P(cid:126)q ∂q1 ∂P(cid:126)q ∂qi (cid:18) P(cid:126)qH(cid:126)q (easy to see since tr formula for the pump power ∂(E1 − E2) ∂t 1 2 = ω1ω2 22 (cid:126)q tr , (D1) = 0). This leads to a (cid:21)(cid:19) , ∂H(cid:126)q ∂q2 . (D2) Eq. (D2) bears close similarity to the quantum pump- ing formula [48], except that instead of the S-matrix, the derivatives in the commutator are of a Hamiltonian. In- deed the photons that enter and leave the driven system do not have a conservation law (since they have different frequencies), and therefore they can not be assigned an S-matrix. This equation may suggest that a generaliza- tion of the quantum pumping formula for conserved but non-quantized quantities, such as energy, may exist. Appendix E: Floquet vs Instantaneous eigenstate initialization In Figure 9 we consider two rational approximations to the Golden mean: ω2 = 3/2ω1 and ω2 = 144/89ω1, for η = 1 and ω1 = 0.1. First of all, even in the case of small commensuration, ω2/ω1 = 2/3 there is significant pumping. The rate of pumping is similar to the one for a nearby incommensurate ratio. As mentioned in Section IV E, for the commensurate drives, the Floquet zone is not fully sampled, and hence the results can depend on the relative phase between the drives. Indeed, varying the initial phase shift, does lead to different pumping rates, Fig. 9c. In Fig. 9a,b,d we have compared the initializations into the instantaneous eigenstates (solid lines) and the Flo- quet eigenstates (dotted lines). In all cases we considered, we found that initialization into the Floquet eigenstates leads to faster pumping. For incommensurate drive fre- quencies, Fig. 9d, the initialization was into the Floquet eigenstate of ω2 = 144/89ω1 (approximation to the in- commensurate FE discussed in Appendix C). That lead to faster and more linear pumping, with the value near the theoretical upper limit Eq. (19). The reason for faster pumping can be traced to higher average fidelity in the FE. In Figures 10 and 11 we consider the energy pumping, for the same parameters as in Figure 9, but for reduced Hamiltonian scales, η = 0.5 and η = 0.2, respectively. √ FIG. 8. Convergence of Floquet eigenfunctions for ω1/ω2 = pi/qi → r, where r is irrational [the Golden mean, γ = 2 (orange line) ]. ∆Ψ is calculated 1.6180... (blue line) and √ relative to the best rational approximation used (2584/1597 for γ and 3363/2378 for 2); it is approximately linear in p/q − r. Parameters of the Hamiltonian are described in the text. U (T ) = T exp−i (cid:90) T 0 H(t(cid:48))dt(cid:48). For a TLS, this is a 2x2 matrix, that can be easily computed and diagonalized to find FE. a. Observations 1+ 1+ 1 2+ 1 2+ 1 To test convergence of Floquet eigenstates, we stud- (20) with parameters η = 1 and ied Hamiltonian Eq. √ m = 1.5. We considered ω2/ω1 = limi→∞ pi/qi = r, with irrational r chosen as γ (with ω1 = 0.3) and 2 (with ω1 = 0.5). We used continued fraction repre- sentations of γ = 2 = 1 + and √ 1 1 1 1+... 2. √ 2+... terminated at a finite level to obtain the sequences of (the best [47]) rational approximations. The largest- denominator approximations that we considered were 2584/1597 for γ and 3363/2378 for In Figure 8 the deviation, ∆Ψi = Ψpi/qi − Ψpmax/qmax is plotted as a function of pi/qi − r for the two cases [the refer- ence values are Ψ2584/1597 ≈ (0.094 + 0.496i, 0.863)† and Ψ3363/2378 ≈ (−0.230 − 0.446i, 0.865)†]. Both show ap- proximately linear convergence, indicating the existence of the Floquet eigenstates for incommensurate drive fre- quencies. We also find, as expected, that in the limit of small frequencies (ωi (cid:28) η) the FE are nearly the eigen- states of the instantaneous Hamiltonian. Note, that we only tested the FE convergence in the gapped phase. While we have not tested it explicitly, we expect that in the gapless points (m = 0, 2), the limit may not exist, since small changes in frequencies would p/q - r10-610-510-4∆ Ψ10-210-110013 a. c. b. d. FIG. 9. Comparison of evolution starting from instantaneous (solid lines) and Floquet (dotted lines) eigenstates. Rational frequencies in a-c are approximations of the Golden mean γ. Hamiltonian scale is η = 1 and ω1 = 0.1, with ω2 stated on plots. Panel c illustrates how the energy transfer for rational drive depends on the the relative phase, φ2 − φ1 = 0, 1.5, 3, 4.5, starting from the respective instantaneous eigenstates. The variation is due to sampling of different closed lines in the phase Brillouin zone. Panel d illustrates how initializing into a Floquet eigenstate (computed for ω2 = 144/89ω1), enhances pumping at long times for irrational drive. Consistently with the expectation, the energy pumping is suppressed, particularly strongly for larger denominator (and irrational) fractions ω1/ω2. Interestingly, for small denominators, e.g. 2/3, the pumping persists even for relatively small η. after a finite time in parameter ranges close to or beyond the topological transition into the trivial range. This is associated with the system exploring parts of the Floquet zone where the band gap of the underlying BHZ model is comparable to the drive frequencies. Appendix F: Additional numerical results In addition to the numerical results we presented in the text, we explored the intermediate and weak drive regimes of the double-drive BHZ model. We present these numerical results here for completeness. Unless stated otherwise, ω1 = 0.1 and ω2 = γω1, where γ is the Golden mean. Figs. 12c 13c show the power absorbed by the drives averaged over the first 2000 time units of the evolution as a function of the gap parameter m for the η = 0.5, 1. The transition between the pumping regime and the trivial regime is quite abrupt. 2. Weak drives, η ≤ ω1,2 1. Intermediate drives, η = 0.5, 1 For η values that exceed ω1,2, but not by much, we still see a strong pumping effect deep in the topologi- cal regime. The effect subsides, however, well before the phase boundary between the topological and trivial pa- rameter regimes (Fig. 12a and 13a). As can be seen from Fig. 12b and 13b, this is a result of the fidelity being lost To explore the weak-drive regime we considered the η = 0.1. As Figures 14 suggest, initializing the sys- tem with the instantaneous eigenstate of the Hamiltonian H(0) results in negligible pumping. Initialization into a Floquet eigenstate of a periodic approximation does not qualitatively change the result, as can be seen from Fig. 11d. time0500010000energy transfer-30-20-100102030ω2 = 3/2 ω1time0500010000energy transfer-30-20-100102030ω2 = 144/89 ω1time0500010000energy transfer-30-20-100102030ω2 = 3/2 ω1time0500010000energy transfer-30-20-100102030ω2 = γ ω114 a. c. a. c. b. d. FIG. 10. Same as Figure 9, but for η = 0.5. b. d. FIG. 11. Same as Figure 9, but for η = 0.2. time0500010000energy transfer-30-20-100102030ω2 = 3/2 ω1time0500010000energy transfer-30-20-100102030ω2 = 144/89 ω1time0500010000energy transfer-30-20-100102030ω2 = 3/2 ω1time0500010000energy transfer-30-20-100102030ω2 = γ ω1time0500010000energy transfer-30-20-100102030ω2 = 3/2 ω1time0500010000energy transfer-30-20-100102030ω2 = 144/89 ω1time0500010000energy transfer-30-20-100102030ω2 = 3/2 ω1time0500010000energy transfer-30-20-100102030ω2 = γ ω115 a. b. c. a. b. c. FIG. 12. Work (a) and fidelity (b) in the intermediate drive regime for η = 1. We initialized the system with the instan- taneous eigenstate of H(0), where the initial phases for these plots are φ1 = π/10, φ2 = 0. (c) Power pumped averaged for t < 2000 as a function of gap parameter m. FIG. 13. Work (a) and fidelity (b) in the intermediate drive regime for η = 0.5. We initialized the system with the instan- taneous eigenstate of H(0), where the initial phases for these plots are φ1 = π/10, φ2 = 0. (c) Power pumped averaged for t < 2000 as a function of gap parameter m. Appendix G: Thouless pump, reloaded In this section we will provide the reinterpretation of the Thouless pump [49] using the Floquet lattice con- struction. Thouless pump is a classic example of a driven system that shows quantized transport behav- ior. Consider a one-dimensional tight binding lattice, xi = i, with spatially- and time-dependent onsite po- tential, V (x, t) = A cos(ω0t − kxi). The potential is time-periodic, and thus we can take advantage of the Floquet transformation described in Section III. As a re- sult we obtain a two-dimensional lattice – one spatial dimension that no longer has spatially varying potential, and one frequency dimension, with the superimposed lin- ear potential corresponding to an applied electric field of strength ω0. The hopping in the positive (negative) frequency direction is Ae±ikxi . The phase factor corre- sponds to an effective linear in x vector potential point- ing in the y (frequency) direction. This is equivalent to a uniform magnetic field of strength k piercing the lattice. We therefore see that the Thouless pump in 1D maps onto a quantum Hall problem in 2D! Indeed [49] is nothing but the the topological invariant in Ref. Chern number in the mixed coordinates of spatial mo- mentum and 'time-momentum' (phase). Due to the pres- ence of crossed "electric" and "magnetic" fields, a particle placed in the lattice will experience drift, with velocity orthogonal to both, of the magnitude given by their ratio, vdrif t = ω0/k. This is nothing but the speed of the po- tential in the original problem, which yields the pumping result of Thouless. 16 [1] T. Oka and H. Aoki, Phys. Rev. B 79, 081406 (2009). [2] J.-i. Inoue and A. Tanaka, Phys. Rev. Lett. 105, 017401 prints (2016), arXiv:1609.08684 [quant-ph]. [24] P. A. Kalugin, A. I. Kitaev, and L. S. Levitov, ZhETF (2010). Pisma Redaktsiiu 41, 119 (1985). [3] N. H. Lindner, G. Refael, and V. Galitski, Nat Phys 7, [25] D. Levine and P. J. Steinhardt, Physical Review Letters 490 (2011). 53, 2477 (1984). [4] N. H. Lindner, D. L. Bergman, G. Refael, and V. Galit- [26] D. Shechtman, I. Blech, D. Gratias, and J. W. Cahn, ski, Phys. Rev. B 87, 235131 (2013). [5] T. Kitagawa, T. Oka, A. Brataas, L. Fu, and E. Demler, Phys. Rev. B 84, 235108 (2011). [6] M. S. Rudner, N. H. Lindner, E. Berg, and M. Levin, Phys. Rev. X 3, 031005 (2013). [7] P. Titum, E. Berg, M. S. Rudner, G. Refael, and N. H. Lindner, Physical Review X 6, 021013 (2016), arXiv:1506.00650 [cond-mat.mes-hall]. [8] N. H. Lindner, G. Refael, and V. Galitski, Nature Physics 7, 490 (2011). [9] F. Nathan and M. S. Rudner, New Journal of Physics 17, 125014 (2015), arXiv:1506.07647 [cond-mat.mes-hall]. [10] R. Roy and F. Harper, ArXiv e-prints (2016), arXiv:1603.06944 [cond-mat.str-el]. [11] R. Roy and F. Harper, ArXiv e-prints (2016), arXiv:1610.06899 [cond-mat.str-el]. [12] C. W. von Keyserlingk and S. L. Sondhi, Phys. Rev. B 93, 245145 (2016), arXiv:1602.02157 [cond-mat.str-el]. [13] A. C. Potter, T. Morimoto, and A. Vishwanath, Physical Review X 6, 041001 (2016). [27] B. A. Phys. Rev. Lett. 53, 1951 (1984). L. 314, T. Science Bernevig, S.-C. http://science.sciencemag.org/content/314/5806/1757.full.pdf. Zhang, Hughes, 1757 and (2006), [28] S. Mukamel, Principles of nonlinear optical spectroscopy, 6 (Oxford University Press on Demand, 1999). [29] G. Floquet, in Annales scientifiques de l' ´Ecole normale sup´erieure, Vol. 12 (1883) pp. 47–88. [30] F. Bloch, Zeitschrift fur physik 52, 555 (1929). [31] T.-S. Ho, S.-I. Chu, and J. V. Tietz, Chemical Physics Letters 96, 464 (1983). [32] G. Casati, I. Guarneri, and D. L. Shepelyansky, Phys. Rev. Lett. 62, 345 (1989). [33] It is important here that ω1/ω2 is irrational; for rational case see below. [34] T. Nakanishi, T. Ohtsuki, and M. Saitoh, Journal of the Physical Society of Japan 62, 2773 (1993), http://dx.doi.org/10.1143/JPSJ.62.2773. [35] G. Sundaram and Q. Niu, Phys. Rev. B 59, 14915 (1999). [36] A. Abragam, The principles of nuclear magnetism, 32 [14] C. W. von Keyserlingk and S. L. Sondhi, ArXiv e-prints (Oxford university press, 1961). (2016), arXiv:1602.06949 [cond-mat.str-el]. [15] C. W. von Keyserlingk, V. Khemani, and S. L. Sondhi, Phys. Rev. B 94, 085112 (2016), arXiv:1605.00639 [cond- mat.dis-nn]. [16] D. V. Else, B. Bauer, and C. Nayak, ArXiv e-prints (2016), arXiv:1607.05277 [cond-mat.stat-mech]. [17] I.-D. Potirniche, A. C. Potter, M. Schleier-Smith, A. Vishwanath, and N. Y. Yao, ArXiv e-prints (2016), arXiv:1610.07611 [cond-mat.quant-gas]. [18] D. Abanin, W. De Roeck, and F. Huveneers, ArXiv e- prints (2014), arXiv:1412.4752 [cond-mat.dis-nn]. [19] A. Lazarides, A. Das, and R. Moessner, Physical Re- view Letters 115, 030402 (2015), arXiv:1410.3455 [cond- mat.stat-mech]. [20] L. Zhang, V. Khemani, and D. A. Huse, ArXiv e-prints (2016), arXiv:1609.00390 [cond-mat.dis-nn]. [37] E. G. Spencer, R. C. LeCraw, and A. M. Clogston, Phys. Rev. Lett. 3, 32 (1959). [38] A. M. Tyryshkin, S. A. Lyon, A. V. Astashkin, and A. M. Raitsimring, Phys. Rev. B 68, 193207 (2003). [39] V. E. Manucharyan, J. Koch, L. and M. H. Devoret, Science 326, http://science.sciencemag.org/content/326/5949/113.full.pdf. I. Glazman, 113 (2009), [40] B. van Heck, S. Mi, and A. R. Akhmerov, Phys. Rev. B 90, 155450 (2014). [41] R.-P. Riwar, M. Houzet, J. S. Meyer, and Y. V. Nazarov, Nature communications 7 (2016). [42] L. Fu, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 98, 106803 (2007). [43] Y. Li and C. Wu, Phys. Rev. Lett. 110, 216802 (2013). [44] Y. Li, S.-C. Zhang, and C. Wu, Phys. Rev. Lett. 111, 186803 (2013). [21] Y. H. Wang, H. Steinberg, P. Jarillo-Herrero, and [45] B. A. Bernevig, J. Hu, N. Toumbas, and S.-C. Zhang, N. Gedik, Science 342, 453 (2013). [22] 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), letter. [23] J. Zhang, P. W. Hess, A. Kyprianidis, P. Becker, A. Lee, J. Smith, G. Pagano, I.-D. Potirniche, A. C. Potter, A. Vishwanath, N. Y. Yao, and C. Monroe, ArXiv e- Phys. Rev. Lett. 91, 236803 (2003). [46] C. Janot, Quasicrystals (Springer, 1994). [47] A. Hurwitz, Math. Ann. 39, 279 (1891). [48] P. W. Brouwer, Phys. Rev. B 58, R10135 (1998). [49] D. J. Thouless, Phys. Rev. B 27, 6083 (1983). [75] B. I. Halperin, hbd (1941). 17 a. b. c. (a) FIG. 14. Work and fidelity in the weak drive regime. Energy transfers for η = 0.1. (b) Fidelities of initial state, an instantenous eigenstate of H(0), and the instantaneous eigenstates of H(t). The initial phases for these plots are φ1 = π/10, φ2 = 0. (c) Power pumped averaged for t < 2000 as a function of gap parameter m.
1112.3665	2	1112	2011-12-28T20:01:12	Electronic structure and chemical bonding of nc-TiC/a-C nanocomposites	[ "cond-mat.mes-hall", "physics.chem-ph" ]	The electronic structure of nanocrystalline (nc-) TiC/amorphous C nanocomposites has been investigated by soft x-ray absorption and emission spectroscopy. The measured spectra at the Ti 2p and C 1s thresholds of the nanocomposites are compared to those of Ti metal and amorphous C. The corresponding intensities of the electronic states for the valence and conduction bands in the nanocomposites are shown to strongly depend on the TiC carbide grain size. An increased charge-transfer between the Ti 3d-eg states and the C 2p states has been identified as the grain size decreases, causing an increased ionicity of the TiC nanocrystallites. It is suggested that the charge-transfer occurs at the interface between the nanocrystalline TiC and the amorphous C matrix and represents an interface bonding which may be essential for the understanding of the properties of nc-TiC/amorphous C and similar nanocomposites.	cond-mat.mes-hall	cond-mat	PHYSICAL REVIEW B 80, 235108 (2009) Electronic structure and chemical bonding of nc-TiC/a-C nanocomposites Martin Magnuson1, Erik Lewin2, Lars Hultman1 and Ulf Jansson2 1Department of Physics, Chemistry and Biology (IFM), Linköping University, SE- 58183 Linköping, Sweden. 2Department of Materials Chemistry, The Ångström Laboratory, Uppsala University, P.O. Box 538 SE-75121 Uppsala Abstract The electronic structure of nanocrystalline (nc-) TiC/amorphous C nanocomposites has been investigated by soft x-ray absorption and emission spectroscopy. The measured spectra at the Ti 2p and C 1s thresholds of the nanocomposites are compared to those of Ti metal and amorphous C. The corresponding intensities of the electronic states for the valence and conduction bands in the nanocomposites are shown to strongly depend on the TiC carbide grain size. An increased charge-transfer between the Ti 3d-eg states and the C 2p states has been identified as the grain size decreases, causing an increased ionicity of the TiC nanocrystallites. It is suggested that the charge-transfer occurs at the interface between the nanocrystalline TiC and the amorphous C matrix and represents an interface bonding which may be essential for the understanding of the properties of nc-TiC/amorphous C and similar nanocomposites. 1 Introduction Nanocomposites comprise materials with two or more phases for which at least one has nanometer-size crystallites [1]. Carbon-based nanocomposites of nanocrystalline metal-carbides (nc-MC) embedded in an amorphous carbon (a-C) matrix, are materials with inherent design possibilities for applications utilizing mechanical, tribological, and/or electrical properties. By tuning the TiC grain size and fraction of a-C matrix, the properties can be controlled as the matrix phase increases the toughness of the material and softens the nanocomposite compared to pure carbides but also provides a source of solid lubricant [2,3,4,5,6,7,12]. The nc-TiC has a NaCl crystal structure and its bonding is a mixture of covalent, ionic and metallic bonds where the covalent contribution consists of Ti eg - C 2p (pdσ), Ti t2g - C 2p (pdπ), and Ti - Ti t2g (ddσ) bonds [9], which are all found in the valence band. Previous experimental investigations of the electronic structure of nc-TiC/a-C nanocomposites have used core-level x-ray photoelectron spectroscopy (XPS) [10,11,12,13]. Core-level C 1s XPS measurements of TiC are known to exhibit two different spectral components originating from C-C and C-Ti bonding. The spectral component of the C-Ti bonding has a rather large (∼3 eV) chemical shift towards lower binding energy in comparison to the C-C component in a-C, which is a signature of electronic charge transfer from Ti to C. For the Ti 2p core levels, the 1 PHYSICAL REVIEW B 80, 235108 (2009) TiC (200) TiC (220) TiC (311) chemical shift of the XPS peaks of TiC is small compared to the large shift at the C 1s core level. An interesting observation is that the C 1s XPS spectrum of the nc-TiC/a- C nanocomposite is different compared to single phase TiC. In the nanocomposite C 1s spectrum, an additional shoulder is observed at about ∼ 283 eV binding energy between the C-C and Ti-C peaks, which we denote C-Ti* [11,14,3,7,15,16,17,18]. The intensity of this additional feature increases upon sputter-etching [19]. However, recent studies with high energy XPS have shown that it is present also in unsputtered samples. Furthermore, the relative intensity of the C-Ti* feature increases with reduced TiC grain size [11]. The fact that it is almost only observed in nanocomposites and related to the grain size suggest that it can be attributed to an interfacial state at the TiC-matrix interface. The presence of such interfacial state of varying quantity may strongly affect the physical, chemical and mechanical properties of the nanocomposites. For a-C films, the relative intensity of the ∼ 283 eV feature is also known to depend on the differences in sample structure or composition caused by different deposition methods [20,21]. XRD TiC (111) ) s t i n u . b r a ( y t i s n e t n I nc-TiC(15 nm)/a-C nc-TiC(10 nm)/a-C nc-TiC(8 nm)/a-C nc-TiC(2 nm)/a-C 50 60 Scattering angle 2(cid:101) (degrees) Additional information about the unoccupied states can be achieved with absorption x-ray soft spectroscopy (SXA) in either surface sensitive total electron yield (TEY) mode or bulk-sensitive fluorescence yield (TFY) mode. A previous SXA of C:V, investigation C:Co, C:Cu films nanocomposite grown by ion beam cosputtering indicate that Figure 1: (Color Online) XRD from the nc-TiC/a-C films with the interfaces are indeed different grain size (in parenthesis). important into take to account [22]. The C 1s x- ray absorption spectra of carbon materials usually exhibit a rather sharp and characteristic π* absorption peak and a broad well-separated σ* shape resonance at more than 5 eV higher energy. However, the origin of a C 1s absorption feature located between the π* and σ* absorption peaks in transition metal carbide materials has been controversial [23,24] and may be due to C-O bonding, sp2-sp3 hybridization or hybridized C 2p - transition metal 3d - 4sp states at the interfaces [22]. The aim of this study is to increase the understanding of the nature of the electronic structure, the interface state and chemical bonding in nc-TiC/a-C nanocomposites using SXA spectroscopy in TEY and TFY modes in combination with soft x-ray emission (SXE) spectroscopy. The SXA technique probes the unoccupied electronic states, while the SXE technique probes the occupied electronic states in the materials. For probing the occupied states, the SXE technique is more bulk sensitive than electron-based spectroscopic techniques which is useful when investigating internal and embedded electronic structures and interfaces. The combination of SXA and SXE measurements on nc-TiC/a-C nanocomposites with different carbide grain sizes (i.e. varying interface/bulk ratio) gives valuable information on possible intermediate interface states. In addition, the different types 70 80 30 40 2 PHYSICAL REVIEW B 80, 235108 (2009) of C contributions are selected by tuning the excitation energies for the SXE measurements which give additional insight into the controversial nature of the ∼ 283 eV absorption peak feature between the π* and σ* C 1s absorption resonances. 2 Experimental C 1s XPS C-C 2.1 Deposition of nanocrystalline films The deposition of the nc-TiC/a-C and a-C films were carried out in an ultra high vacuum chamber (base pressure 10 Torr) by non-reactive, unbalanced DC-magnetron sputtering from separate 2 in. elemental targets, supplied by Kurt J. Lesker Company Ltd. (purity specified as 99,995% and 99,999% for Ti and C, respectively). Through tuning of magnetron currents, samples with a carbon content between 35 and 100 at. % were deposited. The sample thicknesses were kept constant at about 0.2 µm. The Ar-plasma was generated at a constant pressure of 3.0 mTorr, with a flow rate of Ar into the chamber of 150 sccm. Substrates were placed about 15 cm below the magnetrons on a rotating substrate holder. Deposition was carried out simultaneously on amorphous SiO and Si(111)-substrates. The films deposited on SiO-substrates were used in the SXA/SXE experiments and XRD, while the films deposited on Si- substrates were used for XPS analysis. The substrates were heated to 300C by a BN- plate with W-wires, situated about 5 mm below the substrates. The temperature was monitored using a Mikron M90-0 infrared pyrometer, calibrated against a TiC thin film using a thermo-element. Prior to deposition, the substrates were preheated for 30 minutes, and the targets were presputtered for 10 minutes. ) s t i n u . b r a ( y t i s n e t n I nc-TiC(15 nm)/a-C nc-TiC(10 nm)/a-C nc-TiC(8 nm)/a-C nc-TiC(2 nm)/a-C a-C were samples The analyzed X-ray with diffraction (XRD) using a Philips MRD X’pert diffractometer with parallel beam geometry and grazing incidence (GI-XRD) scans using a 2 incident angle. XPS using was performed a Physical Electronics Quantum ESCA 2000 Microprobe. The chemical composition (C and Ti content) was determined from sputter the XPS depth profiles, using sensitivity Figure 2: (Color Online) A series of C 1s XPS spectra of the factors calibrated against a nc-TiC/a-C films with different grain size (in parenthesis), plotted on a binding energy scale. The spectra were measured bulk TiC reference sample to after sputter-etching. calculate composition the with regards to Ti and C. The C bonding was analysed by high resolution XPS acquisitions after sputter etching to a depth of 150 Å. Etching was performed with low ion energy (200 V Ar) to minimise sputter damage; ion beam was rastered over a 1x1 mm area, and the analysis spot was 200 µm in diameter. 285 283 284 Binding Energy (eV) 282 281 C-Ti C-Ti* 287 286 3 PHYSICAL REVIEW B 80, 235108 (2009) 2.2 X-ray absorption and emission measurements The SXA and SXE measurements were performed at the undulator beamline I511-3 at MAX II (MAX-lab National Laboratory, Lund University, Sweden), comprising a 49-pole undulator and a modified SX-700 plane grating monochromator [25]. The SXE spectra were measured with a high-resolution Rowland-mount grazing-incidence grating spectrometer [26] with a two-dimensional detector. The Ti L and C K SXE spectra were recorded using a spherical grating with 1200 lines/mm of 5 m radius in the first order of diffraction. The SXA spectra at the Ti 2p and C 1s edges were measured in both TEY and TFY modes with 0.1 eV resolution at 90 and 20 incidence angles, respectively. The SXA spectra were normalized to the step edge below and far above the thresholds while the SXE spectra were normalized to the peak height. During the Ti L and C K SXE measurements, the resolutions of the beamline monochromator were 0.5, and 0.2 eV, respectively. The SXE spectra were recorded with spectrometer resolutions of 0.5 and 0.2 eV, respectively. All the measurements were performed with a base pressure lower than Torr. In order to minimize self- absorption effects [27], the angle of incidence was 20 from the surface plane during the SXE and SXA-TFY measurements. The x-ray photons were detected parallel to the polarization vector of the incoming beam in order to minimize elastic scattering. For comparison of the spectral shapes, the SXA spectra were normalized to the step edge below and far above the Ti 2p and C 1s thresholds in Fig. 3 and Fig. 4 (490 eV and 310 eV, respectively). The Ti L2,3 and C K SXE spectra were normalized to the main peak heights and were plotted on a photon energy scale (bottom) and an energy scale relative to the (top) in Fig. 3 and Fig. 5. 3 Results 3.1 GI-XRD and C1s core-level XPS analysis Results from GI-XRD and C 1s core-level XPS analysis are presented in Figure 1 and 2, respectively, and are in agreement with previous studies [10, 11]. When the total carbon-content increases the relative amount of a-C phase increases and the TiC grain size decreases, see summary in Table 1 and details below. In the diffractograms (Figure 1) all observed reflections can be indexed to TiC [28, 29] with a lattice parameter of 4.33-4.37 Å. The TiC grain sizes were estimated using Scherrer’s equation [30, 31] to vary between 2 and 15 nm, decreasing with carbon content. This decrease in grain size represents an increase of the surface/volume ratio of the TiC grains by a factor of 7.5. Figure 2 shows a series of C 1s core-level XPS spectra of the nc-TiC/a-C films. The two main peaks at 284.6 and 281.9 eV are associated with C-C and C-Ti bonds, respectivly. The smaller feature at 282.8 eV denoted C-Ti* is partly a feature of the nanocomposite, and is partly due to sputter damage which in this case should be relatively small [19]. The observed increase of the C-Ti* intensity in Fig. 2 with decreasing grain size is consistent with the possible behavior of an interface contribution. Through curve fitting analysis, the relative intensities of the three contributions were extracted and the relative amount of a-C and TiC phases estimated (see Table 1). 4 PHYSICAL REVIEW B 80, 235108 (2009) -20 -10 20 Ti 3d - C 2s L3 TEY TFY ) s t i n u . b r a ( y t i s n e t n I Ti 2p x-ray absorption Energy (eV) 0 10 t2g eg 2p3/2 eg t2g 2p1/2 3.2 Ti 2p x-ray absorption Figure 3 (top and middle) shows Ti 2p TEY- and TFY-SXA spectra following the 2p3/2,1/2->3d4s dipole transitions of the nc-TiC/a-C films with different TiC grain sizes in comparison to Ti metal. The Ti 2p SXA intensity is proportional to the unoccupied 3d states and gives complementary information in the TEY and the TFY modes which depend on the incidence angle, difference in probe depths, transition processes, self-absorption and detection methods [32,33]. The SXA spectrum of pure Ti metal (black curve) has single 2p3/2 and 2p1/2 absorption peaks whereas the SXA the carbide spectra of containing materials exhibit t2g - eg crystal- field split double peaks due to the octahedral symmetry around the Ti in TiC. The eg site orbitals point toward the nearest C-atom sites, t2g orbitals the while point towards the tetrahedral holes in the fcc Ti-atom lattice. The t2g - eg absorption peaks at 457.6 and 459.2 eV are associated with transitions from the core shell while the t2g - eg peaks at 463.2 and 464.8 Figure 3: (Color Online) Top and middle: Ti 2p TEY- and TFY- are associated with the SXA spectra of films with different grain size (in parenthesis) in core shell. Although the comparison to bulk Ti metal. Bottom: nonresonant Ti SXE spectra of nc-TiC/a-C and Ti metal excited at 490 eV. peak intensities vary, the 2p Ti crystal-field splitting (1.6 ± 0.1 eV) is the same in all spectra which is consistent with what has previously been observed in Ti 2p SXA spectra of crystalline bulk TiC [34]. The crystal field-splitting of TiO is known to be much larger (∼ 3 eV) [35]. As observed, the intensity of the Ti 2p absorption increases as the grain size decreases, which shows that the number of empty Ti 3d states increases. 3.3 Ti x-ray emission Figure 3 (bottom) shows Ti L2,3 SXE spectra of nc-TiC/a-C in comparison to Ti metal, nonresonantly excited at 490 eV. Starting with the Ti L2,3 SXE spectra of Ti metal, the main emission line is observed at -1.2 eV on the energy scale relative to the E of the component on the energy scale at the top of Fig. 3. For the Ti L2,3 SXE spectra of nc-TiC/a-C, the main L3 peak observed at -2.1 eV is due to Ti 3d - C 2p hybridization while the weak shoulder at -10 eV is due to Ti 3d - C 2s hybridization [36, 37]. The small L2 emission line is found at +5.0 eV for Ti metal and +4.1 eV for the nc-TiC/a-C relative to the E of the 2p3/2 component. The low-energy shift (-0.9 eV) and the increased broadening of the L3 and L2 peaks for decreasing grain size are L2 460 450 Photon Energy (eV) Ti L2,3 x-ray emission h(cid:105)=490 eV 470 480 Ti 3d - C 2p 430 440 5 PHYSICAL REVIEW B 80, 235108 (2009) an indication of stronger Ti 3d - C 2p interaction and bonding than for the more well- defined Ti 3d - Ti 3d hybridization region in Ti metal [34]. As expected, the t2g - eg crystal field splitting does not appear for the occupied 3d valence band in the solid state consisting of overlapping bonding σ and π bands. The larger L2,3 spin-orbit splitting of 6.2 eV in SXE in comparison to SXA (5.6 eV) is consistent with earlier observations for TiC [34]. The trend in the L3/L2 branching ratio in transition metal compounds is a signature of the degree of ionicity in the systems [38]. For conducting systems, the / ratio is usually significantly higher than the statistical ratio 2:1 due to the additional Coster-Kronig process [27, 40]. The observed L3/L2 ratio systematically decreases as the grain size decreases, see Table I. The Ti atoms in the TiC nanocrystallites with small grain size thus appear to be more ionic and less metallic than those with larger grains. For Ti metal, the observed L3/L2 ratio of 3.74 is not directly comparable with the trend in the nanocomposite compounds. 20 25 ) s t i n u . b r a ( y t i s n e t n I nc-TiC(15 nm)/a-C nc-TiC(10 nm)/a-C nc-TiC(8 nm)/a-C nc-TiC(2 nm)/a-C a-C Energy (eV) 15 10 5 0 C 1s x-ray absorption (1) (2) 3.4 C 1s x-ray absorption Figure 4 shows experimental C 1s TEY- and TFY-SXA spectra of nc-TiC/a-C and a- C where the intensity is proportional to the unoccupied C 2p states. Contrary to the case of the Ti spectra, the C spectra of the nanocomposites consists of superimposed contributions from both the nc-TiC carbide and the a-C matrix. The SXA spectra were measured to identify the absorption features and peak maxima for the excitation energies of the SXE spectra presented in Fig. 5. The SXA energy region below 289 eV to contain π* is known resonances whereas the region above 289 eV contains σ* resonances [41, 42]. The increased absorption above to 1s-> σ* 289 eV due forms a broad transitions shape resonance due to multielectron excitations. The C 1s SXA spectra of nc-TiC/a-C the nanocomposites in Fig. 4 mainly exhibit two π* peaks at 284.9 eV and 288.2 eV, denoted (1) and (2), respectively. The first C 1s SXA peak (1) at 284.9 eV, has two contributions, partly due to C 2p - Ti 3d - t2g hybridization in the carbide nc-TiC nanocrystallites [43, 44] and partly due to C=C bonding contribution from the Figure 4: (Color Online) SXA spectra of nc-TiC/a-C films with different grain size (in parenthesis) and amorphous C with the characteristic and peak regions indicated by the horizontal arrows at the bottom. The dotted vertical lines indicate the excitation energies (1), (2’) and (2) for the SXE measurements shown in Fig. 5. 295 290 300 Photon Energy (eV) 280 285 305 310 TEY TFY (2') (1') (cid:47)(cid:155) (cid:109)(cid:155) 6 PHYSICAL REVIEW B 80, 235108 (2009) -15 -10 5 10 15 C K x-ray emission ) s t i n u . b r a ( y t i s n e t n I (2') h(cid:105)=287.2 eV Energy (eV) 0 -5 a-C nc-TiC(2 nm)/a-C nc-TiC(8 nm)/a-C nc-TiC(10 nm)/a-C nc-TiC(15 nm)/a-C (1) h(cid:105)=284.9 eV a-C matrix [45, 24]. The origin of the second π* peak (2) at 288.2 eV has been controversial [23, 24]. It has been suggested that peak (2) is due to either C-O bonding in an atmospherically oxidized surface layer, sp2-sp3 hybridization or hybridized C 2p - transition metal 3d - 4sp states. As observed in Fig. 4, the intensity of peak (2) is higher in the surface-sensitive TEY spectra than in the more bulk- sensitive TFY spectra [22] and the relative intensity of peak (2) is known to largely depend on the sample structure and composition by different deposition methods [20, 21]. However, the peak intensity in TFY also depends on the incidence angle and is reduced due to self-absorption effects at normal incidence. Peak (1) has a broad and pronounced low-energy shoulder (1’) below the main peak at 281- 283 eV that is solely due to the carbide contribution. The low energy shoulder the reflects gradual carbide formation for increasing grain size as seen in Table I. Peak (2) also has a low- energy shoulder (2’) at 286-288 eV. The observed intensity quenching of the shoulders (1’) and (2’) for decreasing grain size implies of depletion a unoccupied C 2p states as the carbide contribution decreases. Although part of peak (2) in the surface sensitive TEY-XAS spectra can be attributed to C-O to atmospheric bonding due oxidation at the surface, the more bulk sensitive TFY spectra show that it is also due to C 2p - Ti 3d - eg hybridization in TiC with addition of the superimposed a-C contribution. The apparent t2g - eg splitting originating from the unoccupied Ti 3d orbitals is here indirectly observed in the C 1s SXA spectra, but is significantly wider (2-3 eV) than for the Ti 2p SXA spectra in Fig. 3 (1.6 eV). The intensity around 291 eV systematically increases as the grain size decreases with increasing amount of a-C. The a-C spectrum has a well-defined shape-resonance structure around 292 eV observed both in TEY and TFY. Note that the intensity trend is largely opposite at 282 eV in comparison to at the a-C shape-resonance at 292 eV. Figure 5: (Color Online) Resonant and nonresonant C K SXE spectra of nc-TiC/a-C films with different grain size (in parenthesis) and amorphous C excited at peak (1) at 284.9, the shoulder (2’) at 287.2 eV and peak (2) at 288.2 eV in the SXA spectra and nonresonant at 310 eV. The occupied σ and π bands of a-C are indicated at the bottom. 275 285 280 Photon Energy (eV) (2) h(cid:105)=288.2 eV h(cid:105)=310 eV 290 295 (cid:109) (cid:47) C 2p - Ti 3d 270 3.5 C K x-ray emission Figure 5 shows C K SXE spectra excited at peak (1) in the SXA spectra at 284.9 eV, at the shoulder (2’) at 287.2 eV and at peak (2) at 288.2 eV (resonant) and 310 eV 7 PHYSICAL REVIEW B 80, 235108 (2009) (nonresonant) photon energies, probing the occupied C 2p states of the valence bands. As in the case of the C 1s SXA spectra, the C K SXE spectra of the nanocomposites represent a superposition of two contributions from the nc-TiC carbide and the surrounding a-C matrix. As the C 2p intensity is the largest in the case of the smallest TiC grain size, the intensity trend is now opposite from the SXA spectra shown in Fig. 4. As can be seen in Fig. 5, most intensity in the upper valence band is observed for the C K SXE spectra of a-C (black curve). The main peak corresponding to the occupied C 2p σ band is observed at 277 eV photon energy with a broad high-energy band at 281.5 eV with π character. The C K SXE spectra of a-C exhibit a similar spectral shape as sp2 hybridized graphite [46]. However, the intensity of the π band at 281.5 eV is higher for a-C than in graphite, indicating additional influence of hybridized C at the top of the valence band as in the case of hybridized diamond [46]. In comparison to C K SXE spectra of graphite and diamond, the spectral shapes of a-C and the nanocomposites show less excitation energy dependence. For a-C, the σ/π intensity ratio remains constant (2.4) for the resonant photon energies while it decreases (1.4) for the non-resonant excitation at 310 eV. -15 (cid:109) 0 (cid:47) h(cid:105)=288.2 eV (2) Energy (eV) -5 -10 h(cid:105)=287.2 eV (2') ) s t i n u . b r a ( y t i s n e t n I nc-TiC(2 nm)/a-C nc-TiC(8 nm)/a-C nc-TiC(10 nm)/a-C h(cid:105)=284.9 eV (1) The details of the spectral SXE features of the nanocomposites on only depend not the overlapping relative spectral contributions of the π and σ bands of the a-C matrix but also on the addition of the superimposed contributions from nc-TiC and possible interface states. Contrary to the case of a-C, the C K SXE spectra of the nanocrystallites with 15 nm size (red curve), have the characteristic spectral shape of TiC [34]. It has a sharp main peak at -2.5 eV and a low-energy shoulder at -3.8 eV below the E of TiC at the energy scale at the top of Fig. 5. The sharp -2.5 eV peak is characteristic of the strong covalent Ti 3d - C 2p bonding in TiC [39, 34]. Note that the -3.8 eV low-energy shoulder is sharpest for the spectra excited at 284.9 eV 288.2 and eV corresponding to the main SXA peaks (1) and (2) in Fig. 4. A high-energy shoulder between the σ and the the π bands of nanocomposites at -1.0, is most developed for the spectra excited at 284.9 eV (1), indicating contribution of C 2p - Ti 3d - t2g hybridization in this energy region (0 to -2 eV). For the spectra excited at the other excitation energies, the high-energy shoulder is absent. This implies a different type of bonding symmetry for Figure 6: (Color Online) Difference plots of resonant and nonresonant C K SXE spectra from Fig. 5 of three nc- TiC/a-C films with different grain size (in parenthesis) obtained by subtraction from a superposition of the spectra with the 15 nm grain size and a-C excited at peak (1) at 284.9, at the shoulder (2’) at 287.2 and peak (2) at 288.2 eV in the SXA spectra and nonresonant at 310 eV. 275 Photon Energy (eV) 265 270 280 285 h(cid:105)=310.0 eV 8 PHYSICAL REVIEW B 80, 235108 (2009) excitation at peak (1) than at peak (2). In Fig. 5, the SXE spectra of the nanocomposites consist of a sum of the C contributions from the nc-TiC carbide and the surrounding a-C matrix. In addition, there is a possible contribution from an interface state at the surfaces of the nc-TiC crystallites as suggested by the C-Ti* peak in XPS (Fig. 2). A way to resolve the interface contribution is to make difference plots based on the C K SXE spectra in Fig. 5. This was done by a weighted superposition of the spectra from the a-C sample and the 15 nm sized nanocomposite, which has a very similar spectral shape as bulk TiC. The same weight factors were used for all four excitation energies and the results are shown in Fig. 6. For the 2 nm sample, weight factors of 0.60 and 0.40 were used. For the 8 nm sample, weight factor of 0.82 and 0.18 were used. For the 2 nm sample, weight factors of 0.93 and 0.07 were used. The intensity and integrated area of the difference spectra increases as the grain size decreases. This would be consistent with an interface state or component, which should grow for smaller grain sizes. The shape of the difference spectra largely depends on the excitation energy. For the spectra excited at peak (1) at 284.9 eV, the intensity difference is negative in the π energy region 280-282 eV just below the E of TiC. The negative intensity is a signature of broken Ti-C bonds with orbitals of t2g symmetry in TiC. This observation is consistent with theoretical studies of C adsorbed on TiC [47]. The difference spectrum excited at the shoulder (2’) at 287.2 eV has a main peak at -1.8 eV (279.7 eV), which is much lower for the other excitation energies. This is an indication of an additional spectral component in the SXA spectrum at 287.2 eV (2’) and probably a result of additional hybridization with the Ti 3d orbitals of eg symmetry which hybridize with the C 2p states at this excitation energy in the C 1s SXA spectra. Table 1: The first column gives the estimated grain size of the TiC nanocrystallites from XRD and the second column gives the total C content of the samples. The third and fourth column gives the relative amount of bonds from peak-fitting of the XPS spectra in the C 1s region in Fig. 2. The last column shows the L3/L2 ratio in the Ti SXE spectra in Fig. 3. C-Ti + C-Ti* C tot Grain size 93% 35% 15 nm 10 nm 42% 76% 60% 51% 8 nm 2 nm 65% 34% L3/L2 6.56 5.12 4.72 4.08 C-C 7% 24% 40% 66% 4 Discussion In the SXA and SXE spectra of Ti and C in Figs. 3-6, there are several interesting observations. For Ti in the nc-TiC carbide phase in Fig. 3, the number of unoccupied Ti 3d states increase with decreasing grain size both in TEY and TFY. The relative intensities between the t2g - eg ligand field peaks observed in Ti 2p SXA show that the unoccupied 3d states of the Ti atoms in the smallest nanocrystalline TiC grains are most affected by charge-transfer from the Ti 3d to the C 2p orbitals. This charge- transfer may occur within the TiC nanocrystals but more likely across the interface to the surrounding a-C phase. Comparing the Ti 2p SXA and SXE spectra in Fig. 3 of the nc-TiC/a-C nanocomposites with those of Ti metal, the spectral intensities are largely affected by the grain size of the TiC nanocrystallites. At the same time as the number of unoccupied Ti 3d states increase, we observe that the number of occupied Ti 3d states decreases as the grain size decreases. This is consistent with an increased 9 PHYSICAL REVIEW B 80, 235108 (2009) charge-transfer from the Ti 3d to the C 2p orbitals as the grain size decreases. The increased charge-transfer is also reflected in the / branching ratio which shows that the ionicity increases as the grain size decreases. For the superimposed C contributions from the nc-TiC phase and the a-C matrix in Figs. 4 and 5, the number of unoccupied states decreases and the number of occupied states increases as the grain size decreases. In C 1s SXA, the intensities of two pronounced peaks due to C 2p - Ti 3d - t2g and C 2p - Ti 3d - eg hybridization show strong variation with grain size. The origin of the absorption peak (2) between the and C 1s absorption resonances has been controversial [22, 23, 24] and has been assigned to either C-O bonding, sp2-sp3 hybridization or hybridized C 2p - transition metal 3d - 4sp states at the interface. Although part of peak (2) in surface sensitive TEY-XAS can be attributed to C-O bonding due to atmospheric oxidation at the surface, the more bulk sensitive TFY spectra show that it is also due to C 2p - Ti 3d - eg hybridization in TiC with addition of the superimposed a-C contribution. The apparent t2g-eg splitting originating from the unoccupied Ti 3d bands is here indirectly observed in the C 1s SXA spectra, but is significantly broader (2-3 eV) than for the Ti 2p SXA spectra in Fig. 3 (1.6 eV). For the C K emission spectra, an increased number of occupied states is observed as the C content is increased for the smaller TiC crystallites. Note that this trend is opposite from the case of the Ti SXA/SXE spectra in Fig. 3 and implies an increased charge-transfer from Ti to C as the grain size decreases. The largest C K emission intensity is observed for amorphous C which contains the most occupied C 2p electronic states with distinguishable σ and π bands. From the spectral features in the SXE spectra in Figs 3 and 5, two main types of bonds were also identified; the strong Ti 3d - C 2p carbide bonding in the TiC phase and the weaker C 2p - C 2p bonding in the a-C phase. The Ti 3d - C 2p covalent bond region is concentrated to a specific energy region -2.5 eV below E while the C 2p - C 2p hybridization generally occurs in a much wider energy region with overlapping but distinguishable σ and π bands. From the C K SXE difference spectra of weighted superpositions in Fig. 6, the intensity and the integrated area increases as the grain size decreases. The SXE intensity is also largely quenched just below E for the excitation energy at the main peak (1) and additional SXE intensity occurs between the occupied σ and π C bands for excitation at the C 1s SXA shoulder (2’). These observations are correlated with a possible interface state or phase contribution as previously observed as a feature denoted C-Ti in the C 1s XPS spectra in Fig. 2. When the grain size decreases from 15 nm to 2 nm, it implies that the interface/bulk ratio increases for the nc-TiC carbide phase by a factor of 7.5. Although an interface component is not directly observed in the C K SXE spectra, it is identified in the difference spectra in Fig. 6. The C K SXE difference spectra reveal a spectral component with more broken Ti-C bonds of orbitals with t2g character and additional bonds with orbitals with eg character for increasing interface/bulk ratio as the nc-TiC grains become smaller. 5 Conclusions The electronic interface component of structure, chemical bonding and nanocomposites of TiC crystallites embedded in a matrix of amorphous carbon has been investigated using x-ray absorption and x-ray emission spectroscopy. A strong intensity dependence of the TiC grain size is observed. The trend in the L3/L2 branching ratio of Ti L2,3 x-ray emission indicates an increased ionicity of the Ti 10 PHYSICAL REVIEW B 80, 235108 (2009) atoms as the grain size decreases. This is caused by increased charge-transfer from the Ti 3d to the C 2p states as the grain size decreases. Analysis of C K x-ray emission difference spectra by weighted spectral superposition reveals increased spectral intensity as the grains size decreases. This is consistent with an additional component at the TiC/amorphous carbon interface as previously suggested by C 1s photoelectron spectra. It is found that the additional intensity involves more Ti 3d orbitals of eg symmetry than t2g symmetry. This suggests a distinctly different bonding at the surface of the TiC nanocrystallites embedded in amorphous carbon, where an increased charge-transfer from the Ti 3d eg orbitals at the interface to the C 2p orbitals in the amorphous C phase causes an increased ionicity of the entire TiC nanocrystallites. An interface bonding in the nanocomposites may affect the physical properties such as electrical conductivity and possible hardness or elasticity. Since understanding the interfacial bonding in a nanocomposite is essential to enable the design of the materials properties, this will be the subject of further spectroscopic and theoretical studies. 6 Acknowledgements We would like to thank the staff at MAX-lab for experimental support. This work was supported by the Swedish Research Council, the Göran Gustafsson Foundation, the Swedish Strategic Research Foundation (SSF), Strategic Materials Reseach Center on Materials Science for Nanoscale Surface Engineering (MSE). References [1] Nanostructured Coatings, Eds. A.Cavaleiro and J. Th. M. De Hosson (2006). [2] A. A. Voevodin and J. S. Zabinsky; J. Mater. Sci. 33, 319 (1998). [3] M. Stüber, H. Leiste, S. Ulrich, H. Holleck and D. Schild; Surf. Coat. Technol. 150 218 (2002). [4] C. Benndorf, E. Boettiger, M. Fryda, H. G. Haubold, C.-P. Klages and H. Korbele; Synth. Met. 41-43, 4055 (1991). [5] S. Zahng, X. L. Bui, J. Jiang and X. Li; Surf. Coat. Technol. 206 (2005). [6] A. A. Voevodin, S. V. Prasad and J. S. Zabinski; J. Appl. Phys. 82, 855 (1997). [7] W. Gulbinski, S. Mathur, H. Shen, T. Suszko, A. Gilewicz and B. Warcholinski; Appl. Surf. Sci. 239, 302 (2005). [8] T. Zehnder and J. Patscheider; Surf. Coat. Technol. 133/134, 134 (2000). [9] K. Schwarz; CRC Critical Reviews in Solid State and Materials Sciences, 13, 211 (1987). [10] E. Lewin, O. Wilhelmsson and U. Jansson; J. Appl. Phys. 100, 054303 (2006). [11] E. Lewin, P. O. Å. Persson, M. Lattemann, M. Stüber, M. Gorgoi, A. Sandell, C. Ziebert, F. Schäfers, W. Braun, J. Halbritter, S. Ulrich, W. Eberhardt, L. Hultman, H. Siegbahn, S. Svensson and U. Jansson; Surf. Coat. Technol. 202, 3563 (2008). [12] T. Zehnder and J. Patscheider; Surface and Coatings Technology; 133-134, 134 (2000). [13] A. A. Voevodin and J. S. Zabinski; Journal of Materials Science; 33, 319 (1998). [14] M. Grischke; Entwiklung eines Strukturmodells Metallhaltiger Kohlenstoff- Schichten [a-C:H(Me)] mit Carbidbildener Metallkomponente, VDIVerlag 11 PHYSICAL REVIEW B 80, 235108 (2009) GmbH, Dusseldorf, 1989. [15] V. Schier, H.-J. Michel, J. Halbritter and J. Fresenius; J. Anal. Chem. 346, 227 (1993). [16] B. Hornetz, H.-J. Michel, J. Halbritter and J. Frenelius; Anal. Chem. 349, 233 (1994). [17] G. Li and L. F. Xia; Thin Solid Films; 396, 16 (2001). [18] A. Mani, P. Aubert, F. Mercier, H. Khodja, C. Berthier and P. Houdy; Surf. Coat. Technol. 194, 190 (2005). [19] E. Lewin, M. Gorgio, F. Schäfers, S. Svensson and U. Jansson; Surface Coatings and Technology, in press (2009). [20] Y. Muramatsu, S. Hirono, S. Umemura, Y. Ueno, T. Hayashi, M. M. Grush, E. M. Gullikson and R. C. C. Perera; Carbon 39, 1403 (2001). [21] Y. Muramatsu, E. M. Gullikson, R. C. C. Perera; Phys. Scr. T115, 501 (2005). [22] G. Abrasonis, M. Berndt, M. Krause, K. Kuepper, F. Munnik, A. Kolitsch and W. Möller; J. Phys. Chem. C 112, 17161 (2008). [23] J. Diaz, S. Anders, X. Zhou, E. J. Moler, S. A. Kellar and Z. Hussain; Phys. Rev. B 64, 125204 (2001). [24] C. Lenardi, P. Piseri, V. Briois, C. E. Bottani, A. Li Bassi and P. Milani; J. Appl. Phys. 85, 7159 (1999). [25] R. Denecke, P. Vaterlein, M. Bassler, N. Wassdahl, S. Butorin, A. Nilsson, J.-E. Rubensson, J. Nordgren, N. Mårtensson and R. Nyholm; J. Electron Spectrosc. Relat. Phenom. 101-103, 971 (1999). [26] J. Nordgren and R. Nyholm; Nucl. Instr. Methods A246, 242 (1986); J. Nordgren, G. Bray, S. Cramm, R. Nyholm, J.-E. Rubensson and N. Wassdahl; Rev. Sci. Instrum. 60, 1690 (1989). [27] M. Magnuson, N. Wassdahl and J. Nordgren; Phys. Rev. B 56, 12238 (1997). [28] M. C. Morris, H. F. McMurdie, E. H. Evans, B. Paretzkin, H.Parker, N. P. Pyrros, and C. R. Hubbard, National bureau of standards (U.S.) monograph 25 1981, section 18, 73. [29] Phase Diagrams of Binary Titanium Alloys, J. L. Murray, ASM International, 1989. [30] Elements of X-ray diffraction, B. D. Cullity, Addison-wesley Publishing company, Inc., Reading, MA, 1978. [31] P. Scardi, M. Leoni, and R. Delhez, J. Appl. Crystallogr. 37, 381 (2004). [32] S. Eisebitt, T. Boske, J.-E. Rubensson and W. Eberhardt; Phys. Rev. B 47, 14103 (1993). [33] F. Reinhardt, B. Beckhoff, H. Eba, B. Kanngiesser, M. Kolbe, M. Mizusawa, M. Müller, B. Pollakowski, K. Sakurai and G. Ulm; Anal. Chem 81, 1770 (2009). [34] M. Magnuson, O. Wilhelmsson, J. -P. Palmquist, U. Jansson, M. Mattesini, S. Li, R. Ahuja and O. Eriksson, Phys. Rev. B. 74 195108 (2006). [35] M. Matsubara, T. Uozumo and A. Kotani; J. Phys. Soc. Jpn. 69, 1558 (2000). [36] V. A. Gubanov, E. Z. Kurmaev and G. P. Shveikin; J. J. Phys. Chem. Solids 38, 201 (1977). [37] M. Magnuson, M. Mattesini, S. Li, C. Höglund, M. Beckers, L. Hultman, and O. Eriksson; Phys. Rev. B. 74 195127 (2007). [38] J. Kawai, T. Konishi, A. Shimohara and Y. Goshi; Spectrochemica Acta, Part B 49B, 725 (1994). [39] M. Magnuson, J. -P. Palmquist, M. Mattesini, S. Li, R. Ahuja, O. Eriksson, J. Emmerlich, O. Wilhelmsson, P. Eklund, H. Högberg, L. Hultman, U. Jansson; Phys. Rev. B 72, 245101 (2005). 12 PHYSICAL REVIEW B 80, 235108 (2009) [40] M. Magnuson, L.-C. Duda, S. M. Butorin, P. Kuiper and J. Nordgren; Phys. Rev. B 74, 172409 (2006). [41] M. Magnuson, L. Yang, J.-H. Guo, C. Såthe, A. Agui, J. Nordgren, Y. Luo, H. Ågren, N. Johansson, W. R. Salaneck, L. E. Horsburgh and A. P. Monkman; Chem. Phys. 237, 295 (1998). [42] M. Magnuson, J.-H. Guo, S. M. Butorin, A. Agui, C. Såthe, J. Nordgren and A. P. Monkman; J. Chem. Phys. 111, 4756 (1999). [43] A. T. Paxton, M. van Schilfgaarde, M. MacKenzie and A. J. Craven; J. Phys. Cond. Mat. 12, 729 (2000). [44] Pflüger, J. Fink, G. Crecelius, K. P. Bohnen and H. Winter; Solid State Commun. 44, 489 (1982). [45] C. Adelhelm, M. Balden, F. Kost, A. Herrmann and S. Lindig; J. Physics; Conf. Series 100 , 062033 (2008). [46] J. Guo and J. Nordgren; J. Electr. Spectr. 110-111, 105 (2000). [47] C. Ruberto and B. I. Lundqvist; Phys. Rev. B 75, 235438 (2007). 13
1603.02881	1	1603	2016-03-09T13:34:16	Nonmagnetic band gap at the Dirac point of the magnetic topological insulator (Bi$_{1-x}$Mn$_x)_2$Se$_3$	[ "cond-mat.mes-hall" ]	Magnetic doping is expected to open a band gap at the Dirac point of topological insulators by breaking time-reversal symmetry and to enable novel topological phases. Epitaxial (Bi$_{1-x}$Mn$_{x}$)$_{2}$Se$_{3}$ is a prototypical magnetic topological insulator with a pronounced surface band gap of $\sim100$ meV. We show that this gap is neither due to ferromagnetic order in the bulk or at the surface nor to the local magnetic moment of the Mn, making the system unsuitable for realizing the novel phases. We further show that Mn doping does not affect the inverted bulk band gap and the system remains topologically nontrivial. We suggest that strong resonant scattering processes cause the gap at the Dirac point and support this by the observation of in-gap states using resonant photoemission. Our findings establish a novel mechanism for gap opening in topological surface states which challenges the currently known conditions for topological protection.	cond-mat.mes-hall	cond-mat	Nonmagnetic band gap at the Dirac point of the magnetic topological insulator (Bi1−xMnx)2Se3 J. S´anchez-Barriga,1 A. Varykhalov,1 G. Springholz,2 H. Steiner,2 R. Kirchschlager,2 G. Bauer,2 O. Caha,3 E. Schierle,1 E. Weschke,1 A. A. Unal,1 S. Valencia,1, M. Dunst,5 J. Braun,5 H. Ebert,5 J. Min´ar,5,6, E. Golias,1 L. V. Yashina,7 A. Ney,2 V. Hol´y,4 O. Rader1 1Helmholtz-Zentrum Berlin fur Materialien und Energie, Elektronenspeicherring BESSY II, Albert-Einstein-Strasse 15, 12489 Berlin, Germany 2Institut fur Halbleiter- und Festkorperphysik, Johannes Kepler Universitat, Altenbergerstr. 69, 4040 Linz, Austria 3CEITEC and Department of Condensed Matter Physics, Masaryk University, Kotlarska 2, 61137 Brno, Czech Republic 4Department of Condensed Matter Physics, Charles University, Ke Karlovu 5, 12116 Prague, Czech Republic 5Department Chemie, Ludwig-Maximilians-Universitat Munchen, Butenandtstr. 5-13, 81377 Munchen, Germany 6New Technologies Research Centre, University of West Bohemia, Univerzitni 8, 306 14 Pilsen, Czech Republic 7Department of Chemistry, Moscow State University, Leninskie Gory 1/3, 119991, Moscow, Russia 6 1 0 2 r a M 9 ] l l a h - s e m . t a m - d n o c [ 1 v 1 8 8 2 0 . 3 0 6 1 : v i X r a 1 Abstract Magnetic doping is expected to open a band gap at the Dirac point of topological insulators by breaking time-reversal symmetry and to enable novel topological phases. Epitaxial (Bi1−xMnx)2Se3 is a prototypical magnetic topological insulator with a pronounced surface band gap of ∼ 100 meV. We show that this gap is neither due to ferromagnetic order in the bulk or at the surface nor to the local magnetic moment of the Mn, making the system unsuitable for realizing the novel phases. We further show that Mn doping does not affect the inverted bulk band gap and the system remains topologically nontrivial. We suggest that strong resonant scattering processes cause the gap at the Dirac point and support this by the observation of in-gap states using resonant photoemission. Our findings establish a novel mechanism for gap opening in topological surface states which challenges the currently known conditions for topological protection. 2 The topological surface state (TSS) of the three-dimensional topological insulator (3D TI) Bi2Se3 forms a Dirac cone in the band dispersion E(k(cid:107)) of electron energy E versus wave vector component k(cid:107) parallel to the surface [1]. Differently from graphene, the electron spin is nondegenerate and locked to k(cid:107), and the Dirac crossing point is protected by time-reversal symmetry (TRS) against distortions of the system [1, 2]. This stability has been theoretically investigated demonstrating that nonmagnetic impurities at the surface and in the bulk can form a resonance in the surface-state density of states (DOS) without opening a band gap [3–5]. A magnetic field breaks TRS lifting the Kramers degeneracy E(k(cid:107),↑) = E(−k(cid:107),↓) between up (↑) and down (↓) spins, and can open a band gap at the Dirac point [1]. For the two-dimensional quantum-spin-Hall system HgTe, the effect of a perpendicular magnetic field on the topologically-protected edge state has been demonstrated successfully [6], and similar effects are expected from magnetic impurities in this system [7]. In subsequent studies on HgTe, the effect of the magnetic field was much smaller, and recently in an inverted electron- hole bilayer from InAs/GaSb, helical edge states proved robust in perpendicular magnetic fields of 8 T [8]. At the surface of a 3D TI, calculations show that magnetic impurities can open a gap at the Dirac point and exhibit ferromagnetic order with perpendicular anisotropy mediated by the TSS through Ruderman-Kittel-Kasuya-Yosida (RKKY) exchange coupling [4, 9, 10]. 3D TIs with magnetic impurities have been studied by angle-resolved photoelectron spec- troscopy (ARPES). The magnetic impurities have been employed in the bulk [11] and at the surface of Bi2Se3 [10]. For Fe impurities, the opening of large surface band gaps of ∼100 meV was reported in both cases [10, 11]. In a previous work, however, we found that Fe deposited on Bi2Se3 does not open a gap for a wide range of preparation conditions, reveal- ing a surprising robustness of the TSS towards magnetic moments [12]. This conclusion also extends to Gd/Bi2Se3 [13] and Fe/Bi2Te3 [14]. The absence of a surface band gap is consistent with the recent finding that Fe on Bi2Se3 does not favor a magnetic anisotropy perpendicular to the surface, at least in the dilute limit [15]. In this respect, bulk impurities in TIs have an advantage over surface impurities in that ferromagnetic order and perpendicular magnetic anisotropies have been achieved in the bulk systems [11, 16–20, 23]. Fe incorporated in bulk Bi2Te3 is known to order ferromagnetically with a Curie temperature (TC) of ∼12 K for concentrations of x = 0.04 showing an easy axis perpendicular to the base plane [16, 17]. On the other hand, (Bi1−xFex)2Se3 has not 3 been found to be ferromagnetic at temperatures above 2 K [17]. For x = 0.16 and 0.25, ferromagnetic order was found at 2 K [11]. By ARPES [24], surface-state band gaps in (Bi1−xFex)2Se3 were reported and assigned to a magnetic origin for Fe concentrations from x = 0.05 to 0.25 including nonferromagnetic concentrations such as x = 0.12 with a band gap of 45 meV [11]. These band gaps have been attributed to short-range magnetic order. Concerning Mn incorporation, bulk (Bi1−xMnx)2Se3 with x = 0.02 has been shown to exhibit a surface band gap with an occupied width of 7 meV [11], which was suggested as indication of ferromagnetic order induced by the TSS [11]. However, much larger surface band gaps were observed for n-doped (Bi1−xMnx)2Se3 films [20], where ferromagnetic order of Mn at the surface was found to be strongly enhanced with TC up to ∼45 K. The ferromagnetic order resulted in an unusual spin texture of the TSS. The TC at the surface was much higher than in the bulk [20], which was partially attributed to Mn surface accumulation. A strong enhancement of the surface TC was also predicted by mean-field theory for this system [e.g., from 73 K (bulk) to 103 K (surface)] [25]. Magnetically-doped TIs with ferromagnetic order are important as realizations of novel topological phases. When spin degeneracy is lifted by the exchange splitting, bulk band inversion can occur selectively for one spin. If also the Fermi level is in the band gap, as predicted for Cr and Fe in Bi2Se3 [26], this gives rise to an integer quantized Hall conductance σxy in thin films termed the quantized anomalous Hall effect [7, 26]. This has recently been reported for Cr in (Bi,Sb)2Te3 [27]. When a perpendicular magnetization breaks TRS at the surface of a bulk TI, the resulting mass and surface band gap give rise to quantized edge states. In this case σxy is half-integer quantized yielding a half quantum Hall effect [28] which can be probed at ferromagnetic domain boundaries at the surface [29] and leads to exotic topological magnetoelectric effects such as the point-charge-induced image magnetic monopole [30, 31]. Another topological phase predicted for magnetically doped Bi2Se3 is the realization of a 3D Weyl fermion system in which the Dirac-like dispersions become a property of the bulk [32]. Here we present a detailed investigation on the correlation between the TSS and magnetic properties in the Bi-Mn-Se system, where the magnetic impurity Mn is introduced in the bulk of ∼0.4 µm thick (Bi1−xMnx)2Se3 epilayers with concentrations up to x = 0.08. By ARPES we find remarkably large band gaps of ∼200 meV persisting up to temperatures of 300 K far above TC, which is found to be ∼10 K at the surface showing only a limited enhancement by ≤ 4 K over the bulk TC. We find that the magnitude of the band gaps 4 by far exceeds that expected from TRS breaking due to magnetic order. By ARPES from quantum-well states we exclude that Mn changes the inverted bulk band gap. Using reso- nant photoemission and ab-initio calculations we conclude that, instead, impurity-induced resonance states destroy the Dirac point of the TSS. We further support our conclusion by showing that extremely low bulk concentrations of nonmagnetic In do also open a surface band gap in Bi2Se3. Our findings have profound implications for the understanding of the conditions for topological protection. RESULTS Concentration and temperature dependence of surface band gap. Figure 1 shows Mn-concentration-dependent high-resolution ARPES dispersions of the TSS and bulk va- lence band (BVB) states of (Bi1−xMnx)2Se3 with x up to 0.08 measured at 12 K and 50 eV photon energy. At this energy, contributions from the bulk-conduction band (BCB) do not appear due to the dependence of the photoemission transitions on the component of the electron wave vector perpendicular to the surface k⊥ [24], but the BCB is partially occupied as data at 21 eV photon energy reveal (Supplementary Fig. 1). The clean Bi2Se3 film (Fig. 1a) is n-doped and exhibits a well-resolved Dirac point with high photoemission intensity at a binding energy of ED∼0.4 eV that is seen as an intense peak in the energy-distribution curve (EDC) at zero momentum superimposed as red curve on the right-hand side of the panel. The intact and bright Dirac point marked by a horizontal dashed line in Fig. 1a demonstrates that the TSS is gapless in Bi2Se3. At larger binding energies, the lower half of the Dirac cone overlaps with the BVB, which is observed for all (Bi1−xMnx)2Se3 samples (Figs. 1b-d). Increasing Mn concentration, we find a gradual upward shift of the band edges in energy, revealing a progressive p-type doping (hole doping). Most notably, a surface band gap opens at the Dirac point with increasing Mn content. In each EDC, the opening of the gap is also evident from the development of an intensity dip at the binding energy of the original Dirac point. Thus, the energy separation between the upper Dirac band minimum and the energy position of the intensity dip is about half of the gap size. Note that the horizontal dashed lines in Figs. 1b-d highlight the minimum limit of the gap size after tak- ing into account the contribution from the linewidth broadening to the ARPES spectra (see Supplementary Note 1 and Supplementary Fig. 2). The surface band gap rapidly increases with Mn content and exceeds 200 meV for x = 0.08. For x = 0.02, the suppression of 5 the Dirac point intensity as compared to the undoped Bi2Se3 film indicates the existence of a small gap. This is consistent with the fact that the TSS dispersion becomes slightly parabolic and is characterized by an increased effective mass of m∗ ≈ 0.09me for x = 0.02. This value increases further to ≈ 0.15me for x = 0.08, where me is the free-electron mass. The observed surface band gaps are similar to those previously reported [20], which have been attributed to long-range ferromagnetism and TRS breaking of the topologically- protected surface state [20]. Figures 2a-d show high-resolution ARPES dispersions of the TSS as well as the normal-emission spectra through the Dirac point at temperatures of 12 and 300 K for x = 0.08 (see also Supplementary Fig. 3). Strikingly, there is no significant change as temperature is raised, and very clearly the band gap in the TSS persists up to room temperature. Moreover, we find a similar behavior for the whole sample series independently of the Mn content, which challenges the dominant role of ferromagnetic order in inducing the band gap in the TSS, unless the surface TC is above room temperature. Bulk and surface magnetism. We begin in Fig. 3 with characterizing the bulk magnetic properties using superconducting quantum-interference device (SQUID) magnetometry. The comparison of measurements with in-plane and out-of-plane applied magnetic field in Fig. 3a shows that at a temperature of 2 K the bulk easy axis lies in the surface plane. This holds irrespectively of the Mn concentration. Since the opening of a gap at the Dirac point requires a magnetization perpendicular to the surface [9], we concentrate in the following on the out-of-plane component of the magnetization. In the hysteresis loops displayed in Fig. 3b measured with an out-of-plane applied magnetic field at different temperatures, we observe a narrow magnetic hysteresis at 4.2 K, whereas paramagnetic behavior emerges at 7 K once the ferromagnetic transition has been crossed (inset in Fig. 3b). For a better determination of TC we present in Fig. 3c modified Arrott plots according to a 3D Heisenberg ferromagnet, normalized to the mass of the sample, with exponents β = 0.348 and γ = 1.41, from which we deduce TC = 5.5 K in the bulk (see also Supplementary Fig. 4). This is well below the temperature of our ARPES measurements shown in Figs. 1 and 2. The only remaining possibility that long-range ferromagnetism opens the gap at the Dirac point is an enhanced surface magnetism with magnetization perpendicular to the surface [9, 20, 25]. This cannot be verified directly by the bulk-sensitive SQUID magnetometry. Therefore, we perform x-ray magnetic circular dichroism (XMCD) measurements at the Mn 6 L2,3-edges to probe the near-surface ferromagnetic order. The detection by total electron yield leads to a probing depth in the nanometer range. Figure 4a shows for x = 0.04 the normalized intensity of the Mn L2,3 absorption edges obtained upon reversal of the photon helicity at a temperature of 5 K with an out-of-plane applied magnetic field of 3 T. The XMCD difference spectrum is shown in Fig. 4b, with the normalized XMCD difference signal following one part of the out-of-plane hysteresis as a function of the applied magnetic field as inset. The temperature dependence of the XMCD signal measured in remanence (0 T applied magnetic field) is presented in Fig. 4c for Mn concentrations of x = 0.04 and 0.08. This signal is obtained after switching off an applied magnetic field of 5 T perpendicular to the surface, and represents the remanent XMCD, which is proportional to the remanent magnetization of the film surface. We clearly observe a ferromagnetic Mn component in the remanent XMCD appearing around 640 eV which becomes increasingly more pronounced at lower temperatures. The lineshape of the XMCD spectrum is similar to that of Mn in GaAs [21], indicating a predominant d5 configuration. More specifically, the lineshape compares rather well with Mn in Bi2Te3 for which the comparison to an Anderson impurity model recently gave a superposition of 16% d4, 58% d5 and 26% d6 character [22]. The ferromagnetic Mn component shows surface ferromagnetic order in the remanent XMCD only below 10 K for Mn concentrations of x = 0.04 and 0.08 (inset in Fig. 4c). On the one hand, this result is not inconsistent with the predicted strong surface enhancement of TC since the enhancement effect vanishes already when chemical potential and Dirac point differ by 0.2 eV [25]. On the other hand, our XMCD result means that the magnetically induced gap would have to close above 10 K. Having established that the bulk and surface magnetic properties of our samples are very similar, as a next step, we performed field-cooling experiments to re-examine the bulk magnetic properties. Figure 4d shows field-cooled SQUID data for an applied magnetic field of 10 mT parallel and perpendicular to the surface plane. Above TC, there is no preferential orientation of the anisotropy axis perpendicular to the surface. Additional zero-field-cooled SQUID data compare rather well with these results (Supplementary Fig. 5). Further XMCD measurements under an in-plane and out-of-plane applied magnetic field are consistent with the SQUID data and contained in Supplementary Fig. 6. These XMCD and SQUID results strongly suggest that above TC also short-range static magnetic order does not play a role at the surface or in the bulk, respectively (see Supplementary 7 Note 2), and that domains with partial or full out-of-plane magnetic order are absent. We further support this conclusion by additional XMCD measurements carried out by means of x-ray photoelectron emission microscopy (XMCD-PEEM) at room temperature, with a lateral resolution of ∼20 nm. Figure 4e shows the XMCD image which due to the light incidence (16◦ grazing) is sensitve to both in- and out-of-plane components of the magnetization, with additional data given in Supplementary Figs. 7,8. The XMCD image displays no magnetic domains at room temperature thus also excluding short-range static inhomogeneous magnetic order with magnetization partially or fully in or out of the surface plane, at least within our lateral resolution (see Supplementary Note 3 for details). A similar conclusion can be drawn regarding out-of-plane magnetic short-range fluctuations (see discussion in Supplementary Note 4 and Supplementary Fig. 9). All our observations taken together lead us to the important conclusion that the surface band gap is not due to magnetism and, thus, the novel topological phases cannot be realized with (Bi1−xMnx)2Se3. Absence of topological phase transition with concentration. There exists an alter- native nonmagnetic explanation for the surface band gap. It is in principle possible that the incorporation of Mn changes the bulk band structure and even reverts the bulk band inversion, rendering Mn-doped Bi2Se3 topologically trivial. This topological phase transition with concentration is, e.g., the basis for the HgTe quantum spin Hall insulator [6]. Inter- estingly, it has been argued that a gap in the TSS can be a precursor of a reversed bulk band inversion, as discussed for TlBi(S1−xSex)2 [33]. Indium substitution in Bi2Se3 behaves similarly, and leads to a topological-to-trivial quantum phase transition with an inversion point between 3% and 7% In in thin films [34]. Figure 5 shows that the bulk band gap stays constant within 4% of its value for 8% Mn incorporation. This statement is possible because at low photon energies changes in the bulk band gap are traced most precisely from quantum-well states in normal emission (k(cid:107)=0). The simultaneous quantization of BCB and BVB is created by surface band bending after adsorption of small amounts of residual gas [35]. This means that the scenario of reversed bulk band inversion does not hold here. Another remarkable feature is the measured size of the surface band gap. A perpendicu- lar magnetic anisotropy has recently been predicted by density-functional theory (DFT) for Co in Bi2Se3 [36] as well as for 0.25 monolayer Co adsorbed on the surface in substitutional sites, leading to a surface band gap of ∼ 9 meV [36] which, however, is one order of 8 magnitude smaller than the gaps observed here. For (Bi1−xMnx)2Te3, a surface band gap of 16 meV has been calculated under the condition of perpendicular magnetic anisotropy [37]. For Mn in Bi2Se3 at the energetically favored Bi-substitutional sites, an in-plane magnetic anisotropy is predicted [38]. The size of the gap at the Dirac point is only 4 meV [38]. Perpendicular anisotropy and ferromagnetic order are absent in our samples at the temperature of the ARPES measurements. Role of the local magnetic moment and intercalation scenarios. If the large local magnetic moment of the Mn is responsible for the measured band gaps via TRS breaking, the effect should be equal or larger when Mn is deposited directly on the surface. Here we argue in the following way: If TRS is broken only due to the Mn magnetic moment, we do not require hybridization to open a gap. By depositing Mn we create a completely different environment for the Mn at the surface, allowing us to understand whether the Mn magnetic moment is solely responsible for the opening of large surface band gaps via TRS breaking and, ideally, without hybridization playing a role. Mn is very well suited for such comparison, as due to the high stability of its half-filled d5 configuration, its high magnetic moment is to a large extent independent of the atomic environment and the resulting hybridization. DFT obtains 4.0 µB for most substitutional sites (except for the hypothetical Se-substitutional site with still 3.0 µB) and only for interstitial Mn the moment peaks with 5.0 µB [38]. Mn deposition has been performed at ∼30 K in order to keep the Mn atoms isolated from each other and on the surface. As the XMCD showed a predominant d5 configuration for Mn bulk impurities, which is most stable, we can assume the same magnetic moment for Mn impurities deposited on the surface. Figure 6 shows that a similar p-doping occurs as with the Mn in the bulk. In another respect, the magnetic moments do not act in a similar way as in the bulk. Even 30% of a Mn monolayer on Bi2Se3 does not open a band gap at the Dirac point. This result is similar to what we have observed for Fe on Bi2Se3 [12] and Bi2Te3 [14]. Most recently, the appearance of surface band gaps in ARPES at the Dirac point of the TSS has been discussed based on different mechanisms. One is momentum fluctuations of the surface Dirac fermions in real space as observed by scanning tunneling microscopy measurements of Mn-doped Bi2Te3 and Bi2Se3 bulk single crystals [39]. However, these fluc- tuations amount to 16 meV [39], much less than the present band gaps. Another mechanism 9 is top-layer relaxation, i.e., breaking of the van-der-Waals bond between quintuple layers during cleavage of bulk crystals as proposed in Ref. [20]. This would open a gap due to hybridization of TSS's through the layer [40] but such an effect does not occur for epitaxial layers where no sample cleavage is employed for surface preparation. If such a separation of quintuple layers occurs, it is more likely caused by the Mn. Principally, intercalated Mn in the van-der-Waals gaps could separate quintuple layers electronically although such effect has not been seen in experiments yet. At these new interfaces TSS's could form and if they would behave like in ultrathin Bi2Se3 films, they would hybridize and open a band gap [40]. However, band gaps of the order of ∼100 meV would correspond to an unrealistic Mn intercalation pattern grouping three quintuple layers when compared to the films of Ref. [40]. In order to estimate the amount of Mn intercalation, we have analyzed the change of the bulk lattice constant in our samples by x-ray diffraction (see Supplementary Note 5, Supplementary Figs. 10-13 and Supplementary Tables 1,2). We find that the c-lattice constant increases by ∼ 0.15% for 4% Mn and by ∼ 0.45% for 8% Mn. If we would assume that this is completely due to an expansion of the van-der-Waals gaps, the interlayer distance would increase by only ∼ 1.8% for 4% Mn and by less than ∼ 5% for 8% Mn. According to DFT, intercalation of transition metals into the first van-der-Waals gap of Bi2Se3 leads to an expansion between 10 and 20% [41]. New surface-state features can appear inside the bulk band gap but in order to be split remarkably off the BVB and BCB edges, expansions between 20 and 50% are required [41]. In addition, we point out that, despite our observation of a large surface band gap at room temperature, we observe a nearly-zero spin polarization perpendicular to the surface (Supplementary Fig. 14). The absence of a hedgehog spin texture pinpoints a non-TRS breaking mechanism underlying the origin of the gap, a fact which is consistent with our conclusions derived from the magnetic properties. Moreover, it is understood that we cannot verify a small magnetic band gap opening of few meV as cal- culated in Refs. 36, 37 because much larger gaps are present in the whole temperature range. Topological protection beyond the continuum model. As we find that neither ferro- magnetic order in the bulk or at the surface, nor the local magnetic moment of the Mn are causing the large band gaps that we observe, as we can exclude the nonmagnetic explanation of a reversal of the bulk band inversion and as we do not find sufficient indications for the 10 nonmagnetic explanation of surface-state hybridization by van-der-Waals-gap expansion, we point out a different mechanism based on impurity-induced resonance states that locally destroy the Dirac point. In fact, recently the treatment of the topological protection in the continuum model [3–5] has been extended for finite bulk band gaps [42, 43]. As a result, surface [42] and bulk impurities [43] can mediate scattering processes via bulk states and the localized impurity- induced resonance states emerging at and around the Dirac point lead to a local destruction of the topological protection of the Dirac point. The resulting band gap opening depends on the resonance energy, impurity strength U as well as spatial location of the impurities [42, 43]. A typical gap size is of the order of 100 meV [42]. Moreover, the gap does not rely on a particular magnetic property of the impurity and, therefore, the mechanism should apply for magnetic impurities as well. The model can also explain the absence of the gap when Mn impurities are placed on the surface, since bulk-like resonance states do not form when Mn is placed only on the surface. A definite assignment, however, requires realistic electronic structure calculations since a simple atomic Mn d5 configuration does not lead to states at the Dirac point. To identify impurity resonances, we calculated for 8% Mn in Bi2Se3 at the Bi- substitutional sites the corresponding DOS by ab-initio theory. The model structure used in the calculations is shown in Fig. 7a, where Bi atoms (yellow) acquire Mn character (blue). The results of the calculation are shown in Fig. 7b, where it is seen that Mn impurity states (blue) strongly contribute to the total DOS (red) near the BVB maximum. The Mn impu- rity states in the gap are clearly identified by assuming ferromagnetic order and subtracting the minority from majority spin DOS. In experiment, impurity-band states are difficult to observe as the example of Ga1−xMnxAs teaches [44], where similarly to the present case impurity states emerge near the BVB maximum. There are, in fact, indications for the impurity resonances in our data. If we look at the photon-energy dependence from 16 to 21 eV in the ARPES results shown in Supplementary Fig. 15, we see that the surface band gap is only well defined with respect to the minimum of the upper Dirac cone and consistent for 16–21 eV as well as 50 eV (for 50 eV see Figs. 1,2 where the BCB does not appear). The lower Dirac cone exhibits despite its two-dimensional nature a photon-energy (wave vector perpendicular to the surface, k⊥) dependence when Mn is incorporated. The reason for this difference is that the lower Dirac cone strongly overlaps with the BVB while the upper Dirac 11 cone does not overlap with the BCB. This can also be seen in Fig. 5, where an apparent band gap of ∼ 200 meV occurs for 2% Mn at hν = 16 eV. At this small concentration, such an apparent gap seems to be nearly closed at 18 eV, where the suppression of the Dirac point intensity and the more parabolic TSS dispersion as compared to the undoped Bi2Se3 film indicates the existence of a small gap, in agreement with the results of Fig. 1. Note that the minimum band gap in the photon-energy dependence defines the gap size (see Supplementary Fig. 16). The minimum gap size observed at low photon energies agrees well with the one obtained at 50 eV. At 8% Mn, the surface band gap opens at all photon energies and reaches a minimum of ∼ 200 meV, i.e., it is determined rather unambiguously, but some intensity appears in the surface band gap. Such intensity also appears in Ref. 20. The role of the impurity resonance is exactly to couple the TSS to bulk states. This 3D coupling is naturally different for the upper and lower Dirac cone due to the different bulk overlap and thus seen as the photon-energy dependence of the lower half of the Dirac cone. We applied resonant photoemission which fortunately is comparatively strong for Mn due to its half-filled d-shell. Figures 7c,d show for 8% Mn off-resonant (hν = 48 eV) and on-resonant photoemission (50 eV) measurements via the Mn 3p core level, focusing on the region of the surface gap, respectively. The spectra were normalized to the photon flux after taking into account the photon energy dependence of the photoionization cross sections of Bi 6p and Se 4p, respectively. Subtracting off-resonant from on-resonant data allows us to visualize directly the contribution from impurity resonances in the ARPES spectrum, as there is enhanced photoemission from d-like Mn states through decay of electrons that are excited via 3p-3d transitions (Fig. 7e). The difference spectrum shown in Fig. 7f reveals the existence of impurity resonances inside the surface band gap and near the valence- band maximum. The resonances are seen in blue contrast and are marked with arrows, similar to the calculations shown in Fig. 7b. The unexpectedly strong dispersion of the resonant states with wave vector k(cid:107) is in qualitative agreement with our one-step model photoemission calculations (see Supplementary Fig. 17 and Supplementary Note 6), and additionally supports the physical picture of Mn-induced coupling to the bulk states. We should emphasize that the Mn d5 configuration confirmed by our XMCD measurements offers much fewer states in the energy range of the Dirac cone than the other magnetic transition metals. The fact that already Mn breaks the Dirac cone indicates that the present result is of general importance for TIs doped with magnetic transition metals. 12 As nonmagnetic control experiment we have chosen thick In-doped Bi2Se3 bulk samples which give rise to a trivial phase close to ∼ 5% In concentration. Supplementary Fig. 18 shows that a large band gap of the order of ∼ 100 meV appears at the Dirac point already for a much smaller In concentration, namely 2%, far away from the inversion point of the topological quantum-phase transition (see Supplementary Note 7 and Supplementary Fig. 19 for more details). In addition, spin-resolved ARPES measurements around k(cid:107) = 0 for In-doped samples (Supplementary Fig. 14) reveal that the out-of-plane spin polarization is zero, similarly to the result for the gapped Dirac cone in Mn-doped samples. Interestingly, the size of the band gap for 2% In is of the same order as the one for 8% Mn whereas no gap appears for 4% Sn (Supplementary Fig. 18). This indicates that the concentration range at which the large surface gap develops varies from dopant to dopant. Based on the ideas proposed recently [42, 43], this might be associated with the impurity-dependent strength U , regardless whether the dopant is magnetic or not. Our control experiments demonstrate the existence of a novel mechanism for surface band gap opening which is not directly connected to long-range or local magnetic properties. Although it is not possible to directly search for In impurity resonances in the photoemission experiment as there is no resonant photoemission condition available, we point out for completeness that our conclusion on the nonmagnetic origin of the surface band gap in Mn-doped Bi2Se3 films would not have been possible unless several findings in our experiment differed from previous experiments [20]. This refers to the in-plane magnetic anisotropy, the absence of a temperature dependence of the gap, a much lower TC, no out-of-plane spin polarization at the gapped Dirac point, no enhanced surface magnetism, a photon-energy dependence of the band gap, and an available resonant condition via the Mn 3p core level allowing us to directly observe the contribution from in-gap states. To summarize, we revealed the opening of large surface band gaps in the TSS of Mn-doped Bi2Se3 epilayers that strongly increase with increasing Mn content. We find ferromagnetic hysteresis at low temperatures, with the magnetization oriented parallel to the film surface. Magnetic surface enhancement is absent. Even above the ferromagnetic transition, we observe surface band gaps which are one order of magnitude larger than the magnetic gaps theoretically predicted and, moreover, they do not show notable temperature dependence. No indication for a Mn-induced reversal of the bulk band inversion and no significant enhancement of the surface magnetic ordering transition with respect to the 13 bulk are found. Control experiments with nonmagnetic bulk impurities, conducted in the topological phase, reveal that surface band gaps of the order of ∼100 meV can be created without magnetic moments. In line with recent theoretical predictions, we conclude that the band gap opening up to room temperature in Mn-doped films is not induced by ferromagnetic order and that even the presence of magnetic moments is not required. Our results are important in the context of topological protection and provide strong circumstantial evidence that Mn-doped Bi2Se3 is not suited for observing the quantized anomalous Hall effect or the half quantum Hall effect. METHODS Sample growth and structural characterization. The samples were grown by molecular beam epitaxy on (111)-oriented BaF2 substrates at a substrate temperature of 360 ◦C using Bi2Se3, Mn, and Se effusion cells. The Mn concentration was varied between 0 and 8%, and a two-dimensional growth was observed by in situ reflection high-energy electron diffraction in all cases (see Supplementary Fig. 20 and Supplementary Note 8). All samples were single phase as indicated by x-ray diffraction. After growth and cooling to room temperature, the ∼0.4 µm thick epilayers were in situ capped by an amorphous 50 nm thick Se layer, which was desorbed inside the ARPES and XMCD chambers by carefully annealing at ∼ 230 ◦C for ∼1 h. The Mn concentrations determined by core-level photoemission were in good agreement with those obtained from energy dispersive microanalysis (EDX), indicating no significant Mn accumulation at the surface of the samples (see Supplementary Fig. 21 and Supplementary Note 9). High-resolution and spin-resolved ARPES. Temperature-dependent ARPES mea- surements were performed at the UE112-PGM2a beamline of BESSY II at pressures better than 1 · 10−10 mbar using p-polarized undulator radiation. Photoelectrons were detected with a Scienta R8000 electron energy analyzer and the spin-resolved spectra were obtained with a Mott-type spin polarimeter coupled to the hemispherical analyzer. For the spin-resolved measurements of Mn-doped Bi2Se3samples, a magnetic field of +0.3 T was applied perpendicular to the surface plane at 20 K right before the acquisition of the spectra. Overall resolutions of ARPES measurements were 5 meV (energy) and 0.3 ◦ (angular). Resolutions for spin-resolved ARPES were 80 meV (energy) and 0.75 ◦ (angular). 14 Magnetic characterization. The characterization of the bulk magnetic properties was done by SQUID magnetometry. The bulk magnetization was recorded as a function of tem- perature and applied magnetic field applied either perpendicular (out-of-plane) or parallel (in-plane) to the surface of the films. The diamagnetic contribution of the BaF2 substrate was derived from the field-dependent magnetization curves at room temperature and subtracted from all data. The characterization of the surface magnetic properties was done by XMCD and XMCD-PEEM measurements at the UE46-PGM1 and UE49-PGM1a beam- lines of BESSY II, respectively. The experiments were performed using circularly-polarized undulator radiation and under the same pressure conditions as the ARPES measurements. The XMCD absorption spectra were taken using a high-field diffractometer as a function of temperature and applied magnetic field, and the XMCD-PEEM measurements were performed at room temperature and under grazing incidence (16◦) with a lateral resolution of ∼20 nm. Theoretical calculations. The one-step model photoemission calculations were performed using the results of ab-initio theory as an input, and taking into account wave vector and energy-dependent transition matrix elements. The ab-initio calculations were performed within the coherent-potential approximation using the KKR program package [45], with spin-orbit coupling included by solving the four-component Dirac equation. [1] Hasan, M. Z. & Kane, C. L. Topological insulators. Rev. Mod. Phys. 82, 3045-3067 (2010). [2] Qi, X.-L. & Zhang, S.-C. Topological insulators and superconductors. Rev. Mod. Phys. 83, 1057-1110 (2011). [3] Lee, W. C., Wu, C., Arovas, D. P. & Zhang, S. C. Quasiparticle interference on the surface of the topological insulator Bi2Te3. Phys. Rev. B 80, 245439 (2009). [4] Biswas, R. R., & Balatsky, A. V. Impurity-induced states on the surface of three-dimensional topological insulators. Phys. Rev. B 81, 233405 (2010). [5] Lu, J., Shan, W.-Y., Lu, H.-Z. & Shen, S.-Q. Non-magnetic impurities and in-gap bound 15 states in topological insulators. New J. Phys. 13, 103016 (2011). [6] Konig, M. et al. Quantum spin Hall insulator state in HgTe quantum wells. Science 318, 766-770 (2007). [7] Liu, C.-X., Qi, X.-L., Dai, X., Fang, Z. & Zhang, S.-C. Quantum anomalous Hall effect in Hg1−yMnyTe quantum wells. Phys. Rev. Lett. 101, 146802 (2008). [8] Du, L., Knez, I., Sullivan, G. & Du, R.-R. Robust helical edge transport in gated InAs/GaSb bilayers. Phys. Rev. Lett. 114, 096802 (2015). [9] Liu, Q., Liu, C. X., Xu, C. K., Qi, X. L. & Zhang, S. C. Magnetic impurities on the surface of a topological insulator. Phys. Rev. Lett. 102, 156603 (2009). [10] Wray, L. A. et al. A topological insulator surface under strong Coulomb, magnetic and disorder perturbations. Nature Phys. 7, 32-37 (2011). [11] Chen, Y. L. et al. Massive Dirac fermion on the surface of a magnetically doped topological insulator. Science 329, 659-662 (2010). [12] Scholz, M. R. et al. Tolerance of topological surface states towards magnetic moments: Fe on Bi2Se3. Phys. Rev. Lett. 108, 256810 (2012). [13] Valla, T., Pan, Z.-H., Gardner, D., Lee, Y. S. & Chu, S. Photoemission Spectroscopy of magnetic and nonmagnetic impurities on the surface of the Bi2Se3 topological insulator. Phys. Rev. Lett. 108, 117601 (2012). [14] Scholz, M. R. et al. Intact Dirac cone of Bi2Te3 covered with a monolayer Fe. Phys. Status Solidi RRL 7, 139-141 (2013). [15] Honolka, J. et al. In-plane magnetic anisotropy of Fe atoms on Bi2Se3(111). Phys. Rev. Lett. 108, 256811 (2012). [16] Kulbachinskii, V. A. et al. Ferromagnetic transition in the new diluted magnetic semiconductor p-Bi2−xFexTe3. Phys. Lett. A 285, 173-176 (2001). [17] Kulbachinskii, V. A. et al. Ferromagnetism in new diluted magnetic semiconductor Bi2−xFexTe3. Physica B 311, 292-297 (2002). [18] Hor, Y. S. et al.. Development of ferromagnetism in the doped topological insulator Bi2−xMnxTe3. Phys. Rev. B 81, 195203 (2010). [19] Dyck, J. S., H´ajek, P., Lost´ak, P. & Uher, C. Diluted magnetic semiconductors based on Sb2−xVxTe3 (0.01 < x < 0.03). Phys. Rev. B 65, 115212 (2002). [20] Xu, S.-Y. et al. Hedgehog spin texture and Berry's phase tuning in a magnetic topological 16 insulator. Nature Phys. 8, 616-622 (2012). [21] Rader, O. et al. Identification of extrinsic Mn contributions in Ga1−xMnxAs by field-dependent magnetic circular x-ray dichroism. J. Electron Spectr. Relat. Phenom. 144, 789-792 (2005). [22] Watson, M. D. et al. Study of the structural, electric and magnetic properties of Mn-doped Bi2Te3 single crystals. New J. Phys. 15, 103016 (2013). [23] Checkelsky, J. G., Ye, J., Onose, Y., Iwasa, Y. & Tokura, Y. Dirac-fermion-mediated ferro- magnetism in a topological insulator. Nature Phys. 8, 729-733 (2012). [24] Hufner, S. Photoelectron Spectroscopy: Principles and Applications. Springer, Berlin (1995). [25] Rosenberg, G. & Franz, M. Surface magnetic ordering in topological insulators with bulk magnetic dopants. Phys. Rev. B 85, 195119 (2012). [26] Yu, R. et al. Quantized anomalous Hall effect in magnetic topological insulators. Science 329, 61-64 (2010). [27] Chang, C.-Z. et al. Experimental observation of the quantum anomalous Hall effect in a magnetic topological insulator. Science 340, 167-170 (2013). [28] Qi, X. L., Hughes, T. L. & Zhang, S. C. Topological field theory of time-reversal invariant insulators. Phys. Rev. B 78, 195424 (2008). [29] Pankratov, O. A. Supersymmetric inhomogeneous semiconductor structures and the nature of a parity anomaly in (2 + 1) electrodynamics. Phys. Lett. A 121, 360-366 (1987). [30] Qi, X.-L., Li, R., Zang, J. & Zhang, S.-C. Inducing a magnetic monopole with topological surface states. Science 323, 1184-1187 (2009). [31] Zang, J. & Nagaosa, N. Monopole current and unconventional Hall response on a topological insulator. Phys. Rev. B 81, 245125 (2010). [32] Cho, G. Y. Possible topological phases of bulk magnetically doped Bi2Se3: turning a topo- logical band insulator into the Weyl semimetal. Preprint at http://arxiv.org/abs/1110.1939 (2011). [33] Sato, T. et al. Unexpected mass acquisition of Dirac fermions at the quantum phase transition of a topological insulator. Nature Phys. 7, 840-844 (2011). [34] Brahlek, M. et al. Topological-metal to band-insulator transition in (Bi1−xInx)2Se3 thin films. Phys. Rev. Lett. 109, 186403 (2012). [35] Bianchi, M., Hatch, R. C., Mi, J., Iversen, B. B. & Hofmann, P. Simultaneous quantization of bulk conduction and valence states through adsorption of nonmagnetic impurities on Bi2Se3. 17 Phys. Rev. Lett. 107, 086802 (2011). [36] Schmidt, T. M., Miwa, R. H. & Fazzio, A. Spin texture and magnetic anisotropy of Co impurities in Bi2Se3 topological insulators. Phys. Rev. B 84, 245418 (2011). [37] Henk, J. et al. Topological character and magnetism of the Dirac state in Mn-doped Bi2Te3. Phys. Rev. Lett. 109, 076801 (2012). [38] Abdalla, L. B., Seixas, L., Schmidt, T. M., Miwa, R. H. & Fazzio, A. Topological insulator Bi2Se3(111) surface doped with transition metals: An ab-initio investigation. Phys. Rev. B 88, 045312 (2013). [39] Beidenkopf, H. et al. Spatial fluctuations of helical Dirac fermions on the surface of topological insulators. Nature Phys. 7, 939-943 (2011). [40] Zhang, Y. et al. Crossover of the three-dimensional topological insulator Bi2Se3 to the two- dimensional limit. Nature Phys. 6, 584-588 (2010). [41] Eremeev, S. V., Vergniory, M. G., Menshchikova, T. V., Shaposhnikov, A. A. & Chulkov, E. V. The effect of van der Waal's gap expansions on the surface electronic structure of layered topological insulators. New J. Phys. 14, 113030 (2012). [42] Black-Schaffer, A. M. & Balatsky, A. V. Strong potential impurities on the surface of a topo- logical insulator. Phys. Rev. B 85, 121103(R) (2012). [43] Black-Schaffer, A. M. & Balatsky, A. V. Subsurface impurities and vacancies in a three- dimensional topological insulator. Phys. Rev. B 86, 115433 (2012). [44] Gray, A. X. et al. Bulk electronic structure of the dilute magnetic semiconductor Ga1−xMnxAs through hard x-ray angle-resolved photoemission. Nature Mat. 11, 957-962 (2012). [45] Ebert, H., Kodderitzsch, D. & Min´ar, J. Calculating condensed matter properties using the KKR - Green's function method - recent developments and applications. Rep. Prog. Phys. 74, 096501 (2011). Acknowledgements. Financial support from SPP 1666 of the Deutsche Forschungs- gemeinschaft and the Impuls-und Vernetzungsfonds der Helmholtz-Gemeinschaft (Grant No. HRJRG-408) is gratefully acknowledged. J. M. is supported by the CENTEM (CZ.1.05/2.1.00/03.0088) and CENTEM PLUS (LO1402) projects, co-funded by the ERDF as part of the Ministry of Education, Youth and Sports OP RDI programme. V. H. acknowledges the support of the Czech Science Foundation (project 14-08124S). 18 Author contributions. J. S.-B and A. V performed photoemission experiments; G. S., G. B. and L. V. Y. provided samples and performed growth and characterization; H. S., R. K. and A. N. performed SQUID measurements; O. C. and V. H. performed x-ray diffraction measurements; J. S.-B, E. S., E. W. performed XMCD measurements; J. S.-B, A. A. U. and S. V. performed XMCD-PEEM measurements; J. M., M. D., J. B and H. E. carried out calculations; J. S.-B and E. G. performed numerical simulations; J. S.-B performed data analysis and figure planning; J. S.-B. and O. R. performed draft planning and wrote the manuscript with input from all authors; J. S.-B. and O. R. were responsible for the conception and the overall direction. Competing financial interests. The authors declare no competing financial interests. Correspondence and requests for materials should be addressed to J. S.-B. (Email: [email protected]) and O. R. (Email: [email protected]). 19 FIGURES AND CAPTIONS FIG. 1: Effect of Mn doping on the electronic band dispersions of (Bi1−xMnx)2Se3. (a-d) Mn-concentration dependent high-resolution ARPES dispersions measured at 50 eV photon energy and a temperature of 12 K for x values of (a) 0, (b) 0.02, (c) 0.04 and (d) 0.08. The red lines represent energy-distribution curves (EDCs) obtained in normal emission (k(cid:107)=0). A surface band gap opens at the Dirac point of the TSS which increases in size with increasing Mn content. The horizontal white-dashed lines highlight the minimum limit of the gap size after taking into account the contribution from the linewidth broadening. The opening of the gap leads to an intensity dip in the EDCs around the energy region of the original Dirac point. For x = 0.02, the suppression of intensity at the Dirac point and the more parabolic surface-state dispersion are signatures of a small gap. Besides the increased linewidth broadening due to the Mn impurities, these effects become more pronounced with increasing Mn content. 20 0.80.60.40.20.00.20.10.0-0.1-0.2Electron Binding Energy (eV)WaveVector(Å)k-1WaveVector(Å)k-1WaveVector(Å)k-1WaveVector(Å)k-10.20.10.0-0.1-0.20.20.10.0-0.1-0.20.20.10.0-0.1-0.2abcdFIG. 2: Temperature dependence of the large energy scale surface band gap. (a, b) Energy-momentum ARPES dispersions obtained for Mn-doped Bi2Se3 epilayers with 8% Mn concentration. The red lines are the energy-distribution curves obtained in normal emission (k(cid:107)=0). The horizontal white-dashed lines highlight the surface gap. Measurements are taken at 50 eV photon energy and a temperature of (a) 12 K and (b) 300 K. The surface band gap does not show a remarkable temperature dependence. 21 0.60.40.20.0aElectron Binding Energy (eV)0.10.0-0.1WaveVector(Å)k-1b0.10.0-0.1WaveVector(Å)k-10.60.40.20.0FIG. 3: Characterization of the bulk magnetic properties of Mn-doped Bi2Se3 films. (a) In-plane (ip) and out-of-plane (oop) magnetization curves for Mn concentrations of x = 0.04 and x = 0.08, measured at a temperature of 2 K using SQUID magnetometry. The magnetic anisotropy lies in the surface plane irrespective of the Mn concentration. (b) Corresponding hysteresis loops for x = 0.08 measured at a temperature of 4.25 K (squares) and 7 K (triangles), respectively. The applied magnetic field is perpendicular to the (111) sample surface (see sketch). Inset: A zoom-in around zero magnetic field showing hysteresis at 4.25 K and a paramagnetic state at 7 K. (c) Arrott plots at various temperatures from which a Curie temperature TC of 5.5 K is deduced. Between 4.25 and 6.75 K, data are shown for temperature increments of δT = 0.25 K. 22 (emug)-1(Oe emug)-1-80-60-40-20020406080-20-1001020M(kA m-1)?0H (mT)4% Mn8% MnaT=2 Kipoopipoop800600400200c8% MnB[111]b8% Mn-1000-50005001000M (emu mol)-1H (kOe)-10-50510024861012141601000T=4.25 KT=7.0 KT=4.25 KT=6.75 KT=9 KT=11 KT=13 KT =5.5 KcdT=0.25 KB[111]FIG. 4: XMCD in total electron yield at the Mn L2,3-edges. (a) X-ray aborption spectra measured for opposite helicities of the circularly-polarized incident light at 5 K with an out-of- plane applied magnetic field of 3 T. A sketch of the experimental geometry is also shown. (b) Corresponding XMCD difference spectrum. The inset shows XMCD measurements for different applied magnetic fields. The error bars are estimated from the average noise level of the XMCD curves. (c) Temperature dependence of the remanent XMCD for a Mn concentration of x = 0.04. Inset: Detailed temperature dependence for x = 0.04 and 0.08. (d) To re-examine the bulk magnetic properties, field-cooling measurements are obtained under an in-plane (ip) and out-of-plane (oop) applied magnetic field of 10 mT using SQUID magnetometry. The temperature dependence for x = 0.04 and 0.08 compares well to the surface-sensitive measurements shown in the inset of (c). (e) XMCD-PEEM image obtained using photoelectron microscopy for x = 0.08 at room temperature, revealing the absence of magnetic domains with partial or full out-of-plane magnetic order. The scale bar (horizontal white-solid line) is 500 µm. 23 Intensity (arb.units)nBhna4% MnXMCD (%)-200660655650645640635201510503210XMCD (%)B (T)-10Photon Energy (eV)b4% MnT = 5 KB = 3 TXMCD (%)Circ. Pos.Circ. Neg.660655650645640635Photon Energy (eV)c6436426416406395 K7 K10 KPhoton Energy (eV)-40-2-64% Mn4% Mn8% Mn-4-22520151050T (K)4% Mn8% Mn0B= 0 TXMCD (%)-5Magnetization (10 emu)ip8% Mnoopipoop4% Mn246810121402468101214Temperature (K)d16e-4-2024XMCD (%)FIG. 5: Tracing changes in the bulk band gap of Mn-doped Bi2Se3 films. Changes in the bulk band gap are traced most precisely from quantum-well states (QWS) in normal emission (red curves on the right-hand side of each panel).(a-c) Energy-momentum dispersions acquired at a photon energy of hν = 18 eV for (a) Bi2Se3, (b) 2% and (c) 8% Mn doping. The simultaneous quantization of bulk-conduction band (BCB) and valence band (BVB) is created by surface band bending after adsorption of small amounts of residual gas [35]. It is found that Mn doping does not change the bulk band gap of Bi2Se3. (d-f) Similar results as in (a-c), respectively, but at a photon energy of hν = 16 eV, where large apparent surface band gaps are observed. This unexpected photon-energy dependence of a surface state is interpreted as coupling to bulk states mediated by the Mn impurities. 24 0.80.60.40.20.0BCBQWSBVBQWS500meV500meV500meVh= 18 eVnElectron Binding Energy (eV)BCBQWSBVBQWS500meV500meV500meVh= 16 eVnWaveVector()kÅ-1WaveVector()kÅ-1WaveVector()kÅ-10.10.0-0.1abcdef0.80.60.40.20.00.10.0-0.10.10.0-0.1FIG. 6: Effect of Mn deposition on the surface of Bi2Se3 films. (a) Energy-momentum ARPES dispersion showing a gapless Dirac cone in pure Bi2Se3 films after deposition of 0.3 mono- layer Mn at 30 K on the surface. Measurements are performed at the same temperature using 50 eV photon energy. (b) Energy-distribution curves (EDCs) extracted from the measurements shown in (a). The red curves in (a) and (b) emphasize the EDC in normal emission (k(cid:107)=0). 25 a0.40.20.00.50.40.30.20.10.0Intensity (arb. units)bElectron Binding Energy (eV)WaveVector()kÅ-10.10.0-0.1FIG. 7: Resonant photoemission and ab-initio calculations. (a) The structure of Bi2Se3 doped with 8% Mn at the Bi-substitutional sites used in the calculations. Within the coherent- potential approximation, the Bi atoms (yellow) acquire Mn character (blue). Se atoms are shown in gray color. (b) Calculated density of states (DOS). The total DOS (red) contains contributions from impurity resonances (vertical blue arrows) of strong d-character, a seen in the partial Mn DOS (blue). For the purpose of the calculation, ferromagnetic order at T = 0 K was assumed. (c, d) Resonant photoemission at the Mn M -edge, focusing on the region of the surface gap (8% Mn). (c) Off-resonant (48 eV photon energy) and (d) on-resonant data (50 eV). (e) Schematics of the resonant process. At the resonance energy, the excitation occurs via 3p-3d transitions. The exciting photon energy (hν) corresponds to a transition from an occupied core level to a valence band empty state between the Fermi (EF ) and vacuum (Evac) levels. The relaxation of excited electrons leads to enhanced photoemission from d-like Mn states. (f) Intensity difference obtained after subtracting off-resonant from on-resonant data. The resonances are seen in blue contrast and are marked with arrows, similar to the calculations shown in (b). 26 abTotal DOSMn DOS0.0-0.2-0.4-0.60.20.4BVBBCBE-E(eV)Normalized DOS (arb.units)F0.40.30.20.10.50.400.350.300.250.200.150.0-0.10.00.1Intensity (arb.units)LowHighIntensity (arb.units)LowHighcdfeMn 3pMn 3dEvacEFe-hnInitial stateFinal stateElectron Binding Energy (eV)WaveVector(Å)k-1WaveVector(Å)k-10.400.350.300.250.200.15Electron Binding Energy (eV)-0.10.00.1Electron Binding Energy (eV)-0.10.00.1WaveVector(Å)k-1
1112.2526	1	1112	2011-12-12T12:28:47	Impact of electron heating on the equilibration between quantum Hall edge channels	[ "cond-mat.mes-hall" ]	When two separately contacted quantum Hall (QH) edge channels are brought into interaction, they can equilibrate their imbalance via scattering processes. In the present work we use a tunable QH circuit to implement a junction between co-propagating edge channels whose length can be controlled with continuity. Such a variable device allows us to investigate how current-voltage characteristics evolve when the junction length d is changed. Recent experiments with fixed geometry reported a significant reduction of the threshold voltage for the onset of photon emission, whose origin is still under debate. Our spatially resolved measurements reveal that this threshold shift depends on the junction length. We discuss this unexpected result on the basis of a model which demonstrates that a heating of electrons is the dominant process responsible for the observed reduction of the threshold voltage.	cond-mat.mes-hall	cond-mat	Impact of electron heating on the equilibration between quantum Hall edge channels Nicola Paradiso, Stefan Heun,∗ Stefano Roddaro, Lucia Sorba, and Fabio Beltram NEST, Istituto Nanoscienze-CNR and Scuola Normale Superiore, Pisa, Italy Istituto Officina dei Materiali CNR, Laboratorio TASC, Basovizza (TS), Italy (Dated: November 5, 2018) Giorgio Biasiol When two separately contacted quantum Hall (QH) edge channels are brought into interaction, they can equilibrate their imbalance via scattering processes. In the present work we use a tun- able QH circuit to implement a junction between co-propagating edge channels whose length can be controlled with continuity. Such a variable device allows us to investigate how current-voltage characteristics evolve when the junction length d is changed. Recent experiments with fixed ge- ometry reported a significant reduction of the threshold voltage for the onset of photon emission, whose origin is still under debate. Our spatially resolved measurements reveal that this threshold shift depends on the junction length. We discuss this unexpected result on the basis of a model which demonstrates that a heating of electrons is the dominant process responsible for the observed reduction of the threshold voltage. PACS numbers: 73.43.-f, 72.10.Fk I. INTRODUCTION The renewed interest in integer quantum Hall (QH) systems is principally motivated by the peculiar features of edge states.1 They give rise to chiral one-dimensional channels that behave as perfectly collimated beams of electrons, whose trajectory,2 phase,3 -- 7 back-scattering probability,8 -- 11 and energy distribution12 can be accu- rately controlled. QH circuits are used as flexible build- ing blocks for coherent transport devices, e.g. the elec- tron analogue of the Fabry-P´erot,3 Mach-Zehnder4 -- 7 or Hanbury-Brown-Twiss13 interferometer. In recent years a number of experiments14 -- 16 focused on a particularly promising scheme: two co-propagating edge channels are imbalanced by means of selector gates,17,18 then brought into close proximity along a path of finite length, and are finally separated. The junction so defined allows co-propagating edges to exchange either energy and/or charge. In particular the inter-channel charge transfer allows equilibrating the initial electro-chemical potential imbalance. The amount of scattered charge depends on the sample characteristics, on the length of the inter- action path, and on the inter-channel bias.16 For small bias, the relevant equilibration process is elastic scat- tering induced by impurities17,18 that provide the re- quired momentum difference between initial and final edge states. This hypothesis has been confirmed by spatially resolved measurements16 that related the local backscattering map to the specific impurity distribution. For large bias, when the inter-channel imbalance ex- ceeds the energy difference between Landau levels, also radiative transitions are observed.19 This effect has been recently exploited to implement an innovative converter from phase-coherent electronic states to photons in the THz region.20 While the occurrence of this radiative emission is well established, the interpretation of the threshold value is actually not clear. In fact, several pa- pers showed14,17,18,20,21 that the threshold voltage is con- siderably smaller than the nominal Landau level gap ¯hωc. Some gap reduction mechanisms have been suggested,14 but spectroscopic studies evidenced no deviation of the photon energy from ¯hωc.22 Thus a convincing explana- tion for such a shift is missing so far. In the present work we investigate how a finite imbal- ance is equilibrated along the junction length d, by study- ing how current-voltage characteristics change when d is varied. To this end, we exploited the scanning gate mi- croscopy technique described in Ref. 16. The spectral analysis reported in section II reveals that the thresh- old voltage is lowered when the junction length increases and, at the same time, the transition in smoothened. In section III we analyze the relevant inter-channel scatter- ing processes and develop a simple model which accounts for electron heating due to hot carrier injection. The elec- tron temperature increase produces a reduction of the threshold value due to thermal broadening of the Fermi distribution. Finally, in section IV we quantitatively dis- cuss the experimental data on the basis of this model and extract the electron temperature profile along the junction. II. EXPERIMENTAL RESULTS The experimental setup is described in detail in Ref. 16. Devices were realized starting from a high- mobility AlGaAs/GaAs heterostructure. The 2DEG is confined 55 nm under the sample surface. By Shubnikov- de Haas measurements we determined both the elec- tron sheet density (n = 3.2 × 1015 m−2) and mobility (µ = 4.2 × 102 m2/V s). A 1D channel (6 µm-long, 1 µm-wide) was defined by two Schottky gates patterned on the sample. Measurements were performed at a base temperature of about 300 mK (electron temperature of 2 resulting output current is δIB = G0δV1, and therefore δVB = (h/2e2)δIB = δV1/2. Thus, beyond the thresh- old, any excess of imbalance between the two edges is perfectly equilibrated. The most interesting feature in Fig. 2 concerns the de- tail of the transition between the two regimes, whose po- sition and shape clearly depends on the interaction path length d. The dependence of the actual threshold volt- age Vth on the junction length d is shown in the inset of Fig. 2. It is always smaller than ¯hωc and is consis- tently reduced by increasing d. At the same time the transition becomes smoother, as shown in Fig. 2. This is the main experimental finding of the present paper. It crucially depends on the opportunity, given by the SGM technique, to tune the junction length, keeping all the other parameters constant. III. MODEL FOR THE INTER-CHANNEL SCATTERING To discuss our model we will refer to the scheme shown in Fig. 1. The two edge channels meet at x = 0 with an imbalance µi(0) − µo(0) = eV1. Along the junction length d the imbalance ∆µ(x) ≡ µi(x)−µo(x) ≡ e∆V (x) will decrease due to scattering events. In the model we assume an immediate intra-edge relaxation, so that both the chemical potential and the electron temperature T (x) are well defined for each position x. In general, in each junction interval dx the scattered current is given by dI = Φ(∆V (x), T (x))dx, (1) where Φ is a general function of ∆V (x) and T (x) de- pending on the details of the equilibration model (edge 50 0 ) A n ( -50 B I -100 -150 -200 -8 d: 1.5 m 2.4 m 3.2 m 4.0 m 4.8 m 5.6 m 6.3 m Vth ) V e m ( y g r e n e d l o h s e r h t 4.0 3.6 3.2 2.8 1 7 3 2 interaction length d ( m) 6 4 5 -6 -2 -4 bias V1 (mV) 0 2 FIG. 2: Current-voltage characteristics for different values of the junction length d. The threshold points Vth (colored dots) have been determined by extrapolating both the zero- bias and the saturation linear behavior (explicitly shown for d = 1.5 µm), and taking the intersection point. The inset shows the dependence of the threshold voltage on d. FIG. 1: Scheme of the experimental setup. about 400 mK) and bulk 2DES filling factor νb = 4 at B = 3.32 T, which corresponds to a cyclotron gap ¯hωc = 5.74 meV. Figure 1 schematically illustrates our experiment: two cyclotron-split edge channels originate from two distinct voltage contacts at potential V1 and 0, respectively. The channels meet at the entrance of a 1D channel and travel in close proximity for a distance d be- fore they are separated by the action of the electrostatic potential induced by the biased tip of a scanning gate microscope (SGM), as shown in detail in Ref. 16. We label the two channels as inner (i) and outer (o) channel with chemical potential µi = eV1 and µo = 0, respec- tively. After being separated by the SGM tip, the outgo- ing channels are guided to two detector contacts IA and IB. For any value of the interaction distance d compat- ible with device dimensions, i.e. from 0 to 6 µm, we can measure the current-voltage (IB-V1) characteristics of the inter-channel charge transfer. Experimental data are shown in Fig. 2. The first relevant feature concerns the zero-bias differential conductance which monotoni- cally increases with the interaction length d. This is consistent with the differential conductance SGM plots reported in Ref. 16. The curves are asymmetric around zero. While the scattered current displays a non-linear but featureless dependence on V1 for positive bias,23 we will focus on the analysis of the negative bias range (V1 < 0, i.e. µi > µo), where a clear transition between two distinct linear regimes occurs. Two linear curve sec- tions with different slope are separated by a kink, which occurs at a certain threshold voltage Vth. We evaluate Vth for each individual curve by extrapolating straight lines for both the small bias and the saturation regime and taking the abscissa of the intersection point, as ex- plicitly shown in Fig. 2 for the d = 1.5 µm curve. For bias smaller than Vth, the current-voltage character- istics are linear. The junction resistance between the two channels increases when d is lowered. On the other hand, for V1 > Vth the differential conductance satu- rates to G0 ≡ e2/h, i.e. half of the total conductance, so that an increase δV1 of the input bias produces a volt- age increase δV1/2 in both output edges. In fact the 3 In order to estimate expressions as the one on the right hand side of Eq. 5, a model for the edge dispersion is needed. In this paper we will assume the simplest case, i.e. a linear dispersion, a choice that will be justified in Section IV on the basis of the observed temperature ef- fects. In this approximation, we can assume both D and T as constant in the energy window e∆V . In this case the density of states is D(ǫ) = 2dx/(hvd), where vd is the drift velocity. Thus (see appendix A) dI = dx = dx 2eT0 hvd Z ∞ −∞ 2e2T0 hvd (fµi,T (ǫ) − fµo,T (ǫ))dǫ ∆V (x), (6) where T0 is the constant transmission probability. For this process Φ is linear in ∆V (x) and does not depend on T . IB-V1 curves can thus be calculated by solving the ordinary differential equation 3 for ∆V (x) with boundary condition ∆V (0) = V1, which gives an exponential decay of the edge imbalance ∆V (x) = V1e− 2T0 vd x. (7) was assumed behavior exponential in This literature17,18,24 to describe the zero-bias inter-channel scattering in the limit of a uniform distribution of scattering centers. The characteristic length in this case is ℓeq = vd/(2T0), i.e. the average distance between two scattering events. We experimentally verified this exponential decay in our previous work.16 Furthermore, the output current IB is linear in V1 (ohmic behavior): IB = 2e2 h V1 − ∆V (d) 2 = V1 2e2 h 1 − e− d ℓeq 2 . (8) At higher imbalance, comparable to the Landau level gap ¯hωc, other equilibration processes become possible. When µi > µo vertical radiative transitions from the in- ner edge to the outer one are enabled, as depicted in Fig. 3(b). Non-vertical relaxation could in principle oc- cur via phonon-assisted transitions. However, this is a second-order effect that can in first approximation be dis- regarded, at least for low temperatures. The infinitesimal scattered current due to vertical transitions is then given by dI =Z ∞ −∞ eD(ǫ)T1(ǫ)[fµi,T (ǫ)(1−fµo,T (ǫ−¯hωc))]dǫ, (9) where T1 is the probability per unit time for the tran- sition ǫ → ǫ − ¯hωc. Since the Landau level bands are parallel, the transition probability is constant in energy. Therefore we can simplify Eq. 9 [fµi,T (ǫ)(1 − fµo,T (ǫ − ¯hωc))]dǫ dI = dx = dx −∞ 2eT1 hvd Z ∞ kB T (x) ! hvd e∆V (x) − ¯hωc ¯hωc−e∆V (x) 2eT1 1 − e (10) FIG. 3: (a) Scheme of the impurity-induced elastic scattering for a non-interacting electron system. (b) When the chemi- cal potential of the inner edge becomes higher than the outer one by at least the cyclotron gap ¯hωc, vertical radiative tran- sitions can occur. Notice that for opposite polarity vertical transitions are suppressed. dispersion, scattering mechanisms, electron heating etc.). The corresponding changes in the edge potentials are Vi(x + dx) = Vi(x) − Vo(x + dx) = Vo(x) + h 2e2 dI h 2e2 dI (2) where the factor 2 accounts for the spin degeneracy. From equations 1 and 2 we obtain: dI dx = − e2 h d dx ∆V (x) = Φ(∆V (x), T (x)). (3) The output edge currents are IA = IB = 2e2 h 2e2 h V1 + ∆V (d) 2 V1 − ∆V (d) 2 , (4) whose sum equals the total input current Itot = IA+IB = 2(e2/h)V1. Inter-channel scattering can originate from several pro- cesses. For low bias, the inter-edge electron transfer can be either induced by impurity or phonon scattering.17,18 The latter, however, was shown17,18 to be less important when the base temperature is smaller than 1 K. The rel- evant process (sketched in Fig. 3(a)) is thus the elastic scattering induced by sharp impurity potentials which provide the change in momentum needed for the inter- channel transition. The infinitesimal scattered current in the interval dx is dI =Z ∞ −∞ eD(ǫ)T (ǫ)(fµi,T (ǫ) − fµo,T (ǫ))dǫ, (5) where D(ǫ) is the density of states around the energy ǫ and T (ǫ) is the elastic scattering probability per unit time. where the integration is explicitly shown in appendix A. In the Φ function we also have a non-linear addendum, thus the integration of Eq. 3 has to be performed numer- ically. At low temperature, due to the exponential term, the effect of the term in Eq. 10 is negligible for ∆V (x) below the threshold ¯hωc. For ∆V (x) > ¯hωc the avail- ability of empty states in the lower Landau level gives rise to a strong radiative relaxation. As shown in recent experiments,20 the photons emitted in this process can be collected with a suitable waveguide and detected. So far we completely neglected the effect of the elec- tron heating due to the injection of hot carriers. In order to obtain a quantitative estimate of the amount of energy transferred to the electron system, we need to first esti- mate the total energy increase of an edge channel when we increase its chemical potential from the ground level µ = µ0 to µ = µj and its temperature from T = 0 to T = Tj IV. DISCUSSION 4 Figure 4(a) shows the IB-V1 characteristics (red dots) for the d = 2.4 µm case. The behavior is clearly ohmic, as confirmed by a linear fit (blue line, adjusted R2 = 0.997). This agrees with the predictions of our model for low bias, when radiative emission is negligible and Eq. 8 ap- plies. The zero-bias differential conductance depends on the distribution of scattering centers inside the constric- tion. Equation 8 allows us to obtain the equilibration length ℓeq by fitting the IB-V1 curves in the linear region. Figure 4(b) displays the different ℓeq values obtained for each junction length d. The average ℓeq value (21 µm) is consistent with the one reported in Ref. 16 (15 µm), considering that those results were obtained from dif- ferent samples. The graph evidences that ℓeq depends on d. As shown in Ref. 16, the actual impurity den- sity is highly sample-dependent and can fluctuate along Ej = Z ∞ −∞ ≈ 1 2 2τ h 2d hvd (ǫ − µ0)(fµj ,T (ǫ) − fµ0,0(ǫ))dǫ (µj − µ0)2 + 2τ h π2 6 BT 2 k2 j , (11) where in the second line we approximated the integral with the first order Sommerfeld expansion (as shown in detail in appendix B) and τ ≡ d/vd. To calculate explicitly the output temperature T (x) we will assume energy conservation in each infinitesimal element dx Ei(x) + Eo(x) = Ei(x + dx) + Eo(x + dx) (12) together with three additional approximations: (i) the two edges immediately restore the thermal equilibrium after each scattering event; (ii) the temperature is ap- proximately the same in both edges Ti(x) = To(x) = T (x), with T (0) = Tin, where Tin is the bulk electron temperature; (iii) in each element dx only the ohmic part of the scattered current dI contributes to the electron heating. In fact, only the elastic process transfers hot carriers between the two edges, while the radiative term allows electrons to relax by photon emission. With these assumptions, after substituting the expression in Eq. 11 into Eq. 12 as shown in Appendix C, we can deduce an equation which relates the change in temperature with the local imbalance d dx T (x) = 3e2 4π2k2 Bℓeq ∆V 2(x) T (x) . (13) Thus Eq. 3 must be solved together with Eq. 13 to ex- tract both T (x) and V (x). Due to the electron heating, the onset of radiative transitions is shifted below the cy- clotron gap value ¯hωc since thermally excited electrons leave available states in a range of about kBT around the chemical potential of the lower level. The transition itself becomes smoother, since the expression in Eq. 10 is less steep at higher temperatures. ) m ( q e 25 20 15 10 5 0 (a) ) A n ( B I 1.0 0.0 -1.0 -2.0 -3.0 -4.0 -5.0 -6.0 -7.0 -1.5 -1.0 bias V1 (mV) -2.0 35 (b) 30 -0.5 0.0 1 2 3 4 d ( m) 5 6 7 FIG. 4: (a) Detail of the IB-V1 characteristics in the range -2 mV< V1 < 0, for d = 2.4 µm (red dots). The behavior is ohmic as evidenced by the linear fit (blue line). (b) Plot of ℓeq for different junction lengths d. the inter-channel junction. The monotonical decrease ob- served in Fig. 4(b) could however indicate that for short d the scattering centers are somewhat less effective, due to the fact that the edges are smoothly brought into inter- action and separated. Therefore the inter-channel sepa- ration is larger at the constriction ends than at the inner points. These boundary effects are more important for smaller d. The previous results provide the first of the two free parameters of our model, namely ℓeq and T1. Therefore we fit the experimental curves in Fig. 2 with the functions obtained solving Eqs. 3 and 13, with the only free param- eter T1. The fit for d = 2.4 µm is displayed in Fig. 5(a), together with the experimental data. The agreement be- tween the two curves is remarkable: our simple model re- produces both the shift and the smoothing at the thresh- old, i.e. the two main features observed in Fig. 2. The threshold shift can be better seen in Fig. 5(b), where we plot the fits for all experimental curves of Fig. 2 (solid lines), together with the corresponding experimen- tal data (dotted lines). In the inset we show a compari- son between the threshold voltage values extracted from the fitting curves and the ones directly estimated from the IB-V1 characteristics. This graph clearly indicates that the present model suitably describes the observed threshold reduction. The value for the Landau level gap (¯hωc = 5.74 meV) was kept constant in these fits. This value turns out to be optimal once both ℓeq and T1 have been determined, because then a further adjustment of the gap only decreases the fit quality. This result explains the reduction of the threshold for photon emission observed in several experiments.14,20 The significant deviation from ¯hωc/e is an effect due to the electron heating induced by the injection of hot car- riers in the outer edge via elastic scattering. To quantitatively estimate the electron temperature in- crease, we solved Eq. 13, using the parameters ℓeq and T1 provided by the previous fits, with the initial con- dition Tin = 400 mK. Figure 6 shows the solutions for the d values corresponding to the experimental data in Fig. 2. For very small bias the temperature increases al- most quadratically with the imbalance, while for interme- diate values the behavior is approximatively linear, with a slope proportional to d. Finally, the temperature tends to saturate at the onset of radiative emission, which sup- presses further injection of hot electrons into the outer edge. At saturation, the output edge temperatures are by far larger than the base temperature. In section III we considered a linear edge disper- sion, which neglects effects of edge reconstruction due to electron-electron interaction.25 We have also developed alternative models, which take into account the effect of the compressible and incompressible stripes at the sample edge. While such more complex analysis correctly pre- dicts the linear behavior at low bias, it is less satisfactory in describing the threshold evolution, although it con- tains more adjustable parameters (as the compressible and incompressible stripe widths). We interpreted such (a) 0 -25 -50 ) A n ( 5 -1 0 1 Experiment Fit B I -75 -100 -125 (b) 50 0 ) A n ( -50 B I -100 -150 -200 -8 -7 -6 -5 -3 -4 -2 bias V1 (mV) Experiment Fit ) V e m ( y g r e n e d l o h s e r h t 4.0 3.6 3.2 2.8 1 d: 1.5 m 2.4 m 3.2 m 4.0 m 4.8 m 5.6 m 6.3 m -6 7 5 6 interaction length d ( m) 2 0 2 3 4 -2 -4 bias V1 (mV) FIG. 5: (a) Fit (blue line) of the IB-V1 curve for d = 2.4 µm (red dots) using solutions of Eqs. 3 and 13, with the parameter ℓeq obtained from the previous linear fits. (b) Complete set of fitting curves (solid lines) for all measured d values, together with the corresponding experimental data of Fig. 2 (dotted lines). (Inset) Threshold voltages plotted as a function of d as deduced from the fitting curves (red dots), together with the values directly extracted from Fig. 2 (black squares). discrepancy as the effect of the high electron temperature induced by the elastic scattering processes and present on most part of the edge junction. As edge reconstruction is known to be quickly washed out by temperature,21,26,27 we have therefore rather chosen a simple model with a linear edge dispersion, which indeed captures the relevant features observed in the experiment. V. CONCLUSION We demonstrated a tunable-length junction between highly imbalanced edge channels in the quantum Hall regime. The measurements of its current-voltage char- acteristics clearly evidence that the threshold voltage for d: 6.3 m 5.6 m 4.8 m 4.0 m 3.2 m 2.4 m 1.5 m 12 10 ) K ( T 8 6 4 2 0 -7 -6 -5 thus 6 Z ∞ −∞(cid:18) 1 1 + ex−xi − 1 1 + ex−xo(cid:19) kBT dx = = lim x→+∞ − lim x→−∞ kBT (cid:2)− ln(1 + ex−xi) + ln(1 + ex−xo)(cid:3) + kBT (cid:2)− ln(1 + ex−xi) + ln(1 + ex−xo)(cid:3) = = kBT (xi − xo) − 0 = µi − µo = e∆V . (A4) In Eq. 10 we evaluated the integral Z ∞ −∞ [fµi,T (ǫ)(1 − fµo,T (ǫ − ¯hωc))]dǫ. (A5) Defining x ≡ ǫ/kBT , xi ≡ µi/kBT and xo ≡ (¯hωc + µo)/kBT , we have A primitive of the expression in square brackets is -1 0 -3 -4 -2 bias V1 (mV) FIG. 6: Bias dependence of the outgoing electron tempera- ture plotted for different d values. The curves are obtained from Eq. 13, with the initial condition T (0) = 400 mK. The parameters ℓeq and T1 are given by the previous fits of the experimental data. the onset of radiative emission depends on the junction length d. We show how this behavior can be explained by a simple model accounting for the heating effect due to the elastic scattering of hot carriers. Acknowledgments We acknowledge financial support from the Ital- of Research (MIUR-FIRB projects RBID08B3FM, ian Ministry RBIN045MNB, RBIN067A39 002, and RBIN06JB4C). RBIN048ABS, Z ∞ −∞(cid:20) 1 1 + ex−xi (cid:18)1 − exi exi − exo thus 1 exi x→+∞ = lim Z ∞ −∞(cid:20) 1 + ex−xi (cid:18)1 − kBT (cid:20) kBT (cid:20) = kBT (cid:20)0 − exi − exo exi exi − exo exi − exo − lim x→−∞ exi (A6) (A7) 1 1 1 + ex−xo(cid:19)(cid:21) kBT dx. ln(cid:18) exo + ex exi + ex(cid:19) , 1 + ex−xo(cid:19)(cid:21) kBT dx = ln(cid:18) exo + ex ln(cid:18) exo + ex (xo − xi)(cid:21) = exi + ex(cid:19)(cid:21) + exi + ex(cid:19)(cid:21) = = kBT xi − xo 1 − exo−xi = e∆V − ¯hωc ¯hωc−e∆V 1 − e kB T . (A8) Appendix B: First order approximation to the edge energy Appendix A: Integration of expressions containing Fermi functions In order to evaluate the first line of Eq. 11 we exploit the Sommerfeld expansion In Eq. 6 we evaluated the integral −∞ Z ∞ −∞ =Z ∞ (fµi,T (ǫ) − fµo,T (ǫ))dǫ = 1 − ǫ−µi kB T 1 ǫ−µo kB T ! dǫ. 1 + e (A1) 1 + e Defining x ≡ ǫ/kBT , xi ≡ µi/kBT and xo ≡ µo/kBT , we have Z ∞ −∞(cid:18) 1 1 + ex−xi − 1 1 + ex−xo(cid:19) kBT dx. (A2) A primitive of the expression in brackets is − ln(1 + ex−xi) + ln(1 + ex−xo), (A3) −∞ Z ∞ = Z µ −∞ g(ǫ) 1 + e ǫ−µ kB T dǫ = g(ǫ)dǫ + π2 6 k2 µ (cid:19)4 BT 2g ′(µ) + O(cid:18) kBT (B1) where g(ǫ) is a generic function of ǫ and g ′(µ) is its first derivative evaluated at ǫ = µ. By applying this relation to Eq. 11 we obtain −∞ Z ∞ ≈ Z µj 2(cid:18) 2τ µ0 1 = 2d hvd ǫ − µ0 1 + e ǫ−µ kB T dǫ −Z µ0 −∞ 2d hvd (ǫ − µ0)dǫ + 2d hvd π2 6 2d hvd (ǫ − µ0)dǫ ≈ k2 BT 2 j = h (cid:19) (µj − µ0)2 +(cid:18) 2τ h (cid:19) π2 6 BT 2 k2 j . (B2) Appendix C: Determination of T (x) After evaluating Eq. 12 with Eq. C1, using the substi- tutions C2 we obtain When the electron temperature is non-zero, the ex- pression for the total edge energy has an extra term pro- portional to T 2, as seen in Eq. B2. We can thus define the electrostatic and the thermal component of the total edge energy: 2π2 3 k2 BT (x)dT = e2 2 1 ℓeq ∆V 2(x)dx (C3) (where Ti(x) = To(x) = T (x)) from which Eq. 12 easily follows. 7 1 Eel ≡ h (cid:19) (µj − µ0)2 = 2(cid:18) 2τ h (cid:19) π2 Eth ≡ (cid:18) 2τ k2 BT 2 j 6 1 2(cid:18) 2τ h (cid:19) e2V 2 j (C1) where Vj is the edge voltage referred to the ground. Equation C1 allows us to evaluate Eq. 12. As already explained in the paper, only elastic scattering processes transfer hot carriers between the edges, while the radia- tive term allows electrons to relax by photon emission. Thus we modify Eqs. 2 as follows Vi(x + dx) = Vi(x) − = Vi(x) − Vo(x + dx) = Vo(x) + = Vo(x) + e2 h h 2e2 dI elast. 1 h 2e2 ℓeq h 2e2 dI elast. 1 h 2e2 ℓeq e2 h ∆V (x)dx ∆V (x)dx. (C2) ∗ Electronic address: [email protected] 1 B. I. Halperin, Phys. Rev. B 25, 2185 (1982). 2 N. Paradiso, S. Heun, S. Roddaro, L. N. Pfeiffer, K. W. West, L. Sorba, G. Biasiol, and F. Beltram, Physica E 42, 1038 (2010). 3 F. E. Camino, W. Zhou, and V. J. Goldman, Phys. Rev. Lett. 95, 246802 (2005). 4 Y. Ji, Y. Chung, D. Sprinzak, M. Heiblum, D. Mahalu, and H. Shtrikman, Nature 422, 415 (2003). 5 I. Neder, M. Heiblum, Y. Levinson, D. Mahalu, and V. 13 I. Neder, N. Ofek, Y. Chung, M. Heiblum, D. Mahalu, and V. Umansky, Nature 448, 333 (2007). 14 A. Wurtz, R. Wildfeuer, A. Lorke, E. V. Deviatov, and V. T. Dolgopolov, Phys. Rev. B 65, 075303 (2002). 15 T. Nakajima, Y. Kobayashi, S. Komiyama, M. Tsuboi, and T. Machida, Phys. Rev. B 81, 085322 (2010). 16 N. Paradiso, S. Heun, S. Roddaro, D. Venturelli, F. Tad- dei, V. Giovannetti, R. Fazio, G. Biasiol, L. Sorba, and F. Beltram, Phys. Rev. B 83, 155305 (2011). 17 S. Komiyama, H. Hirai, S. Sasa, and S. Hiyamizu, Phys. Umansky, Phys. Rev. Lett. 96, 016804 (2006). Rev. B 40, 12566 (1989). 6 I. Neder, F. Marquardt, M. Heiblum, D. Mahalu, and V. 18 S. Komiyama, H. Hirai, M. Ohsawa, Y. Matsuda, S. Sasa, Umansky, Nat. Phys. 3, 534 (2007). 7 P. Roulleau, F. Portier, D. C. Glattli, P. Roche, A. Ca- vanna, G. Faini, U. Gennser, and D. Mailly, Phys. Rev. B 76, 161309(R) (2007). 8 S. Roddaro, V. Pellegrini, F. Beltram, G. Biasiol, L. Sorba, R. Raimondi, and G. Vignale, Phys. Rev. Lett. 90, 046805 (2003). 9 S. Roddaro, V. Pellegrini, F. Beltram, G. Biasiol, and L. and T. Fujii, Phys. Rev. B 45, 11085 (1992). 19 K. Ikushima, H. Sakuma, S. Komiyama, and K. Hirakawa, Phys. Rev. B 76, 165323 (2007). 20 K. Ikushima, D. Asaoka, S. Komiyama, T. Ueda, K. Hi- rakawa, Physica E 42, 1034 (2010). 21 T. Machida, H. Hirai, S. Komiyama, T. Osada, and Y. Shiraki, Phys. Rev. B 54, R14261 (1996). 22 S. Komiyama, H. Sakuma, K. Ikushima, and K. Hirakawa, Sorba, Phys. Rev. Lett. 93, 046801 (2004). Phys. Rev. B 73, 045333 (2006). 10 S. Roddaro, V. Pellegrini, F. Beltram, L. N. Pfeiffer, and 23 The analysis of this behavior, already observed in other K. W. West, Phys. Rev. Lett. 95, 156804 (2005). 11 S. Roddaro, N. Paradiso, V. Pellegrini, G. Biasiol, L. Sorba, and Fabio Beltram, Phys. Rev. Lett. 103, 016802 (2009). 12 C. Altimiras, H. le Sueur, U. Gennser, A. Cavanna, D. experiments,14 is beyond the scope of the present paper. 24 G. Muller, D. Weiss, A. V. Khaetskii, K. von Klitzing, S. Koch, H. Nickel, W. Schlapp, and R. Losch, Phys. Rev. B 45, 3932 (1992). 25 D. B. Chklovskii, B. I. Shklovskii, and L. I. Glazman, Phys. Mailly, and F. Pierre, Nature Phys. 6, 34 (2010). Rev. B 46, 4026 (1992). 26 K. Lier and R. R. Gerhardts, Phys. Rev. B 50, 7757 (1994). 27 T. Suzuki and T. Ando, J. Phys. Soc. Jpn. 62, 2986 (1993). 8
1706.09589	1	1706	2017-06-29T06:02:32	Bilinear magneto-electric resistance as a probe of three-dimensional spin texture in topological surface states	[ "cond-mat.mes-hall" ]	Surface states of three-dimensional topological insulators exhibit the phenomenon of spin-momentum locking, whereby the orientation of an electron spin is determined by its momentum. Probing the spin texture of these states is of critical importance for the realization of topological insulator devices, however the main technique available so far is the spin- and angle-resolved photoemission spectroscopy. Here we reveal a close link between the spin texture and a new kind of magneto-resistance, which depends on the relative orientation of the current with respect to the magnetic field as well as the crystallographic axes, and scales linearly with both the applied electric and magnetic fields. This bilinear magneto-electric resistance can be used to map the spin texture of topological surface states by simple transport measurements. For a prototypical Bi2Se3 single layer, we can map both the in-plane and the out-of-plane components of the spin texture - the latter arising from hexagonal warping. Theoretical calculations suggest that the bilinear magneto-electric resistance originates from the conversion of a non-equilibrium spin current into a charge current under the application of the external magnetic field.	cond-mat.mes-hall	cond-mat	Bilinear magneto-electric resistance as a probe of three- dimensional spin texture in topological surface states Pan He1†, Steven S.-L. Zhang2†, Dapeng Zhu1, Yang Liu1, Yi Wang1, Jiawei Yu1, Giovanni Vignale2 and Hyunsoo Yang1* 1Department of Electrical and Computer Engineering, and NUSNNI, National University of Singapore, 117576, Singapore 2Department of Physics and Astronomy, University of Missouri, Columbia Missouri 65211, USA †These authors contributed equally to this work. e-mail: [email protected] Surface states of three-dimensional topological insulators exhibit the phenomenon of spin-momentum locking, whereby the orientation of an electron spin is determined by its momentum. Probing the spin texture of these states is of critical importance for the realization of topological insulator devices, however the main technique available so far is the spin- and angle-resolved photoemission spectroscopy. Here we reveal a close link between the spin texture and a new kind of magneto-resistance, which depends on the relative orientation of the current with respect to the magnetic field as well as the crystallographic axes, and scales linearly with both the applied electric and magnetic fields. This bilinear magneto-electric resistance can be used to map the spin texture of topological surface states by simple transport measurements. For a prototypical Bi2Se3 single layer, we can map both the in-plane and the out-of-plane components of the spin texture – the latter arising from hexagonal warping. 1 Theoretical calculations suggest that the bilinear magneto-electric resistance originates from the conversion of a non-equilibrium spin current into a charge current under the application of the external magnetic field. The discovery of three-dimensional (3D) topological insulators (TIs) has triggered an outburst of research activities aimed at understanding the physical properties of this new state of matter1-4, and exploring its applications to optoelectronics and spintronics5, 6. The most remarkable property of TIs is their topologically protected surfaces states characterized by a tight correlation between spin orientation and momentum, known as spin-momentum locking1-4. The resulting spin texture has been mapped by spin- and angle- resolved photoemission spectroscopy (SR-ARPES)7-9. When the topological surface states (TSS) are hexagonally deformed (Fig. 1a)10, 11, the spin texture acquires a momentum- dependent out-of-plane component10. This hexagonally warped helical spin texture, depicted in Fig. 1b, has been observed in various TI materials12-16 by SR-ARPES. In addition to the optical methods13-17, electrical detection of the spin-polarized TSS has been achieved with the use of ferromagnetic contacts as spin detectors18-22. However, only the in-plane spin component could be detected by this method18-22. The magneto-transport properties of 3D TIs have also been extensively studied23- 29. Various types of magneto-resistance in non-magnetic 3D TIs have been identified including the weak (anti)localization induced magneto-resistance23, 24, the non-saturating linear magneto-resistance25-27, and the anisotropic magneto-resistance28, 29, which all appear in the linear response of the current to the applied electric field. Recently, a large unidirectional magneto-resistance (UMR) was measured in heterostructures consisting of 2 non-magnetic and magnetic topological insulators30, which exhibits a linear dependence on the current density, however, the presence of a magnetic layer was deemed to be essential for this nonlinear magneto-resistance. So far, no correlation between the magneto- resistance and the spin texture of the TSS has been established, even though a tight link between the charge and spin currents is expected due to spin-momentum locking. Here, we report the observation, in a single layer of nonmagnetic 3D TI Bi2Se3, of a new magneto-resistance effect which scales linearly with the applied electric current and magnetic field: we thus name it "bilinear magneto-electric resistance". We show that a mapping between momentum-dependent spin textures and the angular dependence of the bilinear magneto-electric resistances can be established, which enables transport measurements of the 3D spin texture in the TSS with hexagonal warping. Observation of the bilinear magneto-electric resistance The samples studied in this work are 20 quintuple layer (QL) Bi2Se3 films, which are capped by MgO (2 nm)/Al2O3 (3 nm) and patterned into Hall bars of length l = 100 µm and width w = 20 µm. The current channels (x axis) of the Hall bar devices were designed along different crystal directions on the same films. To measure the current-independent first harmonic Rω and the current-dependent second harmonic R2ω of the longitudinal resistance31 (see Supplementary Information S1), a.c. currents were injected into the devices, and the first (Vω) and second (V2ω) harmonic longitudinal voltages (Vxx) were recorded simultaneously. The measurements were performed while rotating the applied magnetic field H in the xy, zy, and zx planes as schematically shown in Fig. 2a-c, respectively. 3 In Figs. 2d-f, we show the angular dependences of Rω (blue) and R2ω (red) when a fixed magnetic field of 9 T is swept in the xy, zy, and zx planes, respectively. For scans in all three planes, the linear longitudinal resistance Rω is anisotropic28, 29, and exhibits a sinusoidal angular dependence with a period of 180. This angular dependence shows that Rω is invariant under the reversal of the magnetic field. More interestingly, we observe a novel anisotropic behavior in the nonlinear resistance R2ω. For the three scans shown in Fig. 2d-f, R2ω exhibits sinusoidal angular dependences with a period of 360, but with different phases in different planes. Contrary to Rω, R2ω changes its sign when reversing either the current or the magnetic field polarity (see Supplementary Information S3). The measured R2ω indicates the existence of a current-dependent unidirectional resistance in Bi2Se3, which has also been confirmed by dc measurements30, 32, 33 (see Supplementary Information S4,5). We further investigate R2ω by varying the direction of the current with respect to the -axis of TI. We show that there exists a mapping between the spin canting angle at a given k-point on the Fermi contour of the TSS and the magnetic field canting angle for which R2ω reaches its maximum value in the zy scan, for a current direction parallel to k. This provides a new way, alternative to SR-ARPES, to probe the 3D spin texture of the TSS via transport measurements. Transport measurements of the helical spin texture with hexagonal warping In Fig. 3a-c, we note that in the xy plane R2ω retains the same angular dependence regardless of the orientation of the current along three typical directions in k-space, namely , and . This is consistent with the rotational symmetry of the xy 4 KKMKplane projection of the surface spin texture. R2ω approaches zero when the magnetic field is aligned with the direction of the current (i.e., φ = 0 or 180), has a peak when the current and the magnetic field are orthogonal (i.e., φ = 90 or 270), and changes sign when the magnetic field is reversed. These observations are consistent with the fact that the spin s orientation on the Fermi contour of the TSS is always perpendicular to the corresponding wave vector k7-16, i.e., s (k) ⊥ k, and that the spin directions are opposite for opposite wave vectors, i.e., s (-k) = -s (k), due to the spin-momentum locking. More remarkably, we observe that the out-of-plane field scans of R2ω, in contrast to the in-plane ones, exhibit a striking dependence on the direction of the current. For the zx plane scans, a sizable angular dependence is found when the current is along the and line in Figs. 3a and 3c, with peaks occurring when the magnetic field is perpendicular to the surface (i.e., θ = 0 or 180). Moreover, R2ω switches the sign when the current direction is rotated from to . When the current direction is along the line (forming an angle of 30 with the line, see Fig. 3b), R2ω becomes negligible in the zx scan. These observations are in qualitative agreement with a hexagonally warped helical spin texture, in which the out-of-plane component of the spin changes sign when k is rotated from to and vanishes when k is along 10, 12-16 (see Fig. 1b). In the presence of hexagonal warping, the full rotational symmetry of the spin texture on the Fermi contour around the z-axis is reduced to a lower C3v symmetry, as first predicted theoretically10 and later observed by SR-ARPES12-16. Thus, our R2ω results indicate the existence of hexagonal warping on the Fermi surface. For a quantitative comparison, we compare the magnetic field canting angle with the spin canting angle in the TSS. We define as the angle between the xy 5 KKKKMKKKMHcscHcplane and the direction of the magnetic field for which R2ω reaches its maximum value in the zy scan for a given current direction. is defined as the angle between the xy plane and the orientation of the spin in the TSS with k parallel to the current. The reason for selecting field scans in the zy plane is that the spin of the TSS with k parallel to the current (which is chosen as the x-axis) always lies in the zy plane due to spin-momentum locking. The field canting angle is obtained from , where stands for the field angle corresponding to the positive peak position of , which is quite different for the three current directions in Figs. 3a-c. In Fig. 4a, we show the field canting angle as a function of the angle between the direction of the current and the line (denoted by ). The angular dependence is sinusoidal and has a period of 120, tallying with the C3v symmetry of the spin texture in Bi2Se3 with hexagonal warping13, 16. On the other hand, the out-of-plane spin polarization derived from perturbation theory is10 (1) where is the angle between the line and the k direction that coincides with the current direction, is the magnitude of the Fermi wave vector, is the Dirac velocity, and characterizes the strength of the hexagonal warping10-16. The k-dependent spin canting angle can be calculated by . In Fig. 4b, we show the as a function of as a best fit to the experimental data given in Fig. 4a, obtained by using the following material parameters: m/s, eVÅ3 and the Fermi energy eV ( ), which are in quantitative agreement with the values 6 scppHc90ppR2wyzHcKKpk222)/()3(cos3cos)(FkkkzkskKFk)(arcsin)(zkscssck510518028.0F/FFkobtained by recent measurements11, 13, 16. Thus, we have demonstrated a direct correlation between the angular dependence of R2ω in the TI and the 3D spin texture of the TSS with hexagonal warping. Physical origin of bilinear magneto-electric resistance Nonlinear UMR has been recently observed in heavy metal/ferromagnet or TI/magnetic-TI bilayers30-34. The presence of a ferromagnetic layer has been deemed to play an essential role as a source of spin-dependent scattering30-34. In our system, however, the nonlinear R2ω occurs in a nonmagnetic Bi2Se3 single layer, with no ferromagnetic material involved and no reliance on spin-dependent scattering. In order to further investigate the physical origin of the observed magneto-electric resistance, we have measured R2ω as a function of the amplitudes of the current I (proportional to the applied electric field) and the magnetic field H. As shown in Fig. 5a, the amplitude ΔR2ω of the angular dependent R2ω exhibits a linear dependence on the current amplitude up to 0.75 mA. This feature is shared by the UMR in the bilayers with magnetic materials. In Fig. 5b, ΔR2ω increases linearly with the amplitude of the magnetic field H up to 9 T and displays a negligible value at zero magnetic field. The linear H dependence of R2ω is different from that of UMR, which is constant as a function of H for a saturated FM layer31-33. The linear dependences of R2ω on both the current (electric field) and the magnetic field is a new effect reported in TIs, which we refer to as the bilinear magneto-electric resistance. To further prove that the bilinear magneto-electric resistance is linked to spin- momentum locked TSS with hexagonal warping, we have theoretically calculated the longitudinal resistivity by solving the homogeneous Boltzmann equation with the model 7 Hamiltonian for an electron in a spin-momentum locked band of the form , with the momentum (k)-dependent field given by (2) where and are the unit vectors denoting the directions along and normal to the surface respectively, are the Pauli spin matrices, and the term cubic in momentum describes the hexagonal warping effect10. We obtain an analytical expression of R2ω by expanding the solution of the Boltzmann equation to second order in the electric field and to first order in the magnetic field (see Supplementary Information S6 for detailed derivation). The result is , (3) where and are the y and z components of the magnetic field defined in Fig. 2a-c, and is the angle of a given direction with respect to the x-axis (i.e., the direction of the electric field E), and are the coefficients of the contributions to R2ω from the in-plane and out-of-plane components of the magnetic field, respectively. The theoretical formula for R2ω shows a linear dependence on both the electric field and magnetic field, in line with our experimental findings. In Fig. 3d-f, theoretical fits to the experimental data are shown obtained by using Eq. (3). The excellent agreement between theory and experiment confirms that the spin-momentum locked TSS with hexagonal warping is indeed shaping the magneto-electric resistance. Microscopically, the bilinear magneto-electric resistance can be interpreted as a partial conversion of a pure spin current of second order in the applied electric field E into 8 ])([HkhσBTIgH),3()(22yxkkykzkkhyzMσ)3cos(2KzOPyIPHaHawlERyHzHKK65236egaBFIPFBOPega326a charge current in the TSS. In the absence of the magnetic field, the electron group velocity v is odd in k whereas the second order electron distribution function is even in k, i.e., and (see Fig.1c). On the other hand, electrons at k and carry opposite spins due to spin-momentum locking. As a result, in the absence of a magnetic field, there are equal numbers of electrons carrying opposite spins and moving in opposite directions, which gives rise to a pure spin current at the second order in the electric field E, as schematically shown in Fig. 1d. When a magnetic field is applied, however, both the group velocity and the second order distribution are shifted in k-space and then the two fluxes of electrons with opposite spin orientations no longer compensate each other, causing the spin current to be partially converted into a charge current , as shown in Fig. 1e,f. This extra charge current, which changes with the relative orientation of the magnetic field with respective the electric field, is the physical reason for the bilinear magneto-electric resistance. Finally, we note that the spin-charge current conversion will not take place when the band energy is linear in k, since in this case, the magnetic field changes both the electron velocity and the second order distribution through a constant displacement in momentum space: the two changes cancel against each other leaving the current unaltered. Summary and outlook We have shown that measurements of a magneto-electric resistance that depends linearly on both the electric and the magnetic field provide a simple way to probe the 3D spin texture of TSS with hexagonal warping by purely electrical means and offers an alternative to SR-ARPES. Our experimental observations, together with the theoretical 9 )(2kf)()(kvkv)()(22kkffk)(2sEJ)(2eEJcalculations, reveal the critical role of spin current to charge current conversion occurring at the TSS under a magnetic field in shaping the bilinear magneto-electric resistance. Our method can also be applied to materials with giant Rashba spin splitting at the surface or in the bulk35, as well as two-dimensional transition metal dichalcogenides with spin- polarized valley states36. The magneto-electric resistance is an effect of potential technological interest, similar to the UMR recently observed in magnetic heterostructures, but significantly different from it for being linear in the magnetic field and not dependent on the presence of ferromagnetic elements and spin-polarization-dependent scattering rates. Methods Sample preparation. The high-quality 20 QL Bi2Se3 films were grown on Al2O3 (0001) substrates in a molecular beam epitaxy system with a base pressure < 2×10-9 mbar using the two-step deposition procedure37. Before sample growth, the sapphire substrates were annealed at 750 C for 30 min in a vacuum after transferring into the growth chamber. High-purity Bi (99.9999%) and Se (99.999%) were evaporated from standard Knudsen cells using Se/Bi flux ratio of ~20. To reduce Se vacancies in Bi2Se3, initial 2-3 QL Bi2Se3 were first deposited at 150 C, and then the substrate temperature was ramped to 250 C at 5 C/min under Se flux for the second step growth. For the transport measurements, a capping layer of MgO (2 nm)/Al2O3 (3 nm) was deposited on the Bi2Se3 films as a protection layer. Hall bar devices with channel dimensions of 20×100 µm2 were fabricated using the standard photolithography method. 10 Electrical measurements. Hall bar devices were wire-bonded to the sample holder and installed in a physical property measurement system (PPMS, Quantum Design) for transport measurements with the temperature range of 2-400 K. We performed the measurements of a.c harmonic longitudinal resistance, using a Keithley 6221 current source and two Stanford Research SR830 lock-in amplifiers. During the measurements, a constant amplitude sinusoidal current with a frequency of 251 Hz is applied to the devices, and the in-phase first harmonic Vω and out-of-phase second harmonic V2ω longitudinal voltage signals were measured simultaneously by two lock-in amplifiers, while rotating the magnetic field H in the xy, zy and zx planes. For a.c longitudinal resistance measurements under a dc bias current, Keithley 6221 provided a dc current with a certain polarity and an a.c. current with a small amplitude (50 µA) simultaneously, and Stanford Research SR830 lock-in amplifier was used to measure the in-phase first harmonic voltage Vω. For dc measurements of the longitudinal resistance, Keithley 2400 and 2002 were used. References Fu, L., Kane, C. L. & Mele, E. J. Topological insulators in three dimensions. Phys. Rev. Lett. 98, 106803 (2007). Hasan, M. Z. & Kane, C. L. Colloquium: Topological insulators. Rev. Mod. Phys. 82, 3045- 3067 (2010). Qi, X.-L. & Zhang, S.-C. The quantum spin Hall effect and topological insulators. Phys.Today 63, 33-38 (2010). Yazyev, O. V., Moore, J. E. & Louie, S. G. Spin polarization and transport of surface states in the topological insulators Bi2Se3 and Bi2Te3 from first principles. Phys. Rev. Lett. 105, 266806 (2010). Garate, I. & Franz, M. Inverse spin-galvanic effect in the interface between a topological insulator and a ferromagnet. Phys. Rev. Lett. 104, 146802 (2010). Pesin, D. & MacDonald, A. H. Spintronics and pseudospintronics in graphene and topological insulators. Nat. Mater. 11, 409-416 (2012). Hsieh, D. et al. A tunable topological insulator in the spin helical Dirac transport regime. Nature 460, 1101-1105 (2009). Hsieh, D. et al. Observation of unconventional quantum spin textures in topological insulators. Science 323, 919-922 (2009). 11 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. Nishide, A. et al. Direct mapping of the spin-filtered surface bands of a three-dimensional quantum spin Hall insulator. Phys. Rev. B 81, 041309 (2010). Fu, L. Hexagonal warping effects in the surface states of the topological insulator Bi2Te3. Phys. Rev. Lett. 103, 266801 (2009). Kuroda, K. et al. Hexagonally deformed fermi surface of the 3D topological insulator Bi2Se3. Phys. Rev. Lett. 105, 076802 (2010). Alpichshev, Z. et al. STM imaging of electronic waves on the surface of Bi2Te3: Topologically protected surface states and hexagonal warping effects. Phys. Rev. Lett. 104, 016401 (2010). 17. 14. 15. 16. in topological insulators: A view 13. Wang, Y. H. et al. Observation of a warped helical spin texture in Bi2Se3 from circular dichroism angle-resolved photoemission spectroscopy. Phys. Rev. Lett. 107, 207602 (2011). Souma, S. et al. Direct measurement of the out-of-plane spin texture in the Dirac-cone surface state of a topological insulator. Phys. Rev. Lett. 106, 216803 (2011). Xu, S.-Y. et al. Realization of an isolated Dirac node and strongly modulated spin texture in the topological insulator Bi2Te3. Preprint at https://arxiv.org/abs/1101.3985 (2011). Nomura, M. et al. Relationship between Fermi surface warping and out-of-plane spin polarization from spin- and angle-resolved photoemission. Phys. Rev. B 89, 045134 (2014). McIver, J. W., Hsieh, D., Steinberg, H., Jarillo-Herrero, P. & Gedik, N. Control over topological insulator photocurrents with light polarization. Nat. Nanotech. 7, 96-100 (2012). Li, C. H., van 't Erve, O. M. J., Robinson, J. T., Liu, Y., Li, L. & Jonker, B. T. Electrical detection of charge-current-induced spin polarization due to spin-momentum locking in Bi2Se3. Nat. Nanotech. 9, 218-224 (2014). Ando, Y. et al. Electrical detection of the spin polarization due to charge flow in the surface state of the topological insulator Bi1.5Sb0.5Te1.7Se1.3. Nano Lett. 14, 6226-6230 (2014). Tang, J. et al. Electrical detection of spin-polarized surface states conduction in (Bi0.53Sb0.47)2Te3 topological insulator. Nano Lett. 14, 5423-5429 (2014). Tian, J., Miotkowski, I., Hong, S. & Chen, Y. P. Electrical injection and detection of spin- polarized currents in topological insulator Bi2Te2Se. Sci. Rep. 5, 14293 (2015). Dankert, A., Geurs, J., Kamalakar, M. V., Charpentier, S. & Dash, S. P. Room temperature electrical detection of spin polarized currents in topological insulators. Nano Lett. 15, 7976-7981 (2015). Chen, J. et al. Gate-voltage control of chemical potential and weak antilocalization in Bi2Se3. Phys. Rev. Lett. 105, 176602 (2010). Steinberg, H., Laloë, J. B., Fatemi, V., Moodera, J. S. & Jarillo-Herrero, P. Electrically tunable surface-to-bulk coherent coupling in topological insulator thin films. Phys. Rev. B 84, 233101 (2011). Tang, H., Liang, D., Qiu, R. L. J. & Gao, X. P. A. Two-dimensional transport-induced linear magneto-resistance in topological insulator Bi2Se3 nanoribbons. ACS Nano 5, 7510-7516 (2011). 23. 24. 25. 18. 19. 20. 21. 22. 26. Wang, X., Du, Y., Dou, S. & Zhang, C. Room temperature giant and linear magnetoresistance in topological insulator Bi2Te3 nanosheets. Phys. Rev. Lett. 108, 266806 (2012). He, H. et al. High-field linear magneto-resistance in topological insulator Bi2Se3 thin films. Appl. Phys. Lett. 100, 032105 (2012). 27. 12 28. Wang, J. et al. Anomalous anisotropic magnetoresistance in topological insulator films. 29. 30. 31. 32. 33. 34. 35. Nano Res. 5, 739-746 (2012). Sulaev, A. et al. Electrically tunable in-plane anisotropic magnetoresistance in topological insulator BiSbTeSe2 nanodevices. Nano Lett. 15, 2061-2066 (2015). Yasuda, K., Tsukazaki, A., Yoshimi, R., Takahashi, K. S., Kawasaki, M. & Tokura, Y. Large unidirectional magnetoresistance in a magnetic topological insulator. Phys. Rev. Lett. 117, 127202 (2016). Avci, C. O., Garello, K., Ghosh, A., Gabureac, M., Alvarado, S. F. & Gambardella, P. Unidirectional spin Hall magnetoresistance in ferromagnet/normal metal bilayers. Nat. Phys. 11, 570-575 (2015). Kim, K. J. et al. Current-induced asymmetric magnetoresistance due to energy transfer via quantum spin-flip process. Preprint at https://arxiv.org/abs/1603.08746 (2016). Olejník, K., Novák, V., Wunderlich, J. & Jungwirth, T. Electrical detection of magnetization reversal without auxiliary magnets. Phys. Rev. B 91, 180402 (2015). Zhang, S. S. L. & Vignale, G. Theory of unidirectional spin Hall magnetoresistance in heavy- metal/ferromagnetic-metal bilayers. Phys. Rev. B 94, 140411 (2016). Manchon, A., Koo, H. C., Nitta, J., Frolov, S. M. & Duine, R. A. New perspectives for Rashba spin-orbit coupling. Nat. Mater. 14, 871-882 (2015). 36. Wang, Q. H., Kalantar-Zadeh, K., Kis, A., Coleman, J. N. & Strano, M. S. Electronics and optoelectronics of two-dimensional transition metal dichalcogenides. Nat. Nanotech. 7, 699-712 (2012). Li, H. D. et al. The van der Waals epitaxy of Bi2Se3 on the vicinal Si(111) surface: an approach for preparing high-quality thin films of a topological insulator. New J. Phys. 12, 103038 (2010). 37. Acknowledgments This work was partially supported by the ASTAR's Pharos Programme on Topological Insulators, Ministry of Education-Singapore Academic Research Fund Tier 1 (R-263-000- B47-112) and NSF Grant DMR-1406568. Author contributions P.H. and H.Y. planned the study. D.Z. fabricated devices. P.H. and D.Z. measured transport properties. Y.L., Y.W. and J.Y. helped characterization. S.S.-L.Z and G.V did the theory. All authors discussed the results. P.H., S.S.-L.Z, D.Z., G.V and H.Y. wrote the manuscript. H.Y. supervised the project. Additional information 13 Supplementary information is available in the online version of the paper. Reprints and permissions information is available online at www.nature.com/reprints. Correspondence and requests for materials should be addressed to H.Y. Competing financial interests The authors declare no competing financial interests. 14 Figure captions Figure 1 Schematic of the spin texture of surface states of a 3D topological insulator with hexagonal warping and conversion of spin current to charge current by applying an external magnetic field. a, Hexagonally warped energy dispersion for the surface states with Fermi surface lying in the conduction band. b, Hexagonally warped spin texture at the Fermi contour of the surface states. c, Schematics of the variation of the electron distribution along the k-axis parallel to the applied electric field E: δf1 (blue curve) and δf2 (yellow curve) are the corrections to the equilibrium distribution at the first and second order of the electric field, respectively; the solid arrows represent excess of electrons with their spins along the arrow direction, while the hollow arrows represent depletion of the same. d, When an electric field E is applied along a certain direction in k-space (dash- dotted line), a non-equilibrium spin current is generated at the second order of the electric field, due to spin momentum locking. e-f, When an external magnetic field is applied, the nonlinear spin current is partially converted into a charge current : a high resistance state is reached when the magnetic field is parallel (e) to the spin direction of the spin current, while a low resistance state is reached when it is antiparallel (f). Figure 2 First (Rω) and second (R2ω) harmonic magneto-resistances in three different geometries. Geometries of longitudinal magneto-resistance measurements with rotating H in the xy (a), zy (b) and zx (c) planes. Typical first (blue) and second (red) harmonic resistances measured for a 20 QL Bi2Se3 device under H = 9 T, T = 125 K and I = 0.55 mA along the direction. The solid lines are fits to the data. A vertical offset in R2ω was subtracted for clarity. 15 )(2sEJ)(2eEJKFigure 3 Detection of the hexagonally warped helical spin texture of TSS. a-c, Second harmonic resistance for all three scans for the devices with the current applied at an angle of 0 (a), 30 (b) and 60 (c) with respect to the direction, respectively. These scans were performed under H = 9 T, T = 60 K and I = 0.55 mA. Blue hexagons in a-c represent the Fermi surface of Bi2Se3. The red lines are along the direction and the black arrow denotes the current direction in k-space. d-f, The calculated angular dependences of R2ω corresponding to the three cases in a-c, respectively. The parameters used in the calculation are m/s, eVÅ3, eV, , H = 9 T, E = 100 V/cm, l = 100 µm and w = 20 µm. Figure 4 Quantitative comparison between the field and spin canting angles. a, The measured field canting angle as a function of the angle between the current direction and the line. b, The theoretical spin canting angle for a k-direction that coincides with the direction of the current. Figure 5 Current and magnetic field dependence of R2ω. a,b, Current (a) and magnetic field (b) dependence of the amplitude ΔR2ω in the zy scan at 60 K for a device with the current along . The solid black lines are fits to the data. 16 KK5105165256.0F2gKK Figure 1 17 eHigh resistanceLow resistancefcdabKM Figure 2 18 09018027036018101815182018251830 ()Rzx()-0.020.000.02Rzx2()zxyHθIzxyHθIzxyHϕIzxyHϕbac09018027036018111812181318141815 ()Rxy()-0.020.000.02Rxy2()0901802703601810181518201825 ()Rzy()-0.020.000.02Rzy2()edfxyscanzyscanzxscanFigure 3 19 -40-2002040KI30°M xy zy zx R2(m)090180270360-40-2002040 xy zy zxI60°KK R2(m),()090180270360-40-2002040I60°KKR2(m),() xy zy zx abcdfeExperimentTheory-40-2002040 xy zy zxI0°K R2(m)-40-2002040I0°K xy zy zx R2(m)-40-2002040KI30°M xy zy zxR2(m)Figure 4 20 ab0120240360-60-3003060 Canting angle Hc() ()Experiment0120240360-60-3003060Theory Canting angle sc()k () Figure 5 21
1504.01200	1	1504	2015-04-06T04:05:47	Temperature dependent screened electronic transport in gapped graphene	[ "cond-mat.mes-hall" ]	We report our theoretical calculations on the temperature and energy dependent electrical conductivity of gapped graphene within the framework of Boltzmann transport formalism. Since screening effects have known to be of vital importance in explaining the conductivity of gapless graphene therefore we first worked out the behaviour of the temperature dependent polarization function for gapped graphene as a function of wave vector and band gap, respectively. Polarization of gapped graphene has been compared with that of gapless graphene, bilayer graphene and 2DEG to see the effects of gap. It is found that the gapped graphene polarization function exhibits a strong dependence on temperature, wave vector and band gap and the effect translates to the conductivity of gapped graphene. The nature of conductivity in gapped graphene is observed to be non monotonic ranging from good to poor to semi conducting. We also find that the conductivity computed as a function of temperature by averaging over quasi-particle energy significantly differs from that computed at Fermi energy, suggesting that a notable contribution to temperature dependent conductivity is made by electrons close to the Fermi level.	cond-mat.mes-hall	cond-mat	Temperature dependent screened electronic transport in gapped graphene Digish K Patel,1, A C Sharma,1 and S S Z Ashraf*,2 1 Physics Department, Faculty of Science, The M.S. University of Baroda, Vadodara-390002, Gujarat, India 2 Physics Department, Faculty of Science, Aligarh Muslim University, Aligarh-202002, Uttar Pradesh, India Keywords Boltzmann equation, electronic transport, gapped graphene. e-mail [email protected] * e-mail [email protected] ** Corresponding author: e-mail [email protected] We report our theoretical calculations on the temperature and energy dependent electrical conductivity of gapped graphene within the framework of Boltzmann transport formalism. Since screening effects have known to be of vital importance in ex- plaining the conductivity of gapless graphene therefore we first worked out the behaviour of the temperature dependent polari- zation function for gapped graphene as a function of wave vector and band gap, respectively. Polarization of gapped graphene has been compared with that of gapless graphene, bilayer graphene and 2DEG to see the effects of gap. It is found that the gapped graphene polarization function exhibits a strong dependence on temperature, wave vector and band gap and the effect translates to the conductivity of gapped graphene. The nature of conductivity in gapped graphene is observed to be non mono- tonic ranging from good to poor to semi conducting. We also find that the conductivity computed as a function of temperature by averaging over quasi-particle energy significantly differs from that computed at Fermi energy, suggesting that a notable contribution to temperature dependent conductivity is made by electrons close to the Fermi level. 1 Introduction Graphene is a two dimensional novel material of many fascinating attributes. It is an intellectually stimulating object from both fundamental physics point of view and utility wise. The exceptional novel properties displayed by graphene make it a promising material for numerous applications in disparate fields [1]. One of the key features of single layer graphene (SLG) is its mass less Dirac type of low energy electron excitations having a linear energy dispersion relation; where is the Fermi velocity - which is independent of electron density in graphene and is about 300 times smaller than velocity of light in vacuum, and the subscript sk identifies the spin and wave vector of a state. The SLG is a gapless semiconductor with its valance and conduction bands touching at Dirac points. The absence of bandgap makes it challenging to create graphene-based transistors with large on/off ratios, which are required for logic applica- tions. This instigated research on possible ways to open a gap between valance and conduction band in graphene for de- vice applications. Graphene systems which have extensively been studied are single layer gapless graphene (SLG), single layer gapped graphene (SGG) and bilayer graphene (BLG). Recent studies have demonstrated that a gap between valance band and conduction band can be opened in different ways, such as: (i) graphene placed on suitable substrate like silicon carbide 2 (gap about 0.26 eV) [2-3] and boron nitride (gap about 53 meV) [4] , (ii) application of magnetic field to generate a dynamic gap [5], (iii) a small gap (~10-3 meV) is opened due to spin-orbit coupling or Rashba effect [6-7]. On opening of gap, the relativistic massless Dirac particle dispersion relation changes to The extra intrinsic mass, , is introduced due to a break in the graphene’s sublattice symmetry. Here is the chemical potential at absolute zero temperature with ( is band parameter) and , is normalised gap having values between 0 and 1 when Fermi level is considered to be lying above energy gap. In contrast to chiral lin- ear energy dispersion of SLG, the BLG displays an energy dispersion that is chiral parabolic; fective mass, . , with interlayer hopping matrix element, [8] and Fermi velocity , where ef- The charge transport in SLG, BLG and SGG display novel chirality that has attracted much theoretical and experi- mental attention [9-16]. Charge transport in graphene systems sharply differ from that of 2D electron gas (2DEG) ob- served in doped semiconductor heterostructures [15-16]. Though the energy dispersion in BLG is similar to that of 2DEG but the carrier density dependence of conductivity in BLG seems to manifest the same linear behaviour as that ob- served in SLG. Various theories of charge transport in graphene systems appear to suggest that considering a scattering mechanism based on screened charged impurities; one can obtain from the Boltzmann equation to give that agrees with the experimental results. The intrinsic parameters that goven and electron-impurity scattering rate in SLG, BLG and SGG are quasi particle energy , temperature and carrier concentration , and additionally energy gap ( ) in BLG and SGG. Theoretical understanding of requires detailed investigations on how polarization function and depends on , and . There have been several calculations of the polarization function and its properties for SLG & BLG [11-12, 15-20]. Also the conductivity as a function of various influencing parameters for SLG and BLG has been reported in several publications [9-16, 15-17]. In this paper, we report our theoretical investigations on SGG polarization function and SGG conductivity and its de- pendence on various governing intrinsic parameters. The conductivity in SGG has been dealt within the Boltzmann transport theoretical approach. Most of the existing theoretical studies on SGG have been performed at zero temperature by calculating static and dynamical polarization functions at zero temperature and zero magnetic field [21-28]. It has been shown that the screening of charged impurities at large distances in SGG differs from that in SLG by slower decay- ing of Friedel oscillations ( instead of ), similar to that observed in conventional 2D system. As implied above, the main motive of our work has been to investigate how the polarization function of SGG and hence the conductivity of gapped graphene varies with temperature and gap. A strong temperature dependent conductivity has been reported in high mobility SLG samples [15- and references there in]. Also, at very low temperatures, the Fermi function behaves like step function and hence there is no significant difference between conductivities computed by averaging over and at Fermi energy. However, as temperature increases, conductivity computed by averaging over significantly differs from that computed at Fermi energy. The conductivity computed by averaging over and not the conductivity at Fermi energy should be compared with experimental results. The paper is organized as follows: Section 2 briefly discusses the formalism and section 3 contains a detailed discus- sion on temperature dependent polarization function of SGG. In section 4 we report the results on conductivity of SGG as 3 a function of temperature, energy bandgap, carrier density and coupling constant with their discussion, and finally we conclude our findings in section 5. 2 Essential formalism As outlined in the introduction, in the calculation of conductivity at , all the charge carriers are at the Fermi level and the averaging of the relaxation time can be ignored. At finite temperature the carrier concentration is expressed in terms of distribution function therefore the relaxation time should be taken as the average of the relaxation time of indi- vidual charge carriers. Therefore it is not appropriate to ignore the averaging, since at Fermi level can significantly dif- fer from particularly at moderate and high temperatures. In terms of the average relaxation time , the temperature dependent conductivity for SGG within the Boltzmann transport formalism can be expressed as [9]; (1) where is the average value of over all possible values of quasiparticle energy, . The energy averaged finite tem- perature scattering time is given by [9], . (2) Where is the Fermi distribution function, is the finite temperature and energy dependent scattering time of an electron scattered by disorder or statically screened Coulomb potential, given by [15] . (3) Where is the concentration of impurity centers, is the scattering angle between the scattering in and out wave vectors and , is the norm square of matrix element of scattering potential averaged over configurations of impurities expressed in terms of charge impurity potential, . (4) Where, is the two dimensional Fourier transform of the bare charge impurity Coulomb potential, where is the electronic charge, is the average background dielectric constant ( 5.5 for graphene placed on SiC with other 4 side being exposed to air), is the momentum transferred to a scattered electron, is the temperature dependent static dielectric function, which within linear response theory is given by . Substituting Eqs. (4) & (5) into Eq. (3), and then simplifying the integrals, we obtain . (5) (6) Where, the factor which is invariably present in the Boltzmann transport equation weighs the amount of backward scattering of the electrons by impurity. In case of SLG, factor reduces to that suppresses the large angle scattering in SLG. However, in case of SGG there is breaking of the sub lattice symmetry because of the bandgap, which contributes to the large angle scattering with increasing band gap. At lower temperatures, the derivative of Fermi distribution function, behaves like step function and therefore it is assumed that for all practical purposes there should not be any significant difference in temperature dependent conductiv- ity calculated at Fermi energy or with the use of Eq. (1). Rest of the paper, we used following scaled parameters: , , , , (dimensionless coupling constant, which has value for graphene sheets on SiC (BN) substrate) and (the scaled carrier concentration). The normalized conductivity at finite for SGG is then obtained from the ratio; where has been computed by taking in Eq. (6). 3 Temperature dependent polarizability of gapped graphene The polarization function is an important quantity in calculating scattering rate of screened electron gas. The polarization function for the cases of both doped and undoped gapped graphene has been calculated at zero temperature [24-25, 27] as well as at finite temperature [21] in the recent past and the effect of band gap on the ground state properties of Dirac electrons in a doped graphene at zero temperature has been studied [24]. We present here the detailed analysis of the numerically computed temperature dependent polarization function for SGG and its comparison with that of SLG, BLG and 2DEG. Earlier, analytical results only in the asymptotic limits have been presented and detailed analysis of the polarization function at all -values as well as temperature and gap values is missing [21]. The temperature dependent polarization function for SGG in the random phase approximation (RPA) is given by [21], . (7) Where, 5 (8) and . (9) Here is the finite temperature chemical potential determined by the conservation of the total electron density, defined by . Where , and is given as; . (10) (11) With the use of the scaled variables define in section 2, Eq. (7) using Eqs. (8) & (9) takes the form; , (12) where . The above equation yields the prior reported results [15, 17, 22] for polarizability at for SGG and at finite temperature for SLG when . The computed normalized polarizability as a function of from Eq. (12) is plotted in Fig. 1(a) at for different gap values and in Fig. 1(b) at for different temperature values. As can be seen from Fig. 1(a), the effect of introducing gap in electronic spectrum at is almost unnoticeable for , 6 which means that the intraband and interband transitions almost cancel at zero temperature and the total polarizability is static, similar to that in SLG [18]. But when (large momentum transfer regime); (i) the magnitude of polarizabil- ity versus wave vector curve decreases on increasing the gap, at all -values, and (ii) for the behaviour of po- larizability versus wave vector curve for SGG resembles to a great extent with that of 2DEG polarizability which is in stark contrast to SLG where the polarizability increases for . This means that in SGG the interband transitions dominate over the intraband transition for large wavevectors, suggesting that the scattering by the screened coulomb po- tential is much reduced due to enhanced screening in this limit. This also implies that for , polarizability (SGG) shows relativistic characteristics while at it reflects the non-relativistic nature of 2DEG caused by breaking of sub- lattice symmetry. Unlike the case of and where the polarizability stays constant for the polariza- bility shows a monotonically increasing behaviour for at for values even in the range of similar to the behaviour of SLG as is evident from Fig. 1(b) and also corroborated from Fig. 1(h). Also, the magnitude of polarizability versus wave vector curve enhances on increasing values of . This is due to the enhanced electronic transi- tions to the conduction band from valance band with the increase of temperature. Variation of polarizability with wave vector at larger gap value and at low temperatures are shown in Figs. 1(c) & 1(d). The sharp decline seen in polarizability from a constant value at and , changes to a smooth variation on increasing the tem- perature. This abrupt decline in polarizability at is associated with Friedel oscillations, which at finite and reasona- bly higher temperatures wash out. (a) (b) (c) (d) (e) (f) (g) (h) 7 Figure 1 Wave vector dependent polarizability (a) for different values of gap at and (b) for different values of temperature at . Figures (c) and (d) show the wave-vector dependence of polarizability for different values of tem- 8 perature at gap values of and respectively. The curves in this case are similar to polarization function for a Si(001) inversion layer with electron per cm2 [29]. Temperature dependent polarizability for different gap val- ues at; (e) and (f) . Comparative plot of polarization function of SLG, BLG, SGG & 2DEG at zero and finite temperature are shown in Figs. 1(g) and 1(h), respectively. Here . The computed polarizability as a function of temperature using Eq. (12) for different values of at two values of are plotted in Figs. 1(e) & 1(f). A slight change (decrease) in nature of polarizability versus wave vector curve can be noticed on enhancing from 1 to 2. Also, as can be noticed from the figures, the polarizability first declines with temperature and after hitting a minimum increases almost linearly for all nonzero values of . A reverse trend observed at higher temperatures compared to that at low temperature regime is in- dicative of phase transition taking place in SGG. Finally in Figs. 1(g) & 1(h) the comparative plot of the polarization function of SLG, SGG, BLG & 2DEG are shown at zero and finite temperature , respectively. The interplay of linear energy band dispersion relation, chirality, bandgap and temperature endow SGG with overall strange screening properties which are a mixture of SLG, BLG and 2DEG screening properties. It is known that SLG also exhibits strange screening properties which arise because of a combination of metallic screening due to intraband transitions and insulat- ing screening due to interband transitions; that all ultimately stems from the chiral relativistic dispersion relation [18]. 4 Temperature-dependent Conductivity of gapped graphene This section reports our results on energy averaged conductivity as a function of temperature calculated using Eq. (1), which is then compared with that calculated at Fermi energy. It is found that when we incorporate the temperature de- pendence in dielectric function formalism, we observe significant difference between these two conductivities as function of temperatures. We thus find that it is grossly misleading to calculate temperature dependent conductivity at Fermi en- ergy for comparison with experimental results. Our computed numerical results on normalized conductivity as a function of temperature, with the use of and , are plotted in Figs. 2(a) to 2(d) for different values of . The figures clearly demonstrate that the difference between two values of conductivity is insignificant only for temperatures very close to zero and the difference grows with increasing temperatures. For ), using initially remains almost constant and then it increases with , which is not the case when used to compute . A phase transition on changing temperature is also indicated by conductivity computed using energy averaged scattering rate for higher values of , as can be seen from Figs. 2(a) to 2(d). Elaborated behaviour of y verses near the transition point is shown insets of figures. The green dot of curve indicates turning points of set curves. (a) (b) (c) (d) 9 Figure 2 Conductivity as a function of temperature calculated at Fermi energy (solid black line) and at average energy (dashed brown line) for different gap values; (a) , (b) , (c) and (d) . Here we used coupling constant, (for SiC). The green spot in insets show minimum value of conductivity. Change in nature of conduction, from poor conductor to semiconductor, takes place when for . Strikingly opposite behaviour of conductivity with temperature in low and high temperature regimes is indicative of phase transition that can be obtained by selecting appropriate values of and in SGG. Curves exhibit a minimum at and the change from poor metallic nature for to good conductor nature at in low temperature regime. The behaviour of the conductivity can be understood as follows; the increase in band gap reduces the conductivity in low temperature regime due to electron has insufficient excitation energy to jump from valance band to conduction band but increase in temperature increases the excitations and hence the raises the magnitude of conductivity. The Fig. 3(a) showing the variation of conductivity with bandgap at different temperatures, corroborates the above behaviour. In Fig. 3(b) shows the variation of conductivity as a function of temperature for which corre- sponds to graphene on BN (Boron Nitride) substrate. The higher values of indicate a strong coupling in terms of elec- 10 tron-electron interaction. A slight change (decrease) in nature of conductivity versus temperature curve can be noticed on enhancing coupling constant, , from 1 to 2. (a) (b) Figure 3 Conductivity (a) as a function of band gap for different values of temperature , using , (b) as a function of temperature calculated at average energy for different gap values gap with coupling constant, . 5 Conductivity as a function of carrier concentration Our computed conductivity using Eq. (1) as a function of carrier concentration, at different temperatures is plotted in Figs. 4(a) & 4(b) for two values of . Almost a linear enhancement of conductivity with is seen for both values of , similar to the experimentally observed behaviour in case of SLG [9]. Also the magnitude of conductivity in- creases with the increase in temperature. The Figs. 4(c) & 4(d) depict the effect of variation of conductivity with band gap at and , respectively, for three values of carrier concentration . As can be no- ticed from the figure; this reconfirms that the enhancement in the carrier concentration and temperature increases the conductivity whereas the increase in band gap reduces the conductivity. 11 Figure 4 Conductivity as (i) a function of for different temperatures of with gap values; (a) , (b) , (ii) a function of bandgap for at (c) and (d) . 6 Conclusions We computed temperature dependent polarization function and the conductivity of doped gapped graphene considering electron-impurity scattering as the dominant source and neglecting all other scattering phenomena (e.g. electron-phonon, electron-electron), within the Boltzmann transport theory. The interplay of linear energy band dispersion relation, chiral- ity, bandgap and temperature endow SGG with overall strange screening properties which are a mixture of SLG, BLG and 2DEG screening properties. We find that the nature of conduction in gapped graphene changes from good to poor semiconducting on varying values of and Our results show that conductivity behaves like that of poor metal when and very good conducting metallic behaviour when whereas it displays an insulating behav- iour at higher temperatures for both gapless and gapped graphene. Metallic and semiconducting behaviour at low temperatures and high temperatures is indicative of phase transition that can occur by selecting appropriate values of and in SGG. We also notice that the metallic nature can be enhanced by increasing the coupling constant value. We 12 also obtained numerical results of -dependent conductivity as a function of carrier concentration which shows linear be- haviour as observed experimentally, and also shows an increase in magnitude with the increase in temperature and de- crease in magnitude with the increase in bandgap. Our results on conductivity of gapped graphene are of significance as any experimental work on graphene begins with a characterization of its electrical conductivity. References [1] R. S. Edwards and K. S. Coleman, Nanoscale 5, 38 (2013). [2] F. Varchon, R. Feng, J. Hass, X. Li, B. Ngoc Nguyen, C. Naud, P. Mallet, J. Y. Veuillen, C. Berger, E. H. Conrad, and L. Magaud, Phys. Rev. Lett. 99, 126805 (2007). [3] 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, A. Lanzara, Nature Materials 6, 770 (2007). [4] G. Giovannetti, P. A. Khomyakov, G. Brocks, P. J. Kelly, J. van den Brink, Phys. Rev. B 76, 073103 (2007). [5] V. P. Gusynin, S. G. Sharapov, J. P. Carotte, Int. J. Mod. Phys. B 21, 4611 (2007). [6] Y. Yao, F. Ye, X. L. Qi, S. C. Zhang, Z. Fang, Phys. Rev. B 75, 041401 (2007). [7] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). [8] R. C. Tatar, S. Rabii, Phys. Rev. B 25, 4126 (1982). [9] E. H. Hwang, S. Adam, S. D. Sarma, Phys. Rev. Lett. 98, 186806 (2007). [10] K. Nomura, A. H. Mac Donald, Phys. Rev. Lett. 98, 076602 (2007). [11] T. Ando, J. Phys. Soc. Jpn. 75, 074716 (2006). [12] K. Nomura, A. H. Mac Donald, Phys. Rev. Lett. 96, 256602 (2006). [13] V. V. Cheianov, V. I. Falko, Phys. Rev. Lett. 97, 226801 (2006). [14] Y. Zheng and T. Ando, Phys. Rev. B 65, 245420 (2002). [15] E. H. Hwang, S. D. Sarma, Phys. Rev. B 79, 165404 (2009). [16] M. Lv and S. Wan, Phys. Rev. B 81, 195409 (2010). [17] B. Wunsch, T. Stauber, F. Sols, F. Guinea, New J. Phys. 8, 318 (2006). [18] E. H. Hwang, S. Das Sarma, Phys. Rev. B 75 (2007) 205418 [19] R. Sensarma, E. H. Hwang, S. D. Sarma, Phys. Rev. B 82, 195428 (2010). [20] K. Shiyuza, Phys. Rev. B 75, 245417 (2007). [21] J. N. Zhang, Phys. Scr. 83, 035002 (2011). [22] P. K. Pyatkovskiy, V. P. Gusynin Phys. Rev. B 83, 075422 (2011). [23] O. V. Gamayun, Phys. Rev. B 84, 085112 (2011). [24] A. Qaiumzadeh, R. Asgari, Phys. Rev. B 79, 075414 (2009). [25] P. K. Pyatkovskiy, J. Phys. Condens. Matt. 21, 025506 (2009). [26] A. Qaiumzadeh, R. Asgari, New. J. Phys. 11, 095023 (2009). [27] V. N. Kotov, V. M. Pereira, B. Uchoa, Phys. Rev. B 78, 075433 (2008). [28] A. Scholz, J. Schliemann, Phys. Rev. B 83, 235409 (2011). [29] F. Stern, Phys. Rev. Lett. 44, 1469 (1980).
1605.08133	2	1605	2016-08-02T08:09:41	Broadband Transverse Electric Surface Wave in Silicene	[ "cond-mat.mes-hall" ]	Transverse electric (TE) surface wave in silicine is theoretically investigated. The TE surface wave in silicene is found to exhibit better characteristics compared with that in graphene, in terms of a broader frequency range and more confinement to the surface which originate from the buckled structure of silicene. We found that even undoped silicene can support the TE surface wave. We expect to obtain the similar characteristics of the TE surface wave in other two-dimensional materials that have slightly buckled honeycomb lattice.	cond-mat.mes-hall	cond-mat	Broadband Transverse Electric Surface Wave in Silicene M. Shoufie Ukhtary,1, a) Ahmad R. T. Nugraha,1 Eddwi H. Hasdeo,1 and Riichiro Saito1 Department of Physics, Tohoku University, Sendai 980-8578, Japan (Dated: 13 November 2018) Transverse electric (TE) surface wave in silicine is theoretically investigated. The TE surface wave in silicene is found to exhibit better characteristics compared with that in graphene, in terms of a broader frequency range and more confinement to the surface which originate from the buckled structure of silicene. We found that even undoped silicene can support the TE surface wave. We expect to obtain the similar characteristics of the TE surface wave in other two-dimensional materials that have slightly buckled honeycomb lattice. 6 1 0 2 g u A 2 ] l l a h - s e m . t a m - d n o c [ 2 v 3 3 1 8 0 . 5 0 6 1 : v i X r a PACS numbers: 72.20.Pa,72.10.-d,73.50.Lw Surface electromagnetic waves, or simply surface waves are electromagnetic (EM) waves that propagate on the surface of a material1. Surface waves recently have at- tracted a lot of interest, because of their capability to transport the EM energy on the surface1 -- 6. There are two kinds of surface waves based on their polarizations; the transverse magnetic (TM) and transverse electric (TE) surface waves. In the case of TM surface wave, the com- ponent of magnetic field is transverse to the propagation direction, while the electric field has a component paral- lel to the propagation direction. The TM surface wave that also refers to a surface plasmon, can be seen as an electric dipole wave on the surface of material due to spatial perturbation of charge density7,8. On the other hand, the TE surface wave has the component of electric field transverse to the propagation direction while the magnetic field has a component parallel to the direction of propagation. The TE surface wave can be seen as a magnetic dipole wave on the surface of material due to the self-sustained surface current oscillation7,8. It is important to note that the radiation loss of mag- netic dipole is much smaller than that of electric dipole9. Therefore, the TE surface wave can propagate longer than TM surface wave10,11, which makes the TE surface wave desirable for the transporting EM energy over long distance7,11. However, the TE surface wave cannot ex- ist on the surface of an conventional bulk metal because condition for generating the TE mode is limited which means that the induced surface current is not available in the conventional bulk metal1,2,7,12. Some efforts have been made for designing artificial materials so that the TE surface wave can be generated, such as metamate- rials and a cluster of nanoparticles, which are generally complicated, hence making them less viable and accessi- ble3,7,11,13. The difficulties of generating the TE surface wave can be alleviated by using two-dimensional (2D) mate- rials such like graphene, which is a monolayer of car- bon atoms arranged in honeycomb lattice14,15. Mikhailov and Ziegler have shown that, when the imaginary part of optical conductivity of 2D material is negative (posi- a)Electronic mail: [email protected] FIG. 1. (a) Honeycomb lattice of silicene. (b) Silicene lattice from side view. Sublattice A and B are separated vertically by d. Sublattice A (B) is depicted by red (green) atom. (c) Sil- icene is sandwiched between two dielectric media. TE surface wave propagates on the surface of silicene with wave vector q. The electric field Ey is perpendicular to q. tive), the TE (TM) surface wave can propagate on the surface of the 2D materials15. Due to the presence of the Dirac cone in its electronic structure, the imaginary part of optical conductivity of graphene can be nega- tive at a certain frequency range. This is in contrast to usual 2D electron gas systems, which have a positive imaginary part of optical conductivity10,15. This unusual property has also enabled graphene to have the TE sur- face wave8,10,15,16. However, it was predicted that the TE surface wave in doped graphene may only exist for a narrow frequency range of 1.667EF < ω < 2EF 8,10,15, where EF is the Fermi energy. Moreover, the TE surface wave in graphene is less confined in the direction perpen- dicular to the surface in comparison with the TM surface wave10,15. In this letter, we propose that silicene is a better 2D material rather than graphene to support the TE sur- face wave. Silicene is a monolayer of silicon atoms ar- ranged in honeycomb lattice and the stable structure of silicene is not purely planar, but slightly buckled17 -- 20, i.e., the two sublattices are separated by vertical distance d = 0.46 A due to the sp3-like hybridization20,21. The schematic structure of silicene can be seen in Figs. 1(a) and 1(b). The buckling of the atoms creates potential dif- ference between two sublattices when an external electric field is applied in the direction perpendicular to the sur- face17 -- 20. The induced potential difference, along with the non-negligible spin orbit (SO) coupling in silicene, will give a tunable energy gap17,18,20,22. We will show that the tunable energy gap of silicene affects a unique optical conductivity and the properties of TE surface wave which makes it a key difference from graphene. Suppose that a silicene layer, or generally any mono- dxyxz(a)AB(b)(c)HxHzEyqε1ε22 points (hexagonal corners of Brillouin zone)17 -- 20. How- ever, we should consider the following two factors: (1) the SO coupling in silicene is much larger than that of graphene17,18,22, and (2) the potential difference between the sublattices A and B can be induced by an external 17 -- 20. The Hamiltonian of perpendicular electric field Ez silicene can be written in the following matrix form, 2 ςη∆SO + 1 vF(ηkx + iky) 2 ∆z vF(ηkx − iky) 2 ∆z 2 ςη∆SO − 1 1 , (2) (cid:20) − 1 (cid:98)Hςη = (cid:21) (cid:114) where vF is the Fermi velocity of electron and it is 5.5 × 105m/s for silicene22. The Hamiltonian is spin and valley dependent, labeled by ς = +1(−1) for spin up (spin down) and η = +1(−1) for K (K(cid:48)) valley. ∆SO ≈ 3.9 meV represents the SO coupling for silicene20 and ∆z = eEzd denotes the potential difference between sublattices. d = 0.46 A denotes the vertical distance between the A and B atoms shown in Fig. 1(b). The eigenvalues of Eq. (2) are expressed by sςη(k) = (−1)s+1ςη(k), with s is 1 and 2 for the conduction and valence band, respectively. ςη(k) is the energy dispersion for electron with ς spin and at η valley, which is given by (cid:113) ςη(k) = (vFk)2 + 1 4 ∆2 ςη , (3) −1 k2 x + k2 y and ∆ςη(∆z) = ∆z − ςη∆SO de- where k = notes the energy gap which is tunable by applying the Ez up to 2.6 V A where the structure of silicene becomes unstable23. The energy gap is defined as the energy sep- aration from the top of the valence band to the bottom of the conduction band with the same spin sign. There are only two distinct values of ∆ςη(∆z) for four possi- ble combination of ∆ςη, since ∆++(∆z) = ∆−−(∆z) and ∆+−(∆z) = ∆−+(∆z). The optical conductivity σ(ω) of silicene can be ob- tained by the Kubo formalism for current-current cor- relation function24,25. The electron scattering is ignored here. Following the derivation of graphene's conductivity by Falkovsky and Varlamov, σςη can be expressed by24, (cid:90) (cid:40)(cid:88) (cid:40)(cid:90) s d2k σςη(ω) = − ie2 4ωπ2 + iωe2 2π2 2 (cid:41) d2k[vx ss(k)]2 df [sςη] dsςη f [1ςη(k)] − f [2ςη(k)] 2ςη(k) − 1ςη(k) (cid:41) 12(k)vx vx 21(k) 2ω2 − [2ςη(k) − 1ςη(k)]2 , (4) where f [] is the Fermi distribution function and vss(cid:48) U−1 (∂Hςη(k)/∂k) U in the x direction, where U is the is the matrix element of velocity matrix (cid:98)v(k) = unitary matrix which diagonalize Hςη. The(cid:98)v(k) matrix is explicitly given as follows: FIG. 2. Electronic energy dispersions of silicene for [(a)-(c)] K and [(d)-(f)] K' valleys for several ∆z's (2∆SO,4∆SO, and 8∆SO). The solid (dash-dotted) lines correspond to spin up (down) electron dispersion. Positions of EF = 7.8 meV are indicated by the horizontal dotted lines. layer 2D material, in the x − y plane is sandwiched be- tween two dielectric media with dielectric constant ε1 and ε2 as shown in Fig. 1(c). The dispersion of the TE surface wave can be obtained by employing the Maxwell equations with boundary conditions of TE wave near the surface of the layer. Here we assume that the 2D mate- rial is negligibly thin and it is characterized by its opti- cal conductivity σ which will appear as a surface current density in the boundary conditions for magnetic field as shown below. The TE surface wave has an electric field Ey in the y direction and the wave vector q in the x di- rection. There are two magnetic field components Hx and Hz in TE surface wave as shown in Fig. 1(c). Due to the confined nature of the surface wave, the EM fields should decay in the direction perpendicular to the sur- face (z). Thus, we can write the magnetic fields in the x (x, z) = A1 exp(−κ1z + iqx) and media 1 and 2 as H (1) H (2) x (x, z) = A2 exp(κ2z +iqx), respectively. The electric field in the k-th medium (k = 1, 2) is obtained through y (x, z) = −iωµ0 E(k) x (x, z)dz. The decay constant κk is given by κk =(cid:112)q2 − εk(ω/c)2. The boundary con- (ε1 = ε2 = 1, thus κ1 = κ2 = (cid:112)q2 − (ω/c)2 ≡ κ), we ditions at the surface are (i) E(2) y = 0 and (ii) (H (2) x ) = J, where J = σEy is defined as sur- face current density. Employing the boundary conditions and assuming that the two dielectric media as vaccum (cid:82) H (k) x − H (1) y − E(1) obtain the TE surface wave dispersion15,16, 2 − iσ(ω)ωµ0 κ = 0 . (1) Since ω is a positive value, Eq. (1) requires a negative value of Im σ. Next, we derive the σ(ω) of silicene. Similar to graphene, the behavior of electrons at low-energy can be described by the Dirac Hamiltonian near the K and K(cid:48) ϵsςη/EFK-valley(a)(b)(c)Δz = 2ΔSOSpinupSpin downΔ++Δ-+Δz = 4ΔSOΔ++Δ-+Δz = 8ΔSOΔ++Δ-+K-valleyK-valleyK'-valleyħvFkx/EFϵsςη/EFΔz = 2ΔSOΔ--Δ+-Δz = 4ΔSOΔ--Δ+-Δz = 8ΔSOΔ--Δ+-K'-valleyK'-valleyħvFkx/EFħvFkx/EF(d)(e)(f)-4-2024-4-2024-4-2024-4-20246-6-4-20246-6(cid:34) (cid:98)v(k) = v2 Fk ςη (cid:98)x cos θ +(cid:98)y sin θ Z+{(cid:98)xη(Γ+ + A+) −(cid:98)yi(B+ + I+)} −Z−{(cid:98)xη(−Γ− + A−) −(cid:98)yi(B− − I−)} −(cid:98)x cos θ −(cid:98)y sin θ (cid:35) , 3 (5) where we define β± = ςη ± 1/2∆ςη, Z± = (β±/β∓)1/2, Γ± = ∆ςη cos θ/β±, A± = i2ςη sin θ/β±, B± = 2ςη cos θ/β±, and I± = i2∆ςη sin θ/β±. Here θ is the angle between kx and ky, while vx nm denotes the x- component of the n − m element of (cid:98)v matrix. The first (second) term in Eq. (4) corresponds to the intraband (interband) conductivity, which is later labeled as σA ςη (σE ςη) . By using Eqs. (3) and (5), we can calculate σ in Eq. (4) for silicene at T = 0 K17,18. Here, σ is the total conduc- tivity for both spin and valley degrees of freedom. For simplicity, we fix the EF = 2∆SO = 7.8 meV, and vary the ∆z. Then, we get σ as follows (cid:88) ςη σ(ω, ∆z) = σA ςη(ω, ∆z) =i 4E2 e2 16π EFω (6) ςη(ω, ∆z) + σE Θ [2EF − ∆ςη(∆z)] (7) (cid:8)σA F − [∆ςη(∆z)]2 1 + (cid:32) 1 + (cid:32) (cid:12)(cid:12)(cid:12)(cid:12)ω + g(∆z) ςη(ω, ∆z)(cid:9) , (cid:33)2 Θ [ω − g(∆z)] (cid:33)2 (cid:12)(cid:12)(cid:12)(cid:12) + i e2 [∆ςη(∆z)]2 82πωg(∆z) ω − g(∆z) ∆ςη(∆z) ω ∆ςη(∆z) ω e2 16π , σE ςη(ω, ∆z) = e2 16 − i × ln (8) where Θ(x) is the Heaviside function and g(∆z) = max[2EF, ∆ςη(∆z)]. If we set ∆ςη = 0, we get the optical conductivity of graphene10,15. In Fig. 2, we plot the electron energy dispersions for K and K' valleys based on Eq. (3) for several ∆z's. In varying ∆z, we choose three cases for both the K and K' valleys depending on the position of EF relative to the energy gap, which are shown in Fig. 2. The first case is ∆z = 2∆SO , in which EF is higher than bottoms of the two conduction bands for spin up and spin down (EF > ∆++/−− and ∆−+/+−) [Figs. 2(a) and 2(d)]. The second case is ∆z = 4∆SO, in which EF lies between two bottoms of the conduction bands (∆++/−− < EF < ∆−+/+−) [Figs. 2(b) and 2(e)] and the third case is ∆z = 8∆SO, in which EF exists in energy gaps [Figs. 2(c) and 2(f)]. In Fig. 3, we plot the optical conductivity σ of sil- icene as a function of frequency where the solid and dashed lines are the imaginary and real parts of σ, re- spectively, for the three ∆z. We also plot the σ of graphene in black lines for a comparison. The loga- rithmic singularities in Im σ in Eq. (8) correspond to the lowest excitation energies for interband transitions FIG. 3. Optical conductivity (σ) of silicene for three different ∆z values and compared with that of graphene. The solid lines represent the imaginary part of σ and the dashed lines represent the real part of σ. Position of ω = 2EF is pointed at f = 3.77 THz. of electrons between energy bands having the same spin directions and the same valleys. Im σ is singular for any frequency which satisfies condition ω = g(∆z). Since there are two distinct values of ∆ςη(∆z), there are two possible singularity points, ω1 = 2EF/ and ω2 = ∆−+/ if ∆++/2 < EF < ∆−+/2 [∆z = 4∆SO] or ω1 = ∆++/ and ω2 = ∆−+/ if EF < ∆++/2 and ∆−+/2 [∆z = 8∆SO]. When EF > ∆−+/2 and ∆++/2 there is only one singularity point at ω = 2EF/. Since σ of silicene depends on ∆z, σ of silicence can be tuned not only by EF but also by Ez. As mentioned in Eq. (1), the negative value of Im σ correspond to the condition for TE surface wave. The TE surface wave cannot exist for the region that Im σ > 0. In the following discus- sion, we call the frequency range of Im σ < 0 as the TE frequency range. Furthermore we focus only on the frequency range where Re σ = 0 in which the TE sur- face wave is not damped15. For graphene (∆ςη = 0), the TE frequency range is fixed at 1.667EF < ω < 2EF (3.14 < f < 3.77 THz), which reproduces the previous results10,15. −1 In general, the TE frequency range in silicene is wider than that in graphene for the same EF and it is tun- able by ∆z as shown in Fig. 3. For example, for ∆z = 2∆SO (Ez = 16.96 mVA ), the TE frequency range lies within 1.4EF < ω < 2EF (2.64 < f < 3.77 THz). By increasing ∆z, ∆ςη increases. From Eq. (7)-(8) we know that increasing ∆ςη not only makes Im σE ςη more negative, but also reduces Im σA ςη whose value is al- ways positive15. Altogether, Im σ decreases, hence the TE frequency range becomes wider when we increase ∆z. The Im σA ςη can be suppressed when ∆ςη > 4∆SO, or the Fermi level is located in ∆ςη [Figs. 2(b) -- (c)]. 024681012σ (10-4S)Graphene 8ΔSOEF = 7.8 meVħω = 2EFFrequency (THz) 4Δ SOΔz = 2Δ SORe σIm σ-3-2-1012344 graphene, 1/κ = 13994 µm, while in case of silicene, 1/κ = 2906.2 µm for ∆z = 2∆SO, 1/κ = 1747.7 µm for ∆z = 4∆SO, and 1/κ = 3146.7 µm for ∆z = 8∆SO. In the case of ∆z = 8∆SO, we might get a larger 1/κ. This is because Im σ is singular at higher frequency, which makes 1/κ for ∆z = 8∆SO slowly diverge. By solving Eq. (1) for λ = 2π/q, we can define the dif- ference between the wavelength of TE surface wave λ and the wavelength of freely propagating EM wave in vacuum λ0 = 2πc/ω as ∆λ = λ−λ0 . In the inset of Fig. 4 we plot ∆λ as a function of frequency for graphene and silicene with ∆z = 4∆SO. We can see that ∆λ is sufficiently small which means that λ is almost the same as λ0 (3 THz corresponds to λ0 = 100µm). However, ∆λ for sil- icene is more negative compared with that for graphene, which is almost zero. Negative ∆λ means that there is shrinkage of the wavelength of TE surface wave which is the preferable feature of surface wave since more infor- mation can be compressed in the wave. From the inset of Fig. 4, we can see more shrinkage of the wavelength in silicene compared with that in graphene. In conclusion, silicene is theoretically proved to be a versatile platform for utilizing TE surface wave. We have shown that silicene supports the TE surface wave propagation and it exhibits more preferable surface wave properties compared with those of graphene, such as the tunable broadband frequency and smaller confinement length. The TE surface wave in silicene is tunable by the Fermi energy as well as by the external electric field. These characteristics originate from the two-dimensional buckled honeycomb structure. M.S.U. and E.H.H. are supported by the MEXT schol- arship. A.R.T.N. acknowledges the Leading Graduate School Program in Tohoku University. R.S. acknowledges MEXT (Japan) Grants No. 25107005 and No. 25286005. 1W. T. Hill III and C. H. Lee, Light-Matter Interaction (Wiley- VCH, Weinheim, 2007). 2S. A. Maier, M. L. Brongersma, P. G. Kik, S. Meltzer, A. A. G. Requicha, and H. A. Atwater, Adv. Mater. 13, 1501 (2001). 3H. Liu, D. A. Genov, D. M. Wu, Y. M. Liu, J. M. Steele, C. Sun, S. N. Zhu, and X. Zhang, Phys. Rev. Lett. 97, 243902 (Dec 2006). 4R. F. Oulton, V. J. Sorger, D. A. Genov, D. F. P. Pile, and X. Zhang, Nature Photon. 2, 496 (2008). 5A. Vakil and N. Engheta, Science 332, 1291 (2011). 6M. Jablan, H. Buljan, Y. Yin, and M. Soljaci´c, Phys. Rev. B 80, 245435 (2009). 7Z. Sun, X. Zuo, T. Guan, and W. Chen, Opt. Express 22(4), 4714 (2014). 8S. Menabde, D. Mason, E. Kornev, C. Lee, and N. Park, Sci. Rep. 6, 21523 (2016). 9J. D. Jackson, Classical Electrodynamics (Wiley-VCH, Wein- heim, 1999). 10X. Y. He and R. Li, IEEE J. Sel. Top. Quantum Electron. 20, 62 (2014). 11N. Liu, S. Mukherjee, K. Bao, Y. Li, L. V. Brown, P. Nordlander, and N. J. Halas, ACS Nano 6, 5482 (2012). 12D. Sarid and W. Challener, Modern Introduction to Surface Plas- mons: Theory, Mathematica Modeling, and Applications (Cam- bridge University Press, Cambridge, 2010). 13R. Ruppin, Phys. Lett. A 277, 61 (2000). 14K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. FIG. 4. Confinement length 1/κ for silicene and graphene as a function of frequency. Inset : ∆λ as a function of frequency for graphene and silicene. −1 −1 This occurs for ∆z = 4∆SO (Ez = 33.92 mVA ) and ∆z = 8∆SO (Ez = 67.84 mVA ) (see Figs. 2(b) and (c) respectively). For ∆z = 4∆SO, only Im σA−+ and Im σA +− are suppressed, therefore we still have Im σ > 0 at certain frequency and Re σ (cid:54)= 0 for ω ≥ 2EF (f ≥ 3.77 THz, see Eqs. (7) and (8)). Hence, the TE frequency range becomes 1.61 < f < 3.77 THz. But in the case of ∆z = 8∆SO, all Im σA ςη vanish and Im σ has nega- tive value at all frequency. Re σ (cid:54)= 0 for ω ≥ ∆++ (f ≥ 6.60 THz). Therefore, the TE frequency range becomes 0 < f < 6.60 THz. Re σ appears at higher frequency than that for ∆z = 4∆SO, because the Fermi level exists in all of the energy gaps, in which we need a higher excitation energy for interband transition. Another interesting finding is that the undoped sil- icene (EF = 0) may also support TE surface wave. From Eqs. (6) -- (8), for EF = 0 we get Im σ: (cid:40)(cid:34) (cid:88) (cid:12)(cid:12)(cid:12)(cid:12)ω + ∆ςη(∆z) ω − ∆ςη(∆z) ςη (cid:19)2(cid:35) (cid:18) ∆ςη(∆z) (cid:41) (cid:12)(cid:12)(cid:12)(cid:12) − 2∆ςη(∆z) ω ω . Im σ(ω, ∆z) = − e2 16π 1 + × ln (9) The TE frequency range lies within 0 < ω < ∆++/−−. It is noted that Im σ(ω) vanishes at EF = 0 in graphene15, hence the TE surface wave does not exist for undoped graphene. From Eq. (1), we can define a confinement length of TE surface wave 1/κ, as follows 1 κ = 2 iωσ(ω, ∆z)µ0 . (10) A smaller value of 1/κ corresponds to better confine- ment. In Fig. 4, we plot 1/κ of the TE surface wave in graphene and silicene for comparison. The plot starts at f = 3.15 THz (ω = 1.667EF), which is the lower bound of TE frequency range in graphene. We can see that the TE surface wave in silicene is much more con- fined than in graphene and tunable by ∆z. For ex- ample, at f = 3.25 THz (ω = 1.725EF), in case of EF = 7.8 meVGrapheneΔz = 8ΔSO 4ΔSO2ΔSO34567Frequency (THz)1031041/κ (μm)-0.03-0.02-0.010Frequency (THz)3.13.33.53.7Δλ (μm) 4ΔSOGrapheneKatsnelson, I. V. Grigorieva, S. V. Dubonos, and A. A. Firsov, Nature 438, 197 (2005). 15S. A. Mikhailov and K. Ziegler, Phys. Rev. Lett. 99, 016803 075131 (2007). 20M. Ezawa, New J. P 14, 033003 (2012). 21C.-C. Liu, W. Feng, and Y. Yao, Phys. Rev. Lett. 107, 076802 (2007). 16M. Jablan, H. Buljan, and M. Soljaci´c, Opt. Express 19, 11236 (2011). (2011). 22C.-C. Liu, H. Jiang, and Y. Yao, Phys. Rev. B 84, 195430 (2011). 23N. D. Drummond, V. Zolyomi, and V. I. Fal'Ko, Phys. Rev. B 17L. Stille, C. J. Tabert, and E. J. Nicol, Phys. Rev. B 86, 195405 85, 075423 (2012). (2012). 18C. J. Tabert and E. J. Nicol, Phys. Rev. B 89, 195410 (2014). 19G. G. Guzm´an-Verri and L. C. L. Y. Voon, Phys. Rev. B 76, 24L. A. Falkovsky and A. A. Varlamov, EPJ B 56, 281 (2007). 25H. Bruus and K. Flensberg, Many-body Quantum Theory in Condensed Matter Physics (Oxford University Press, New York, 2004). 5
1312.3044	1	1312	2013-12-11T06:28:35	Magnonic Band Structure, Complete Bandgap and Collective Spin Wave Excitation in Nanoscale Two--Dimensional Magnonic Crystals	[ "cond-mat.mes-hall" ]	We present the observation of a complete bandgap and collective spin wave excitation in two-dimensional magnonic crystals comprised of arrays of nanoscale antidots and nanodots, respectively. Considering that the frequencies dealt with here fall in the microwave band, these findings can be used for the development of suitable magnonic metamaterials and spin wave based signal processing. We also present the application of a numerical procedure, to compute the dispersion relations of spin waves for any high symmetry direction in the first Brillouin zone. The results obtained from this procedure has been reproduced and verified by the well established plane wave method for an antidot lattice, when magnetization dynamics at antidot boundaries is pinned. The micromagnetic simulation based method can also be used to obtain iso--frequency countours of spin waves. Iso--frequency contours are analougous of the Fermi surfaces and hence, they have the potential to radicalize our understanding of spin wave dynamics. The physical origin of bands, partial and full magnonic bandgaps has been explained by plotting the spatial distribution of spin wave energy spectral density. Although, unfettered by rigid assumptions and approximations, which afflict most analytical methods used in the study of spin wave dynamics, micromagnetic simulations tend to be computationally demanding. Thus, the observation of collective spin wave excitation in the case of nanodot arrays, which can obviate the need to perform simulations may also prove to be valuable.	cond-mat.mes-hall	cond-mat	Magnonic Band Structure, Complete Bandgap and Collective Spin Wave Excitation in Nanoscale Two -- Dimensional Magnonic Crystals D. Kumar,1, a) J. W. K(cid:32)los,2, a) M. Krawczyk,2 and A. Barman1, b) 1)Thematic Unit of Excellence on Nanodevice Technology, Department of Condensed Matter Physics and Material Sciences, S. N. Bose National Centre for Basic Sciences, Block JD, Sector III, Salt Lake, Kolkata 700 098, India. 2)Faculty of Physics, Adam Mickiewicz University in Poznan, Umultowska 85, Pozna´n, 61-614, Poland. (Dated: 8 November 2018) We present the observation of a complete bandgap and collective spin wave excita- tion in two-dimensional magnonic crystals comprised of arrays of nanoscale antidots and nanodots, respectively. Considering that the frequencies dealt with here fall in the microwave band, these findings can be used for the development of suitable magnonic metamaterials and spin wave based signal processing. We also present the application of a numerical procedure, to compute the dispersion relations of spin waves for any high symmetry direction in the first Brillouin zone. The results ob- tained from this procedure has been reproduced and verified by the well established plane wave method for an antidot lattice, when magnetization dynamics at anti- dot boundaries is pinned. The micromagnetic simulation based method can also be used to obtain iso -- frequency countours of spin waves. Iso -- frequency contours are analougous of the Fermi surfaces and hence, they have the potential to radicalize our understanding of spin wave dynamics. The physical origin of bands, partial and full magnonic bandgaps has been explained by plotting the spatial distribution of spin wave energy spectral density. Although, unfettered by rigid assumptions and approximations, which afflict most analytical methods used in the study of spin wave dynamics, micromagnetic simulations tend to be computationally demanding. Thus, the observation of collective spin wave excitation in the case of nanodot arrays, which can obviate the need to perform simulations may also prove to be valuable. 3 1 0 2 c e D 1 1 ] l l a h - s e m . t a m - d n o c [ 1 v 4 4 0 3 . 2 1 3 1 : v i X r a a)These authors have contributed equally to this work. b)Electronic mail: [email protected] 1 I. INTRODUCTION Analogous to photonic crystals, magnonic crystals (MCs)1 are magnetic meta-materials designed for the propagation of spin waves (SWs).2 -- 4 Based on their design, MCs exhibit a characteristic SW dispersion relation complete with bands and, sometimes, band gaps which can be tuned by controlling material and structural parameters as well as the strength and orientation of the bias magnetic field.5,6 This phenomenon makes MCs useful as potential candidates for the design of SW based signal processing and logic devices.7 The knowledge of dispersion relation of a wave propagating through a medium is neces- sary to understand its transmission characteristics. Although MCs have been a subject of intense study lately,8 -- 13 reports on a time domain numerical calculation of dispersion rela- tions of SWs propagating in two-dimensional (2D) MCs are very rare.14 -- 16 As other analytical methods are available, the use of time domain simulations and spatial Fourier transform to obtain the dispersion relation in a photonic or phononic crystal is rarely seen17 as well. We hope to fill that gap in research with this work. The underlying principles, over which the procedure described here is used, has been discussed more generally by us in Ref. 18. Here too, we essentially use a micromagnetic simulator called Object Oriented Micromagnetic Framework19 (OOMMF) to obtain magnetization M, as a function of position r, and time t. Then we use a multi-domain discrete Fourier transform to obtain the desired dispersion relation: SW power as a function of wavevector k = (kx, ky), and frequency f . However, while simulating the magnetization dynamics in large (ideally infinite20) 2D crystals, one can be expected to need far greater computational resources than during the simulations of their one-dimensional (1D) counterparts.10 Using a finite sample size may produce some spurious modes in the obtained dispersion relation.20 Thus, the use of 2D periodic bound- ary condition21 (PBC) becomes mandatory in order to obtain good numerical resolution in wavevector and frequency domains while consuming finite computational resources. Also, 2D crystals have more high symmetry directions when compared to their 1D analogues. Different techniques will be required to obtain the results for different directions in the 2D reciprocal space covering the entire irreducible part of the Brillouin zone (BZ).22 Moreover, the signal which generates the waves will have to be carefully designed so that the resulting spectrum represents the physical dispersion relation of plane propagating SWs. Due to all these complications, a need to validate the results obtained here with a well established 2 method, such as the plane wave method (PWM)23 becomes very clear. Both OOMMF and PWM solve the following Landau-Lifshitz-Gilbert (LLG) equation:24 dM dt = −¯γM × Heff − α¯γ Ms M × (M × Heff) . (1) Here, Ms is the saturation magnetization, ¯γ is the gyromagnetic ratio and α is the Gilbert damping constant. Heff is the sum of the bias field Hbias, the excitation signal Hsig (used in micromagnetic simulations (MS) only), the exchange field, the demagnetizing field and the magneto-crystalline anisotropy. Other factors like magnetostriction, should also be consid- ered here if applicable. The details of MC considered here are presented in Sec. II A. Simulation parameters and PWM are described further in Sec. II B. OOMMF uses the finite difference method (FDM) to solve Eq. (1) as an ordinary differential equation in time and space (derivatives with respect to space are hidden away in Heff). PWM is based on the Bloch wave formalism. As these two methods are fundamentally different in approach, some quantitative differences in results are to be expected. The results from both the methods and their differences have been discussed in Sec. III for the antidot lattice (ADL). Due to small lattice constant, the considered system is an exchange dominated one and consequently, the differences in dis- persion relations along the bias magnetic field and perpendicular to it are subtle. These differences have been explored by calculating the iso-frequency contours in the wavevector space using both MS and the PWM. The iso-frequency contours are the curves of the con- stant frequency plotted in the wavevector space, they are wave counterparts of the Fermi surfaces known from the theory of the solid state physics.22 The iso-frequency contours are very important tool for the analysis of the wave propagation phenomena, giving a deep insight into direction and velocity of propagating, reflected and refracted waves in artificial crystals. Such type of analysis, while widely explored in photonic and phononic crystals for designing their metamaterials properties,25 -- 27 is almost absent in magnonics. Thus, develop- ing the ability to compute these iso-frequency contours using MS can be a breakthrough in exploring magnonic metamaterials based on MC; because the MSs can be performed without approximation limited applicability of the PWM (or other analytical methods28), and thus yields experimentally realizable results even with complex magnetic configurations. We also plot the energy spectral density and phase distributions associated with different modes in the SW spectrum in order to understand their physical origin and explain any 3 observed partial or complete bandgaps. Finally, we use the method described here to obtain the SW dispersion relations in the case of 2D dot array where the SW propagation is me- diated by inter-dot stray magnetic field as opposed to dipole-exchange interaction in ADL. This brings about an interesting change in the spectra, which is discussed in Sec. III along with their effective properties. II. METHOD A. Magnonic crystal lattice and material parameters The structure considered here is an infinitely large square array of square antidots with their ferromagnet-air interface under pinned boundary conditions.10 The geometrical struc- ture of the sample is shown in Fig. 1 (a). The lattice constant a = 30 nm and the an- tidots are square holes of edge length, l = 12 nm. The material parameters of permalloy (Py: Ni80Fe20) are used during simulations and in PWM calculations: exchange constant, A = 13×10−12 J/m, saturation magnetization, Ms = 0.8×106 A/m, gyromagnetic ratio, ¯γ = 2.21×105 m/As and no magnetocrystalline anisotropy. A saturating bias magnetic field of µ0Hbias = 1 T points in x direction. B. Micromagnetic simulations and the plane wave method The micromagnetic simulations involve solving the LLG equation (1) using a finite dif- ference method based ordinary differential equation solver; and then, Fourier transforming the obtained space and time dependent magnetization data to get SW spectral density in wavevector and frequency domains.18 Cell size (d, d, s) = (1.5, 1.5, 3) nm along (x, y, z) axis was used during the FDM based simulations. The pinning in micromagnetic simulations was introduced by fixing magnetization vector in all cells of the discretization mesh, which border the antidots, i.e., in regions marked with different texture in Fig. 1. Figures 1 (b), (c) and (d) show parts of the elements over which 2D PBC are used to simulate the dis- persion relation for different directions of the wave vector. These elements extend over 100 (up to 300) repetitions of unit cells in the horizontal direction to yield good resolution in the wavenumber domain. The 2D PBC are also implemented in order to improve the re- sults with finite computational resources.21 Figure 1 (e) shows the first BZ, the path in its 4 FIG. 1. (a) The 2D antidot lattice under consideration. The lattice constant of the square lattice is a = 30 nm. The thickness s of the film is 3 nm. The antidots are square (white) air holes of edge l = 12 nm in ferromagnetic Py (black) medium. Dynamics is pinned at the edge of holes. The pinned region is marked with a different texture. Element geometry used in micromagnetic simulations extends to over hundred repetitions in length (horizontal dark arrows in (b), (c) and (d)) for good wavenumber resolution. 2D PBC is applied over these elements to mimic the infinite geometry. White arrows in (b), (c) and (d) show the direction of bias field used for simulations of SW dispersion for backward volume and Damon-Eshbach configuration. (d) shows the first BZ in the inverse lattice with typical symmetry point labels. irreducible part and typical symmetry points: Γ = (0, 0), X = π/a(1, 0), Y = π/a(0, 1) and M = π/a(1, 1).22 Note that when the bias field is in the plane, an asymmetry is expected between the two orthogonal directions of SW propagation: Hbiask (backward volume) and Hbias⊥k (Damon-Eshbach).12 Thus, the triangle ΓXM is no longer the irreducible BZ. How- ever, in the forward volume arrangement when Hbias is perpendicular to the plane of the 2D MC, the symmetry is restored and dispersion is the same in the two orthogonal directions. The technique described here can be used independent of the direction of Hbias. In order to get the results in the Γ − X and Y − M directions, we use different excitation signals of the form Hsig = (0, 0, Hz), on elements shown in Figs. 1 (b) and (c), respectively. Hbias is horizontal along the x axis (dashed white arrows). Similarly, dispersion along the Γ − Y and X − M directions can be obtained when Hbias is across the width of the elements 5 xylaa(c)Hbias(b)HbiasHbias(d)ΓMXY(e)PyAir(a)(vertical arrows along y axis). Here, Hz = H0NtN (x)ny with µ0H0 = 5 mT and Nt, N (x) and ny as given by Eqs. (2), (3) and (4), respectively: Nt = sin(2πfc(t − t0)) 2πfc(t − t0) , N (x) = sin(kcx) kcx , ny = cos(2πy/ymax) + sin(2πy/ymax). (2) (3) (4) See Eq. (3) in Ref. 13 for the detailed description of the terms involved in these equations. Here, the origin of coordinates is at the center of the considered geometry. It is due to Nt and N (x) that the signal contains power between ±fc and ±kc in frequency and wavevector domains respectively.18 ny should be asymmetric to ensure that both symmetric and anti- symmetric modes are present in the resulting spectrum.2 In Eq. (4), y goes from 0 to ymax. While computing dispersion along Γ − X and Γ − Y directions (Fig. 1 (b)), ymax = a. How- ever, for Y − M and X − M directions (Fig. 1 (c)), ymax = 2a. Both the elements in Figs. 1 (b) and (c) will span the same infinite 2D geometry under a 2D PBC; except, in the later case we can control whether the dynamics in the neighboring rows will be in phase or out of phase. Thus we can fix the wavevector component ky or kx to 0 or π/a in the simulations. This is necessary to differentiate between the parallel directions Γ − X and Y − M or Γ − Y and X − M. Also, nmn given by the expression y nmn y = Cm cos(2mπy/ymax) + Cn sin(2nπy/ymax) (5) can be used instead of ny to selectively alter the amplitude of m-th symmetric or n-th an- tisymmetric mode. The freedom of choice of amplitudes Cm and Cn allows us to artificially control the statistical temperature of the magnons in the crystal and also helps in isolating a single mode in the case of a degeneracy. We can also sum over m and n to alter multiple modes in a single dynamic simulation. We also attempt to obtain the dispersion in Γ − M direction by using the element shown in Fig. 1 (d). However, as there are two scattering cen- ters (antidots) per cell in this arrangement, we can obtain the dispersion relations correctly only up to half of the BZ in that direction.15 Until now, we could use a signal similar to the one we did in the case of an 1D lattice.13 But, this limitation forced us to come up with a new signal Hz = H0NtN (x)N (y)n(x)n(y), (6) 6 which has to be used in a larger 2D lattice of 100×100 antidot array (with the cell size d increased to 3 nm to decrease time of computations). Here, n(x) is given by: 5(cid:88) n(x) = (sin(2πmx/a) − cos(2πmx/a)) , (7) m=1 with analogous formula for n(y). This signal is a point like source with the amplitude decay with distance as described by sinc function (in N (x) Eq. (3) along x axis and in similar form for N (y) for y dependence), having sharp cut-off in Fourier domain and able to excite symmetric and antisymmetric modes with respect to x− or y− axis. This signal was arrived upon largely by intuition, nevertheless, its agreement with the results obtained from PWM validates the usefulness of this signal. Spectral density, periodicity and asymmetry of the excitation signal (or source) should also be considered while developing similar techniques for other kinds of crystals (e.g. photonic or phononic crystals). Three fold (one in time and two in space) Fourier transforms was needed to obtain the SW dispersion here. Magnetization was assumed to be uniform across the thickness of the film. We can now easily generalize that in the case of three-dimensional MCs, a signal of the form Hz = H0NtN (x)N (y)N (z)n(x)n(y)n(z) will be required followed by a four fold discrete Fourier transform. We have also calculated the spatial distribution of energy spectral density (ESD), Sf and phase, θ from the following equations: Sf = m(r, f )2 ; θ = tan−1 (cid:18)Im ( m(r, f )) Re ( m(r, f )) (cid:19) . (8) (9) Here, m(r, f ) is the time domain Fourier transform, of a dynamical magnetization data. Unlike the new method used in Ref. 13, this gives us power from the entire wavevector domain for a selected frequency f . However, if power is present for just one particular wavevector then both methods yield qualitatively identical results. The PWM is a spectral method in which the eigenproblem is numerically solved in the frequency and wavevector domains by the standard numerical routines. We solve here LLG equation (1) in linear approximation without damping. The PWM calculations are per- formed with the assumption of the full magnetic saturation of the ADL along the bias magnetic field. As pinning during simulation will occur at the cell's center, a hole size of 7 l + d was assumed during PWM calculations. Due to small thickness of the ADL, uniform SW profile across the thickness is assumed. The PWM in this formulation was already used in the calculations of the SW dynamics in 2D ADL and proved to give correct results.10,29 -- 31 The detailed description of the method can be found in Refs. 30 and 32. III. RESULTS AND DISCUSSIONS The dispersion along the path in the first BZ shown in Fig. 1 (e) calculated with MSs by using the elements shown in Fig. 1 (b)-(d) is assembled as Fig. 2 (a) using solid lines. An overlay of dashed lines representing the SW dispersion relation obtained from the PWM is √ √ √ √ 2a(cid:1) here (the spatial periodicity is total BZ extent. This is because we set kc to π/(cid:0)√ provided for comparison. Both these results appear to agree with each other except for the Γ − M direction where the numerical method was able to yield results for only half of the 2a). Compared to the element shown in Fig. 1 (b), which can be used to produce results for Γ−X or Γ − Y directions, the one in Fig. 1 (d) features two scatter centers per unit cell. And, √ 2π/a, both scattering centers will be activated to produce if we artificially increase kc to additional spurious modes.15 To demonstrate the same we plot Sf (normalized between 0 and 1) and θ (given by Eqs. (8) and (9), respectively), for frequency f ≈ 62 GHz in Figs. 2 2a (b) to (e). Note that the horizontal separation between regions of high ESD is about 2a(cid:1). This reduces to a/ in Fig. 2 (b) for kc = π/(cid:0)√ about π and π/2 radians out of phase with each other in former (kc = π/(cid:0)√ both scattering centers in the unit cell (of the element shown in Fig. 1 (d)) are activated at once. The phase distributions also confirm that neighbouring locations of high ESD are 2a(cid:1): Fig. 2 2 in Fig. 2 (c) for kc = 2π/a when (d)) and later (kc = 2π/a: Fig. 2 (e)) cases, respectively. Apart from incomplete result for the Γ − M direction, we can also see that the modes here (shown by solid lines) do not match with those for Γ − Y direction at the Γ point. This is because (cell size) d = 2 nm was used while simulating for the Γ − M direction as opposed to d = 1.5 nm, which was used in the case of Γ − Y direction. Also, there are additional modes of lower amplitudes visible in the case of Γ − M direction. This is due to the fact that N (x) becomes a stepped approximation of the right hand side of (3) by the use of the FDM; thus compromising the 2a(cid:1), and exciting the second scattering centre to effectiveness of the cut off at kc = π/(cid:0)√ √ √ some extent (but not as well as kc = √ 2π/a). 8 FIG. 2. (a) SW dispersion calculated using MSs (solid line) and PWM (dashed lines). ESD Sf , distribution for the horizontal line (f ≈ 62 GHz) shown in (a) in parts of the sample when the propagation direction is along Γ − M for (b) kc = π/(cid:0)√ 2π/a. Corresponding 2a(cid:1) and (c) kc = √ phase θ, distribution is shown in (color online) (d) and (e), respectively. In pursuit of our quest to close the gap in the Γ − M direction we eventually decided to simulate the SW dynamics in a large 2D MC with signal defined by Eq. (6) and perform a three-fold Fourier transform in contrast with the two-fold transforms done earlier. We transformed time to frequency domain and x− and x− dimensions to the 2D wavevector domain. The resulting dispersion relation as calculated from numerical method is shown in Fig. 3 (a) using solid lines. The agreement with the PWM results (shown by dashed lines) is 9 (b)(d)0(c)(e)01 (rad)S (arb. units)f0.5XYMMf (GHz)1601401201008060(a)f62 GHzFIG. 3. (a) SW dispersion calculated using MS (solid line) and the PWM (dashed lines). The full and partial magnonic bandgaps are marked and numbered by Roman numerals. The circled Arabic numerals indicate the points on the dispersion for which the mode profiles are calculated in Fig. 4. Iso-frequency lines from (b) 63 GHz to 67 GHz (c) 86 GHz to 107 GHz (it is around the top and bottom of the first and second magnonic band, respectively) using the PWM is shown with dashed lines. Iso-frequency lines for (b) f ≈ 67 GHz and (c) f ≈ 100 GHz calculated by the numerical method is shown using solid lines. poorer in comparison with Fig. 2 (a). This is due to the fact that cell size in the later attempt was increased from d = 1.5 nm to d = 3 nm. The complete and partial bandgaps width 10 XYMMIXVIIIIVIIIIVVIIXVIIIXI12345123451234512345f (GHz)1601401201008060(a)XXYYMMMM899295981011041079592898686-0.4-0.20.20.4-0.4-0.20.20.4(c)k (/2a)xk (/2a)yXXYYMMMM65656767676763-0.4-0.20.20.4-0.4-0.20.20.4(b)k (/2a)xk (/2a)yand center frequency, as seen from the dashed lines in Fig. 3 (a), are extracted in Tab. I. Here, values for partial bandgaps depend upon the path, which has been used to plot the dispersion. Bandgap I is the only complete bandgap observed here with the maximum width of 15.37 GHz. TABLE I. Magnonic bandgap widths and center frequencies across different directions as calculated by the PWM and labeled in Fig. 3 (a). Label Extent Center (GHz) Gap Width I Complete Bandgap 76.39 II III IV V VI VII VIII IX X XI Γ − X Γ − X Γ − X − M − Γ X − M X − M Γ − Y Γ − Y Γ − Y Y − M Y − M (GHz) 15.36 9.24 7.7 1.9 5.9 8.4 100.18 114.45 155.85 107.75 149.6 100.68 10.24 114.2 85.80 108.1 150.1 8.2 3.01 4.6 9.4 Most bands observed in Fig. 3 (a) increase or decrease almost monotonously along any high symmetry direction. Consequently, the width of bandgap I too, appears to decrease monotonously as we move either along Γ → X → M or Γ → Y → M. Both upper and lower limits of bandgap I are present at point M which suggests an anti-crossing of bands at that point. This can also be regarded as the cause of the gap formation. Narrower bandgap widths have been observed by different techniques before.33 The relatively high width of 15.37 GHz of bandgap I here can be attributed to small lattice dimensions and edge pinning.10 Bandgaps II to XI (Fig. 3 (a) and Tab. I) are direction dependent partial bandgaps. This is mainly because bands approaching point M from other high symmetry directions (with the exception of the band starting at Γ 5(cid:13)) tend to show greater slopes. As X 5(cid:13)→ Y 5(cid:13) is a relatively flatter line, bandgap IV survives for three high symmetry directions. In a more 11 isotropic forward volume arrangement,15 bandgap IV might also have qualified as a complete bandgap if the dispersion in the X − M direction was also calculated. On the other hand, if wavevector dependent anisotropy is overlooked,14 partial bandgaps (e.g. bandgap IV, or II and VII, or III and VIII) will appear as a complete bandgap. Partial bandgaps IV, V, VIII, X and XI are direct, while II, III, VI, VII and IX are indirect. Direct bandgaps are formed when the minimal and the maximal frequency of the magnonic bands surrounded the bandgap, from the top and bottom, respectively, are characterized by the same wavevector. While two different wavevectors are involved in the formation of indirect bandgap. In Fig. 3 (a) the minimal and maximal frequencies appear at high symmetry points. Occasionally, a bandgap may form between two high symmetry points due to anti-crossing of modes in a folded BZ,31 but that is not observed here. FIG. 4. (Color online) ESD Sf , and phase θ, for high symmetry points Γ, X, M and Y at points 1(cid:13) through 5(cid:13) marked on Fig. 3 (a). Now we calculate mode profiles ESD Sf and phase θ, at the high symmetry points, using the PWM, for the first five modes as marked in Fig. 3 (a). The results are tabulated as 12 MXY12345SfFig. 4 where Sf is represented by color saturation and θ is represented by hue. A general trend of higher frequency mode profiles limiting themselves to smaller regions in real space is observed. This trend has been seen for 1D systems as well.31 Here, mode profiles appear similar in size at points X 5(cid:13), M 5(cid:13) and Y 5(cid:13). Although, the distribution at Y 5(cid:13) is vastly different due to a (nearby) mode -- crossing in the Y−M direction (see Fig. 3 (a)). Mode profile at Y may be obtained by rotating the mode profiles at X by 90◦. Modes with negligible group velocity are trapped and forbidden to move in specific high symmetry directions. Also, the number of nodal lines, which controls the spatial quantization of modes, generally increases i(cid:13): i ∈ {1, 2, 3, 4, 5}. No nodal lines are evident for Γ 1(cid:13). Vertical and with mode number horizontal nodal lines are seen at X 1(cid:13) and Y 1(cid:13), respectively; while M 1(cid:13) features both vertical and horizontal nodal lines. From Fig. 3 (a), we can see that points Γ 2(cid:13), X 2(cid:13), M 3(cid:13) and Y 3(cid:13) belong to the same mode and points Γ 3(cid:13), X 3(cid:13), M 2(cid:13) and Y 2(cid:13) belong to a different mode. As the crossing between these modes occurs along the X− M direction, the mode profiles at X 2(cid:13) and M 3(cid:13) are comparable. Similarly, mode profiles at X 3(cid:13) and M 2(cid:13) are also comparable except, X 3(cid:13) has higher frequency and consequently, is more confined is space. In general, vertical and horizontal nodal lines dominate at points X and Y, respectively; while a more isotropic distribution is observed at point Γ and M. Modes 1(cid:13), 2(cid:13), and 5(cid:13) are isotropic along x− and y−axes for the Γ point. However, modes 3(cid:13) and 4(cid:13) are disposed along rows and columns, respectively. There local shape and size is comparable and accordingly, they 4(cid:13) maintains its size and frequency; except the Damon-Eshbach34 geometry is evident in the later case. Similarly, the expanses of mode profiles at M 2(cid:13) and 3(cid:13) are comparable (as there frequencies are within 5 GHz of each other), and yet their orientations are mutually orthogonal. are also degenerate as seen in Fig. 3 (a). Going from Γ to either X or Y, Iso-frequency lines are shown in Figs. 3 (b) and (c), using both the PWM (dashed lines) and the MSs (solid lines). Iso-frequency contours calculated using the proposed method are thicker because small a yields a low wavevector resolution. The agreement between the results obtained from the two methods as 67 GHz line calculated using the MSs and the 65 GHz line calculated using the PWM is clear, but the 2 GHz difference in frequencies is due to the shift of the dispersion curves calculated with both methods shown in Fig. 3 (a). In contrast to Fig. 3 (b) the two methods appear to give identical results for the 100 GHz iso-frequency line, where the results of MS and PWM coincide. The shapes of iso-frequency lines control the direction of the propagating waves and consequently also alter the shapes 13 of their wavefronts. Thus, although the dispersion along Γ − X and Γ − Y directions may appear comparable, the wavefronts of the propagating SWs from the first band will quickly uncover the underlying anisotropy, because of slightly different group velocity and curvature of different iso-frequency contours in two orthogonal directions, which is easily noticeable by the inspection of the contours for 63 and 65 GHz in Fig. 3 (b). This anisotropy is a manifestation of dipolar interactions hardly visible in this size and frequency regime in the magnonic band structure shown in Fig. 3 (a). Backward volume modes are characterized by negative group velocity in the case of dipole dominated or dipolar-exchange SW propagating in a ferromagnetic thin film.3 This is not seen in Fig. 3 (a) as due to weakness of the dipolar interactions the exchange field makes a significant contribution with increasing wavevector k already near the BZ center. FIG. 5. First mode in a permalloy nano -- dot array with varying angle φ, between the bias field Hbias (µ0Hbias = 1 T), and wavevector k, showing the transition from magnetostatic backward volume mode to Damon-Eshbach configuration. The dashed lines are calculated using the analytic expressions for these two configuration with a reduced saturation magnetization. The structure considered here is given in the top left corner with a = 9 nm, l = 6 nm and thickness s = 3 nm. Material parameters remain the same as before. The developed method is not limited to the antidot lattices nor exchange dominated SWs. To prove this and better understand the properties of dipolar waves in MCs we take a look at the dispersion of SWs in the case of 2D MC composed of a square array (of lattice constant a = 9 nm) of square dots (of edge l = 6 nm and 3 nm thick). This structure is shown in the top left panel of Fig. 5 along with the dispersion relations of the first mode with increasing angle φ (from φ = 0 to φ = 90◦), between Hbias and k in the subsequent panels. Here the wave propagation is mediated by the dipolar field only. We note how the mode's group 14 =15=30=45=60=75=90=0333630-1-1-1-111110000ka/ka/ka/ka/f (GHz)333630f (GHz)alHbiaskvelocity gradually increases from negative (backward volume) to positive (Damon-Eshbach) as φ goes from 0◦ to 90◦.16 The transition appears to occur at a critical angle φ = φc ≈ π/3. Note that here the direction of Hbias is being changed as opposed to that of k in the previous case. It is interesting to note that the dispersion relations obtained here for the array of nano -- dots reminds us of the dispersion of magnetostatic waves in thin ferromagnetic film. To verify this hypothesis we calculate the dispersion relation of magnetostatic waves in the thin ferromagnetic magnetic film (3 nm thick) with reduced magnetization, i.e. with the effective value of the saturation magnetization Ms,eff. The dashed lines overlaid in Fig. 5 are computed using the analytical expression for backward volume and Damon-Eshbach configuration in the case of thin film3 with a reduced saturation magnetization Ms,eff = Msl2/a2. A good agreement between the dispersion in the array and the effective thin film is found. A minor disagreement is introduced by the presence of the BZ boundaries but only near these boundaries. Further, the critical angle, φc = tan−1(cid:112)Hbias/Ms,eff 35 in the case of such thin film is also 56.24◦ ≈ π/3. This implies that one should also be able to use the analytical expression to calculate the SW manifold between backward volume and Damon- Eshbach geometries. This also shows, that a thin film MC composed of an array of saturated ferromagnetic nanodots can be used as a magnonic metamaterial, i.e., an artificial crystal with tailored effective properties of spin wave dynamics.36 -- 39 Further studies are necessary to elucidate the limits of the effective saturation magnetization approach presented here, the influence of the dot -- shape, their arrangement and inter -- dot separation (mode-splitting has been experimentally demonstrated for nano -- dot arrays40), but this is outside the scope of this paper. IV. CONCLUSIONS We have described a numerical algorithm to calculate the dispersion of plane propagating SWs in a 2D MC using multi-domain Fourier transform of results obtained from micromag- netic simulations. At the core of this technique is a new excitation signal, which is capable of generating SWs whose energy spectral density corresponds to the characteristic dispersion relation of the 2D MC. The lack of such signal has been discussed before in the case of 1D MCs.41 The results obtained from this procedure were verified by the plane wave method when magnetization dynamics at antidot boundaries is pinned. We noted that both methods 15 were in qualitative agreement with each other. The fact that better quantitative agreement was observed while using 2D PBC over 1D elements was due to lower cell size.31 Apart from a new numerical algorithm to compute the dispersion relation in any given direction of a two- or three-dimensional inverse lattice, this method will also allow for the numerical computation of iso-frequency contours from micromagnetic simulations. Thus the numerical tool for study metamaterials properties of MCs was provided. It gives the possi- bility for design the properties of SWs relevant to technological applications and potentially exceeding these known from the homogeneous ferromagnetic thin films. The negative re- fraction, unidirectional media or caustic propagations are only some of examples here.42 -- 44 Further, this method can be generalized to aid the numerical computation of dispersion or iso-frequency contours in the case of two- or three-dimensional phononic45 and photonic46 crystals as well. The dispersion here appeared to be similar in Γ − X and Γ − Y directions. However, a noticeable anisotropy between the backward volume and Damon-Eshbach geometries was very evident from the study of the mode profiles and the iso-frequency contours. As dipole field mediates the SW propagation in a 2D dot array we were able to obtain the negative group velocity associated with the first mode in the case of a backward volume magnetostatic configuration. We were also able to analyze the nature of bands and complete and partial bandgaps that were obtained from the dispersion calculations in the case of an MC. This can be useful in the design of attenuators,47 phase-shifters,9 filters48 and logic gates.49 Recent advances in lithography techniques11,50 -- 52 have made it possible to fabricate dot and antidot lattices with a resolution below 10 nm. Thus, one can fabricate samples with dimensions comparable to the systems considered here. Experimental techniques similar to Brillouin light scattering spectroscopy53 can be used to explore the SW dispersion relation. ACKNOWLEDGMENTS We acknowledge the financial support from the Department of Science and Technology, Government of India (Grant nos. INT/EC/CMS (24/233552), Department of Information Technology, Government of India (Grant no. 1(7)/2010/M&C), the European Community's FP7/2007-2013 (GA nos. 233552 (DYNAMAG) and 228673 (MAGNONICS)) and NCN of Poland (DEC-2-12/07/E/ST3/00538). D.K. would like to acknowledge financial support 16 from CSIR - Senior Research Fellowship (File ID: 09/575/(0090)/2011 EMR-I) and fruitful discussions with A. Mookerjee at the S. N. Bose National Centre for Basic Sciences. REFERENCES 1S. Nikitov, P. Tailhades, and C. Tsai, J. Magn. Magn. Mater. 236, 320 (2001); S. O. Demokritov and A. N. Slavin, Magnonics: From Fundamentals to Applications (Springer, 2013); V. V. Kruglyak, S. O. Demokritov, and D. Grundler, J. Phys. D: Appl. Phys. 43, 264001 (2010). 2B. Hillebrands and K. Ounadjela, Spin Dynamics in Confined Magnetic Structures I , Anglistische Forschungen (Springer, 2002). 3D. D. Stancil and A. Prabhakar, Spin Waves (Springer US, 2009). 4S. Neusser and D. Grundler, Adv. Mater. 21, 2927 (2009); B. Lenk, H. Ulrichs, F. Garbs, and M. Munzenberg, Physics Reports 507, 107 (2011). 5S. Mamica, M. Krawczyk, and J. W. K(cid:32)los, Adv. Cond. Matt. Phys. 2012, 6 (2012). 6H. Yang, G. Yun, and Y. Cao, J. Phys. D: Appl. Phys. 44, 455001 (2011). 7J. Ding, M. Kostylev, and A. O. Adeyeye, Appl. Phys. Lett. 100, 073114 (2012). 8S. Mamica, M. Krawczyk, M. L. Sokolovskyy, and J. Romero-Vivas, Phys. Rev. B 86, 144402 (2012). 9Y. Au, M. Dvornik, O. Dmytriiev, and V. V. Kruglyak, Appl. Phys. Lett. 100, 172408 (2012). 10J. W. K(cid:32)los, D. Kumar, J. Romero-Vivas, H. Fangohr, M. Franchin, M. Krawczyk, and A. Barman, Phys. Rev. B 86, 184433 (2012). 11R. Mandal, S. Saha, D. Kumar, S. Barman, S. Pal, K. Das, A. K. Raychaudhuri, Y. Fukuma, Y. Otani, and A. Barman, ACS Nano 6, 3397 (2012); F. Ciubotaru, A. V. Chumak, N. Y. Grigoryeva, A. A. Serga, and B. Hillebrands, J. Phys. D: Appl. Phys. 45, 255002 (2012); S. Tacchi, B. Botters, M. Madami, J. W. K(cid:32)los, M. L. Sokolovskyy, M. Krawczyk, G. Gubbiotti, G. Carlotti, A. O. Adeyeye, S. Neusser, and D. Grundler, Phys. Rev. B 86, 014417 (2012); G. Gubbiotti, S. Tacchi, M. Madami, G. Carlotti, S. Jain, A. O. Adeyeye, and M. P. Kostylev, Appl. Phys. Lett. 100, 162407 (2012); S. Saha, S. Barman, J. Ding, A. O. Adeyeye, and A. Barman, ibid. 102, 242409 (2013). 12G. Venkat, D. Kumar, M. Franchin, O. Dmytriiev, M. Mruczkiewicz, H. Fangohr, A. Bar- 17 man, M. Krawczyk, and A. Prabhakar, IEEE Trans. Mag. 49, 524 (2013). 13D. Kumar, P. Sabareesan, W. Wang, H. Fangohr, and A. Barman, J. Appl. Phys. 114, 023910 (2013). 14S.-K. Kim, J. Phys. D: Appl. Phys. 43, 264004 (2010). 15T. Schwarze, R. Huber, G. Duerr, and D. Grundler, Phys. Rev. B 85, 134448 (2012). 16M. Dvornik, Y. Au, and V. Kruglyak, in Magnonics, Topics in Applied Physics, Vol. 125, edited by S. O. Demokritov and A. N. Slavin (Springer Berlin Heidelberg, 2013) p. 101. 17A. Jafarpour, C. Reinke, A. Adibi, Y. Xu, and R. K. Lee, IEEE J. Quant. Electron. 40, 1060 (2004). 18D. Kumar, O. Dmytriiev, S. Ponraj, and A. Barman, J. Phys. D: Appl. Phys. 45, 015001 (2012). 19M. Donahue and D. G. Porter, "Oommf user's guide," (2002), http://math.nist.gov/oommf/doc/userguide12a3/userguide/. 20V. V. Kruglyak, P. S. Keatley, A. Neudert, R. J. Hicken, J. R. Childress, and J. A. Katine, Phys. Rev. Lett. 104, 027201 (2010). 21W. Wang, C. Mu, B. Zhang, Q. Liu, J. Wang, and D. Xue, Comput. Mater. Sci. 49, 84 (2010). 22C. Kittel, Introduction to solid state physics (Wiley, 1976). 23M. Krawczyk and H. Puszkarski, Phys. Rev. B 77, 054437 (2008). 24L. D. Landau and E. M. Lifshitz, Phys. Zeitsch. der Sow. 8, 153 (1935); T. Gilbert, IEEE Trans. Mag. 40, 3443 (2004). 25C. Luo, S. G. Johnson, J. D. Joannopoulos, and J. B. Pendry, Phys. Rev. B 65, 201104 (2002). 26S. Enoch, G. Tayeb, and B. Gralak, IEEE Trans. Antennas Propag. 51, 2659 (2003). 27X.-F. Li, X. Ni, L. Feng, M.-H. Lu, C. He, and Y.-F. Chen, Phys. Rev. Lett. 106, 084301 (2011). 28M. Grimsditch, L. Giovannini, F. Montoncello, F. Nizzoli, G. K. Leaf, and H. G. Kaper, Phys. Rev. B 70, 054409 (2004). 29S. Neusser, H. G. Bauer, G. Duerr, R. Huber, S. Mamica, G. Woltersdorf, M. Krawczyk, C. H. Back, and D. Grundler, Phys. Rev. B 84, 184411 (2011). 30S. Tacchi, G. Duerr, J. W. Klos, M. Madami, S. Neusser, G. Gubbiotti, G. Carlotti, M. Krawczyk, and D. Grundler, Phys. Rev. Lett. 109, 137202 (2012). 18 31J. W. K(cid:32)los, D. Kumar, M. Krawczyk, and A. Barman, Sci. Rep. 3, 2444 (2013). 32S. Neusser, G. Duerr, S. Tacchi, M. Madami, M. L. Sokolovskyy, G. Gubbiotti, M. Krawczyk, and D. Grundler, Phys. Rev. B 84, 094454 (2011). 33Z. K. Wang, V. L. Zhang, H. S. Lim, S. C. Ng, M. H. Kuok, S. Jain, and A. O. Adeyeye, ACS Nano 4, 643 (2010). 34R. Leite and R. C. Filho, J. Magn. Magn. Mater. 226-230, Part 1, 482 (2001). 35M. Hurben and C. Patton, J. Magn. Magn. Mater. 139, 263 (1995). 36V. V. Kruglyak, M. Dvornik, R. V. Mikhaylovskiy, O. Dmytriiev, G. Gubbiotti, S. Tacchi, M. Madami, G. Carlotti, F. Montoncello, L. Giovannini, R. Zivieri, J. W. K(cid:32)los, M. L. Sokolovskyy, S. Mamica, M. Krawczyk, M. Okuda, J. Eloi, S. W. Jones, W. Schwarzacher, T. Schwarze, F. Brandl, D. Grundler, D. Berkov, E. Semenova, and N. Gorn, "Metama- terial," (Intech, 2012) Chap. Magnonic Metamaterials, pp. 1 -- 31. 37R. Zivieri and L. Giovannini, Metamaterials 6, e127 (2012). 38M. Mruczkiewicz, M. Krawczyk, R. V. Mikhaylovskiy, and V. V. Kruglyak, Phys. Rev. B 86, 024425 (2012). 39O. Dmytriiev, M. Dvornik, R. V. Mikhaylovskiy, M. Franchin, H. Fangohr, L. Giovannini, F. Montoncello, D. V. Berkov, E. K. Semenova, N. L. Gorn, A. Prabhakar, and V. V. Kruglyak, Phys. Rev. B 86, 104405 (2012). 40B. Rana, D. Kumar, S. Barman, S. Pal, Y. Fukuma, Y. Otani, and A. Barman, ACS Nano 5, 9559 (2011). 41K. Di, H. S. Lim, V. L. Zhang, M. H. Kuok, S. C. Ng, M. G. Cottam, and H. T. Nguyen, Phys. Rev. Lett. 111, 149701 (2013); K.-S. Lee, D.-S. Han, and S.-K. Kim, ibid. 111, 149702 (2013). 42V. Veerakumar and R. E. Camley, Phys. Rev. B 74, 214401 (2006). 43M. P. Kostylev, A. A. Serga, and B. Hillebrands, Phys. Rev. Lett. 106, 134101 (2011). 44S.-K. Kim, S. Choi, K.-S. Lee, D.-S. Han, D.-E. Jung, and Y.-S. Choi, Applied Physics Letters 92, 212501 (2008). 45D. Alleyne and P. Cawley, J. Acoust. Soc. Am. 89, 1159 (1991). 46G. Burt, S. V. Samsonov, K. Ronald, G. G. Denisov, A. R. Young, V. L. Bratman, A. D. R. Phelps, A. W. Cross, I. V. Konoplev, W. He, J. Thomson, and C. G. Whyte, Phys. Rev. E 70, 046402 (2004); H. Gersen, T. J. Karle, R. J. P. Engelen, W. Bogaerts, J. P. Korterik, N. F. van Hulst, T. F. Krauss, and L. Kuipers, Phys. Rev. Lett. 94, 073903 (2005). 19 47K. Sekiguchi, T. N. Vader, K. Yamada, S. Fukami, N. Ishiwata, S. M. Seo, S. W. Lee, K. J. Lee, and T. Ono, Appl. Phys. Lett. 100, 132411 (2012). 48S.-K. Kim, K.-S. Lee, and D.-S. Han, Appl. Phys. Lett. 95, 082507 (2009). 49A. Khitun, M. Bao, and K. Wang, IEEE Trans. Mag. 44, 2141 (2008). 50F. H. M. Rahman, S. McVey, L. Farkas, J. A. Notte, S. Tan, and R. H. Livengood, Scanning 34, 129 (2012). 51V. Sidorkin, E. van Veldhoven, E. van der Drift, P. Alkemade, H. Salemink, and D. Maas, J Vac Sci Technol B 27, L18 (2009). 52T. Siegfried, Y. Ekinci, H. H. Solak, O. J. F. Martin, and H. Sigg, Appl. Phys. Lett. 99, 263302 (2011). 53K. Perzlmaier, M. Buess, C. H. Back, V. E. Demidov, B. Hillebrands, and S. O. Demokri- tov, Phys. Rev. Lett. 94, 057202 (2005). 20
1009.0207	1	1009	2010-09-01T15:43:02	Probing spin relaxation in an individual InGaAs quantum dot using a single electron optical spin memory device	[ "cond-mat.mes-hall" ]	We demonstrate all optical electron spin initialization, storage and readout in a single self-assembled InGaAs quantum dot. Using a single dot charge storage device we monitor the relaxation of a single electron over long timescales exceeding 40{\mu}s. The selective generation of a single electron in the quantum dot is performed by resonant optical excitation and subsequent partial exciton ionization; the hole is removed from the quantum dot whilst the electron remains stored. When subject to a magnetic field applied in Faraday geometry, we show how the spin of the electron can be prepared with a polarization up to 65% simply by controlling the voltage applied to the gate electrode. After generation, the electron spin is stored in the quantum dot before being read out using an all optical implementation of spin to charge conversion technique, whereby the spin projection of the electron is mapped onto the more robust charge state of the quantum dot. After spin to charge conversion, the charge state of the dot is repeatedly tested by pumping a luminescence recycling transition to obtain strong readout signals. In combination with spin manipulation using fast optical pulses or microwave pulses, this provides an ideal basis for probing spin coherence in single self-assembled quantum dots over long timescales and developing optimal methods for coherent spin control.	cond-mat.mes-hall	cond-mat	Probing spin relaxation in an individual InGaAs quantum dot using a single electron optical spin memory device D. Heiss,∗ V. Jovanov,∗ F. Klotz, D. Rudolph, M. Bichler, G. Abstreiter, M. S. Brandt, and J. J. Finley† Walter Schottky Institut, Technische Universitat Munchen, Am Coulombwall 3, D-85748 Garching, Germany (Dated: July 9, 2021) We demonstrate all optical electron spin initialization, storage and readout in a single self- assembled InGaAs quantum dot. Using a single dot charge storage device we monitor the relaxation of a single electron over long timescales exceeding 40µs. The selective generation of a single elec- tron in the quantum dot is performed by resonant optical excitation and subsequent partial exciton ionization; the hole is removed from the quantum dot whilst the electron remains stored. When subject to a magnetic field applied in Faraday geometry, we show how the spin of the electron can be prepared with a polarization up to 65% simply by controlling the voltage applied to the gate electrode. After generation, the electron spin is stored in the quantum dot before being read out using an all optical implementation of spin to charge conversion technique, whereby the spin projec- tion of the electron is mapped onto the more robust charge state of the quantum dot. After spin to charge conversion, the charge state of the dot is repeatedly tested by pumping a luminescence recy- cling transition to obtain strong readout signals. In combination with spin manipulation using fast optical pulses or microwave pulses, this provides an ideal basis for probing spin coherence in single self-assembled quantum dots over long timescales and developing optimal methods for coherent spin control. PACS numbers: 78.66.Fd, 78.67.De, Keywords: Quantum Dots, GaAs, InGaAs, Readout, Charge, Spin The spin of charges trapped in semiconductor quan- tum dots (QDs) is one of the most promising solid state qubits,1 primarily due to the robust quantum coherence and excellent scaling prospects.2 -- 11 Several groups have demonstrated electrical methods to initialize, control and readout QD spin qubits.2,5,6,12 However, optically active dots are particularly attractive since pulsed lasers can be used to selectively address spin qubits over ultrafast timescales.9,13 -- 16 Optical readout of a single spin is ex- tremely challenging and sensitive techniques based on resonant light scattering17 and Faraday/Kerr15,18 rota- tion have been applied. Such measurements are typ- ically repeated at high frequencies (>10MHz) to pro- vide enough signal, limiting their potential to probe slow spin dynamics occurring over timescales >1µs. Recent experiments using time-resolved resonance fluorescence showed spin detection19 in a single QD over much longer timescales. Nevertheless, in such approaches the QDs are tunnel coupled to a Fermi reservoir and co-tunneling leads to enhanced spin relaxation close to the edge of the charging thresholds. Thus, the electric field regime where the device can be operated is rather limited poten- tially posing problems when moving from a single dot to few QD systems where electric field also allows tuning of dot-dot tunnel coupling.20 Here, we demonstrate an all optical spin memory device that allows us to prepare a single spin in an individual QD, store it over millisecond timescales and read it out with high fidelity. Readout is based on spin to charge conversion, whereby spin is first mapped onto the charge status of the dot before charge is repeatedly measured via a luminescence recycling transi- tion. We apply our technique to optically monitor the relaxation of an individual spin in a single dot. Our methods open the way to probe slow spin dynamics, for example due to coupling to nuclear spins or via dipolar interactions with proximal spins. We performed our experiments using a single dot charge FIG. 1: (color online) Schematic representation of the device bandstructure biased in the three operating regimes (a) dis- charging, (b) optical charging, and (c) readout. (d) Photolu- minescence intensity for the neutral exciton (X 0), negatively charged exciton (X−1, X−2) and biexciton (2X 0, 2X−1) emis- sion from a single quantum dot as a function of emitted pho- ton energy and applied electric field with DC bias applied (upper panel) and with application of a periodic discharge pulse (lower panel). The inset shows the voltage sequence. storage device (see Refs. 21,22). These devices are grown by molecular beam epitaxy and consist of a single layer of self-assembled In0.5Ga0.5As quantum dots embedded into the intrinsic region of an n-type GaAs Schottky pho- 0 1 0 2 p e S 1 ] l l a h - s e m . t a m - d n o c [ 1 v 7 0 2 0 . 9 0 0 1 : v i X r a todiode. After n+ buffer layers (n = 1 × 1018cm−3) the intrinsic region is formed from the following epitaxial lay- ers in order of growth: 15nm of nominally undoped GaAs followed by a 20nm thick Al0.45Ga0.55As blocking barrier to facilitate charge storage.21 This is followed by a 5nm of undoped GaAs onto which a single layer of quantum dots is grown. The epitaxial layer sequence is completed by 100nm of undoped GaAs. An Ohmic contact is es- tablished to the buried n-contact by annealing a GeAu- NiAu contact at 420◦C in a forming gas atmosphere. Subsequently, 200x300µm2 sized windowed photodiodes are fabricated using photolithography and electron beam metalization with a semitransparent 5nm thick Ti-layer. Shadow mask apertures are then defined to optically iso- late single dots by evaporating a 30nm thick Ti and 200nm thick Au layer over randomly distributed 1µm diameter polystyrene balls (Polybeads). These beads are removed after lift-off to produce circular 1µm diameter apertures in the opaque Au film. As depicted schematically in Fig. 1, the sample can be biased in three different modes of operation: discharging (Fig. 1a), optical electron charging (Fig. 1b) or read- out (Fig. 1c). In the discharging mode the electric field across the QD FQD is tuned such that optically gener- ated electrons and holes are rapidly removed from the dot by tunneling, allowing it to be initially prepared in an uncharged state (Fig. 1a). The lower electric field present in the charging mode (Fig. 1b) ensures that elec- trons can be selectively created in the dot via optical excitation. Here, the 20nm thick AlGaAs tunneling bar- rier immediately below the QD layer inhibits electron tunneling escape whilst holes tunnel out rapidly over a timescale determined by Vbias the voltage applied. In contrast, during readout (Fig. 1c) Vbias is tuned such that the tunneling escape rate of both electrons and holes is much lower than the radiative recombination rate (Γrad ≈0.7ns−1) and carriers can recombine giving rise to photoluminescence (PL). We identified the discharge, charging and readout modes of operation of our device using optical spectroscopy performed as a function of Vbias, laser excitation energy and intensity.21,22 Figure 1d shows Vbias dependent PL spectra recorded from a single dot as a function of the electric field applied when the laser is switched on (FQD=Fread). The upper and lower panels of Fig. 1d compare data recorded whilst Fread is held constant (upper panel) or with a periodic discharging pulse (lower panel), respectively. Without the discharging pulse a number of negatively charged exciton transitions, labeled X−1 and X−2 on Fig. 1d, are observed over the entire range of Fread investigated. The strong emission from negatively charged excitons and complete absence of X 0 emission shows that op- tical charging takes place for the laser power density and frequency used in the experiment.21 In contrast, the lower panel of Fig. 1d shows PL recorded from the same dot whilst periodically polling the electric field from FQD=Fread to Freset=200kV/cm for 200ns prior to each optical excitation cycle. In this case, the negatively 2 charged exciton transitions X−1 and X−2 are accompa- nied by strong emission from the charge neutral exciton X 0 for Fread <30kV/cm. As shown on Fig. 1d the X 0, X−1 and X−2 transitions from the single dot each have characteristic transition energies and, thus, the charge state of the QD (0e, 1e or 2e) can be uniquely deter- mined from the PL yield from different excitonic species. FIG. 2: (color online) (a) Electrical and optical pulse sequence used for the single spin measurements. It consists of five dis- tinct measurement phases: (i) discharging, (ii) spin gener- ation, (iii) spin storage, (iv) spin to charge conversion and (v) charge readout. (b) Demonstration of sequential optical charging of the QD with 1e and 2e. The luminescence in the readout phase from X−1 is generated as the electric field during the first and the second charging step is varied. Using this information we constructed the electrical and optical pulse sequence used for our single spin mea- surements. As depicted schematically in Fig. 2a it con- sists of five distinct phases making up a single measure- ment cycle: (i) discharging, (ii) spin generation, (iii) spin storage, (iv) spin to charge conversion and (v) charge readout. Initially, in phase (i) of the measurement the dot is emptied of all stored charge.22 During the spin- generation phase a single electron is optically created in the dot using a single frequency diode laser resonant with X 0. In practice, the laser frequency is fixed close to the X 0 resonance and we tune the QD transition into reso- nance using the DC Stark effect. As depicted schemat- ically on Fig. 2c this resonance occurs at the 1e charg- ing field F 0 1e. Multiple charging is prevented by the large X 0−X−1 energy splitting of 4.2meV in our QDs (see Fig. 1d). After retention of the optically generated electron for a storage time ∆t, the spin information is mapped onto the charge occupancy of the dot, either 1e or 2e, by the spin-to-charge conversion phase of the measure- ment (iv). Hereby, the negatively charged trion transi- tion (X−1) is tuned into resonance with the laser at the 2e charging field F 0 2e. Finally, the sample is biased into the readout mode (Fig. 1c) whereupon the charge occu- pancy of the dot (1e or 2e) can be probed by measuring timeFQD(t)Laser(i)(ii)(iii)(iv)Readout(v)LaserF1eF2eDtX0X-1EF2e02e chargingX0X-11e chargingF1e0Laserab90100110 F1e (kV/cm)Ibg X-1 Intensity (cps) 5 cpsI1eF1e607080 F2e (kV/cm) 77%I2eF2ecthe relative luminescence yield from X−1. Control mea- surements show that luminescence can be pumped for at least 100µs when optically exciting the system via an ex- cited orbital state without changing the charge state of the QD.21 To perform our measurements we use a magneto-micro photoluminescence setup that provides a magnetic field up to 12T applied along the QD growth axis and typical sample temperatures of T =10K. Resonant optical exci- tation of the QD is achieved using one of two tunable external cavity single frequency lasers (Sacher GmbH) operating around ≈1320meV. The second laser is used to quasi resonantly excite the QD during the readout phase of the measurement via an excited state transition of the QD, approximately 23meV above the neutral exci- ton transition. The laser pulses used during the charging phases of the measurement typically had a duration of 300ns and power density of 0.3W/cm2. During the read- out phase the second laser is gated on for duration of 1µs with a power density of 5W/cm2, while simultaneously turning on a single photon counter (Perkin Elmer SPCM- AQR-16). The control voltage sequence was applied to the sample by an arbitrary waveform generator (Agilent 33250A) that is synchronized to the laser charging 1e, 2e and readout pulses (see Fig. 2a). The switching speed of the device is limited to 100ns by stray capacities and impedance mismatch in the experimental setup. Never- theless, the timing of the voltage sequence and the laser pulses has been adapted to allow voltage stabilization before each of the laser pulses are turned on. The data presented in this manuscript is typically integrated over 106 measurement cycles. In first experiments we begin by applying the pulse sequence depicted in Fig. 2a whilst scanning only the electric field applied during the 1e charging phase of the measurement (F1e) and blocking the laser during the 2e charging phase. As F1e is swept, we monitor the inte- grated intensity of the singly charged states (X−1 and 2X−1) during the readout phase of the measurement. As F1e is tuned through F 0 1e=99kV/cm we observe a clear peak in the X−1 emission intensity (Fig. 2b, left panel), whilst keeping the readout phase of the mea- surement completely unchanged, showing that the dot is successfully charged with a single electron. We then test whether a second charge can be subsequently added. To do this, we unblock the laser during the 2e charging phase of the measurement (iv) and scan F2e to achieve resonance between the excitation laser and X−1. Figure 2b shows a measurement where the pulse sequence dis- played in Fig. 2a is used including both the 1e and 2e charging steps. The first charging field F 0 1e=99kV/cm is fixed to charge with a single electron, while F2e is scanned from 50-85kV/cm to search for the resonance between X−1 and the charging laser. A clear dip is ob- 2e=70kV/cm with a reduced X−1 PL served close to F 0 signal showing that a second electron has been added to the dot (Fig. 2b, right panel). The relative intensity r = (I1e − I2e)/(I1e − Ibg)=0.77 (I1e and I2e being the 3 X−1 intensities at 1e and 2e charging fields and Ibg being the background signal) reflects the efficiency with which we add a second electron to the dot. Until now the spin of the first and second electrons were not considered. At high magnetic fields and linear polarized excitation, the Zeeman splitting of the X 0 line can be spectrally resolved (gex=0.6) providing a con- venient way to select the spin of the first electron gen- erated in the dot and probe its spin projection after a storage time ∆t. Each Zeeman branch can be selectively excited simply by fixing the energy of the charging laser and identifying the charging fields for spin up (F1e,↑) and spin down (F1e,↓) electrons, respectively. A typical mea- surement performed at a magnetic field B=12T is pre- sented in Fig. 3a, clearly showing two charging peaks corresponding to the two different stored electron spin states. The spectrum can be fitted with two Lorentzian peaks, having an energy linewidth of 0.4meV, centered at F1e,↑=87kV/cm and F1e,↓=91kV/cm, respectively. The identification of the spin states is verified by the spin dynamics presented below. FIG. 3: (color online) (a) Measurement of the 1e charging characteristics with a magnetic field B=12T applied in growth direction, for which the Zeeman components of the neutral exciton can be clearly resolved. The Zeeman split absorption lines can be attributed to the creation of spin up (down) elec- trons at electric fields of F1e,↑ and F2e,↑ (F1e,↓ and F2e,↓). (b) A clear suppression of the charging efficiency is detected for addition of parallel spins in the 2e charging step, demonstrat- ing spin to charge conversion. The selective generation of a spin orientated electron in the QD is confirmed by measuring the spectral char- acteristics of the 2e charging dip (see Fig. 3b) and its evolution with the storage time (∆t) between the 1e and 2e charging phases of the measurement (Fig. 2a). Figure 3b shows the 2e charging dips measured following exci- tation of a spin up (F1e,↑=87kV/cm - squares) or spin down (F1e,↓=91kV/cm - circles) electron in the dot, re- spectively. Upon exciting the first electron spin up and storing it for 800ns, we observe a pronounced 2e charg- ing dip at F2e,↓=58kV/cm (squares - Fig. 3b). In strong contrast, upon exciting the first electron spin down we observe the 2e charging dip at F2e,↑=56.5kV/cm (circles - Fig. 3b). Mapped onto an energy scale, the split- ting between the F2e,↑ and F2e,↓ 2e charging dips cor- respond precisely to the observed Zeeman splitting of the 1e charging peaks. This is expected since the Zee- 8090100406080100505560651.00.80.60.40.20.0 Relative X-1 Intensity F2e (kV/cm) X-1 Intensity (cps) F1e (kV/cm)abF1e,↑F1e,↓F2e,↑F2e,↓B=12T4 We now apply our methods to monitor the spin re- laxation of a single, optically prepared electron. Again, the pulse sequence depicted in Fig. 2a is used and ∆t is varied from 0.8µs to 16µs by changing the delay between the 1e and 2e charging pulses. The resulting data is pre- sented in Figs. 5a and 5b. The 1e charging electric field F1e is chosen such as to initialize spin down (Fig. 5a) or spin up (Fig. 5b) electrons, respectively. The data is presented as probabilities p↑,↓ = r↑,↓/(r↓ + r↑) to find the electron in a particular spin orientation after the storage time ∆t. For short storage times, a dominance of the spin orientation that was chosen for initialization is ob- served. As the storage time increases the system evolves towards thermal equilibrium, which can be described by the Boltzmann statistics of non-interacting spins. In all four traces of Figs. 5a and 5b a mono-exponential de- cay can be fitted with a single time constant, the spin relaxation time T1=3.1±0.4µs. This value is in excellent agreement with lifetime measurements performed under similar conditions on ensembles of self-assembled QDs,11 showing that relaxation is mediated by spin-orbit cou- pling mediated by single phonon scattering.4,10,11 man splitting of X 0 and X−1 are identical.23 These ob- servations clearly show that the spin projection of the first electron generated in the dot can be directly mea- sured via the relative amplitudes of the 2e charging dips. Moreover, they demonstrate that the spin projection is retained across the storage phase of the measurement, before the 2e charging step is attempted. To gain more insight on the selectivity of the spin gen- eration, we performed measurements of the initial de- gree of spin polarization ρ = r↑ − r↓ as a function of F1e. The degree of polarization is presented in Fig. 4. The measurement was performed using a storage time of ∆t=0.8µs. To compensate for relaxation effects the de- gree of polarization has been extrapolated to ∆t=0 using the relaxation dynamics detected in time resolved experi- ments presented later. The minimum and maximum of ρ are found around F1e,↑=87kV/cm and F1e,↓=91kV/cm, with values of 60% and 65%, respectively. Between these two maxima a sharp transition is observed with a change of sign, whilst towards lower and higher electric fields, the degree of polarization decreases to zero. With increasing detuning from the charging resonance, the 1e charging efficiency reduces rapidly and, thus, the fidelity of the measurement degrades. The efficiency of spin up and spin down generation can be estimated by fitting the 1e charg- ing peak with two Lorentzian peaks (Fig. 3a). The max- imum degree of polarization for undisturbed spin gener- ation is then calculated from these efficiencies and shown as full line in Fig. 4. Since the calculated values coin- cide well with the measured degree of polarization, we conclude that pure spin initialization is only limited by the selective addressing of a single Zeeman level. Spin initialization can, therefore, either be improved by us- ing quantum dots with larger Zeeman splitting or by the use of circular polarized light. These observations clearly demonstrate that the device operates as an optically ad- dressable single electron spin memory. FIG. 4: (color online) The degree of polarization ρ = r↑ − r↓ measured as a function of the first charging field F1e. The maximum values of ρ are limited by the spectral separation of the two excitonic Zeeman levels. The full line shows the ex- pectation based on two Lorentzian transitions with a spectral linewidth of 0.4meV, determined by the tunneling time of the hole out of the dot. The good agreement indicates that the fidelity of the spin preparation is limited only by the relatively low excitonic g-factor in this sample. FIG. 5: (color online) (a, b) Time evolution of the probabili- ties for a particular electron spin orientation after spin down (a) and spin up (b) initialization. An electron spin relax- ation time of T1=3.1±0.4µs is extracted from this data. (c) Temperature dependence of the 2e charging efficiency after initialization of a spin up electron. The data is presented as (r↑ − r↓) normalized to the initial value on a semi-logarithmic scale. (d) Extracted spin relaxation times fitted to the data presented in (a). As expected, relaxation times decrease with increasing temperature. Similar measurements performed as a function of lat- tice temperature reveal that T1 decreases with increasing temperature, from 9.9±1.1µs at T =1.4K to 1.4±0.2µs at T =20K (Fig. 5c). Note that in the case of the low tem- perature measurements an extremely long storage time of 34µs is used, during which reliable spin detection is F1e,↓F1e,↑80859095100-100-75-50-250255075100  (%)F1e (kV/cm)0102030400.010.11110110051015020406080100 Probability %Storage Time t (µs)051015p pppT=10 KT=20 K Normalized Intensity (arb. u.)Storage Time t (s)T=1.4 KT (K)Lifetime (µs) m=-0.70.1cdab5 still possible. A power T1 ∝ T m law with m=-0.7±0.1 is fitted (Fig. 5d), in agreement with the expected linear behavior for phonon mediated spin-orbit coupling of the two spin states in the dot.5 In summary, we have demonstrated a single QD op- tical spin memory device and used it to monitor spin relaxation of an individual electron over timescales ap- proaching a millisecond. We showed that the spin of the optically prepared electron can be selected simply by choosing the gate voltage applied to the device and, moreover, read out the spin state using an all optical im- plementation of a spin to charge conversion technique. When compared with the spin storage technique em- ployed previously,11 where each spin generated one pho- ton per measurement cycle, our results represent an in- crease of four orders of magnitude in photons per spin, per measurement cycle. This was made possible by map- ping the spin information onto the more robust charge status of the QD and then measuring charge. Whilst our methods already allow us to probe an individual spin in a single QD, with further optimization of the opti- cal collection efficiency and sample design, single shot spin readout is conceivable. Finally, we note that per- turbations arising from illumination, DC currents and co-tunneling phenomena are absent in our experiment whilst the spin is stored in the QD. When used in combi- nation with fast optical spin manipulation or microwave pulses, this builds an ideal basis for probing spin decoher- ence in single self-assembled QD over long timescales and developing optimal methods for coherent spin control. The authors gratefully acknowledge financial support by the DFG via SFB631, the German Excellence Initia- tive via the Nanosystems Initiative Munich (NIM) and the European Union via SOLID. ∗ These authors contributed equally to this work † Electronic address: [email protected] 1 D. Loss and D. P. DiVincenzo, Phys. Rev. A 57, 120 (1998). 2 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). 3 S. Amasha, K. MacLean, Iuliana P. Radu, D. M. Zumbuhl, M. A. Kastner, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 100, 046803 (2008). 4 M. Atature, J. Dreiser, A. Badolato, A. Hogele, K. Karrai, and A. Imamoglu, Science 312, 5773, (2006). 5 R. Hanson, L. P. Kouwenhoven, J. R. Petta, S. Tarucha, and L. M. K. Vandersypen, Rev. Mod. Phys. 79, 1217 (2007). 6 K. C. Nowack, F. H. L. Koppens, Yu. V. Nazarov, and L. M. K. Vandersypen, Science 30, 318, 1430 (2007). 7 M. H. Mikkelsen, J. Berezovsky, N. G. Stoltz, L. A. Col- dren, and D. D. Awschalom, Nature Physics 3, 770 (2007). 8 B. D. Gerardot, D. Brunner, P. A. Dalgarno, P. Ohberg, S. Seidl, M. Kroner, K. Karrai, N. G. Stoltz, P. M. Petroff, and R. J. Warburton, Nature 451, 441 (2008). 9 D. Press, T. D. Ladd, B. Zhang, and Y. Yamamoto, Nature 456, 218 (2008). 10 D. Brunner, B. D. Gerardot, P. A. Dalgarno, G.Wust, K. Karrai, N. G. Stoltz, P. M. Petroff, and R. J. Warburton, Science 325, 70 (2009). 11 M. Kroutvar, Y. Ducommun, D. Heiss, M. Bichler, D. Schuh, G. Abstreiter, and J.J. Finley, Nature 432, 81 (2004). 12 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). 13 A. J. Ramsay, S. J. Boyle, R. S. Kolodka, J. B. B. Oliveira, J. Skiba-Szymanska, H. Y. Liu, M. Hopkinson, A. M. Fox, and M. S. Skolnick, Phys. Rev. Lett. 100, 197401 (2008). 14 X. Xu, Y. Wu, B. Sun, Q. Huang, J. Cheng, D. G. Steel, A. S. Bracker, D. Gammon, C. Emary, and L. J. Sham. Phys. Rev. Lett 99, 097401 (2007). 15 M. Atature, J. Dreiser, A. Badolato, and A. Imamoglu, Nature Physics 3, 101 (2007). 16 A. Greilich, D. R. Yakovlev, A. Shabaev, Al. L. Efros, I. A. Yugova, R. Oulton, V. Stavarache, D. Reuter, A. Wieck, and M. Bayer. Science 313, 341 (2006). 17 A. Hogele, M. Kroner, S. Seidl, K. Karrai, M. Atature, J. Dreiser, A. Imamoglu, and R.J. Warburton, Appl. Phys. Lett. 86, 221905 (2005). 18 J. Berezovsky, O. Gywat, F. Meier, D. Battaglia, X. Peng, and D. D. Awschalom, Nature Physics 2, 831 (2006). 19 C.-Y. Lu, Y. Zhao, A. N. Vamivakas, C. Matthiesen, S. Falt, A. Badolato, and M. Atature, Phys. Rev. B 81, 035332 (2010). 20 H. J. Krenner, M. Sabathil, E. C. Clark, A. Kress, D. Schuh, M. Bichler, G. Abstreiter, and J. J. Finley, Phys. Rev. Lett. 94, 057402 (2005). 21 D. Heiss, V. Jovanov, M. Bichler, G. Abstreiter, and J. J. Finley, Phys. Rev. B 77, 235442 (2008). 22 D. Heiss, V. Jovanov, M.Caesar, M. Bichler, G. Abstreiter, and J.J. Finley, Appl. Phys. Lett. 94, 7 (2009). 23 M. Bayer, G. Ortner, O. Stern, A. Kuther, A. A. Gor- bunov, A. Forchel, P. Hawrylak, S. Fafard, K. Hinzer, T. L. Reinecke, S. N. Walck, J. P. Reithmaier, F. Klopf and F. Schafer, Phys. Rev. B 65, 195315 (2002).
1406.3649	3	1406	2015-06-01T19:38:38	Spontaneous emission from a quantum dot in a structured photonic reservoir: phonon-mediated breakdown of Fermi's golden rule	[ "cond-mat.mes-hall", "physics.optics", "quant-ph" ]	We describe how a structured photonic medium controls the spontaneous emission rate from an excited quantum dot in the presence of electron-phonon coupling. We analyze this problem using a polaron transformed master equation and we consider specific examples of a photonic crystal cavity and a coupled cavity waveguide. We find that when the relaxation times of the photon and phonon baths are comparable, phonons influence spontaneous emission in a non-trivial way. We demonstrate why and how the broadband frequency dependence of the local photon density of states determines the photon emission rate, manifesting in a complete breakdown of Fermi's golden rule. For a single cavity resonance, we generalize Purcell's formula to include the effects of electron-phonon coupling. For a waveguide, we show a suppression and a 200-fold enhancement of the photon emission rate.	cond-mat.mes-hall	cond-mat	Spontaneous emission from a quantum dot in a structured photonic reservoir: phonon-mediated breakdown of Fermi's golden rule Kaushik Roy-Choudhury and Stephen Hughes Department of Physics, Queen's University, Kingston, Ontario, Canada, K7L 3N6 Compiled June 28, 2018 Quantum dots in semiconductor photonic reservoirs are important systems for studying and exploiting quantum optics on a chip, and it is essential to understand fundamental concepts such as spontaneous emission. According to Fermi's golden rule, the spontaneous emission rate of a quantum emitter weakly coupled to a structured photonic reservoir is proportional to the local density of photon states (LDOS) at the emitter's position and frequency. Coupling to lattice vibrations or phonons however significantly modifies the emission properties of a quantum dot compared to an isolated emitter (e.g., an atom). In the regime of phonon-dressed reservoir coupling, we demonstrate why and how the broadband frequency dependence of the LDOS determines the spontaneous emission rate of a QD, manifesting in a dramatic breakdown of Fermi's golden rule. We analyze this problem using a polaron transformed master equation and consider specific examples of a semiconductor microcavity and a coupled cavity waveguide. For a leaky single cavity resonance, we generalize Purcell's formula to include the effects of electron-phonon coupling and for a waveguide, we show a suppression and a 200- fold enhancement of the photon emission rate. These results have important consequences for modelling and understanding emerging QD experiments in a wide range of photonic reservoir systems. c(cid:13) 2018 Optical Society of America OCIS codes: (270.0270); (350.4238); (160.6000). Recent developments in chip-scale quantum optical technologies [1] have generated substantial interest in quantum dots (QDs) which act as "artificial atoms" in solid-state media. However, electron-phonon coupling in solid-state media has been shown to significantly modify the emission properties of a QD as compared to an iso- lated atom [2]. Studying phonon interactions in govern- ing the emission properties of QDs has been an intense area of research, leading to a number of effects beyond a simple pure-dephasing model [3]. For driven QD excitons yielding Rabi oscillations, phonon coupling manifests in damping and frequency shifts [4–6]. In QD-cavity sys- tems, phonons cause intensity-dependent broadening of Mollow side-bands [7], off-resonant cavity feeding [8] and asymmetric vacuum Rabi doublets [9, 10]. Semiconductor quantum dots (QDs) coupled to struc- tured photonic reservoirs, provide a promising plat- form for tailoring light-matter interaction in a solid- state environment. One of the primary interests in cou- pling QDs to structured reservoirs is for modifying the spontaneous emission rate (SE), γ, via the Purcell ef- fect [11, 12]. Photonic crystals are a paradigm exam- ple of a structured photonic reservoir, and both pho- tonic crystal cavities (Fig.1(a)) and coupled-cavity op- tical waveguide (CROW, Fig.1(b)) structures have been investigated for modifying QD SE rates [12, 13]. For an unstructured reservoir, γ remains unchanged in the pres- ence of phonons [14]. For structured reservoirs, previ- ous theories have assumed phonon processes to be much faster than all relevant system dynamics [15, 16], thus restricting them to structures with sharp variations of photon local density of states (LDOS) (e.g., high-Q cav- ity or photonic band edge). A primary example of a structured reservoir is a microcavity, and existing the- Fig. 1. Two photonic reservoir systems under considera- tion and an energy level picture of the various quantum states. Schematic of a semiconductor cavity (a) and waveguide (b) using a photonic crystal platform, containing a single QD. (c) Energy level diagram of a neutral QD (electron-hole pair) inter- acting with a phonon bath and a photon bath. The operator f † creates a photon and the b† q operator creates a phonon. k ories [15,17,18] treat the cavity mode as a system opera- tor and find that phonons modify the QD-cavity coupling rate, through g → hBig [17, 19], where hBi is the ther- mal average of the coherent phonon bath displacement operators B± [17]. Hence the Purcell factor is believed to scale as g2 → hBi2g2 [16]. However, such theories do not apply to large κ cavities, where κ is the cavity decay rate, and one would expect to recover the result that γ-and thus g-are not affected by phonons. More- over, for an arbitrary photonic bath medium, it is not known how phonons affect the SE rates, yet clearly such an effect is of significant fundamental interest and also important for understanding emerging QD experiments. 1 In this Letter we introduce a self-consistent ME ap- proach with both phonon and photon reservoirs included and we explore in detail the influence of a photon reser- voir on the phonon-modified SE rate. When the relax- ation times of the photon and phonon baths compare, the frequency dependence of the LDOS is found to dic- tate how phonons modify the SE rates, causing a clear breakdown of Fermi's golden rule. Such an effect arises due to dressing of QD excitations by phonons in a solid state medium. Importantly, our theory can be applied to any general LDOS medium, and as specific examples we consider a semiconductor microcavity and a slow-light coupled cavity waveguide. We model the QD as a two-level system interacting with an inhomogeneous semiconductor-based photonic reservoir and an acoustic phonon bath [17] (see Fig. 1 (c)). Assuming the QD of dipole moment d = dnd at spa- tial position rd, the total Hamiltonian of the system in a frame rotating at the QD exciton frequency ωx, is [20] πǫ0 , with Jph(τ ) = R ∞ Jpn(ω) is the phonon spectral function [21]. The pho- ton bath correlation function can be expressed in terms of the photon-reservoir spectral function Jph(ω) = d·Im[G(rd,rd;ω)]·d 0 dωJph(ω)ei(ωx−ω)τ . In the Markov limit (t → ∞), equation (2) generalizes Fermi's golden rule for QD SE, since the LDOS at various frequencies can now contribute to the phonon-modified SE rate, γ. Similar expressions for the SE rate in the fre- quency domain have been used to explain mode pulling effects in QD cavities [22]. In the absence of phonon cou- pling, the SE decay rate of the QD in a structured pho- 0 Re[Jph(τ )]dτ , where γ(t → ∞) ∝ LDOS(ωx). McCutcheon and Nazir [14] use a similar approach to show that γ → γ for a free-space bath function. ton reservoir reduces to γ(t) = 2R t To appreciate how phonons modify the SE rate in a structured photonic reservoir, we first consider a sim- ple Lorentzian cavity (cf. Fig. 1(a)). For a single cavity mode, in a dielectric with a dielectric constant ǫ = n2 b, dω f†(r, ω)f(r, ω) + Σqωqb† qbq Jph(ω) = g2 1 π κ 2 (ω − ωc)2 + ( κ 2 )2 , (3) 1 2 d2ω 2ǫ0n2 bVeffi where g = h is the QD-cavity coupling rate, and the QD has its dipole aligned with the cavity mode polarization and is positioned at the field antinode. Defining the long-time SE rate as γ ≡ γ(t → ∞), and similarly for γ, from equations (2)-(3), we obtain γ = 2g2 hBi2 Re(cid:20)Z ∞ 0 eφ(τ )e−i∆cxτ −κτ /2dτ(cid:21) , (4) where hBi = exp[− 1 2 φ(0)] [19] and ∆cx = ωc − ωx. We can now generalize the Purcell factor [PF] for the en- hanced SE rate of a QD in a semiconductor cavity: PF =" 3 4π2 (cid:18) λ0 Veff nb(cid:19)3 Q κ2 4 4 !# χ, ∆2 cx + κ2 (5) where λ0 = ωc/(2πc), Q = ωc/κ is the quality factor, and χ ≡ γ/γ is the phonon-modification factor. κ 2 ∆2 i.e., cx+( κ + 2g2 hBi2 2 hBi2 g2Re[R ∞ To obtain a mean-field approximation in the high in the limit κ ≪ 2π/τpn (where Q limit, τpn ≈ 1 ps is the phonon relaxation time), then γ → γmean = Γa†σ− 2 )2 , where the phonon-mediated cavity scattering rate Γa†σ− = 0 dτ e−i∆cxτ (eφ(τ ) − 1)] [21]. This is ex- actly the expression for SE rate [15, 21] which is derived (See Supplement 1, Sec. 3) with a polaron ME [21] when treating the cavity mode as a system operator, with phe- nomenological damping κ and using the bad cavity limit (κ > g/2). Our theory thus not only recovers previous (polaronic) cavity-QED results in the appropriate limit, but also reveals a fundamental limitation of these ap- proaches for sufficiently large κ (low Q) cavities. Specif- ically, when the cavity relaxation time becomes compa- rable to τpn, or smaller, these formalisms break down. To test this hypothesis, consider the example of three different cavity decay rates, κ = 0.06, 0.6, and 2.4 meV, 2 H = Z drZ ∞ −(cid:20)σ+eiωxtZ ∞ 0 0 dω d · E(rd, ω) + H.c(cid:21) + σ+σ−Σqλq(b† q + bq), (1) where σ+/σ− are the Pauli operators of the exciton (electron-hole pair), bq/b† q are the annihilation and cre- ation operators of the acoustic phonon reservoir and λq is the exciton-phonon coupling strength. The operators f/f† are the boson field operators of the photon reservoir and these satisfy the usual commutation rules for bo- son operators. The interaction between the QD and the photonic reservoir is written using the dipole and the ro- tating wave approximations. The electric-field operator E(r, ω) is related to the Green function of the medium, G(r, r′; ω), which satisfies the Kramers-Kronig rela- tions [20]. To include phonon interactions to all orders, we perform the polaron transform on the Hamiltonian H given by H ′ → eP He−P where P = σ+σ−Σq q − bq) [19]. Assuming weak-to-intermediate coupling with the photon bath, we derive a time-local polaron ME for the QD reduced density operator ρ, using the Born ap- proximation. The usual incoherent terms from the pho- ton reservoir can be written as ∂ρ inc = Lph(ρ). Subse- quently, the phonon-modified SE decay rate can be ob- tained from Re(Lph), yielding the familiar Lindblad su- peroperator, γ(t) 1 2 (2σ−ρσ+ − σ+σ−ρ − ρσ+σ−), where the SE decay rate is derived to be (See Supplement 1, Sec. 2) ∂t ph (b† λq ωq γ(t) = 2Z t 0 Re[Cpn(τ )Jph(τ )]dτ, (2) where Jph(τ ) and Cpn(τ ) are the photon and the phonon bath correlation functions, respectively. The phonon bath correlation function Cpn(τ ) is defined as Cpn(τ ) = e[φ(τ )−φ(0)] where φ(t) = 0 dω Jpn(ω) [coth(ω/2kBT ) cos(ωt) − i sin(ωt)], and ω2 R ∞ 300 a F P 0 30 c F P 0 8 e F P 4 0 −2 0 ∆ xc /κ 30 b χ 15 χ d 0 4 2 0 1.5 f χ 1 0.5 −4 2 ω l 30 a F P 20 10 h p ) ) t ( J ( e R ω u 4 0 0 1 t (ps) 2 3 −4 −2 0 2 4 6 b χ 1.2 1 0.8 −4 −2 0 2 4 0 −6 2 1 0 ) χ ( 0 1 g o l −2 0 (meV) ∆ xc 2 4 −6 −4 −2 0 ω −ω (meV) x 0 2 4 6 Fig. 2. Phonon-modified spontaneous emission of a QD in a dielectric cavity Near-resonant PFs (left panels) and phonon- modified SE factor χ (right panels) for cavities with κ = 0.06 meV (a, b), 0.6 meV (c, d) and 2.4 meV (e, f). We use the continuous form of the phonon spectral function, Jpn(ω) = αpω3 exp[− ω2 2ω2 b for longitudinal acoustic phonon interaction, and adopt experimen- tal numbers for InAs QDs [2]: cavity-QD coupling rate g = 0.08 meV, phonon cutoff frequency ωb = 1 meV, and exciton-phonon coupling strength αp/(2π)2 = 0.06 ps2 [21]. The dark (red) dashed line on left panels is the PF without phonons. The dark and light solid lines correspond to phonon-modified PF and χ at T = 4 K and 40 K, respectively. The dashed lines on the right panels plot χmean (see text) at T = 4 K (light) and 40 K (dark). ], Fig. 3. Phonon-modified spontaneous emission of a QD in a coupled-cavity waveguide (a) Purcell factors and (b) phonon- modified SE factor χ in a log10 scale, for a PC CROW structure, where ω0 marks the band-center. For the waveguide, PF = γ/γh, where γh = d2nbω3/(6πǫ0c3) is the SE rate within the homoge- nous slab material. The solid (dashed) line in (a) represents PF with (without) phonons; the inset shows Jph(t) at the upper mode edge (light line) and band center (dark line) in arbitrary units; the dashed line shows the simple exponential damping function e−λt (see text). The phonon calculations are performed at T = 40 K. Inset in (b) plots χ in a linear scale to show better display χ inside the waveguide band. LDOS, while γnl = hBi2 Re[2R ∞ corresponding to Q factor of around 23000, 2300, and 600, respectively (with ωc/2π = 1440 meV). In Fig. 2 we plot the PF (left panels) and phonon-modified SE factor χ = γ/γ (right panels). For each cavity, we investigate two different bath temperatures (4 K and 40 K), and the dashed lines on the left panels represent PFs without phonon modification. The clear asymmetry in χ about the LDOS peak arises since phonon emission is more probable than absorption [21] at low temperatures. The main results can be explained by writing γ = hBi2 γ +γnl where hBi2 γ is the coherently renormalized bare SE rate and arises due to local (ωx) sampling of photonic 0 (eφ(τ ) − 1)Jph(τ )]dτ ac- counts for the non-local contribution (i.e., frequencies that would not contribute to a Fermi's golden rule ex- pression). Note when κ is small, γnl → Γa†σ− . Due to the non-local component, the reduction of the SE rate is always ≥ hBi2, at zero detuning. At large detunings, the non-local component dominates leading to an overall en- hancement of the SE rate. Figures 2(b, d, f) show that χ varies significantly over several meV. The dashed lines on right panels represent χmean = γmean/γ, which evidently differs from our full calculations in the limit κ ≈ τ −1 pn (Figs.2 (d,f)). This is because the reservoir structure of the photon bath is properly accounted in the present calculations and is not approximated as a high-Q cavity (also see Supplement 1, Sec 1). It is also important to note that low-Q (several hundred) cavities are commonly employed for measuring the vertical emission from QDs 3 in planar cavities [2] and for modifying the SE rates in simple photonic crystal cavities [12], while intermediate Q (≈ 3000) cavities are used for all optical switching [23] and enabling single photon sources [24]. We now depart from the simple cavity, and con- sider the richer case of a photonic crystal CROW (cf. Fig. 1(b)). Photonic crystal waveguides (Fig. 1(b)) can be used for realizing slow-light propagation [25] and for manipulating the emission properties of embedded QDs for on-chip single photon emission [26, 27], with a number of recent experiments emerging. Current theories in this regime either ignore phonon coupling or assume a coherent renormalization factor (hBi2 γ), consistent with a Fermi's golden rule (modified by a mean-field reduc- tion factor). For the photon reservoir function, we adopt a model LDOS for a CROW [28], and use an analytical tight-binding technique to calculate the photonic band structure [29]; the photon reservoir spectral function is obtained analytically, as Jph(ω) = −d2ω 2ǫ0n2 bVeff 1 π Im" l )# , (6) 1 p(ω − ωu)(ω − ω∗ where ωu,l = ωu,l ± iκu,l [29], ωu,l is the mode-edge fre- quencies of the waveguide (see Fig. 3(a)), κu,l are effec- tive damping rates, and Veff is the mode volume of a single cavity. The photonic LDOS has a rich non-trivial spectral structure compared to a smooth Lorentzian cav- ity, especially within the band width of the phonon bath (which spans about 5-10 meV (See Supplement 1, Sec. 1). For our calculations, we use parameters that closely rep- resent a CROW made up of a local width modulation of a line-defect photonic crystal cavity [30] which yields a band structure [31] consistent with experiments [32]. In Fig. 3(a), we show the calculated PF with (solid) and without (dashed) phonons, using a bath tempera- ture of 40 K. We see that in contrast to current theo- ries, phonons significantly influence the spectral shape of the SE rates, causing a reduction at the mode edges and a significant enhancement inside and outside the waveguide band. Figure 3(b) shows a slight asymme- try (which increases with decreasing temperatures) in χ which is again due to unequal phonon emission and absorption rates. The spectral dependence of χ in a the waveguide can be qualitatively understood by treat- ing Jph(ω) approximately as a sum of two Lorentzians located at the mode edges (ωu,l). The corresponding J approx (t) = e−(iωu+λ)t +e−(iωl+λ)t is a sum of two expo- ph nentially damped oscillatory functions, where 2λ is the bandwidth of the waveguide mode edge LDOS. At a sharp mode-edge of the LDOS, the contribu- tion from local (ωx) photonic LDOS dominates and χ ≈ hBi2 [16]. Away from the mode-edge (Figs. 3(a-b)), SE rate is enhanced due to non-local effects. This simple model discussion is, however, approximate as in reality the mode-edge LDOS is non-Lorentzian. For a symmet- ric Lorentzian with the same bandwidth, λ−1 ≈ 50 ps, Jph(t) damps much faster than λ−1 initially and damps very slowly thereafter (Fig. 3(a), inset). The long time decay rate is set by the linewidth of the sharper side of the mode-edge LDOS (≈ 0.1 ns). This non-Lorentzian mode-edge in turn leads to a very strong enhancement of the PF (×200) outside the waveguide band (Fig. 3(b)), compared to a symmetric Lorentzian line-shape (see Fig. 2(b)). In conclusion, we have demonstrated how the fre- quency dependence of the LDOS of a photonic reser- voir determines the extent to which phonons modify the SE of a coupled QD. The relative dynamics be- tween the phonon and the photon bath correlation func- tions is found to play a fundamentally important role; specifically, when the relaxation times are comparable, phonons strongly modify the emission spectra leading to non-Lorentzian cavity lineshapes and even enhanced SE. These effects are not obtained using the usual Fermi's golden rule. Our formalism is important for understand- ing related experiments with QD-cavity systems, such as with photoluminescence intensity measurements with a coherent drive [33], and is broadly applicable to various photonic reservoir systems. We thank D. P. S. McCutcheon and A. Nazir for useful discussions and for sharing their results of [14] prior to publication. Funding Information References 1. H. Kim, R. Bose, T. Shen, G. Solomon and E. Waks, Nature Photon. 7, 373 (2013). 2. S. Weiler, A. Ulhaq, S. M. Ulrich, D. Richter, M. Jetter, P. Michler, C. Roy, and S. Hughes, Phys. Rev. B 86, 241304(R) (2012). 3. B. Krummheuer, V. M. Axt and T. Kuhn, Phys. Rev. B 65, 195313 (2002). 4. J. Forstner, C. Weber, J. Danckwerts and A. Knorr, Phys. Rev. Lett. 91, 127401 (2003). 5. A. J. Ramsay, T. M. Godden, S. J. Boyle, E. M. Gauger, A. Nazir, B. W. Lovett, A. M. Fox, and M. S. Skolnick, Phys. Rev. Lett. 105, 177402 (2010). 6. L. Monniello, C. Tonin, R. Hostein, A. Lemaitre, A. Martinez, V. Voliotis, and R. Grousson, Phys. Rev. Lett. 111, 026403 (2013). 7. S. M. Ulrich, S. Ates, S. Reitzenstein, A. Loffler, A. Forchel, and P. Michler, Phys. Rev. Lett. 106, 247402 (2011). 8. A. Majumdar, E. D. Kim, Y. Gong, M. Bajcsy, and J. Vuckovi´c, Phys. Rev. B 84, 085309 (2011). 9. F. Milde, A. Knorr and S. Hughes, Phys. Rev. B 78, 035330 (2008). 10. Y. Ota, S. Iwamoto, N. Kumagai & Y. Arakawa, arXiv:0908.0788 (2009). 11. E. M. Purcell, Phys. Rev. 69, 681 (1946). 12. D. Englund, E. Waks, G. Solomon, B. Zhang, T. Nakaoka, Y. Arakawa, Y. Yamamoto, and J. Vuckovi´c, Phys. Rev. Lett. 95, 013904 (2005). 13. T. Lund-Hansen, S. Stobbe, B. Julsgaard, H. Thyrrestrup, T. Sunner, M. Kamp, A. Forchel, and P. Lodahl, Phys. Rev. Lett. 101, 113903 (2008) 14. D. P. S. McCutcheon and A. Nazir, Phys. Rev. Lett. 110, 217401 (2013); see also the Supplementary Mate- rial. 15. P. Kaer, T. R. Nielsen, P. Lodahl, A.-P. Jauho and J. Mørk, Phys. Rev. B 86, 085302 (2012). 16. C. Roy and S. John, Phys. Rev. A 81, 023817 (2010). 17. C. Roy and S. Hughes, Phys. Rev. Lett. 106, 247403 (2011). 18. U. Hohenester, Phys. Rev. B 81, 155303 (2010). 19. I. Wilson-Rae and A. Imamoglu, Phys. Rev. B 65, 235311 (2002). 20. S. Scheel, L. Knoll and D.-G. Welsch, Phys. Rev. A 60, 4094 (1999). 21. C. Roy and S. Hughes, Phys. Rev. X 1, 021009 (2011). 22. D. Valente, J. Suffczy´nski, T. Jakubczyk, A. Dousse, A. Lemaıtre, I. Sagnes, Loıc Lanco, P. Voisin, A. Auff´eves, and P. Senellart, Phys. Rev. B 89, 041302(R) (2014). 23. E. Waks and J. Vuckovi´c, Phys. Rev. Lett. 96, 153601 (2006). 24. C. A. Foell, E. Schelew, H. Qiao, K. A. Abel, S. Hughes, F. C. J. M. van Veggel, and J. F. Young, Opt. Express 20, 10453 (2012). 25. M. Notomi, K. Yamada, A. Shinya, J. Takahashi, C. Takahashi, and I. Yokohama, Phys. Rev. Lett. 87, 253902 (2001). 26. V. S. C. Manga Rao and S. Hughes, Phys. Rev. B 75, 205437 (2007). Natural Sciences and Engineering Research Council of Canada. 27. A. Schwagmann, S. Kalliakos, I. Farrer, J. P. Griffiths, G. A. C. Jones, D. A. Ritchie and A. J. Shields, Appl. 4 Phys. Lett. 99, 261108 (2011). 28. A. Yariv, Y. Xu, R. K. Lee and A. Scherer, Opt. Lett. 24, 711 (1999). 29. D. P. Fussell and M. M. Dignam, Phys. Rev. A 76, 053801 (2007). 30. E. Kuramochi, M. Notomi, S. Mitsugi, A. Shinya, T. Tanabe, and T. Watanabe, Appl. Phys. Lett. 88, 041112 (2006). 31. D. P. Fussell, S. Hughes and M. M. Dignam, Phys. Rev. B 78, 144201 (2008). 32. M. Notomi, E. Kuramochi and T. Tanabe, Nature Pho- ton. 2, 741 (2008). 33. K. Roy-Choudhury and S. Hughes, Optics Letters 40, 1838 (2015). 5 5 1 0 2 n u J 1 ] l l a h - s e m . t a m - d n o c [ 3 v 9 4 6 3 . 6 0 4 1 : v i X r a Supplementary material for "Spontaneous emission from a quantum dot in a structured photonic reservoir: phonon-mediated breakdown of Fermi's golden rule" Kaushik Roy-Choudhury and Stephen Hughes Department of Physics, Queen's University, Kingston, Ontario, Canada, K7L 3N6 In this document we present supplementary information for our accompanying manuscript "Spontaneous emission from a quantum dot in a structured photonic reservoir: phonon- mediated breakdown of Fermi's golden rule". First, we show the photon and photon spectral function in the frequency and time domains, using the simple cavity structures introduced in the main text. Second, we show a derivation of the phonon-modified spontaneous emission (SE) rate, equation (2) in our main paper. Third, we derive the cavity-induced SE rate from a previous polaron master equations using cavity-QED equations, and show direct agreement with our results in the high Q limit (i.e., with γmean in the main paper). Last, we discuss how one can include the influence from a coher- ent pump field into the quantum master equation and incoherent scattering terms, while c(cid:13) 2018 Optical Society of America accounting for both photon and phonon bath coupling. OCIS codes: (270.0270) ; (350.4238) ; (160.6000) . S1. Cavity Photon and phonon bath functions For phonon bath temperatures of 4 K and 40 K, Figs. S1(a) and S1(b) show the phonon correlation function versus time and frequency, respectively. For simplicity, a polaron shift ∆P (defined below in Section S2) is implicitly absorbed in the definition of ωx, and we also define a phonon correlation function that decays to pn(τ ) = eφ(τ ) − 1. All phonon parameters and φ(τ ) are given in our main paper. The time zero, through C′ evolution of the real part of the phonon correlation function shows that typical phonon correlation times are very fast (τpn ≤ 3 ps). For comparison, we also show a photon correlation function in Figs. S1(c) and S1(d), for the three different values of κ used in our paper: κ = 2.4 meV (thick dark line), 0.6 meV (thin dark line), and 0.06 meV (thin light line). The cavity bath correlation functions Jph(t) are oscillatory functions damped at the cavity decay rate, with an oscillation frequency is determined by the QD-cavity detuning ∆xc, and we show the case for zero detuning. From equation (2) in our paper, and Fig. S1 below, we expect that (i) phonons should not influence the SE rate γ in a strongly damped cavity and (ii) phonons will reduce the SE rate to its mean-field limit [1] (hBi2γ) only for a weakly damped cavity. Significant deviations from these two limits occur when the damping times of the photon and phonon correlation functions are comparable. This is shown explicitly in Fig. 2 of the main text using the cavity spectral functions shown below. It is also shown explicitly in Fig. 3 for the slow-light photonic crystal waveguide. S2. Derivation of phonon-modified spontaneous emission rate Theoretical descriptions of electron-phonon scattering in QDs systems range from the independent Boson model [2], correlation expansion [3], perturbative master equations (MEs) [4], to polaron MEs [5–8], varia- tional MEs [9] and real-time path integrals [10]. In our Letter we introduce a self-consistent ME approach 1 n p ) ) t ( ' C ( e R ) . u . b r a ( ) ) t ( h p J ( e R a c 2 1 0 8 4 0 0 1 b 0.5 0 1 d 0.5 ) ) ω ( n p ' C ( e R ) ω ( h p J 3 t (ps) 6 0 −5 0 ω−ω (meV) 0 5 Fig. S1. Phonon and photon bath function in time and frequency domain (a) Time evolution of the real part of phonon bath correlation function, C ′ pn(t) for T = 4 K (solid line) and 40 K (dashed line).(b) Corresponding phonon bath function Re[C ′(ω)]. (c) Time evolution of real part of photon bath correlation function, Jph(t) for cavities with κ = 0.06 meV (thin light, Q ≈ 23000), κ = 0.6 meV (thin dark, Q ≈ 2300) and κ = 2.4 meV (thick dark, Q ≈ 600) at ∆xc = 0 meV. (d) Corresponding Jph(ω), where ω0 is ωc. with both phonon and photon reservoirs included. We model the QD as a two-level system interacting with an inhomogeneous semiconductor based photonic reservoir and an acoustic phonon bath. Assuming a QD of dipole moment d = dnd and position at rd, the total Hamiltonian of the coupled system in a frame rotating at the QD exciton frequency ωx, can be written as [11] H = Z drZ ∞ 0 dω f†(r, ω)f(r, ω) − [σ+eiωxtZ ∞ 0 dω d · E(rd, ω) + H.c] + Σq ωqb† qbq + σ+σ−Σq λq(b† q + bq), (S1) where σ+/σ− are the Pauli operators of the exciton (electron-hole pair), bq/b† q are the annihilation and creation operators of the acoustic phonon reservoir, and the exciton-phonon coupling strength λq is assumed to be real. The operators f/f† are the boson field operators of the photon reservoir and these satisfy the usual commutation rules for boson operators. The interaction between the QD and the photonic reservoir is written using the dipole and the rotating wave approximation. The electric-field operator E(r, ω) is given by ǫI (r′, ω)f(r′, ω) [11], where G(r, r′; ω) is the electric field Green's function E(r, ω) = iR dr′G(r, r′; ω)q πǫ0 2 for the medium and E(r, ω) satisfies the Kramers-Kronig relations, with a complex dielectric constant ǫ = ǫR + iǫI. To include phonon interactions nonperturbatively, we perform the polaron transform on the Hamiltonian q − bq) [7, 12]. H ′ consists of the following reservoir and (b† H given by H ′ → eP He−P , where P = σ+σ−Σq interaction terms, λq ωq H ′ =H ′ R + H ′ I, dω f†(r, ω)f(r, ω) + Σqωqb† qbq, H ′ R =Z drZ ∞ I = − [B+σ+eiωxtZ ∞ 0 0 H ′ dω d · E(rd, ω) + H.c], (S2) where the coherent phonon bath displacement operators B± are defined as B± = exp[±Σq polaron frequency shift, ∆P = R ∞ q)]. A is implicitly absorbed in the definition of ωx, where Jpn(ω) is the phonon spectral function [13]. For our calculations, we use the continuous form of the phonon spectral function Jpn(ω) = αpω3 exp[− ω2 ] for longitudinal acoustic (LA) phonon interaction, resulting primarily 2ω2 b from deformation potential coupling in InAs QDs [14]. In the above formula, ωb is the phonon cutoff frequency and αp is the exciton-phonon coupling strength [13]. 0 dω Jpn(ω) ω λq ωq (bq − b† We subsequently transform the Hamiltonian H ′ to the interaction picture using H ′ → U †(t)H ′U (t) with U (t) = exp[−iH ′ Rt/]. Assuming weak coupling with the photon bath, a time-convolutionless [15] polaron master equation for the QD reduced density operator ρ is then derived using second-order Born approximation in H ′ I. The following interaction-picture ME is ontained, ∂ ρ(t) ∂t = − 1 2 Z t 0 dτ TrRphTrRpn {[ H ′ I(t), [ H ′ I(t − τ ), ρ(t)ρR]]}, (S3) where ρ is the reduced density operator of the QD in the interaction picture and TrRph(pn) denotes trace with respect to the photon (phonon) reservoirs which are assumed to be statistically independent, ρR = ρRph ρRpn [16]. The density operator of the photonic reservoir ρRph is assumed to be initially in thermal equilibrium. We use the bath approximation, TrRph[f(r, ω), f†(r′, ω′)] = [n(ω) + 1]δ(r − r′)δ(ω − ω′) and TrRph[f†(r, ω), f(r′, ω′)] = n(ω)δ(r − r′)δ(ω − ω′) and consider the zero temperature limit (n(ω) = 0). Using the relationR dsǫI (s, ω)G(r, s, ω)G∗(s, r′; ω) = Im[G(r, r′; ω)], transforming to the Schrodinger picture, and carrying out the trace over the photon [17] and phonon reservoir, we derive the following generalised ME: ∂ρ ∂t = Lph(ρ), where Lph(ρ) =Z t dτZ ∞ 0 0 pn(τ )ρσ+σ−e−i∆xτ ], − C∗ dω Jph(ω)[−Cpn(τ )σ+σ−ei∆xτ ρ + C∗ pn(τ )σ−ρσ+e−i∆xτ + Cpn(τ )σ−ρσ+ei∆xτ (S4) and ∆x = ωx − ω. The photon-reservoir spectral function Jph(ω), is given by Jph(ω) = d·Im[G(rd,rd;ω)]·d and the phonon bath correlation function Cpn(τ ) is defined as Cpn(τ ) = e[φ(τ )−φ(0)], where φ(t) = 0 dω Jpn(ω) [coth(ω/2kBT ) cos(ωt) − i sin(ωt)]. A simple expression for the phonon-modified SE decay rate can be derived from the real part of Lph, so that Re(Lph) reduces to the Lindblad form, γ(t)L[σ−], where γ is the phonon-modified SE decay rate of a QD given by R ∞ πǫ0 ω2 γ(t) = 2Z t 0 Re[Cpn(τ )Jph(τ )]dτ, (S5) where Jph(τ ) = R ∞ the photon bath correlation function, and L[O] = 2 (2OρO† − O†Oρ − ρO†O). Note that the imaginary part of Lph yield Lamb shifts [13]. This is Equation (2) 0 dωJph(ω)ei(ωx−ω)τ is 1 3 in our main paper. In absence of the phonon coupling, the SE decay rate of the QD in a structured photon 0 Re[Jph(τ )]dτ . A similar expression for γ was derived by McCutcheon and Nazir [18], who then use a free space photon reservoir function to show no phonon-modification to the SE rate. reservoir reduces to γ(t) = 2R t S3. Correspondence with previous work in the mean field cavity-QED regime: Derivation of γmean using a polaron cavity-QED master equation Using the polaron transformed effective Lindblad master equation of an coupled QD-cavity system [13], an expression for the SE rate can be derived in weak excitation approximation (WEA) and bad cavity limit. The effective phonon master equation is defined through, ∂ρ ∂t = 1 i [H eff sys, ρ] + Γa†σL(a†σ−) + Γσ+aL(σ+a) + κL(a), (S6) where a is the cavity lowering operator and H eff sys = ∆cxa†a + hBi g(a†σ− + σ+a), in a frame rotating at the exciton frequency ωx. The QD-cavity coupling strength is g and the cavity damping rate is κ and hBi = hB+i = hB−i is the thermally averaged phonon bath dispacement operator [12]. The cavity (exciton) feeding terms are Γa†σ−/σ+a = 2 hBi2 g2Re[R ∞ are given by 0 dτ e∓i∆cxτ (eφ(τ ) − 1)] [13]. In the weak excitation approximation, the Bloch equations for hai and hσ−i derived from Equation (S6) d hai dt = = − i∆cx + κ + Γσ+a 2 ! hai − i hBi g(cid:10)σ−(cid:11) , d hσ−i dt = = − Γa†σ 2 (cid:10)σ−(cid:11) − i hBi g hai . (S7) The coupled equations can be expressed in a matrix form as V = MV, where V = [hai ; hσi]. In the bad cavity limit, the real part of the eigenvalues of M give the decay rates hai and hσi. The SE rate of the QD is twice the decay rate of hσ−i and is given by γWEA = Re" − Γa†σ− 2 + κ1 + i∆cx! −s(cid:18) Γa†σ− + κ1 2 + i∆cx(cid:19)2 − 4(cid:18)hBi2 g2 + Γa†σ− 2 ( κ1 2 + i∆cx)(cid:19)#, where κ1 = κ + Γσ+a. The above expression can be manipulated to give γWEA = Re" − Γa†σ− 2 + κ1 + i∆cx! −s(cid:18) Γa†σ− − κ1 2 − i∆cx(cid:19)2 − 4 hBi2 g2#, (S8) (S9) which, in the bad cavity limit (κ ≫ g/2), can be further simplified by expanding the second term to the first power of 4hBi2g2 −i∆cx(cid:19)2 . The final simplified expression for the QD SE rate is (cid:18) Γa† σ− −κ1 2 WEA = Γa†σ− γbad + 2 hBi2 g2 Moreover, for κ ≫ Γσ+a − Γa†σ− , 2 (cid:16) κ+Γσ+ a−Γa† σ− (cid:17) cx +(cid:16) κ+Γσ+ a−Γa† σ− 2 ∆2 WEA = Γa†σ− γbad + 2 hBi2 g2 4 ( κ 2 ) ∆2 cx + ( κ 2 )2 , (cid:17)2 . (S10) (S11) which is equivalent to γmean in our paper. Using the same polaronic cavity-QED approach, an identical expression is derived in Ref. [8] in the large detuning limit (∆ ≫ g) by adiabatically eliminating the cavity. Our paper makes it clear that γbad WEA, and indeed the entire polaron cavity-QED master equation, is only valid when κ ≪ 2π/τpn (τpn is around 1 ps for InAs QDs). S4. Comments about how to INCLUDE a coherent pump field The QD can be excited using a weak coherent pulse of the form Hpump = ηx(σ+e−iωLt + σ−eiωLt), where ηx and ωL are the amplitude and central frequency of the pump pulse. For resonant excitation ωL = ωx. Under this situation, the polaron-transformed system Hamiltonian, H ′ (Equation (S2)) has the form S + H ′ R + H ′ I, H ′ =H ′ H ′ H ′ S = hBi ηx[σ+ + σ−], R =Z drZ ∞ I = − [B+σ+eiωxtZ ∞ 0 0 H ′ dω f†(r, ω)f(r, ω) + Σqωqb† qbq, dω d · E(rd, ω) + H.c] + Xgζg + Xuζu, (S12) 2 (B+ + B− − 2 hBi) and ζu = 1 where ζg = 1 2i (B+ − B−) are the phonon induced fluctuation operators [7] and Xg and Xu are defined through Xg = ηx(σ− + σ+) and Xu = iηx(σ+ − σ−). Following the same steps as before (Section II), we subsequently transform the Hamiltonian H ′ to the interaction picture using H ′ → U †(t)H ′U (t) where U (t) = exp[−i(H ′ R)t/]. Assuming weak coupling with the photon bath, a time-convolutionless [15] polaron master equation for the QD reduced density operator ρ is then derived using second-order Born approximation in H ′ I. After carrying out the trace over the photon and phonon baths, the final ME for the QD reduced density operator ρ in the Schrodinger picture is given by dρ dt = 1 i [H ′ S, ρ] + Lph(ρ) + Lpn(ρ). The phonon part Lpn(ρ) remains decoupled from the photon reservoir and, to a very good approximation, can be expressed through the following Lindblad terms (enhanced radiative decay and incoherent excitation [13]): S + H ′ Lpn(ρ) = Γσ− L[σ−] + Γσ+ L[σ+] (S13) 0 dτ e±i(ωL−ωx)τ (eφ(τ )−1)]. The photon part Lph on the other hand is modified where, Γσ+/− by the phonons and is given by = 2 hBi2 η2 xRe[R ∞ dτZ ∞ 0 Lph(ρ) =Z t 0 dω Jph(ω)[−Cpn(τ )σ+σ−(−τ )ei∆xτ ρ + C∗ pn(τ )σ−ρσ+(−τ )e−i∆xτ + Cpn(τ )σ−(−τ )ρσ+ei∆xτ − C∗ pn(τ )ρσ+(−τ )σ−e−i∆xτ ], (S14) Sτ /σ±eiH′ where the time-dependent operators σ±(−τ ) = e−iH′ Sτ /. This indicates that the scattering rates are pump-field dependent in general and for strong pumps, different dressed states (ω = ωx, ωx ± 2ηx) can sample different regions of the photonic LDOS [5,9,19]. Such behavior leads to asymmetric Mollow triplets [20] in the absence of phonons, if the Rabi strength is large enough to sample difference values and an asymmetric LDOS (see Ref. [21] for detailed discussions). Since we assume weak excitation, only the photonic LDOS around ωx is sampled. Thus σ±(−τ ) is replaced by σ±, which reduces Lph(ρ) to Equation (S4), the form used in the main manuscript. The QD can also be excited by a weak incoherent pump by using higher states of the QD. The precise form of the pump is not important for the study of SE [8], as long as it weakly excites the system. However, one can add in a pump field using our general approach as we show above. 5 References 1. C. Roy and S. John Microscopic theory of multiple-phonon-mediated dephasing and relaxation of quan- tum dots near a photonic band gap. Phys. Rev. A 81, 023817 (2010). 2. B. Krummheuer, V. M. Axt and T. Kuhn, Theory of pure dephasing and the resulting absorption line shape in semiconductor quantum dots. Phys. Rev. B 65, 195313 (2002). 3. J. Forstner, C. Weber, J. Danckwerts, and A. Knorr, Phonon-Assisted Damping of Rabi Oscillations in Semiconductor Quantum Dots. Phys. Rev. Lett. 91 127401 (2003). 4. A. Nazir, Photon statistics from a resonantly driven quantum dot. Phys. Rev. B 78, 153309 (2008). 5. C. Roy and S. Hughes, Phonon-Dressed Mollow Triplet in the Regime of Cavity Quantum Electrody- namics: Excitation-Induced Dephasing and Nonperturbative Cavity Feeding Effects. Phys. Rev. Lett. 106, 247403 (2011). 6. U. Hohenester, Cavity quantum electrodynamics with semiconductor quantum dots: Role of phonon- assisted cavity feeding. Phys. Rev. B 81, 155303 (2010). 7. I. Wilson-Rae and A. Imamoglu, Quantum dot cavity-QED in the presence of strong electron-phonon interactions. Phys. Rev. B 65, 235311 (2002). 8. P. Kaer, T. R. Nielsen, P. Lodahl, A.-P. Jauho and J. Mørk, Microscopic theory of phonon-induced effects on semiconductor quantum dot decay dynamics in cavity QED. Phys. Rev. B 86, 085302 (2012). 9. D. P. S. McCutcheon, N. S. Dattani, E. M. Gauger, B. W. Lovett, and A. Nazir, A general approach to quantum dynamics using a variational master equation: Application to phonon-damped Rabi rotations in quantum dots. Phys. Rev. B 84, 081305(R) (2011). 10. A. Vagov, M. D. Croitoru, M. Glassl, V. M. Axt, and T. Kuhn, Real-time path integrals for quantum dots: Quantum dissipative dynamics with superohmic environment coupling. Phys. Rev. B 83, 094303 (2011). 11. S. Scheel, L. Knoll and D.-G. Welsch, Spontaneous decay of an excited atom in an absorbing dielectric. Phys. Rev. A 60, 4094 (1999). 12. G. D. Mahan, Many-Particle Physics (Plenum, New York, 1990). 13. C. Roy and S. Hughes, Influence of Electron−Acoustic Phonon Scattering on Intensity Power Broadening in a Coherently Driven Quantum Dot−Cavity System. Phys. Rev. X 1, 021009 (2011). 14. A. Ulhaq, S. Weiler, C. Roy, S. M. Ulrich, M. Jetter, S. Hughes, and P. Michler, Detuning-dependent Mollow triplet of a coherently-driven single quantum dot. Optics Express 21, 4382 (2013). 15. H.-P. Breuer and F. Petruccione, The Theory Of Open Quantum Systems (Oxford Univ. Press, Oxford, 2002). 16. H. J. Carmichael, Statistical Methods in Quantum Optics 1: Master Equations and Fokker-Planck Equa- tions (Springer, 2003). 17. A. Kowalewska-Kud lask and R. Tana´s, Generalized master equation for a two-level atom in a strong field and tailored reservoirs. J. Mod. Opt. 48, 347 (2001). 18. D. P. S. McCutcheon and A. Nazir, Model of the Optical Emission of a Driven Semiconductor Quantum Dot: Phonon-Enhanced Coherent Scattering and Off-Resonant Sideband Narrowing. Phys. Rev. Lett. 110, 217401 (2013). 19. C. Roy and S. Hughes, Polaron master equation theory of the quantum-dot Mollow triplet in a semi- conductor cavity-QED system. Phys. Rev. B 85, 115309 (2012). 20. M. Florescu and S. John, Single-atom switching in photonic crystals. Phys. Rev. A 64, 033801 (2001). 21. R. C. Ge, C. Van Vlack, P. Yao, J. F. Young, S. Hughes, Accessing quantum nanoplasmonics in a hybrid quantum-dot metal nanosystem: Mollow triplet of a quantum dot near a metal nanoparticle. Phys. Rev. 6 B 87, 205425 (2013). 7
1910.08500	1	1910	2019-10-18T16:50:16	Strong-field-driven dynamics and high-harmonic generation in interacting 1D systems	[ "cond-mat.mes-hall" ]	We explore the roles of electronic band structure and Coulomb interactions in solid-state HHG by studying the optical response of linear atomic chains and carbon nanotubes to intense ultrashort pulses. Specifically, we simulate electron dynamics by solving the single-particle density matrix equation of motion in the presence of intense ultrafast optical fields, incorporating tight-binding electronic states and a self-consistent electron-electron interaction. While linear atomic chains constitute an idealized system, our realistic 1D model readily provides insight on the temporal evolution of electronic states in reciprocal space, both in the absence or presence of electron interactions, which we demonstrate to play an important role in the HHG yield. This model further predicts that doped semiconductors generate high harmonics more efficiently than their metallic and undoped counterparts. To complement this idealized system we also show results for HHG in more realistic quasi-1D structures such as carbon nanotubes, the behavior of which is found to be in good qualitative agreement with the atomic chains. Our findings apply directly to extreme nonlinear optical phenomena in atoms on surfaces, carbon-based structures, linear arrays of dopant atoms in semiconductors, and linear molecules, such as polycyclic aromatic hydrocarbon chains, and can be straightforwardly extended to optimize existing platforms for HHG or identify new solid-state alternatives in the context of nonlinear plasmonics.	cond-mat.mes-hall	cond-mat	Strong-field-driven dynamics and high-harmonic generation in interacting 1D systems Sandra de Vega,1 Joel D. Cox,2, 3, ∗ Fernando Sols,4, 5 and F. Javier Garc´ıa de Abajo1, 6, † 1ICFO-Institut de Ciencies Fotoniques, The Barcelona Institute of Science and Technology, 08860 Castelldefels (Barcelona), Spain 2Center for Nano Optics, University of Southern Denmark, Campusvej 55, DK-5230 Odense M, Denmark 3Danish Institute for Advanced Study, University of Southern Denmark, Campusvej 55, DK-5230 Odense M, Denmark 4Departamento de F´ısica de Materiales, Facultad de Ciencias F´ısicas, Universidad Complutense de Madrid, E-28040 Madrid, Spain 5IMDEA (Instituto Madrileno de Estudios Avanzados) Nanociencia, Cantoblanco 28049, Madrid, Spain 6ICREA-Instituci´o Catalana de Recerca i Estudis Avan¸cats, Passeig Llu´ıs Companys, 23, 08010 Barcelona, Spain (Dated: October 21, 2019) The observation of high-order harmonic generation (HHG) from bulk crystals is stimulating sub- stantial efforts to understand the involved mechanisms and their analogue to the intuitive three-step recollision model of gas phase HHG. On the technological side, efficient solid-state HHG is antici- pated to enable compact attosecond and ultraviolet light sources that could unveil electron dynam- ics in chemical reactions and provide sharper tomographic imaging of molecular orbitals. Here we explore the roles of electronic band structure and Coulomb interactions in solid-state HHG by study- ing the optical response of linear atomic chains and carbon nanotubes to intense ultrashort pulses. Specifically, we simulate electron dynamics by solving the single-particle density matrix equation of motion in the presence of intense ultrafast optical fields, incorporating tight-binding electronic states and a self-consistent electron-electron interaction. While linear atomic chains constitute an idealized system, our realistic 1D model readily provides insight on the temporal evolution of electronic states in reciprocal space, both in the absence or presence of electron interactions, which we demonstrate to play an important role in the HHG yield. This model further predicts that doped semiconductors generate high harmonics more efficiently than their metallic and undoped counterparts. To com- plement this idealized system we also show results for HHG in more realistic quasi-1D structures such as carbon nanotubes, the behavior of which is found to be in good qualitative agreement with the atomic chains. Our findings apply directly to extreme nonlinear optical phenomena in atoms on surfaces, carbon-based structures, linear arrays of dopant atoms in semiconductors, and linear molecules, such as polycyclic aromatic hydrocarbon chains, and can be straightforwardly extended to optimize existing platforms for HHG or identify new solid-state alternatives in the context of nonlinear plasmonics. I. INTRODUCTION High-harmonic generation (HHG) is perhaps the most striking example of a nonlinear optical process and its ability to spectrally and temporally disperse intense laser light.1,2 Initial reports of HHG from atomic gases3 re- vealed a light emission intensity plateau extending over many integer multiples of the fundamental exciting laser frequency and characterized by an abrupt drop at a spe- cific cutoff energy. Concise theoretical explanations of the underlying physics were developed shortly thereafter, culminating in the celebrated three-step model of an atom interacting with a single cycle of an intense im- pinging optical field: tunnel ionization triggered by the driving electric field liberates an electron from the atom that gains additional kinetic energy as it is driven away from and back towards the parent nucleus, ultimately emitting, upon recollision (through coherent interactions of the electron wave function with itself), light at high harmonic orders of the fundamental frequency.4 -- 6 This extreme nonlinear optical phenomenon is a source of co- herent high-frequency electromagnetic radiation, which can be processed to produce attosecond optical pulses, thus garnering significant attention as the means to de- velop micron-scale XUV-light sources7 (i.e., the equiv- alent of table-top synchrotrons) and perform quantum logic operations at optical clock rates,8 while enabling vi- sualization of electronic band structures,9,10 monitoring electron-hole recollisions in real time,11 resolving subfem- tosecond processes governing chemical reactions,12 and recording electron dynamics in molecular orbitals.13,14 Despite the numerous fascinating advances in science and technology that have resulted from atomic HHG, the expense and delicacy of the associated experimental set- ups renders their use hardly practical outside of special- ized laboratory facilities. In contrast, recent observations of HHG from solids15 -- 17 are establishing new paths for attosecond science and strong-field physics, potentially leading towards XUV and attosecond light sources in compact solid-state devices. While the three-step recol- lision model18 offers an intuitive understanding of HHG from atoms in the gas phase, the picture is less clear for solid-state HHG. Early theoretical proposals considered a three-step-like model in which an electron undergoes Bloch oscillations within an electronic band (either va- lence or conduction after interband tunneling) as a con- sequence of the change in direction of acceleration af- ter half an optical cycle of the driving electric field;19 -- 21 9 1 0 2 t c O 8 1 ] l l a h - s e m . t a m - d n o c [ 1 v 0 0 5 8 0 . 0 1 9 1 : v i X r a subsequently the excited electron scatters within its band (i.e., intraband HHG) or recombines with the parent hole or ion (i.e., interband HHG), and finally, it recollides with the first- and second-nearest holes or ions. How- ever, this simplified description does not explain the role of electron-electron correlations, and furthermore, avail- able experiments and numerical simulations often do not elucidate the specific origin of generated harmonics (e.g., from interband or intraband charge-carrier motion); the generation of even-order harmonics, the existence of atto- chirps, the formation of a well-defined high-energy cutoff, and numerous aspects of the electronic band structure still remain underexplored in the context of solid-state HHG.22 -- 25 Further insight into the aforementioned open questions in HHG from condensed-matter systems can be gathered by analyzing one of the simplest models in solid-state physics: the Su-Schrieffer-Heeger (SSH) chain,26 -- 28 con- sisting of a dimerized linear chain of atoms described in the tight-binding approximation, with alternating hop- ping energies assigned to each of the two neighboring atoms on side of any given atom [Fig. 1(a)]. As we discuss below, the SSH model is a convenient system to explore electronic band structure effects in the opti- cal response of materials, as appropriate choices of hop- ping energies reveal either metallic, insulating, or topo- logically insulating behavior. To explore the effect of topology on HHG, recent works29,30 have employed the SSH tight-binding model and its analogue in more rigor- ous time-dependent density functional theory (TDDFT) simulations of atomic chains, predicting improved har- monic yields associated with the topolotical insulator (TI) phases for sub-band-gap photon energies that are ro- bust under distortions, continuous phase transitions, and choice of on-site potentials.31 In a related study,32 the transition from atomic-like systems to solid-state bulk materials was analyzed in the context of HHG, empha- sizing the evolution in cutoff energy as the chain length increases, and concluding that a chain of six atoms con- stitutes the optimal length for this transition to occur as a consequence of changes in the state density. Seeking to optimize HHG yields in condensed-matter systems, we explore the synergy between electronic band structure and optical resonances in finite SSH chains, which constitute a convenient, computationally inexpen- sive model that has already been demonstrated to qual- itatively describe HHG predicted in the more rigorous TDDFT simulations of related 1D systems.29,30 We aug- ment the SSH tight-binding Hamiltonian with a term ac- counting for electron-electron interactions, incorporating a single-electron density matrix description of the optical response and introducing the effect of inelastic charge- carrier scattering through a phenomenological damping rate; this prescription allows us to systematically explore the dependence of HHG yield on the spectral charac- teristics of the impinging optical pulse and identify fre- quencies at which HHG is enhanced by optical resonances associated with the electronic band gap or collective elec- 2 tron motion (i.e., plasmons) in SSH chains. We further explore the effect of electrical doping on HHG by popu- lating the electronic bands with additional charge carri- ers; the added charges can Pauli-block specific electronic transitions and introduce collective resonances, thus fa- cilitating explorations of both. In order to verify the qualitative predictions based on the SSH model in a more realistic condensed-matter platform, we investigate HHG in finite carbon nanotubes (CNTs) of various chiralities that produce similar electronic band features and also display different electronic behavior (metallic, insulat- ing, and topologically insulating). Our findings elucidate the roles of these features intrinsic to different solid-state systems, providing a road map for the identification and engineering of next-generation solid-state nonlinear opti- cal devices, with a view to producing XUV light and/or attosecond pulses. II. ELECTRON DYNAMICS In our SSH model, spin-degenerate electrons occupy the orbitals l(cid:105) located at atomic sites xl = la uni- formly spaced with the lattice constant a. Single-electron states ϕj(cid:105) with associated energy eigenvalues εj satis- fying εj ϕj(cid:105) = H0 ϕj(cid:105) are then obtained by expanding l ajl l(cid:105), where ajl are real-valued expansion coefficients. Following the formalism introduced elsewhere to simulate the optical response of graphene nanoislands,33,34 the electron dy- namics is described by the single-particle density matrix ll(cid:48) ρll(cid:48) l(cid:105)(cid:104)l(cid:48) constructed from time-dependent ma- trix elements ρll(cid:48) and governed by the equation of motion in the site basis according to ϕj(cid:105) = (cid:80) ρ =(cid:80) ρ = − i [H0 − eφ, ρ] − γ 2 (cid:0)ρ − ρ0(cid:1) , (1) ll l(cid:48) vll(cid:48)ρind l(cid:48) l = (cid:80) where ρ0 denotes the equilibrium density matrix to which the system relaxes at a phenomenological rate γ = 50 meV (i.e., a relaxation time τ = γ−1 ∼ 13.2 fs) and φ = φext + φind is the electrostatic potential, which in- cludes contributions from both the impinging light elec- l = −xlE(t), and the electron-electron (e-e) tric field, φext (cid:1) interaction, φind ; the latter quantity ren- ders the equation of motion self-consistent through its dependence on the induced charge ρind (the factor of 2 accounts for spin degeneracy) mediated by the spatial dependence of the Coulomb interaction vll(cid:48) between atoms l and l(cid:48), for which we choose param- eters associated with carbon 2p orbitals.33 In Fig. 1(b) we plot the employed Coulomb interaction compared to that of a point-like charge. The equilibrium density ma- trix is constructed in the state representation according to ρ0 jj(cid:48) = δjj(cid:48)fj, where fj is the occupation factor of state ϕj(cid:105) according to the Fermi-Dirac statistics (we assume zero temperature), and transformed to site representa- l = −2e(cid:0)ρll − ρ0 tion through ρll(cid:48) =(cid:80) jj(cid:48) ajlaj(cid:48)l(cid:48)ρjj(cid:48). Incidentally, linear response theory [obtained by re- placing the [φ, ρ] term by [φ, ρ0] in Eq. (1)] yields a so- lution to Eq. (1) for a monochromatic external electric field Eexte−iωt + c.c. of frequency ω in the form of the harmonic density matrix component ρ(1) e−iωt; the in- duced charge density ρind = −2eρ(1) ll(cid:48) (with a factor of 2 for spin degeneracy) is then self-consistently computed in the so-called random-phase approximation35,36 (RPA) as ρind = χ(0)φ, where χ(0) ll(cid:48) = 2e (fj(cid:48) − fj) ajlaj(cid:48)lajl(cid:48)aj(cid:48)l(cid:48) ω + iγ/2 − (εj − εj(cid:48)) (2) (cid:88) jj(cid:48) where the response function χ = χ(0)(cid:2)1 − χ(0)v(cid:3)−1 is the non-interacting RPA susceptibility. The poles of χ(0) are related to individual electron-hole (e-h) pair exci- tations, so that omission of the induced charge by taking φ → φext yields a response comprised of Lorentzian peaks at the energies (εj − εj(cid:48)); including the self-consistent potential, we isolate the induced charge as ρind = χφext, intro- duces new poles associated with collective charge carrier excitations through the Coulomb interaction. For sim- plicity, we neglect exchange interaction and spin effects. Going beyond linear response, we solve the equation of motion through either of two complementary approaches that allow us to investigate the nonlinear optical response in different regimes. In the first method we resort to direct numerical integration of Eq. (1) in the time domain (TD) to obtain the induced dipole moment p(t) = −2e ρind l xl (3) l produced by various types of external fields E(t) [e.g., continuous wave (cw) illumination or ultrashort pulses], from which Fourier transformation of p(t) reveals its spectral decomposition and characterizes the optical re- sponse. The TD approach does not impose any limit on the strength or type of impinging field, thus enabling the study of the intensity-dependent optical response, in- cluding simultaneously the effects of saturable absorption and high-order harmonic generation. As we are primar- ily interested in the latter phenomenon, we quantify the radiation emitted from the SSH chain by the dipole ac- celeration, p(ω) = ω2p(ω)2.29,30,32,37 In the second approach, we assume monochromatic il- lumination (as in the RPA) and perturbatively expand the density matrix entering Eq. (1) according to ∞(cid:88) n(cid:88) n=0 s=−n ρ = ρnseisωt, where n = 1, 2, . . . denotes the perturbation order and s the harmonic index, such that ρns is defined only when s ≤ n. We then obtain the polarizabilities α(n) sω as (cid:88) sω = − 2e α(n) (E0)n ρns ll xl, (4) (cid:88) l where ρns is computed following the prescription in Ref. 34 that constitutes an extension of the linear RPA to 3 higher perturbation orders. We employ this method to calculate the nonlinear polarizabilities α(3) ω (i.e., the lead- ing nonlinear contribution to the response at the funda- mental frequency, which is associated with the Kerr non- linearity and two-photon absorption). III. SU-SCHRIEFFER-HEEGER MODEL Originally introduced to describe pz electrons in CH monomer chains (polyacetylene), the Su-Schrieffer- Heeger (SSH) model describes a 1D dimerized chain of N atoms through a tight-binding (TB) Hamiltonian,26 and constitutes a simple yet powerful tool to explore non- trivial topological electronic band structure. We consider an SSH chain that contains N/2 unit cells with two sites per cell occupied by one atom from either sublattice A or B [Fig. 1(a)]. Also, we denote the intracell and intercell hoppings as t1 and t2, respectively. The TB Hamiltonian describing the chain is38 N/2(cid:88) N/2−1(cid:88) l=1 l=1 H0 = t1 + t2 (l, B(cid:105)(cid:104)l, A + h.c.) (l + 1, A(cid:105)(cid:104)l, B + h.c.) , (5) which, expressed in a purely spatial representation, takes the form of a tridiagonal N × N matrix containing zeros along the diagonal and hoppings just above and below:  0 . . . 0 0 0 t1 0 t1 0 t2 0 . . . 0 0 0 t2 0 t1 . . . 0 0 0 0 t1 0 . . . 0 0 ... ... 0 0 0 0 0 0 . . . 0 . . . 0 t1 0 . . . t1 0 ... ... ...  . H0 = (6) The choice of hopping parameters determines the phase of the chain:38 the band structure becomes metallic when t1 = t2, insulating if t1 > t2, and a TI (i.e., insulat- ing in the bulk and with a edge states in the gap) when t1 < t238. Throughout this study we consider a chain with N = 50 atoms located at the sites xl = la and having fixed intercell hopping t2 = t0, choosing values a = 0.1421 nm and t0 = −2.8 eV inspired by graphene. From the metallic chain (t1 = t2 = t0), we perturb t1 = t0 + 0.5 eV to enter an insulating phase, whereas t1 = t0 − 0.5 eV yields the band structure of a TI. Di- agonalization of H0 reveals single-electron states charac- terized by the coefficients ajl (i.e., the amount of wave- function j(cid:105) within the orbital at xl) and energies εj; we plot the electronic energies εj obtained for each of the three phases in Fig. 1(c-e). With the chosen param- eters, a band gap of energy Eg = 1.17 eV emerges when t1 (cid:54)= t2, with two quasi-degenerate states appearing in 4 FIG. 1: Characterizing the linear and nonlinear optical response of SSH chains. (a) Schematic representation of the SSH chain comprised of atoms A and B in the unit cell, with intracell (intercell) hopping t1 (t2), and uniform interatomic spacing a. (b) Coulomb interaction of the SSH chain v(x) compared to 1/x. (c-e) Band structure of SSH chains containing N = 50 atoms, fixing the intercell hopping to t2 = −2.8 eV and varying the intracell hopping t1; depending on the choice of t1 the system is (c) metallic (t1 = t2 = −2.8 eV, black curves), (d) insulating (t1 = −3.3 eV, red curves), and (e) a topological insulator (TI) (t1 = −2.3 eV, blue curves). (f-h) Normalized absorption cross section with electron-electron interactions switched on (e-e on) and off (e-e off) for the metal (f), the insulator (g), and the TI (h). (i-k) Harmonic generation, quantified through the dipole acceleration p(ω) = ω2p(ω)2 (where p(ω) denotes the ω component of the induced dipole), produced by pulses of 1013 W/m2 peak intensity, 100 fs FWHM duration, and carrier frequency ω0, with the latter quantities in each panel indicated by the color-coded legends and arrows in (f-h). (l-n) Same as (i-k) but for a fixed pulse carrier energy of 0.2 eV, away from the resonances appearing in (f-h). the middle of the band gap when t1 < t2 [Fig. 1(d)], corresponding to the edge states of the chain and giving the insulator its topological character. IV. SELF-CONSISTENT INTERACTIONS IN THE OPTICAL RESPONSE OF SSH CHAINS Optical resonances, and plasmons in particular, are widely exploited in nano-optics to intensify local elec- tromagnetic fields for a variety of applications, some of which involve the enhancement of nonlinear optical 5 FIG. 2: Effect of doping in the linear response of SSH chains. Normalized absorption cross section as a function of frequency (vertical axis) and additional charge carriers Q (horizontal axis) for the (a) metallic, (b) insulating, and (c) topologically insulating chains considered in Fig. 1. processes.39,40 Here, we explore the ability of optical res- onances in the three phases of the SSH model to drive HHG. In Fig. 1(f-h) we identify through the linear ab- sorption cross-section the available optical resonances in the metallic, insulating, and TI phases, both in situa- tions when the self-consistent electron-electron (e-e) in- teraction is omitted and included. Neglecting e-e inter- actions, peaks in the absorption spectrum are associated with the energies of individual single-electron transitions (see discussion in Sec. I), with amplitudes determined by their transition dipole moments. In contrast, if we include e-e interactions, the dominant transitions un- dergo large spectral blueshifts in all three phases. In Figs. 1(f-h) we show the absorption cross section of the SSH chains, which we calculate through the optical the- ω }. We normalize the cross orem σabs = 4π(ω/c)Im{α(1) section to roughly the geometrical in-plane projection of the chain, N a2. Incidentally, the edge states of the TI [Fig. 1(h)] give rise to a low-energy resonance that does not appear for the insulator. Given their importance in the linear response, it is expected that self-consistent e-e interactions also play a leading role in the nonlinear response. Figs. 1(i-n) show normalized high harmonic spectra (quantified via dipole accelerations) produced by Gaussian pulses of 1013 W/m2 peak intensity and 100 fs full-width-at-half- maximum (FWHM) duration with central frequencies ω0 (i) targeting the dominant resonances in the linear spec- tra [Figs. 1(i-k); see color-coded arrows indicating the energy of ω0] and (ii) off-resonance, with a frequency arbitrarily fixed to 0.2 eV (i.e., away from optical reso- nances) in all cases [Figs. 1(l-n)]. Resonant excitation of the metallic chain yields lower HHG when e-e inter- actions are included, presumably because charge screen- ing in the metal damps the electron motion, while this effect is less important in the gapped systems. The num- ber of observable harmonics is typically larger for lower- energy excitation and associated with more efficient in- terband generation, where the maximum cutoff energy in the non-interacting case is indicated by the largest avail- able single-electron transition energy. Incidentally, the height of the first harmonic can vary widely from on- resonant to off-resonant conditions (e.g., by a factor up to 104 in metallic chains). Additionally, e-e interactions lead to collective opti- cal resonances of higher strength compared with single- electron transitions, thus allowing us to reach HHG with significantly reduced intensity compared with previous studies that neglect those interactions.20,41 Plasmons are associated with the motion of free elec- trons, and thus do not emerge in pristine semiconducting materials. However, in the 50-atom SSH chains that we consider here, the addition of only a few electron charges is sufficient to dramatically change the optical response; this phenomenon is explored in Fig. 2, where we study the linear response in the RPA as a function of the dop- ing charge Q in all three SSH phases. Note that the charge is added in such a way that the free electrons equally populate available degenerate states. In contrast to the almost negligible electrical tuning for the metal- lic chain [Fig. 2(a)], the insulating chains present overall a blueshift with increasing charge carrier density in the low-energy spectral features, which tend to coalesce into a prominent peak associated with intense optical absorp- tion and a concentration of electromagnetic energy within the material. In particular, at Q = 2 in Fig. 2(b) we observe a sharp feature that corresponds to the filling of the low- est unoccupied molecular orbital (LUMO); that is, the insulator gets free carriers in the conduction band and starts behaving as a metal (we note that a Q = 2 dop- ing corresponds to a Fermi level EF ∼0.81 eV, which corresponds to the highest filled state in a one-electron picture, which is larger than the energy of the LUMO, ELUMO = Eg/2 = 0.59 eV). For this reason, for Q > 2 the main resonant feature begins to stabilize and by Q = 5 (EF ∼1.11 eV) it has coalesced in a prominent dipolar plasmon mode of frequency similar to that of the metal- lic chain because of the similar value of the density of states at the Fermi level in both cases. In contrast, for the TI chain [Fig. 1(j)] we observe that quasi-degenerate edge states in the middle of the band gap produce a slight redshift of the main resonance and damp the strength of the absorption cross section, particularly at the LUMO energy (Q = 2 or EF ∼0.63 eV). At Q = 4 the main resonance starts growing and by Q = 5 the response is dominated by the plasmon, just like in the insulating chain. V. INTENSITY-DEPENDENT ABSORPTION The realization of HHG in solid-state systems necessi- tates optical pulses with peak intensities that cannot be sustained for long duration, lest the material be destroyed in the process. However, the interaction of extended pulses or cw fields with matter is appealing for techno- logical applications relying on saturable absorption, an extreme nonlinear optical phenomenon that arises in all photonic materials and, like HHG, cannot be described in a perturbative framework. The enhanced light-matter interaction provided by op- tical resonances produces a more measurable absorption signal that facilitates detection of changes in the dielec- tric environment or the impinging light intensity; the latter effect is intensified by the concentration of elec- tromagnetic energy in the material, which in turn can enhance its nonlinear optical response. Following this approach for HHG, we explore the nonlinear response associated with optical resonances of SSH chains by con- sidering their interaction with intense impinging cw light, characterizing the optical response by the induced dipole moment at the fundamental frequency. Specifically, we extract the effective polarizability αω by computing the induced dipole p(t) in response to a monochromatic field E(t) = E0e−iωt + c.c. of intensity I ext = (c/2π)E02. We then Fourier transform p(t) over a single optical cycle to obtain (cid:90) tcw ω αω = 2πE0 tcw−2π/ω p(t)e−iωtdt, (7) where tcw (cid:29) τ corresponds to a time at which the system has entered a steady state regime. In Fig. 3 we study the dependence on pulse intensity of the main resonances in the absorption spectra (upper rows) of the different SSH chains in relation to their cor- responding Kerr polarizabilities α(3) ω , that is, with per- turbation order n = 3 and harmonic index s = 1 (lower rows), for three different dopings: undoped [Figs. 3(a- f)], LUMO doping (i.e., with two additional electrons, Q = 2) [Figs. 3(g-l)], and plasmonic-regime doping (i.e., with five additional electrons, Q = 5). 6 Independent of doping, the absorption cross section of metallic chains remains relatively unchanged by increas- ing the optical intensity, an observation compatible with their consistently smaller nonlinear polarizabilities. In contrast, the effective polarizabilities of the insulating phases offer a larger nonlinear response, where in partic- ular Figs. 3(c), (h), and (n) present strong saturation and also shifting of the peak energy, which eventually should transition towards a bistable regime. This behavior is corroborated by the large corresponding Kerr polarizabil- ities [Figs. 3(f), (k) and (q), respectively], with the real part determining the peak shift strength and direction (e.g., red- vs blue-shift), while the imaginary part gov- erns its saturation. Within the range of parameters con- sidered here, we conclude that the TI is the most nonlin- ear material without doping, presumably because of the localized spatial and spectral character of the edge states (i.e., intrinsic anharmonicity), while the population of its edge states and subsequent Pauli blocking through dop- ing renders its nonlinearity comparable to that of the insulator. The strong nonlinearity of the two types of doped insulators compared with the metallic chain can be understood in terms of a sparse Fermi sea, where in- teractions are poorly screened. VI. HIGH-HARMONIC GENERATION IN SSH CHAINS We further study the optical response of SSH chains to strong ultrashort laser pulses. In Figs. 4 and 5 we plot the induced dipole acceleration p normalized to the max- imum dipole acceleration pmax for each input frequency ωin considering Gaussian pulses of 100 fs FWHM dura- tion and peak intensity I = 1013 W/m2, with the hor- izontal pulse carrier frequency given on the horizontal axes and the frequency component of p indicated on the vertical axes; each contour plot of p is supplemented by the associated linear optical response of the system under consideration (upper panels). In the insulating phases we use white vertical dashed lines to indicate the electronic band gap energy Eg and Eg/2, with the latter quantity indicating the edge-state-to-LUMO transition energy for the TI. In Fig. 4 we consider undoped SSH chains, which present prominent features along the ω = sωin curves, where s is an odd integer (i.e., harmonic generation for s > 1). In all cases we observe that strong HHG is pro- duced where the optical resonances intersect with the impinging light frequency, particularly in the low-energy feature observed in the absorption spectrum of the TI (see Fig. 4c) lying below its band gap and associated with its topologically protected states, for which more high-order harmonics can couple to interband transitions. In Fig. 5 we study HHG in LUMO-doped [Fig. 5(a-d)] and plasmonic-regime-doped [Fig. 2(e-h)] chains. The general trends for HHG are similar to those observed for Kerr polarizabilities in Fig. 3 and discussed in the previ- 7 FIG. 3: Intensity-dependent nonlinear absorption in SSH chains. (a-c) Normalized absorption cross-section σabs/N a2 simulated in the time domain for monochromatic cw illumination of increasing intensity near the optical resonances of undoped (a) metallic, (b) insulating, and (c) topologically insulating SSH chains. (d-f ) Perturbative solutions of the third-order polarizabilities associated with the optical Kerr nonlinearity α(3) ω , presented in atomic units for the corresponding SSH chains in (a-c), with real and imaginary parts indicated by green and orange curves, respectively. (g-l) Same as (a-f) but for SSH chains doped with 2 additional electrons (Q = 2). (m-r) Same as (g-l) but for SSH chains doped with 5 additional electrons (Q = 5). 8 FIG. 4: High harmonic generation in undoped SSH chains. (a-c) Induced dipole acceleration p in undoped SSH chains as a function of the output frequency ω, as generated by a Gaussian pulse of peak intensity 1013 W/m2, 100 fs FWHM duration, and central frequency ωin for the (a) metal, (b) insulator, and (c) topological insulator 50-atom SSH chains. The inset above each panel shows the respective linear response of the chain, while the vertical dashed lines mark the band gap energy Eg and its half-value Eg/2. ous section. As in the case of monochromatic excitation, we once again observe that the metal nonlinear response remains relatively unaffected by doping, while the insu- lator and the TI present higher HHG yields. In panels (d) and (h) of Fig. 5 we compare p at the dominant op- tical resonance of each SSH [a frequency denoted ω0, see arrows in the upper panels of Figs. 5(a-c) and (e-g)]. For Q = 2 [Fig. 5(d)] the insulator presents the best har- monic yield, producing sizable peaks up to ∼ 25ω0, with harmonics exhibiting a slight blueshift associated with the pulse self-interaction involving the out-of-equilibrium electrons that it excites.42 In contrast, at Q = 5 doping the insulator and TI produce very similar HHG yields that extend up to ∼ 21ω0, with the harmonics pro- duced by the insulator slightly redshifted for ω > 9ω0. Both types of insulators display a harmonic yield much stronger than the metal. As already discussed in the pre- vious section, the strong HHG of the doped insulators can be ascribed to weak screening in a low-density electron gas. VII. BEYOND THE SSH MODEL: HIGH HARMONIC GENERATION IN CARBON NANOTUBES To validate the predictive capabilities of the SSH model, we turn now to a more realistic 1D system in which to study HHG. Carbon nanotubes (CNTs) consti- tute a material platform that can behave as a metal, insu- lator, or TI, depending on chirality, and furthermore, like graphene, their electronic properties can be reasonably well-described by a tight-binding Hamiltonian. CNTs themselves hold great potential for diverse applications43 because of their excellent mechanical, electronic, and op- tical properties, exemplified through the recent demon- stration of a functioning CNT-based transistor.44 In the field of nano-optics, recent experimental studies have explored the low-energy plasmons supported by CNTs when they are electrically doped,45 -- 47 similar to the col- lective excitations in highly doped graphene48, motivat- ing their application for nanophotonic devices and non- linear plasmonic elements.49 From a theoretical perspec- tive, plasmons in CNTs have been extensively studied using both ab initio methods50,51 and also semi-classical approaches52 -- 54 based on the RPA35 to calculate their optical conductivities. Incidentally, it has been experi- mentally proven that electrons in CNTs behave as Lut- tinger liquids,55,56 which have special relevance at low temperatures and are qualitatively corroborated by our methods. √ CNTs are constructed by wrapping a graphene layer into a cylindrical surface; a carbon atom at the origin is then identified with one at the graphene lattice posi- tion na1 + ma2, where a1 and a2 are the conventional graphene lattice vectors, while the pair of integers (n, m) determine the chirality of the tube. The resulting CNT diameter (in nm) is57 d ≈ 78.3 × 10−3 n2 + m2 + nm. We can additionally classify CNTs depending on their topology, which incidentally is determined by n and m: when n = m the CNT is metallic, if n = m + 1 we have an insulator, and otherwise the CNT is a TI.58,59 To explore 1D-like structures more similar to SSH chains, we choose extremely thin CNTs experimentally reported with different chiralities.60 -- 62 Also, to facilitate the com- parison with our 50-atom SSH chains, we take them to be roughly 7 nm long. In particular, we consider CNTs with chiralities (3,3), (4,3), and (5,1), which have di- 9 FIG. 5: High harmonic generation in doped SSH chains. Same as Fig. 4 when the chain is doped with either 2 (a-c) or 5 (e-g) electrons. Panels (d) and (h) show the normalized induced dipole at the resonances pointed by the arrows in the linear response for the metal (black), the insulator (red), and the topological insulator (blue) in (a-c) and (e-g), respectively. ameters of 0.41 nm, 0.48 nm, and 0.44 nm, and corre- spond to a metal, an insulator, and a topological insula- tor, respectively. In Fig. 6 we show the band structures of these three CNTs calculated through a tight-binding model with a phenomenological nearest-neighbors hop- ping of 2.8 eV,57,63 revealing Eg = 1.80 eV for the TI and Eg = 1.68 eV for the insulator. Strong-field driven electron dynamics in CNTs is simu- lated once again by inserting a tight-binding Hamiltonian into the equation of motion (1), adopting a phenomeno- logical damping rate of γ = 50 meV, and applying 100 fs FWHM laser pulses of intensity I = 1013 W/m2 to excite high harmonics in these structures. In Fig. 7 we com- pare the HHG yields of undoped [Fig. 7(a-c)] and doped (EF = 1 eV) CNTs [Fig. 7(d-g)], again supplementing contour plots of p with linear absorption spectra (upper panels). For undoped CNTs we observe weaker HHG for both the insulator and the TI, only exciting up to the 7th order, in contrast to the metallic tube, exhibit- ing a distinctly higher HHG yield, particularly when the impinging light energy coincides with the dominant op- tical resonance near 0.9 eV. This particular trend is not 10 cally different types of electronic structure constitute an underexplored possibility to enhance the electric fields driving HHG, which we address here through the use of an intuitive model that contains much of the relevant physics. More precisely, our main conclusions based on the model SSH 1D chain are corroborated in their more realistic carbon-based analogues. In metals or doped semiconductors, where free electrons are present, self- consistent electron interactions become extremely impor- tant in both the linear and nonlinear response, and not only when dealing with optical resonances. This effect is stronger for doped semiconductors than for metals be- cause weaker screening in the former makes their inter- actions more important. While HHG appears to be most efficient in semiconductors for harmonics generated be- low the band gap, the addition of a small amount of dop- ing charge can produce a poorly screened, low-density electron gas with an intraband plasmon excitation that falls in this regime, therefore concentrating the imping- ing electromagnetic fields and further boosting the HHG efficiency. This finding suggests the exploration of highly doped semiconducting materials as a promising platform for solid-state HHG. Our results pave the way for further investigation on the effects of electron-electron interac- tions in solid-state HHG, elucidating the involved mi- croscopic mechanism and the relation between electronic band structure and the HHG yield, thus supporting its application towards nonlinear plasmonics, topological op- toelectronics, and all-optical time-resolved probing of topological phases. Acknowledgments This work has been supported in part by the Spanish MINECO (MAT2017-88492-R and SEV2015- 0522, and FIS2017-84368-P), the ERC (Advanced Grant 789104- eNANO), the European Commission (Graphene Flag- ship 696656), the Catalan CERCA Program, and the Fundacio Privada Cellex, and Universidad Complutense de Madrid (grant No. 962085). S. d. V. acknowledges financial support through the FPU program from the Spanish MECD. The Center for Nano Optics is finan- cially supported by the University of Southern Denmark (SDU 2020 funding). J. D. C. was supported by VILLUM Fonden (grant No. 16498). FIG. 6: Band structure of CNTs. Single-electron energies for CNTs with chiralities (3,3), (4,3), and (5,1), correspond- ing to a metal (black squares), insulator (red diamonds), and topological insulator (blue circles), respectively. The inset shows the region near zero energy (indicated by the dashed horizontal line) in more detail. followed by SSH chains, which in the undoped configura- tion produce stronger yield in insulators compared with metals [cf. Figs. 4(a-c) and Figs. 7(a-c)]. The HHG yield is strongly enhanced by doping the CNTs to a Fermi energy EF = 1 eV [Fig. 7(d-g)], which introduces localized plasmon resonances. By inspect- ing the high harmonics generated at resonant frequencies when excited at optical resonances [Fig. 7(g)] we see that both the insulator and the TI have higher yields, that their harmonics are quite strongly redshifted beyond the 7th order, and that their cutoff is at the 21st harmonic; this behavior was qualitative predicted by the simpler SSH model. VIII. CONCLUSIONS Despite the impressive pace at which the field of solid- state HHG is developing in both experiment and theory, the ideal material platform in which to realize this ex- treme nonlinear optical process has yet to be identified. Optical resonances supported by materials with intrinsi- ∗ Electronic address: [email protected] † Electronic address: [email protected] 1 T. Brabec and F. Krausz, Rev. Mod. Phys. 72, 545 (2000). 2 P. B. Corkum and F. Krausz, Nat. Phys. 3, 381 (2007). 3 N. H. Burnett, H. A. Baldis, M. C. Richardson, and G. D. Enright, Appl. Phys. Lett. 31, 172 (1977). 4 A. McPherson, G. Gibson, H. Jara, U. Johann, T. S. Luk, I. A. McIntyre, K. Boyer, and C. K. Rhodes, J. Opt. Soc. Am. B 4, 595 (1987). 5 M. Ferray, A. L'Huillier, X. F. Li, L. A. Lompre, G. Main- fray, and C. Manus, J. Phys. B 419, 803 (1988). 6 M. Lewenstein, P. Balcou, M. Y. Ivanov, A. L'Huillier, and P. B. Corkum, Phys. Rev. A 49, 2117 (1994). 7 I. McKinnie and H. Kapteyn, Nat. Photon. 4, 149 (2010). 8 M. Hohenleutner, F. Langer, O. Schubert, M. Knorr, U. Huttner, S. W. Koch, M. Kira, and R. Huber, Nature 11 FIG. 7: High harmonic generation in CNTs (a-c) Dipole acceleration p as a function of the output frequency ω generated by a Gaussian pulse of peak intensity I = 1013 W/m2, 100 fs FWHM duration, and central frequency ωin. The insets above each panel indicate the associated linear response for the (a) metallic, (b) insulating, and (c) topological insulating undoped CNTs. We have marked with white vertical dotted lines both the energy of the band gap Eg and half the energy of the band gap Eg/2. (d-f ) Same as (a-c) but for 1 eV-doped CNTs. The arrows indicate the resonant input frequency ω0 of the Gaussian pulse at which there is a boost in HHG. (g) Normalized dipole acceleration at the resonances pointed by arrows in the linear response on top of panels (d-f). 523, 572 (2015). 9 G. Vampa, T. J. Hammond, N. Thir´e, B. E. Schmidt, F. L´egar´e, C. R. McDonald, T. Brabec, and P. B. Corkum, Nature 522, 462 (2015). 10 C.-M. Wang, T.-S. Ho, and S.-I. Chu, J. Phys. B 49, 225401 (2016). 11 B. Zaks, R. B. Liu, and M. S. Sherwin, Nature 483, 580 (2012). 12 B. Wolter, M. G. Pullen, A.-T. Le, M. Baudisch, K. Doblhoff-Dier, A. Senftleben, M. Hemmer, C. D. Schroter, J. Ullrich, T. Pfeifer, et al., Science 354, 308 (2016). 13 A. H. Zewail, JPCA 104, 5660 (2000). 14 L. Feng, J. Hu, L. W. Clark, and C. Chin, Science 363, 521 (2019). 15 S. Ghimire, A. D. DiChiara, E. Sistrunk, P. Agostini, L. F. DiMauro, and D. A. Reis, Nat. Phys. 7, 138 (2011). 16 O. Schubert, M. Hohenleutner, F. Langer, B. Urbanek, C. Lange, U. Huttner, D. Golde, T. Meier, M. Kira, S. W. Koch, et al., Nat. Photon. 8, 119 (2014). 17 T. T. Luu, M. Garg, S. Y. Kruchinin, A. Moulet, M. T. Hassan, and E. Goulielmakis, Nature 521, 498 (2015). 18 P. B. Corkum, Phys. Rev. Lett. 71, 1994 (1993). 19 U. Huttner, M. Kira, and S. W. Koch, Laser Photon. Rev. 11, 1700049 (2017). 20 E. N. Osika, A. Chac´on, L. Ortmann, N. Su´arez, J. A. 12 P´erez-Hern´andez, B. Szafran, M. F. Ciappina, F. Sols, A. S. Landsman, and M. Lewenstein, Phys. Rev. X 7, 021017 (2017). 21 S. Ghimire and D. Reis, Nat. Phys. 15, 10 (2019). 22 A. F. Kemper, B. Moritz, J. K. Freericks, and T. P. Dev- ereaux, New J. Phys. 15, 023003 (2013). 23 G. Vampa, C. R. McDonald, G. Orlando, D. D. Klug, P. B. Corkum, and T. Brabec, Phys. Rev. Lett. 113, 073901 (2014). 24 T. T. Luu and H. J. Worner, Phys. Rev. B 94, 115164 (2016). 25 C. R. McDonald, G. Vampa, G. Orlando, P. B. Corkum, 45 P.-H. Ho, D. B. Farmer, G. S. Tulevski, S.-J. Han, D. M. Bishop, L. M. Gignac, J. Bucchignano, and A. L. F. P. Avouris, Proc. Natl. Academ. Sci. 115, 12662 (2018). 46 A. L. Falk, K.-C. Chiu, D. B. Farmer, Q. Cao, J. Tersoff, Y.-H. Lee, P. Avouris, and S.-J. Han, Phys. Rev. Lett. 118, 257401 (2017). 47 I. Soto Lamata, P. Alonso-Gonz´alez, R. Hillenbrand, and A. Y. Nikitin, ACS Photon. 90, 125417 (2015). 48 Z. Fei, G. O. Andreev, W. Bao, L. M. Zhang, A. S. McLeod, C. Wang, M. K. Stewart, Z. Zhao, G. Dominguez, M. Thiemens, et al., Nano Lett. 11, 4701 (2011). 49 J. D. Cox and F. J. Garc´ıa de Abajo, Acc. Chem. Res. 52, and T. Brabec, J. Phys. Conf. Ser. 594, 012021 (2015). 2536 (2019). 26 W. P. Su, J. R. Schrieffer, and A. J. Heeger, Phys. Rev. 50 D. S´anchez-Portal, E. Artacho, J. M. Soler, A. Rubio, and Lett. 42, 1689 (1979). 27 H. W. Streitwolf, Phys. Stat. Sol. (b) 127, 11 (1985). 28 F. Gebhard, K. Bott, M. Scheidler, P. Thomas, and S. W. Koch, Philos. Mag. B 75, 1 (1997). 29 D. Bauer and K. K. Hansen, Phys. Rev. Lett. 120, 177401 P. Ordej´on, Phys. Rev. B 59, 12678 (1999). 51 R. Demichelis, Y. Noel, P. D'Arco, M. R´erat, C. M. Zicovich-Wilson, and R. Dovesi, J. Phys. Chem. C 115, 8876 (2011). 52 P. Avouris, M. Freitag, and V. Perebeinos, Nat. Phot. 2, (2018). 341 (2008). 30 C. Jurss and D. Bauer, Phys. Rev. B 99, 195428 (2019). 31 H. Drueke and D. Bauer, Phys. Rev. A 99, 053402 (2019). 32 K. K. Hansen, D. Bauer, and L. B. Madsen, Phys. Rev. A 53 L. Mart´ın-Moreno, F. J. Garc´ıa de Abajo, and F. J. Garc´ıa- Vidal, Phys. Rev. Lett. 115, 173601 (2015). 54 S. de Vega, J. D. Cox, , and F. J. Garc´ıa de Abajo, Phys. 97, 043424 (2018). 33 S. Thongrattanasiri, A. Manjavacas, and F. J. Garc´ıa de Abajo, ACS Nano 6, 1766 (2012). 34 J. D. Cox and F. J. Garc´ıa de Abajo, Nat. Commun. 5, 5725 (2014). 35 D. Pines and D. Bohm, Phys. Rev. 85, 338 (1952). 36 L. Hedin and S. Lundqvist, in Solid State Physics, edited by D. T. Frederick Seitz and H. Ehrenreich (Academic Press, 1970), vol. 23 of Solid State Physics, pp. 1 -- 181. 37 J. C. Baggesen and L. B. Madsen, J. Phys. B 44, 115601 (2011). 38 J. K. Asb´oth, L. Oroszl´any, and A. P´alyi, A short course on topological insulators. Band structure and edge states in one and two dimensions (Springer, Switzerland, 2016). 39 M. Kauranen and A. V. Zayats, Nat. Photon. 6, 737 (2012). 40 J. D. Cox, A. Marini, and F. J. Garc´ıa de Abajo, Nat. Commun. 8, 14380 (2017). 41 K. A. Pronin, A. D. Bandrauk, and A. A. Ovchinnikov, Phys. Rev. B 50, 3473 (1994). 42 M. Baudisch, A. Marini, J. D. Cox, T. Zhu, F. Silva, S. Te- ichmann, M. Massicotte, F. Koppens, L. S. Levitov, F. J. Garc´ıa de Abajo, et al., Nat. Commun. 9, 1018 (2018). 43 M. F. L. D. Volder, S. H. Tawfick, R. H. Baughman, and A. J. Hart, Science 339, 535 (2015). 44 G. Hills, C. Lau, A. Wright, S. Fuller, M. D. Bishop, P. K. Tathagata Srimani, R. Ho, A. Amer, Y. Stein, D. Murphy, et al., Nature 572, 595 (2019). Rev. B 94, 075447 (2016). 55 M. Bockrath, D. H. Cobden, J. Lu, A. G. Rinzler, R. E. Smalley, L. Balents, and P. L. McEuen, Nature 397, 598 (1999). 56 H. Ishii, H. Kataura, H. Shiozawa, H. Yoshioka, H. Ot- subo, Y. Takayama, T. Miyahara, S. Suzuki, Y. Achiba, M. Nakatake, et al., Nature 426, 540 (2003). 57 T. W. Odom, J.-L. Huang, P. Kim, and C. M. Lieber, J. Phys. Chem. C 104, 2794 (2000). 58 R. Okuyama, W. Izumida, and M. Eto, Phys. Rev. B 99, 115409 (2019). 59 E. A. Laird, F. Kuemmeth, G. A. Steele, K. Grove- Rasmussen, J. Nygard, K. Flensberg, and L. P. Kouwen- hoven, Rev. Mod. Phys. 87, 703 (2015). 60 A. C. Torres-Diasab, T. F. T. Cerqueira, W. Cui, M. A. L. Marques, S. Botti, D. Machona, M. A. Hartmann, Y. Sun, D. J. Dunstan, and A. San-Miguel, Carbon 123, 145 (2017). 61 T. Hayashi, Y. A. Kim, T. Matoba, M. Esaka, K. Nishimura, T. T. M. Endo, and M. S. Dresselhaus, Nano Lett. 3, 887 (2003). 62 L. Guan, K. Suenaga, and S. Iijima, Nano Lett. 8, 459 (2008). 63 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009).
1606.09226	1	1606	2016-06-29T19:20:47	All-strain based valley filter in graphene nanoribbons using snake states	[ "cond-mat.mes-hall" ]	A pseudo-magnetic field kink can be realized along a graphene nanoribbon using strain engineering. Electron transport along this kink is governed by snake states that are characterized by a single propagation direction. Those pseudo-magnetic fields point towards opposite directions in the K and K' valleys, leading to valley polarized snake states. In a graphene nanoribbon with armchair edges this effect results in a valley filter that is based only on strain engineering. We discuss how to maximize this valley filtering by adjusting the parameters that define the stress distribution along the graphene ribbon.	cond-mat.mes-hall	cond-mat	a All-strain based valley filter in graphene nanoribbons using snake states L. S. Cavalcante,1, ∗ A. Chaves,1, 2, † D. R. da Costa,1, 3, ‡ G. A. Farias,1, § and F. M. Peeters3, 1, ¶ 1Universidade Federal do Cear´a, Departamento de F´ısica Caixa Postal 6030, 60455-760 Fortaleza, Cear´a, Brazil 2Department of Chemistry, Columbia University, 3000 Broadway, 10027 New York, NY 3Department of Physics, University of Antwerp, Groenenborgerlaan 171, B-2020 Antwerp, Belgium A pseudo-magnetic field kink can be realized along a graphene nanoribbon using strain engineer- ing. Electron transport along this kink is governed by snake states that are characterized by a single propagation direction. Those pseudo-magnetic fields point towards opposite directions in the K and K ′ valleys, leading to valley polarized snake states. In a graphene nanoribbon with armchair edges this effect results in a valley filter that is based only on strain engineering. We discuss how to maximize this valley filtering by adjusting the parameters that define the stress distribution along the graphene ribbon. PACS numbers: 81.05.U-, 72.80.Vp, 73.63.-b I. INTRODUCTION The advent of graphene1,2 not only represented the be- ginning of a new era of atomically thin materials, with potential technological applications in future electronic and photonic devices, but also brought the possibility of observing several novel phenomena due to its unique band structure, consisting of Dirac cones in points la- beled as K and K′ in its first Brillouin zone. In fact, the existence of two inequivalent cones is of special im- portance, since it enables a new degree of freedom to be explored in novel valley-tronic devices. Several suggestions have been made to harvest valley polarization in graphene: Rycerz et al.3 demonstrated that specific combinations of armchair and zigzag edges in a monolayer graphene ribbon lead to efficient valley filtering. Non-uniform substrate induced masses can also be used to obtain valley polarization, as shown in Refs. [4,5]. As for bilayer graphene, valley filtering can be ob- tained by specific configurations of external potentials,6 or boundaries with monolayer graphene regions.7,8 On the other hand, recent studies have demonstrated that pseudo-magnetic fields can be induced in graphene by specific strain configurations and, since these fields point towards opposite directions in different Dirac cones9, sev- eral suggestions of strain-based valley filters have been proposed in the literature. Most of these proposals in- volve combinations of the strain induced fields with ap- plied magnetic and electric fields.10 -- 14 Indeed, strain- based valley filters are specially interesting, because of graphene's ability to withstand large mechanical stress.15 Very large pseudo-magnetic fields have been experimen- tally observed in e.g. naturally formed bubbles in a monolayer graphene system on a Pt substrate.16 In this paper, we propose a very different valley fil- ter device made of a single graphene layer17 that does not depend on specific edge configurations,3 substrate induced masses,4 or external magnetic fields10,18. Yet, it is all based on a particular kind of strain induced in a monolayer graphene nanoribbon, that provides a pseudo-magnetic kink barrier along the ribbon width. Such strain is expected to be attainable by using flexible substrates in combination with e.g. appropriate piezos or contacts that deform when cooled down.10,15,19,20 The valley polarization originates from a combination of (i) the uni-directional motion of snake states along the pseudo-magnetic kink, and (ii) the opposite direction of the pseudo-magnetic field felt in the K and K′ valleys. This combination leads to electrons moving in single or- bits propagating in opposite directions in the two differ- ent valleys. In order to verify the efficiency of such valley filtering device, we calculate the transmission probability of wavepackets through this structure within the tight- binding model. Our results demonstrate that a valley polarization efficiency up to 90% can be reached, pro- vided specific conditions are met by the system, as we will discuss in what follows. II. THEORETICAL MODEL FOR THE STRAIN Our system consists of a monolayer graphene nanorib- bon with width Wy ≈ 6387 A and length Wx ≈ 2214 A corresponding to 1801 × 3000 carbon atoms, that is me- chanically strained in a specific configuration, as sketched in Fig. 1: along a certain region of length β, the ribbon is distorted into two circular arcs of radius R, in opposite directions. Such in-plane circular bending is obtained by defining the displacement of the atomic sites as15 ux(x, y) = (R + x) cos(cid:20) 2y Wy arcsin(cid:18) Wy 2R(cid:19)(cid:21) − y, (1) and uy(x, y) = (R + x) sin(cid:20) 2y Wy arcsin(cid:18) Wy 2R(cid:19)(cid:21) − R − x, (2) where ux,y is the in-plane lattice distortion due to strain and the radius has its sign reversed at the y = 0 axis, i.e. R = R (2θ(y) − 1), with θ(y) being the step func- tion. A sharp transition between strained and unstrained regions of the ribbon would be clearly impossible, since it would lead to unrealistically large atomic distances in the vicinity of the transition region, specially for small R. Therefore, we consider a smooth (Gaussian) variation of the curvature KR = 1/R = (1/R0)e(−j2/β 2), where j is the index of the column to which a given atomic site belongs in the lattice,21 the length is described by the dimensionless parameter β, while R0 provides the maxi- mum radius of the curve (namely, at the central column of atoms, where xi,j = 0 in the absence of strain). In general, strain effects on the electronic properties of graphene can be mapped into the analogous problem of an electron under a pseudo-magnetic field distribution,22 whose magnitude and orientation may oscillate over the space, thus making the production of a local non-zero pseudo-flux challenging.23 However, it has been recently demonstrated9,15 that an in-plane circular distortion, as (a) input lead R y x (b) yW /2 0y -W /2 y Max s p 0B Min Max s p B 0 Min -W /2 x R output lead strained region (~ ) Min Bps 0 Max 1 ( ,R )b 0 2 ( ,R )b 0 0 x -W /2 x FIG. 1: (Color online) (a) Sketch of the strained graphene rib- bon. Strain is defined by two circles of radius R (red dashed), which smoothly decay to zero towards the input and output leads, and the width of the strained region is defined by β. The color map indicates smaller (brighter regions) and larger (darker regions) local displacements. Central atoms (green dashed line along the x-axis) are always unstrained in this configuration. (b) (top) Contour plot of the induced pseudo- magnetic field profile for a representative strain configuration characterized by the parameters (β, R0). (bottom) A cross- view of the pseudo-magnetic field along the lines placed in y > 0 (purple short-dashed) and y < 0 (blue long-dashed) regions of the system in the top panel, for two different max- imum radii R2 0 > R1 0. 2 the one proposed here, deforms the Brillouin zone, shift- ing the Dirac cones with respect to each other, just like when an uniform magnetic field is applied perpendicu- lar to the graphene plane, leading to K→K+2πA/Φ0, where Φ0 = e/h is the flux quantum. Such lattice dis- tortion changes the hopping energies and thus induce an effective vector potential24,25 Ax + iAy = 1 evF Xδaij δτij e−iK·δaij (3) where δaij is the vector distance between the adjacent atoms i and j in the strained lattice, vF is the Fermi velocity, δτij is the difference between the strained and unstrained hopping energies, and the pseudo-magnetic field is given by Bps = ∇ × A. Moreover, the distor- tion in different directions for y > 0 and y < 0 provides a pseudo-magnetic kink barrier, with pseudo-magnetic field regions that change sign at y = 0. A schematic ex- ample of the pseudo-magnetic field distribution induced by such strain configuration is presented in Fig. 1(b) for a representative set of parameters (β, R0). We point out that the sample considered in Fig. 1(b) is much smaller than the one investigated throughout this paper, since calculating and plotting a vector potential distribution along the 1801 × 3000 atomic sites of our actual sam- ple requires high computational costs. Therefore, the pseudo-magnetic field in Fig. 1(b) is discussed here only in a qualitative way. The pseudo-magnetic field is found to be zero at input and output leads, where the lattice displacements vanish, and assume its minimum and max- imum values along the ribbon width around x = 0, where the strain is maximum. Two additional kinks are also consistently observed on the left and right sides of this main central kink. They are however much smaller than the central one and, thus, do not play an important role in the valley filtering process, as we will demonstrate fur- ther on. In fact, we observe that as we increase R0, these additional kinks become even lower as compared to the main kink, so that their importance for the transport properties of the actual sample studied throughout the paper (with larger R0) is negligible. This can be verified by comparing the bottom panels in Fig. 1(b) for two dif- ferent maximum radii R2 0 assuming a fixed width for the strained region β. 0 > R1 III. SNAKE STATES ALONG A MAGNETIC FIELD KINK Keeping with the analogy between this strain config- uration and a magnetic field kink, let us first calculate the energy dispersion along the ribbon in the presence of such a magnetic barrier. We assume an inhomoge- neous magnetic field B = B z that depends only on the transversal coordinate, given by B(y) = B(θ(y)−θ(−y)). Low energy electrons in graphene exhibit a linear energy dispersion, so that they behave as massless Dirac-Weyl fermions, thus, obeying the Dirac equation: σ ·(cid:16)−i∇ + e A(cid:17) Ψ = ¯EΨ, (4) with energy E = vF ¯E around the K valley (a similar analysis2 can be made for electrons in K′). Defining the vector potential in the Landau gauge, A = ¯A(y)ex, with ¯A(y) = −B(y)y, the general solution for the wavefunction with translational invariance in the x-direction is Ψ(x, y) = ψ(y)eikx. Therefore, we obtain from Eq. (4) k − ∂y + A(y) 0 ψ2(y)(cid:19) (cid:19)(cid:18) ψ1(y) (5) 0 k + ∂y + A(y) (cid:18) ψ2(y) (cid:19) , = ¯E(cid:18) ψ1(y) where A = e for each component: component, ¯A. This leads us to decoupled equations for instance, for the upper spinor e e y + of distance, we obtain B(y)yi2 B(y) −hk − (cid:26)∂2 + ¯E2(cid:27) ψ1(y) = 0. (6) If we use the magnetic length lb = p/eB as the unit y + sgn(B(y)) − [klb − sgn(B(y))y]2 + ǫ2o ψ1(y) = 0. n∂2 where ǫ = ¯Elb = Elb(cid:14)vF . The energy dispersion along the y-direction is obtained quasi-analytically by solving this equation in terms of parabolic cylinder functions26, Dp(q). Notice the mag- netic field B(y) is piecewise constant, hence, one can separate solutions for each region as (7) ψB>0(y) =X± a±(cid:18) ψB<0(y) =X± a±(cid:18) Dp(±q) iǫ Dp+1(±q)(cid:19) , √2 ∓ √2 Dp+1(±q) iǫ (p + 1)Dp(±q)(cid:19) , ± (8) (9) where q = √2 [sgn(B)klb], and p = ǫ2 2 − 1. The continuity of the wavefunction and its derivatives at these regions provides boundary conditions that lead to quantization of the energy of the system. Eqs. (8) and (9) represent solutions for the first region of the system, but they also can express solutions for the second region by replacing the coefficients a± → c±. In order that the wavefunctions are normalizable we demand a+ = c− = 0. Then, the boundary condition at y = 0 gives the equation that generates the energy quantization condition. Therefore, we obtain iǫ √2 v2 − √2 iǫ (p + 1)u2! = 0, (10) 3.0 2.5 2.0 1.5 1.0 0.5 0.0 -2 3 K K' -1 0 k (1 / lb) 1 2 FIG. 2: (Color online) Energy states for a Dirac particle in the presence of a (pseudo-)magnetic kink. Results for K and K ′ valleys are the same for an external magnetic field, whereas for a pseudo-magnetic field, the K ′ spectrum (red dashed curves) differs from that from the K valley (black solid curves). √2 where the functions are given by u = Dp(−√2klb), and iǫ Dp+1(−√2klb). Notice that these results for the v = wavefunction are very closely related to the one for a magnetic kink profile in a normal 2D semiconductor.27 Numerical results for this system are illustrated in Fig. 2, where we observe an asymmetry in the en- ergy bands along the kink with respect to the kx = 0 axis. Namely, the energy states (predominantly) mono- tonically decrease with kx, implying a negative velocity v = (1/)dE/dk, that eventually converges to zero as k = kx → +∞. It is straightforward to verify that by inverting the sign of the magnetic field kink, this figure is reflected with respect to the kx = 0 axis, and the prop- agation velocities are then predominantly positive. The physical interpretation of this result has its basis on the existence of snake states28,29 that propagate along the kink, which can also be inferred from a simple classi- cal analysis of this problem, involving Lorentz force, cy- clotron orbits and the right-hand rule. Nevertheless, this result is of special importance in the context of pseudo- magnetic kinks discussed here: since the strain-induced pseudo-magnetic field points towards opposite directions in the different Dirac cones K and K′, electrons in each cone will propagate in opposite directions in the pseudo- magnetic kink proposed here, thus yielding an efficient valley filtering process, as we will verify with our numer- ical results afterwards. IV. WAVEPACKET PROPAGATION METHOD In order to investigate the transport properties in our strained graphene, we will use a wavepacket propaga- tion method. A comparison between this kind of method and those based on Green's function formalism can be 4 found in Ref. [30]. The advantage of using this approach is in the possibility of observing the trajectories of the wave packet describing the electron propagating across the scattering (strained) region, which reveals important information about the physics behind any unusual be- havior of the current through the system, being due to e.g. inter-valley scattering, skipping orbits, snake states, etc., as we will demonstrate in the following Section. We use a Hamiltonian within the tight-binding model HT B = −Xi,j τij c†i cj + h.c., (11) where the operator c†i (ci) creates (annihilates) an elec- tron on site i, and τij is the hopping energy between ad- jacent atoms i and j (nearest-neighbors), that depends on the distance δaij between them according to9 Each of these equations is re-written using the Cayley form for the exponentials, e.g. (cid:18)1 + i∆t 4 Hi(cid:19) ηij ≈(cid:18)1 − i∆t 4 Hi(cid:19) Ψt ij, (20) and the remaining tridiagonal matrix equation is then nu- merically solved by standard computational routines.31 For our study, we used a wavepacket width of σ = 300 A and its wave vector ~k has a modulus of k = 0.06 A−1, unless otherwise explicitly stated in the text. Using the linear spectrum approximation for low-energy electrons in graphene, in which E = vf k, the wavepacket energy is estimated to be E = 343 meV. Besides, as we intend to demonstrate the valley polarization of the wavepacket, we place it in different valleys in reciprocal space by shift- ing the wave vector towards the two inequivalent Dirac points: τij → τij(cid:18)1 + 2δaij a0 (cid:19) . (12) kx ← k, ky ← ± 4π 3√3a , (21) We consider an initial Gaussian wavepacket Ψ(x, y) = N exp(cid:20)− (x − x0)2 + (y − y0)2 2σ2 + ikxx + ikyy(cid:21) , (13) where N is a normalization factor and calculate its time evolution using the split-operator method,18,21 in which the time-evolution operator for the Hamiltonian H = Hi + Hj is split as Ψt+∆t ij = e− i 2 Hi∆te− i Hj ∆te− i 2 Hi∆tΨt ij, (14) where Hi(j) is the term of the tight-binding Hamiltonian H that corresponds to a horizontal (vertical) hopping between atomic sites Hii, ji = τ′iji, j + 1i + τ′′iji, j − 1i Hji, ji = τiji + 1, ji + τiji − 1, ji. (15) (16) Notice that for the horizontal term Hi, one has to differ- entiate between hoppings to the right and left neighbour- ing sites, since in the honeycomb lattice, each site has only horizontal hops to one side. Hence, τ′ij = τij =⇒ τ′′ij = 0 and τ′′ij = τij =⇒ τ′ij = 0. The advantage of such splitting lies in the fact that these operators can be represented by tridiagonal matri- ces, that are easily handled by standard computational routines. The wavefunction after a single time step t+∆t is then obtained in three steps where the positive (negative) sign refers to a displace- ment towards the K (K′) point of the Brillouin zone, and a ≈ 1.42 A is the inter-atomic distance. As the Gaussian wavepacket propagates, we calculate the probability of finding the electron before (P1), within (P2), and after (P3) the strained region, as the integral of the square modulus of the wavepacket, taken within the intervals −3, 000 A ≤ x ≤ −400 A, −400 A ≤ x ≤ 400 A, and 400 A ≤ x ≤ 3, 000 A, respectively. Transmission probabilities are assumed to be the converged value of P3 as t → ∞. Besides, we keep track of the wavepacket tra- jectories by calculating the average value of the position, (hxi, hyi), at each time step. The armchair edges of the ribbon do not support edge states, therefore, modelling the electron propagat- ing through the system as a wavepacket, whose tails do not reach the ribbon edges, is justified. Moreover, any improvement to come from other calculation methods, involving e.g. plane waves, scattering matrices and the Landauer-Buttiker formalism, would lead to rather quan- titative corrections to our results, while the qualitative behavior of the system and the proof-of-concept of val- ley filtering with a pseudo-magnetic kink proposed here, which are the main goals of this work, would still hold, since they are based on more fundamental physical prop- erties of the proposed structure, as we will discuss in what follows. ηij = e− i 2 Hi∆tΨt ij, ξij = e− i 2 Hj ∆tηij , Ψt+∆t ij = e− i 2 Hi∆tξij . (17) (18) (19) V. RESULTS AND DISCUSSION The existence of snake states in such a strained graphene lattice, as due to the induced pseudo-magnetic fields, is confirmed by the trajectories drawn in Fig. 3 of the center-of-mass of a k = 0.06 A−1 wavepacket propa- gated in time through the system described by Fig. 1(a), assuming β → ∞ and R0 = 104 A, as a test case. If ) A 800 600 400 200 0 -200 -400 -600 -800 ( > y < K K' -1000 -500 0 <x> (A) 500 1000 FIG. 3: (Color online) Examples of calculated trajectories of electron wavepackets propagating with momenta around K (black solid curves) and K ′ (red dashed curves) valleys, starting at (x, y) points (indicated by blue solid dots) given by (1250 A, 300 A) and (−1250 A,−300 A), respectively. Arrows indicate the direction of propagation along the trajectories. this wavepacket has momentum around the Dirac cone K (black solid curves) and propagates from left to right, starting at the bottom-half of the ribbon, its trajectory is deflected by the pseudo-magnetic Lorentz force towards the top-half, where it is deflected downwards again by the opposite pseudo-magnetic field, thus performing a snake-like trajectory. If this same packet has momentum around the K′ cone (red dashed curves), it is deflected downwards and eventually repelled from the strained re- gion. If this packet starts from the top-half instead, both 1 time (fs) 300 FIG. 4: (Color online) Probability densities, as a function of time, of finding the electron before (P1), within (P2), and after (P3) the β = 900 and R0 = 10, 000 A strained region, for a wavepacket with k = 0.06 A−1 around the K point of the Brillouin zone. Results for P3 considering a wavepacket around K ′ are shown for comparison. 5 curves are just mirror-reflected with respect to the hyi = 0 axis of Fig. 3, and the situation remains the same. Conversely, if the wavepacket propagates from right to left, it is the K′ packet that draws a snake trajectory, whereas the K packet is reflected. One could think that wavepackets deflected towards the edges of the system (i.e., further away from its center) would be reflected by the ribbon edges, perform skipping orbits, and eventually pass through the strained region. However, since the rib- bon has armchair edges, reflected wavepackets are scat- tered to the other Dirac cone, where the pseudo-magnetic field is opposite, thus the skipping orbit follows the oppo- site direction and the wavepacket comes back anyway.17 This non-propagating edge state is emphasized in Fig. 3 for a wavepacket that started at the bottom-half (top- half) of the ribbon and around K′ (K) Dirac valley. In this way, one completely avoids the problem of having valley mixing of the snake and edge states at the end of the ribbon, which would otherwise occur in the case of propagating edge states e.g. in the presence of an exter- nal applied magnetic field. Valley mixing by scattering at the contacts can also be further suppressed by using graphene electrodes.32 Notice, however, that the present proposal will not work very efficiently for zigzag graphene nanoribbons, where such inter-valley edge scattering does not occur and skipping orbits are allowed to propagate at the zigzag edges. Although this represents a limitation of the proposed system, fabrication techniques have been advancing fast, and armchair graphene nanoribbons with very high edge quality have already been experimentally demonstrated.33 Results in Fig. 3, thus, allow us to conclude that electrons in K (K′) cones in such a strained armchair graphene ribbon can only propagate towards the right (left). Analogously, if the strain configuration is inverted, trajectories drawn by K and K′ packets are switched. Such a picture of snakes states strongly suggest a valley filtering effect. In fact, if one now considers a system with finite strain region β = 900 (in units of the inter-atomic distance a0 = 1.42 A) and R0 = 10, 000 A, a wavepacket with k = 0.06 A−1 around the K cone passes through this region with a high probability P3 ≈ 0.9, whereas the same packet in K′ would have a much lower transmission probability P ′3 ≈ 0.3. Let us now search for an optimization of the val- ley polarization effect. The polarization, as defined by P = 1 − P3/P ′3, where P ′3 is the transmission probabil- ity for a wavepacket in the K′ cone, is shown in Fig. 5 for k = 0.06 A−1, assuming different parameters β and R0. For a fixed strain radius R0, increasing the length of the strain region β increases the polarization, as shown in Fig. 5(a). Besides, results in this panel also sug- gest that decreasing R0 would always improve the po- larization; this would be reasonable, since smaller radii yield stronger distortions in the lattice and, consequently, larger pseudo-magnetic fields. This is however not al- ways the case: Fig. 5(b) shows that even for β as large 6 in Fig. the rest of the wavepacket readily turns back. On the other hand, if the energy is too high, orbit radii may end up being larger than the strained region length, so that the snake-like propagation that leads to valley po- larization no longer occurs. Moreover, the Lorentz orbit radius is inversely proportional to the pseudo-magnetic field intensity, therefore, increasing the strain by reduc- ing R0 would also lead to Lorentz orbits with smaller radius. This classical picture is consistent with our nu- merical findings: 6(a), the largest strain ra- dius R0 = 12, 000 A provides the fastest decay of po- larization as the energy increases, since it yields a lower strain-induced pseudo-magnetic field and, therefore, en- ergies slightly higher than the optimal E ≈ 125 meV already provide orbit radii larger than the strained re- gion length. Conversely, for energies lower than E ≈ 125 meV, R0 = 12, 000 A provides the best polarization, as its weaker pseudo-magnetic field compensates for the low energy and prevents the orbits radius of becoming too small. Also, Fig. 6(b) shows that, for higher energies, where orbit radii are larger, polarization is more efficient for larger length β, and the energy for optimal polariza- tion increases with this parameter. The valley filtering effect by pseudo-magnetic kinks 900 R0 (Å) 10000 FIG. 5: (Color online) Valley polarization of the outgoing wavepacket, with k = 0.06 A−1, as a function (a) of the width of the strained region β, for different radii R0, and (b) as function of the strain radius R0, for different β values. as 900 A, decreasing the strain radius will always lead to a maximum polarization at an intermediate value R0 ≈ 5, 000 A, so that the polarization is reduced as the ra- dius is further decreased. This is due to the fact that the smooth connection between the unstrained ribbon leads and the strained region might end up creating a com- plicated pseudo-magnetic field distribution, with regions with fields pointing to opposite directions, which would harm the polarization effect investigated here. This also explains the negative polarization observed for small β in Fig. 5(a). So far, all results were obtained for k = 0.06 A−1, which corresponds to a wavepacket energy E = 343 meV. It is however important to check how the polarization de- pends on the wavepacket energy. This is shown in Fig. 6, where we verify that the valley filtering process pro- posed here has an optimal range of energies. Indeed, if the energy is too low, the pseudo-magnetic Lorentz or- bits would have a very small radius, so that only portions of the wavepacket that are very close to the y = 0 line would pass through the system as snake states, whereas E (meV) FIG. 6: (Color online) Valley polarization of the outgoing wavepacket as a function of its energy, considering (a) β = 900 and different radii R0, and (b) R0 = 6, 000 A, for different values of β. demonstrated here is also expected to be robust against impurity and defects scattering: as already discussed, electrons in each valley have only one possible direction of propagation (see Fig. 3), due to the monotonic behav- ior of all the energy states as a function of momentum (see Fig. 2), which provides a single direction for the group velocity in each valley. After scattering by impu- rities or defects, the electron must end up in one (or a combination) of the states in Fig. 2. If the electron is already in the valley that allows its propagation through the system (as a snake state), with positive velocity (i.e. monotonically increasing energies as a function of mo- mentum), any component of the scattered electron wave function that ends up in the other valley must be de- flected backwards, since there is simply no energy state in that valley with positive velocity as well. Thus, provided the pseudo-magnetic field kink distribution is preserved, only electrons in one of the valleys are allowed to reach the other end of the ribbon, even after scattering events. VI. CONCLUSIONS We have investigated the wavepacket propagation through a graphene nanoribbon with armchair edges for a specific strain distribution. The latter provides a pseudo- magnetic barrier kink along the ribbon. By following the trajectory of the center-of-mass of the wavepacket, calculated by solving the time-dependent Schrodinger equation for the tight-binding Hamiltonian, one observes snake states, which have a fixed propagation direction, consistent with the pseudo-magnetic kink picture. How- 7 ever, one can analytically verify that, by reversing the magnetic kink, the propagation direction of snake states must be reversed. Since the pseudo-magnetic field points towards oppo- site direction in the different Dirac cones, wavepackets in the different cones can only have fixed opposite direc- tions of propagation. This effect results in an efficient valley filtering process, which does not require either lat- tice defects, edge engineering, or any externally applied fields or potentials. Our numerical results show signifi- cant valley polarization through this system, which can be optimized by the parameters (β, R0) that depend on the electron energy (i.e. the Fermi energy). Notice that the in-plane circular deformation of a graphene nanoribon proposed here is just one particu- lar way of inducing a kink pseudo-magnetic field bar- rier: any other strain distribution that produces such a pseudo-magnetic kink would lead to similar valley filter- ing effect, which requires only a graphene ribbon with armchair boundaries (to avoid edge propagation) and a pseudo-magnetic field that flips its direction across a line parallel to them. Acknowledgments Discussions with R. Grassi are gratefully acknowl- edged. This work was supported by the Brazilian Coun- cil for Research (CNPq), under the PRONEX/FUNCAP and Science Without Borders (SWB) programs, CAPES, the Lemann Foundation, and the Flemish Science Foun- dation (FWO-Vl). ∗ Electronic address: [email protected] † Electronic address: [email protected] ‡ 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 A. Rycerz, J. Tworzydlo, and C. W. J. Beenakker, Nat. Phys. 3, 172 (2007). 4 M. Zarenia, O. Leenaerts, B. Partoens, and F. M. Peeters, Phys. Rev. B 86, 085451 (2012). 6, 30 (2010). 10 Tony Low and F. Guinea, Nano Lett. 10, 3551 (2010). 11 T. Fujita, M. B. A. Jalil, and S. G. Tan, Appl. Phys. Lett. 97, 043508 (2010). 12 F. Zhai, Y. Ma, and Y.-T. Zhang, J. Phys.: Condens. Mat- ter 23, 385302 (2011). 13 N. Myoung and G. Ihm, Curr. Appl. Phys. 14, 1455 (2014). 14 F. Zhai, X. Zhao, K. Chang, and H. Q. Xu, Phys. Rev. B 82, 115442 (2010). 15 F. Guinea, A. K. Geim, M. I. Katsnelson, and K. S. Novoselov, Phys. Rev. B 81, 035408 (2010). 16 N. Levy, S. A. Burke, K. L. Meaker, M. Panlasigui, A. Zettl, F. Guinea, A. H. Castro Neto, and M. F. Crommie, Science 30, 544 (2010). 17 D. R. da Costa, A. Chaves, G. A. Farias, L. Covaci, and 5 M. Ramezani Masir, A. Matulis, and F. M. Peeters, Phys. F. M. Peeters, Phys. Rev. B 86, 115434 (2012). Rev. B 84, 245413 (2011). 6 D. R. da Costa, Andrey Chaves, S. H. R. Sena, G. A. Farias, and F. M. Peeters, Phys. Rev. B 92, 045417 (2015). 7 Takeshi Nakanishi, Mikito Koshino, and Tsuneya Ando, Phys. Rev. B 82, 125428 (2010). 18 A. Chaves, L. Covaci, Kh. Yu. Rakhimov, G. A. Farias, and F. M. Peeters, Phys. Rev. B 82, 205430 (2010). 19 C. S. C. Downs, A. Usher, and J. Martin, J. Appl. Phys. 119, 194305 (2016). 20 Topi Korhonen and Pekka Koskinen, Phys. Rev. B 93, 8 L. Pratley and U. Zulicke, Appl. Phys. Lett. 104, 082401 245405 (2016). (2014). 9 F. Guinea, M. I. Katsnelson, and A. K. Geim, Nature Phys. 21 Kh. Yu. Rakhimov, Andrey Chaves, G. A. Farias, and F. M. Peeters, J. Phys.: Condens. Matter 23, 275801 (2011). 8 22 B. Amorim, A. Cortijo, F. de Juan, A. G. Grushine, F. Guinea, A. Guti´errez-Rubio, H. Ochoa, V. Parente, R. Rold´an, P. San-Jose, J. Schiefele, M. Sturla, M. A. H. Voz- mediano, Phys. Rep. 617, 1 (2016). 23 T. O. Wehling, A. V. Balatsky, A. M. Tsvelik, M. I. Kat- snelson, and A. I. Lichtenstein, Europhys. Lett. 84, 17003 (2008). 24 M. R. Masir, D. Moldovan, and F. M. Peeters, Solid State Commun. 175-176, 76 (2013). 25 D. Moldovan and F. M. Peeters, Phys. Status Solidi RRL, 1-7 (2015) 28 L. Oroszl´any, P. Rakyta, A. Korm´anyos, C. J. Lambert, and J. Cserti, Phys. Rev. B 77, 081403(R) (2008). 29 Y. Liu, R. P. Tiwari, M. Brada, C. Bruder, F. V. Kusmart- sev, and E. J. Mele, Phys. Rev. B 92, 235438 (2015). 30 T. Kramer, AIP Conference Proceedings 1334, 142 (2011). 31 W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes 3rd Edition: The Art of Sci- entific Computing (Cambridge University Press New York, New York, 2007), p. 56. 32 G. Jo, M. Choe, S. Lee, W. Park, Y. H. Kahng, and T. Lee, Nanotechnology 23, 112001 (2012). 26 T. K. Ghosh, A. De Martino, W. Hausler, L. Dell'Anna, and R. Egger, Phys. Rev. B 77, 081404(R) (2008). 27 J. Reijniers and F. M. Peeters. J. Phys.: Condens. Matter 33 A. Kimouche, M. M. Ervasti, R. Drost, S. Halonen, A. Harju, P. M. Joensuu, J. Sainio, and P. Liljeroth, Nat. Comm. 6, 10177 (2015). 12, 9771 (2000).
1601.03212	2	1601	2016-07-07T12:44:30	Low-threshold RNGH instabilities in quantum cascade lasers	[ "cond-mat.mes-hall", "physics.optics" ]	A theoretical study on low-threshold multimode instabilities in quantum cascade lasers (QCLs) and laser diodes (LDs) is presented. Previously, low threshold Risken-Nummedal-Graham-Haken (RNGH) instabilities were reported in several experimental studies of QCLs. They were attributed so far to the combined effect of the induced grating of carrier population and of a built-in saturable absorption feature that may be present in the monolithic single-section cavity of these Fabry-P\'erot lasers. Here we show that low-threshold RNGH instabilities in QCLs occur due to a carrier coherence grating induced in the gain medium and not due to intracavity saturable absorption. We find that QCLs with a few mm long cavity exhibit intermittent RNGH self-pulsations while regular self-pulsations are possible in short-cavity QCLs, with the cavity length of 100 um or smaller. We examine a transient behavior to RNGH self-pulsations in short-cavity QCLs and find features that resemble cooperative superradiance. Our findings open a practical way of achieving ultra-short pulse production regimes in the mid-infrared spectral range. Applying same approach to semiconductor laser diodes (LDs) we explain the absence of RNGH self-pulsation in LDs based on a quantum well gain media, while practically established method for reaching the ultrafast coherent emission regimes in LDs is to incorporate a separately contacted saturable absorber section in the LD cavity.	cond-mat.mes-hall	cond-mat	Low-threshold RNGH instabilities in quantum cascade lasers Nikola Vukovic1, Jelena Radovanovic1, Vitomir Milanovic1 and Dmitri L. Boiko2 1University of Belgrade, School of Electrical Engineering, Bulevar kralja Aleksandra 73, 11000 Belgrade, Serbia e-mail address: [email protected] 2 Centre Suisse d'Électronique et de Microtechnique (CSEM), Jaquet-Droz 1, CH-2002 Neuchâtel, Switzerland e-mail address: [email protected] A theoretical study on low-threshold multimode instabilities in quantum cascade lasers (QCLs) and laser diodes (LDs) is presented. Previously, low threshold Risken- Nummedal-Graham-Haken (RNGH) instabilities were reported in several experimental studies of QCLs. They were attributed so far to the combined effect of the induced grating of carrier population and of a built-in saturable absorption feature that may be present in the monolithic single-section cavity of these Fabry-Pérot lasers. Here we show that low-threshold RNGH instabilities in QCLs occur due to a combined effect of the carrier coherence grating and carrier population grating induced in the gain medium and not due to an intracavity saturable absorption. We find that QCLs with a few mm long cavity exhibit intermittent RNGH self-pulsations while regular self-pulsations are possible in short-cavity QCLs, with the cavity length of 100 µm or smaller. We examine a transient behavior to RNGH self-pulsations in short-cavity QCLs and find features that resemble cooperative superradiance. Our findings open a practical way of achieving ultra-short pulse production regimes in the mid-infrared spectral range. Applying same approach to semiconductor laser diodes (LDs) we explain the absence of RNGH self- pulsation in single-section LDs based on a quantum well gain media, while practically established method for reaching the ultrafast coherent emission regimes in LDs is to incorporate a separately contacted saturable absorber section in the LD cavity. PACS: 42.55.Px, 42.65.Sf, 42.60.Mi 1 I. INTRODUCTION Following the first demonstration at Bell Labs in 1994 by F. Capasso and J. Faist, quantum cascade lasers (QCLs) have experienced rapid and dramatic improvements in power, efficiency and wavelength span. Continuously operating QCLs in the mid infra-red (MIR) spectral range have important applications in security, health, gas sensing and gas analysis [1]. Production of frequency combs in QCLs has been demonstrated and proven to be very promising for spectroscopic applications [2]. Yet, QCL potential as a versatile spectroscopic tool could be significantly enhanced if operation in the ultra- short pulse regime will be possible. Such regime will enable time resolved spectroscopic measurements in MIR and FIR ranges for a wide spread of applications such as LIDARs (LIght Detection and Ranging), Earth observation, environmental remote sensing of molecules, in particular, of greenhouse gases, high-speed QCL based communications utilizing the atmospheric transmission windows of 3-5 μm or 8-12 μm and many more [1]. However ultrafast carrier relaxation at picosecond time scale prohibits passive mode-locking or Q- switching operation in QCLs because of the cavity roundtrip time being longer than the gain recovery time [3]. Therefore gain switched pulse production has been attempted, yielding 120 ps pulse width [4]. On the other hand, active mode-locking was achieved in QCLs utilizing diagonal transition, with the upper state lifetime being increased to 50 ps so as to match the cavity roundtrip time. Yet the diagonal transition renders QCL operation temperature out of the practical use [5]. One more promising approach to generate short MIR pulses has emerged from experimental observations that some of CW operating QCL samples exhibit features of low-threshold [6] Risken- Nummedal-Graham-Haken (RNGH) multimode instabilities [7,8] and hence they might be capable of producing ultra-fast self-pulsations. However, the second-order interferometric autocorrelation measurements have shown that the output optical pulses have significant stochastic constituent [5], if 2 these autocorrelation traces should not be attributed at all to coherence spikes. In order to confirm RNGH instability and to understand conditions for reaching regular self-pulsation in QCLs, one should figure out the origin of such low second threshold in these lasers. In general, the multimode RNGH instabilities are related to a non-adiabatic [9] behavior of the medium polarization, excitation of rapid coupled oscillations of the medium polarization P and population inversion N. They show up as self-pulsations in the output laser emission, whereas the optical spectrum is expected to be split in two mode clusters (sidebands) with frequency separation of the order of the Rabi oscillation frequency. Below the RNGH instability threshold, the optical mode is fully controlled by the laser cavity. The buildup of Rabi oscillations at above the second threshold indicates that now a medium macroscopic polarization (coherence) has the major impact on the optical field in the laser cavity. The original RNGH theory, which was established for a CW operating unidirectional ring laser, requires the pump rates of at least 9 times above the lasing threshold in order for multimode instability to occur. However the experiments revealed a different second threshold behavior. In Ref [10], a ring-cavity dye laser with homogeneously broadened gain line has shown RNGH-like multimode instability at small excess above the lasing threshold. More specifically, quenching of the main lasing mode and excitation of two optical sidebands was observed at 1.2 times above the lasing threshold. The frequency separation of the sidebands was close to the Rabi oscillation frequency or one of its sub-harmonics and it was increasing with the pump rate. Later on there were several claims on RNGH instabilities in CW operating Er3+ doped ring fiber lasers [11-13], and ring Nd:YAG lasers [14]. However, interpretation of these experimental observations remained doubtful. Thus the period of self-pulsations was close to the cavity roundtrip time, so they can be attributed to mode-locking phenomena. Most unexpectedly, the self-pulsations were occurring almost at the lasing threshold, contrary to the theoretical predictions for at 3 least 9-fold excess above the lasing threshold. Note that the relaxation T1 and dephasing T2 times in the active gain media of these lasers are relatively long as compared to their semiconductor counterparts and hence their Rabi oscillation frequencies are low. It was understood that (i) the cavity length must be made very long in order to match the intermodal frequency separation to that one of the Rabi oscillations, and that (ii) the lowering of second threshold in Er3+ doped ring fiber lasers can be attributed to a 3-level structure of the optical transitions in Er3+ ions [15]. Nevertheless the fact of experimental observation of RNGH instability remained debatable. Ten years later, a clear Rabi frequency splitting between two clusters of modes was observed in the lasing spectra of CW operating QCLs [6], indicating multimode RNGH instability in these Fabry-Pérot (FP) cavity lasers. Interestingly, the RNGH instability was observed as in ridge waveguide QCLs as in buried heterostructure QCLs. Although the role of spatial hole burning (SHB) in lowering the second threshold was understood, it was thought [16] that the low-threshold RNGH instability does not occur just as a result of the induced grating of carrier population. Therefore, an additional assumption was made in [16] on a built-in saturable absorber in the cavity of QCL, allowing for a reduction of the 2nd threshold from 9-fold excess to about 1.1 times above the lasing threshold. As a matter of fact, the saturable absorber has been shown to lower the instability threshold in a CW ring laser [17]. However, the nature of saturable absorption in QCLs has never been fully clarified [1]. In particular, the authors in [16] have evoked the Kerr lensing effect as a possible mechanism responsible for the saturable absorption. In case of narrow ridge waveguide lasers, due to overlap of the waveguide mode tails and the metal contact deposited on the waveguide, the Kerr lensing does may lead to saturable absorption. However it cannot produce the saturable absorption effect of the same strength in buried heterostructure QCLs. At the same time low-threshold RNGH instability was observed in both ridge waveguide and buried heterostructure QCLs. 4 In this paper, we develop a model for low-threshold multimode RNGH instability in a QCL without saturable absorber. This is a more realistic representation for a monolithic single-section FP cavity laser. Technically, a set of important corrections to the original treatment [7,8,16] was made. It consists in (i) accounting for the induced carrier coherence grating alongside with the carrier population grating and in (ii) accounting for the carrier diffusion process that leads to relaxation of both gratings. The outcomes of our analysis for the second threshold in QCLs are in perfect agreement with the numerical simulations based on a travelling wave (TW) rate equation model as well as with the experimental data available in the literature [6,16,18]. Interestingly, we find that regular self-pulsations at picosecond time scale are possible in short-cavity QCLs. Our model also explains the nonexistence of multimode RNGH instability in quantum well (QW) laser diodes (LDs). This paper is organized in the following manner: The model is presented in Sec. II. Section III addresses the second threshold conditions and discusses the RNGH instability in long-cavity and short- cavity devices, as well as the importance of the carrier diffusion effects in QCLs and conventional LDs; Sec. IV provides a conclusion. II. MODEL DESCRIPTION A. QCL laser as a model system for RNGH instabilities A quantum cascade laser is a unipolar device based on intersubband transitions. Its active region consists of a large number of periodically repeated epilayers. Each period (stage) comprises a gain and injection/relaxation regions. While the gain region serves to create a population inversion between the two levels of the lasing transition, the purpose of the injection/relaxation region is to depopulate the lower lasing level and to provide injection of electrons in the upper lasing level of the next stage [1]. Although QCLs have a complicated structure of epilayers, it was understood and confirmed 5 experimentally that the internal structure does not alter directly the appearance of multimode RNGH instabilities and that a semiclassical model with effective parameters of the laser gain medium suffices [16]. Alongside with the studies of RNGH instabilities in InGaAs QCLs, we apply our analytic model to QW LDs based on AlGaAs and InGaN alloys. For all of them we assume a simple single-section Fabry- Perot (FP) cavity design. The parameters of our model for three types of lasers considered here are summarized in Table I. TABLE I. Dynamic model parameters for QCLs and QW LDs considered in this paper. Parameter Λ T1 T2 T2_eff Tg T2_g αi D ng R1, R2 Γ ∂g/∂n nt Name Lasing wavelength Carrier lifetime Carrier dephasing time Effective carrier dephasing time in the presence of diffusion Relaxation time of the carrier population grating Relaxation time of the coherence grating Intrinsic material loss Diffusion coefficient Group refractive index Cavity facet reflection coefficients Optical mode confinement factor Differential material gain Transparency carrier density InGaAs QCL 10 µm 1.3 ps 140 fs 138 fs 0.927 ps 0.158 ps 128 fs 24 cm-1 180 cm2/s 3.3 27% 0.5 41 fs 5 cm-1 20 cm2/s (ambipolar) 3.8 27% 0.01 AlGaAs QW LD 850 nm InGaN QW LD 420 nm 1-2 ns 100 fs 86 fs 84 fs 0.13 ps 37 fs 35 cm-1 7 cm2/s (ambipolar) 3.5 18 % 0.02 2.1×10-4 cm3/s 1×10-6 cm3/s 2.2×10-6 cm3/s 7×1014 cm-3 2×1018 cm-3 1.6×1019 cm-3 As a model system for RNGH instability, we consider an InGaAs QCL with a direct transition, having the carrier lifetime T1 of about one picosecond. This is much smaller than the cavity roundtrip time. As in the case of heterostructure epilayers, we do not examine any particular lateral cavity design since it 6 was shown experimentally that the laser design (ridge waveguide [16] or buried heterostructure [6]) has no or little impact on the RNGH instability in QCLs. Although our QCL model parameters are slightly different from the ones in Ref.[16], this difference has no impact on the main conclusions of the present paper, in particular, regarding the occurrence of low-threshold RNGH instability. In the table, the optical mode confinement factor and transparency carrier density for QCL are defined using the overall thickness of epilayers in the period. In Section III.E we apply our model to AlGaAs QW LDs and InGaN QW LDs (last two columns in Table I) with the objective to verify the model agreement with experimental observations that single- section QW LDs do not reveal any feature of RNGH instability. For all considered here model systems, the photon lifetime in the cavity exceeds the dephasing time T2, so as the dynamic behavior reported here is intrinsically different from the Class-D laser dynamics discussed in [19]. B. Travelling wave rate equation model As a starting point of our analysis we use semiclassical Maxwell-Bloch (MB) equations for a two-level system. Following along the lines of the travelling wave model approach from [20] and accounting for the diffusion term in the Schrödinger equation [21], we obtain the following system of rate equations:    ab  i ab  i E      ab T 2  D 2   ab z 2  Ei  2      (   ab * ab   )     D 2   z 2  pump T 1  E 2 z  2 g 2 n c  E 2 t  N  c 2  0  2 t (  ab * ab  )   0 ln g c E  t (1) (2) (3) 7 where * ab   ba is the off-diagonal element of the density matrix, the non-equilibrium carrier density N   N   aa  bb ( ) is represented as the population inversion between the upper (index "bb") and lower (index "aa") levels of the lasing transition, N  ab * ab ( ) is the active medium polarization ("coherence"), ω and µ denote the resonant frequency and the dipole matrix element of the lasing transition, respectively, T1 and T2 are the longitudinal and transverse relaxation times, N  / T 1 is the pump pump rate due to injection of electrons into the upper lasing level, N is defined by the doping density in the injector region and D is the diffusion coefficient for electrons in the plane of active QWs of the QCL structure (for a LD, the ambipolar diffusion coefficient for the holes and electrons is used instead). E, Γ and ng stand for the cavity mode electric field, the overlap between the optical mode and the active region and the group refractive index for the cavity mode, respectively. Note that the differential material gain is g  / n  2  0 gn 2  T 2 / and 0l is the linear loss coefficient. The difference between our MB equations for QCLs and equations used in [16] is the coherence diffusion term / z D ab  2 2   that appears in the first equation. Numerical simulations with the travelling wave rate equation model reported below were accomplished by introducing in (1)-(3) two slowly varying amplitudes for the counter propagating waves in the Fabry-Perot (FP) cavity and distinguishing the medium's polarizations associated with the forward and backward traveling waves [22]. C. Approximation of coupled-mode rate equation model The system of MB equations (1)-(3) is not ideally suited for analytical study on multimode RNGH instability. In Ref [23] a comprehensive approach for analysis of multimode instabilities in a ring laser was established using a truncated set of coupled-mode equations. A similar approach was applied in [16] for the case of FP-cavity QCLs. However the key point of such truncated modal expansion is to take into account all coupled harmonics in a self-consistent way [23]. We argue that previous theoretical 8 treatment of RNGH instabilities in QCLs [16] does not take into account the spatial harmonics of induced macroscopic polarization of the gain medium in the FP-cavity laser or bidirectional ring laser. As a result, in the limit of adiabatic elimination of the medium polarization, the truncated set of coupled- mode equations from [16] does not agree with well-established models for a Class-B laser dynamics [24]. (See discussion to equations (7)-(11) below and also the Appendix A. We stress that by no means our considerations for bidirectional ring lasers and FP cavity lasers can be applied to the unidirectional ring laser considered in Refs. [7,8]). Furthermore, in order to match the experimentally measured second threshold of ~1.1·Ith (Ith is the lasing threshold pump current), the model in [16] assumes that there is a built-in saturable absorber in the monolithic cavity of QCL due to the Kerr lensing effect. As a matter of fact, the saturable absorber does lower the RNGH instability threshold in a CW laser [17]. However in the buried heterostructure QCLs, which also exhibit low second threshold, the proposed Kerr-lensing mechanism for saturable absorption has never been unambiguously confirmed [1]. In order to elucidate another possible origin of the low-threshold RNGH instability, we have expanded the MB equations (1)-(3) in to a truncated set of self-consistent coupled mode equations. Our analysis of the SHB effect has shown that as soon as the spatial grating of carrier population is taken into account (the terms 2 below), the third-order spatial harmonics of the induced macroscopic polarization grating (the terms  and  below) have to be accounted for in the coupled-mode expansion:  i (  t kz  ) E e *     E e  i (  t kz  )    1 2  i (  t kz  ) E e *     E e  i (  t kz  )          2 ikz e  e i (  t kz  )        2 ikz e  e i (  t kz  ) (4) (5) (6) ( , ) E z t  1 2 ( , ) ab z t   ( , ) z t      2 0 e 2 ikz e 2 ikz    2 , 9 where ),( tzE and ),( tz are real-valued variables ( *    2 2 ) and the amplitudes E,  ,  , 0 and  vary slowly in time and space as compared to the plane wave carriers. The two optical carrier 2 waves propagating at frequency ω and wavenumbers ±k stand for the initial lasing mode in the cavity. Substitution of Eqs. (4)-(6) in MB equations (1)-(3) leads to the following coupled-mode equations in the slowly varying envelope approximation: n c  t E     z E   i N  cn   0 g    1 2 l E 0  ,   t     t   i   2  i E  2   E  0    2 E      T 2  k D 2   ,    2   T 2  9 Dk 2   ,     t 0    pump T 1 0  i    E * *       E   . c c  ,    t  2 i    E * *       E   E *     E *        2 T 1  4 k D 2   2 (7) (8) (9) (10) (11) where 0l is the cavity loss coefficient that comprises intrinsic material losses and output coupling losses. In order to verify our coupled-mode expansion (4)-(6), we have performed the adiabatic-following approximation test [9] (see Appendix A) and find that the adiabatic approximation for our set of equations is in excellent agreement with the well-established Class-B laser model. Note that this is not the case for a set of coupled-mode equations used in [16]. The second important difference is that our coupled-mode system (7)-(11) shows high-frequency instabilities at Rabi oscillation frequency while the one of Ref.[16] shows instabilities at much lower frequency (see Section III.F). Next, we carry out a linear stability analysis of our model system (7)-(11). In what follows we introduce the effective relaxation times that account for the contribution from the diffusion terms 10 T g   1  T 1  4 Dk 2   1 , T 2 _g   1 T  2  9 Dk 2   1 and T 2 _ eff   1 T  2  2 Dk   1 . (These relaxation times are quoted in Table 1.) We also define new variables   El e / 0  , l    0 / ( T  2_ th eff ) , n 0   l 0 0 / ( T  2_ th eff ) , n  l   0 / ( T  2_ th eff ) . The pump rate is accounted for by the parameter p   pump /  , which measures the pump rate excess above the lasing threshold in the absence of SHB, th and N  th   0 ln 0 g 2   T _2 eff is the carrier density at the lasing threshold. We obtain the steady-state solution of Eqs.(7)-(11) assuming an arbitrary initial phase θ of the wave propagating in the positive z axis direction (forward wave). Because the optical mode field ),( tzE and the population parameter ),( tz in Eqs.(4)-(6) are the real-valued variables, the phase of the wave propagating in the backward direction is –θ, yielding the following solution for CW lasing regime: e   E0le i , n 0  l  0 0 T 2_eff  e i  , n   ( 0 l 2   0 T 2 _ eff  1) ,    ei 2  l i E0 ,    where T _2 T _2 g eff i 2 (  0  )1 e  i l 3  E0 , E p  0 2 effTT _21 (12) (13) is the normalized field amplitude. Note that variables n n  , e  e  , are complex conjugate while the variables      ,       are anti-conjugate. The SHB increases the effective lasing threshold and reduces the slope efficiency. To account for this effect, we introduce an additional parameter     , which is the ratio of the fundamental 0 0 / th 11 harmonic of the average carrier density NΔ0 to its value at the lasing threshold NΔth (at p=1), when SHB effect does not yet settle–in. We obtain the following expression for ν0:  0  1 2     p 1   T 2 _ T 2 _ eff g  2 TT 1 2 _ T T 2 _ g eff g      1 4     p 1   T 2 _ T 2 _ eff g  2 TT 1 2 _ T T 2 _ g eff g 2      p 1      T 2 _ T 2 _ eff g      2 TT 1 2 _ T T 2 _ g eff g . (14) At lasing threshold, ν0=1 and it increases above the threshold (at p>1). At very high pump rate (p>>1), ν0 asymptotically approaches the value of 1  T 2 _ eff / T 2 _ g . The linear stability analysis of the steady-state solution (12)- (14), is performed by introducing small perturbations e i e      i e r  , ,    i    r , i    ii (     i rr ) , 0n and n n i n      r i  in the corresponding variables of the coupled-mode equations (7)-(11). Taking into account the complex conjugate and anti-conjugate relationships between the variables n n *     , e e *     , *      ,     *   , we obtain the following linearized system of differential equations: d dt   i  e   r n   0  n   r   ii    r  e  i  n   i    rr                                          )1 2cos  E  1  T eff _2 1  4 cos E    ( 0   0 2 T 1 2  2 E   _2 eff i  cos  cos 2 0 1 T 1  E  cos 2 0 0 0 0 0 (  0  0 0  2sin)1  2 T _2 0 eff E   sin 2 0 4 E sin   2 E sin  2 E cos   2 E sin  0  2 E sin cos      E  2 E cos  0  (     )1 0 0 0 T _2 T _2 (  0 (  0 g 3cos     eff 2cos eff )1  2 T _2 2sin)1  2 T _2 0 eff 0 0   E E E   sin 2 0  1 T g cos 2 sin  2 0  1 T _2 0 g 0  0 1 T _2 eff 1   cos       sin  E  2cos  0 0  E  E sin 2 cos 2 0  (     (      g  )1 3sin T _2 T _2 eff 2sin)1 (   0 2 T _2 eff )1   0 2 T 1 2  g 3cos _2 eff i    ( 0 T _2 T _2 eff ( )1   0 2 T _2  eff 2 E sin  0  (  0  )1      T _2 T _2 (  0       sin g 3sin eff 2sin)1  2 T _2 eff  E 0 0 0 2 E sin   2 E  0  )1 E  sin 2 0 0   cos 2cos       E  E  1 T g cos 2  0 0 0 0 0 0                            2 E cos   1 T _2 g    i  e   r n   0  n   r   ii    r  e  i  n   i    rr               (15) where ng /cl 0 is the photon lifetime in the cavity. 12 Using the ansatz from Ref. [8], we recast perturbations to the main lasing mode in the form of propagating waves  exp( / gin z c   , where Λ can be interpreted as Lyapunov exponent, Ωng/c is the t ) detuning of the propagation constant from the one of the lasing mode and  Im(  ) is the frequency offset. Note that among all possible solutions Λ(Ω) of the dispersion equation that follows from Eq.(15), only that solutions with positive increment Re(Λ)>0 will build-up in the cavity after several roundtrips, which fulfill the cavity roundtrip phase self-repetition condition 2ΩLng/c=2πn, where L is the cavity length and n is an integer number. Nevertheless, it is convenient to carry out our stability analysis considering  as a continuous parameter. It can be regarded as the offset frequency from the lasing mode because for all practical situations of interest in this study, the approximation Im( ) suffices. Numerical solution of Eq.(15) reveals that only one Lyapunov exponent of the 9×9 linear stability matrix may have positive real part and thus may lead to a multimode instability in the practical range of pump rates considered here (see Fig. 1 below and related discussion). Via a transformation of variables, the 9×9 matrix in Eq.(15) can always be altered into a block- diagonal form with 5×5 and 4×4 blocks, as depicted in Eq.(16) below. Out of these two matrices, only the smaller-size matrix may exhibit Lyapunov exponent with the positive real part in the practically feasible range of pump currents (see discussion to Fig.1 below). The required transformation and hence the set of variables responsible for instability are both changing with the initial phase of the mode θ. However the instability increment Re(Λ) remains fixed, attesting that the initial phase has no impact on the occurrence of instability. For simplicity, in what follows we assume that initially the laser operates in a CW lasing regime and the phase of the mode is θ=0, yielding us the following eigenproblem: 13   T 1 _2 eff 1  4  E  2 E 0 0 0 0 0                           1 eff _2 i   2 T 1 2   2 E   (   0     1    E g  )1 T _2 T _2 (   0 2 T _2 0 eff )1 eff 0 0 0 E 2 0 1 T 1 0  E 2 0 0  1 T g  0 0 0 0 0  E 2 0 0 0 0 0 0 0 2 E  1 T _2 g 0   0 0 0 0 0 0 0 0 1  T _2 eff 1   2 E  0 0 0 0 0 0  1 2  )1  eff  )1 2(  0 2 T _2 i  T _2 T _2 eff )1 (   0 2 T _2  eff g   1      (   0   0 0 0 0 0 E 2 0 0 0 0 0 0 0 0 E  1 T g   2 E E 2  1 T _2 g                               i  e   r n   0  n   r   ii    r  e  i  n   i    rr                0 (16) We find that only one eigenvalue of the lower 4×4 block may have Re(Λ)>0 and may lead to unstable CW lasing regime observed in experiment. The upper 5×5 block shows only stable solutions in a large range of pump rates above the lasing threshold. Thus for a QCL from Table 1, this behavior is observed up to p~60 times above the lasing threshold (Fig. 1). For an AlGaAs QW LD from Table 1, it continuous up to p~1100. In what follows we limit discussion to this sufficiently large range of pump rates when only one eigenvalue in Eq. (16) may lead to a multimode instability. (a) t n e n o p x e v o n u p a y L t s e g r a l e h t f o t r a p l a e R 1.0x1011 5.0x1010 0.0 -5.0x1010 -1.0x1011 -1.5x1011 -2.0x1011 -2.5x1011 p=50 1, from 4x4 matrix 2, from 5x5 matrix 1x1010 0 -1x1010 2.4x1013 0 1x1013 3x1013 2x1013  2.5x1013 4x1013 (b) t n e n o p x e v o n u p a y L t s e g r a l e h t f o t r a p l a e R 1.0x1011 5.0x1010 0.0 -5.0x1010 -1.0x1011 -1.5x1011 -2.0x1011 -2.5x1011 p=60 1, from 4x4 matrix 2, from 5x5 matrix 1.0x1010 0.0 0 1x1013 -1.0x1010 2.6x1013 2.7x1013 2x1013  3x1013 4x1013 FIG. 1. Spectra of instability increments (largest real parts of the Lyapunov exponents) of the 4×4 (red curves) and 5×5 matrix (blue curves) blocks at p=50 (panel (a)) and p=60 (panel (b)) for QCL with parameters from Table 1. 14 The instability in Eq.(16) is related to variables r, ie , in and rr . When the initial phase of the mode θ is detuned from θ=0, the variables associated with the lower 4×4 block are altered. For example, for / 2  , the instability is due to perturbations i, re , rn and ii. Therefore, it is not possible to identify the nature of instability (RNGH-like or of a different kind) from the set of variables involved. At the same time we find that whatever the initial phase θ, the offset frequency Ωmax at the maximum increment Re(Λ) is close to the Rabi oscillation frequency   Rabi ( p   0 ) / TT 1 2 _ eff (see Section III below). This allows us to attribute this instability to the RNGH-like behavior discussed in Refs.[7,8]. This behavior should be contrasted with the one reported in Ref.[16], where (i) the carrier coherence grating (9) has not been taken into account and (ii) truncated coupled mode equations for carrier population grating and medium polarization contained errors (see our discussion to Eqs. (7)-(11), Appendix A and our analytical solution obtained in Ref. [25]). If one omits the variables rr and ii associated with the carrier coherence grating, the lower 4×4 matrix block in Eq.(16) simply becomes of 3×3 size. It still exhibits an instability at small pump rate above the lasing threshold. According to Ref. [16] the instability occurs at significantly lower frequencies around  SHB   1 T 1 ( p  1 ) / T T ~ 3 1 2  Rabi / T / 1 4 3 . Therefore this instability has been attributed in [16] to the SHB effect. Below we show that this instability occurs in fact at much higher frequency close to ΩRabi. These outcomes of our numerical studies agree very well with our analytic solution reported in Ref. [25]. The upper 5×5 matrix block in (16), which shows only stable solutions in the presence of the carrier coherence grating (9) and in the reasonable range of pump rates, takes the form of a 4×4 matrix after dropping out the variables rr and ii. Our analysis shows that it may lead to RNGH instability at the pump rate approximately 10 times above the lasing threshold, in agreement with the theoretical 15 predictions from [7,8]. The second threshold always remains that high unless a saturable absorber is introduced to the model system [17], the possibility which was studied in [16]. We argue that inclusion of both grating terms (9) and (11) allows one to obtain low-threshold RNGH instability in a FP cavity laser (or bidirectional ring laser) without saturable absorber. This instability is of different origin from the one discussed in Ref.[16]. In our case it originates from the lower 4×4 matrix block in Eq.(16) and not from the upper 5×5 block. The coherence grating (9) renders the second threshold associated with the 5×5 block further more prohibitively higher (at p~60 instead of the initial value of p~9-10 in the original RNGH instability case for a unidirectional ring laser). In the next section we elucidate these points in further details. III. RESULTS AND DISCUSSION A. RNGH instability threshold For QCLs that have the cavity length of 2-4 mm, our linear stability analysis predicts the multimode RNGH instability at a pump rate of a few percent above the lasing threshold. This prediction is in agreement with numerous experimental data available in the literature [6,16,18]. Our numerical simulations based on TW rate equation model also confirm this behavior and the transition to multimode operation at just above the lasing threshold (see Section III.C). In the case of a short-cavity QCL of just 100 µm long, our numerical simulations reveal stable CW operation regime up to the pump rate of 2-2.5 times above the lasing threshold. The RNGH-like self- pulsations occur above the second threshold of pth2~2.5 (see below). However, the linear stability analysis based on Eq.(16) shows that the real part of the Lyapunov exponent Re(Λ) becomes positive at the pump rate p just slightly above 1. Moreover, with increasing p, the instability increment Re(Λ) monotonically increases without revealing any specific behavior at pth2. Therefore we examine the 16 spectral behavior of the increment Re(Λ) and we arrive to a new insight into the second threshold condition, which we introduce below. Figure 2 illustrates the spectral behavior of the increment Re(Λ) in a QCL with the cavity length of 100 µm. Only positive offset frequencies are shown, because the spectral shape of Re(Λ) is an even function of Ω. The instability increment is plotted being normalized to the cavity mode separation, which yields us the spectrum of the roundtrip gain coefficient 2Re(Λ)ngL/c for various cavity modes. In the figure, the instability gain spectra are depicted at three different pump rates of p=1.1, 1.2 and 3 (green, red and blue curves, respectively). We also indicate the locations of a few cavity modes by plotting the Airy function of the "cold cavity" (black curve, not in scale). The zero offset frequency corresponds to the initially lasing mode. main lasing mode (cavity mode n) 1,0 n+1 cavity mode n+2 cavity mode , y t i l i b a t s n i i r o f n a g p i r t d n u o R ) L g / n 2 c ( / }  { e R p=1.1 p=1.3 p=3 FP spectrum max/2>c/2ngL gain maximum 0,5 1,0 1,5 2,0 2,5 Normalized Fourier frequency, (/2c/2ngL 0,5 0,0 0,0 FIG. 2. Positive offset frequency part of the spectrum of the roundtrip gain for instability in a 100 µm long QCL at normalized pump rate of p=1.1 (green curve), p=1.3 (red curve) and p=3. The black curve indicates the Airy function of the cold cavity (not in scale). Other QCL parameters are given in Table I. The RNGH condition [7,8] for multimode instability in a CW operating single-mode laser requires that the instability increment is positive at a frequency of another cavity mode. An example that matches this definition is the instability gain curve plotted in Fig. 2 for p=1.3 (red curve). It shows a positive gain for the first two adjacent cavity modes at the offset frequencies Ω/2π=±c/2ngL. In Fig. 3, this gain coefficient is plotted as a function of the pump rate (dashed blue curve). According to the conventional 17 definition of the second threshold, the multimode RNGH instability should occur at p=1.25, when the gain for instability becomes positive at some of the non-lasing cavity modes. , y t i l i b a t s n i r o f n i a g p i r t d n u o R ) L g n 2 / c ( / )  ( e R 3,0 2,5 2,0 1,5 1,0 0,5 0,0 L=4mm L=100um, this paper L=100 um, [8] 1,5 1,0 3,0 Pump excess above threshold, p 2,0 2,5 FIG. 3. Lyapunov stability analysis: roundtrip gain for RNGH instability vs. pump excess above threshold for the QCLs with the cavity lengths of 4 mm (red curve) and 100 µm (solid blue curve). For explanations on the dashed blue curve, see discussion in the text. In long-cavity QCL, the second threshold is very low, at pth2=1.05, while pth2=2.35 in the short-cavity QCL. Other QCL parameters are given in Table I. In order to verify the second threshold condition, we perform a series of numerical simulations based on the travelling wave rate equation model [22]. As a model system we use a short-cavity single-section sample (L=100 µm). The travelling wave model incorporates Langevin force terms that seed the spontaneous polarization noise into the system (these terms are not indicated in Eqs. (1)-(3), but an example can be found in [22]). We examine several different power levels of the noise injected into the system. Fig. 4 shows that the output power of amplified spontaneous emission Psp varies over 6 orders of magnitude for the range of spontaneous polarization noise used here. ] W p [ E S A n i r e w o p t u p t u O Psp=8 nW Psp=81 pW Psp=0.8 pW Psp=8 fW 1000 10 0,1 1E-3 0 100 Time [ps] 200 18 FIG. 4. Output power variation in amplified spontaneous emission (ASE) regime below the lasing threshold (p=0.9) for different levels of spontaneous polarization noise used in the numerical simulations. QCL cavity length is 100 µm. Other parameters are given in Table I. In Fig 5, for each set of parameters (p, Psp) we perform a series of 20 simulations and calculate the probability of occurrence of the RNGH like self-pulsations (various examples of the waveforms can be found in Figs. 9 (c)-(f) below). These statistical data are displayed in Fig. 5 as vertical bars with the color corresponding to the spontaneous polarization noise power in Fig.4. It follows that there are no systematic correlations between the occurrence of RNGH instability and spontaneous noise power Psp injected into the system. In Fig 5 we also plot the average probability over all 80 realizations at different Psp (solid curve). It has the dispersion of ±0.056, which is indicated in Fig. 5 by the error bars. The data in Fig. 5 attest that there is no multimode RNGH instability at the pump rate p=1.5. This is in contradiction with the conventional definition of the second threshold, foreseeing pth2=1.25 (Fig 3, dashed blue curve). At p=2, only a few realizations have resulted in RNGH self-pulsations. The average probability of these is of 0.04, which is below the uncertainty limit. The RNGH instability develops at p between 2 and 2.5, with p=2.5 being the first point in Fig. 5 for which the probability exceeds the uncertainty range. Obviously, the results of our numerical simulations are not in agreement with the second threshold condition as originally proposed by Risken and Nummedal [7], and Graham and Haken [8]. Analyzing the data in Fig.5, we find that RNGH instability occurs when the instability gain maxima located at the offset frequencies  max Rabi are either on resonance with the nearest-neighbor cavity modes at ±c/4πngL or at a higher frequency offset as in the case indicated by the blue curve in Fig. 2. The RNGH instability gain coefficient corresponding to this refinement of the second threshold is plotted in Fig 3 as a function of the pump rate (solid blue curve). It reveals an abrupt switching-on behavior for the RNGH instability, the feature which is not seen in long-cavity QCLs because of the 19 small frequency separation between the cavity modes (Fig. 3 red curve). The numerical results in Fig 5 are in reasonable agreement with the second threshold value of pth2=2.35 in Fig 3 (solid blue curve) obtained using our modified second threshold condition. s n o i t a s u p - f l l e s H G N R f o y t i l i b a b o r P 0,8 0,6 0,4 0,2 0,0 Psp=8 nW Psp=81 pW Psp=0.8 pW Psp=8 fW average 2 Pump excess above threshold, p 4 6 FIG. 5. Probability of occurrence for RNGH self-pulsations is plotted as a function of the pump excess above lasing threshold p for different levels of spontaneous polarization noise (bars) as well as for the average over all realizations (curve). QCL cavity length is 100 µm. Other parameters are given in Table 1. See further details in the text. The proposed refinement for the second threshold has a clear physical meaning. Recall that RNGH instability in a CW single-mode laser arises due to Rabi splitting of the lasing transition induced by the lasing mode. As a result of such spectral broadening and reshaping of the gain curve, the laser can provide sufficient optical gain for other longitudinal modes [26]. Most importantly, the medium polarization does not follow adiabatically the optical field in the cavity. Therefore in our refinement of the second threshold, we have applied the pulse area theorem [9,27], assuming that the initial perturbation to the lasing mode has a pulse shape. In the most general case, the characteristic time of such perturbation pulse is roughly a half of the cavity roundtrip time τp~Lng/c. From the pulse area theorem, we have noticed that a perturbation in the form of an optical pulse is unstable and the pulse area grows so as the fast (non-adiabatic) medium polarization dynamics becomes possible if, initially, ΩRabi·τp>π. Because the maximum gain for RNGH instability is located at the offset frequency    , we conclude that the multimode RNGH instability occurs only when max Rabi 20  max / 2   c / 2 Ln c g . (17) The blue curve in Fig.2 satisfies this condition while the red one does not. This condition has minor impact on QCLs with long cavities. However it raises the second threshold considerably in short-cavity devices because of the large frequency separation between the cavity modes. In Ref.[25] we have obtained an analytic expression for the pump rate at second threshold. B. Experimental observations of RNGH instability in QCLs In Section II.A we have refined the necessary conditions for occurrence of multimode RNGH instability. The numerical simulations reveal regimes with either regular or chaotic self-pulsations (sections III.C and III.D below). Therefore we should clarify which features provide an evidence for RNGH instability. The experimental data available in the literature attest that RNGH instability does not always lead to observation of a clear Rabi splitting in the optical mode spectrum of a QCL. In some experimental realizations, RNGH instability causes just a broadening of the lasing spectrum to the offset frequencies of the order of the Rabi oscillation frequency. This is not surprising because the notion of the Rabi splitting in the gain spectrum of a laser is introduced considering small-amplitude perturbations to the initial optical field in the cavity. This picture is very different from the situation in the laser undergoing large-amplitude self-pulsations. The appearance of Rabi splitting in the optical spectrum can be tailored by changing the sample temperature [6,16,18]. For example, in [16] at low temperatures (80-150 K) the measured lasing spectra in a buried heterostructure QCL sample are just simply broadened, but with increasing temperature, the Rabi splitting between mode clusters becomes more apparent. At room temperature conditions, two distinct mode clusters emerge in the optical spectrum. The nature of the temperature effect has not been entirely understood. The temperature-dependent saturable absorption is suggested as one of the possible 21 mechanisms. However in the present paper we argue that low-threshold RNGH instability occurs in QCLs even without saturable absorption, due to a combined effect of coherence grating and carrier population grating. The temperature dependent carrier diffusion leading to the contrast change of SHB grating was proposed as another possible temperature mechanism. However, for all considered temperatures, the emission spectra have exactly the same width, of ~40 cm-1 for the pump rate of p~1.7-1.8 times above threshold and of ~60 cm-1 for p~3.5-3.6. Both values are comparable to the Rabi oscillation frequency ΩRabi. The dynamic behavior responsible for such spectral broadening is governed by very fast processes. It cannot be attributed solely to the SHB effect and mode coupling via induced grating of population inversion, whose behavior is defined by slow characteristic time T1. Therefore we conclude that the large spectral broadening on the order of ΩRabi shall be attributed to multimode RNGH instability, while the optical spectrum is not necessarily split in two distinct mode clusters separated by 2ΩRabi. In addition, we notice that an inclusion of the coherence grating, whose relaxation time is on the scale of T2 and subject to the effect of diffusion, may then provide a required mechanism to explain the thermal sensitivity of multimode RNGH instability. Thus in [18], QCLs show a different spectral behavior with the temperature. One sample reveals two distinct lasing mode clusters with the separation of ~40 cm-1 at liquid He temperature (6 K). However at an increased temperature of 77 K, no clear Rabi splitting is observed for all reported samples. At the same time, it was noticed that the spectral broadening and the appearance of the two distinct mode clusters is a function of the sample length. Unfortunately no detailed studies about the effect of the sample length were reported. In the cited works, the complex structure of the lasing transition in QCL is usually not taken into account. The optical gain spectrum is usually assumed to be broad and homogeneous. At the same time 22 inhomogeneous features in the gain curve structure or dispersive characteristics of the cavity (e.g due to backscattering from microcracks [28]) may significantly reshape the spectrum of RNGH instabilities and impact the second threshold. Finally, even if the two distinct mode clusters are observed in the lasing spectrum, their frequency spitting is not necessary equal to 2ΩRabi. The pump rate dependence of the Rabi frequency [see Eq.(17) below] assumes that this splitting is proportional to the square root of the output power. In [6,16], this behavior is confirmed for the range of QCL output power up to 36 mW and the frequency splitting up to 1 THz is observed in 3 µm-wide buried heterostructure lasers. However for 10-15 µm wide ridge waveguide lasers from Refs. [16] and [18], the spectral splitting is clamped at about 0.8 THz frequency for the output power of ~25 mW and higher. Thereby we see that at high pump rates, the overall spectral broadening due to RNGH instability can be smaller than 2ΩRabi. C. RNGH instability in a QCL with long cavity We perform numerical simulations with the TW rate equation model [22], utilizing a QCL with the cavity length of 4 mm as a model system for long-cavity devices. The linear stability analysis indicates the second threshold of pth2=1.05 (Fig.3, red curve). For the pump rate p=1.2, which is slightly above the instability threshold, we do observe a quasi-periodic chaotic behavior in the output power waveform [Fig. 6(a)]. Here the cavity roundtrip time is 88 ps, much larger than the carrier lifetime T1 (Table 1). In Fig. 6(a) and (b), the time domain extends over 7 cavity roundtrips. The laser is initially not pumped and the pump current is switched on at a time t=0 s. The onset of the lasing emission is seen in less than 3 roundtrips in the cavity (at t~200 ps). After ~6 roundtrips, the system reaches a steady regime of quasi- periodic (chaotic) self-pulsations. The inset in Fig. 6(a) shows a zoom at one example of such self- pulsations after 6 roundtrips in the cavity. Figures 6(c) and 6(d) show another realization obtained with 23 the same model parameters but with the simulation domain extended to 100 cavity roundtrips. Comparing the self-pulsations in Fig.6(d) at the cavity roundtrips from 90 to 100 with those in Fig.6(c), one can see that the QCL reaches the steady regime of chaotic self-pulsations just after 6 cavity roundtrips (~530 ps). This time appears too short for establishing fixed phase relationships between many individual modes as that would be in the case of active mode-locking operation [29,30]. The last one does require thousands of cavity roundtrips. However the model laser sample we consider here does not have a built in electroabsorber section in order to produce active mode-locking operation as in [29,30], while the waveforms in Fig.6 clearly attest that this is not the mode-locking regime. Therefore the considerations about thousands roundtrips required to reach the steady operation do not apply for our single-section Fabry-Perot cavity laser. In the original work of Risken and Nummedal [7], the steady regime of self-pulsations in a unidirectional ring laser is reached in less than 160 cavity roundtrips. The carrier population grating and coherence grating in our Fabry-Perot laser not only drastically reduce the second threshold but also reduce the time needed for settlement of the self-pulsations. It is didactic to compare the build-up time of the lasing emission in Fig.6 with an estimate based on the well-known text book expression that reads τc·ln(PLAS/PSE)/(p-1) (see e.g. [31]). For the case considered in Fig.6, the photon lifetime in the cavity is τc=4 ps while the power ratio in the steady lasing regime and in the initial regime of spontaneous emission (at t=0) is of PLAS/PSE=106 [see Fig.6(e)]. For the pump excess above threshold of p=1.2, the estimated build-up time is of 280 ps, which corresponds to 3 roundtrips in the cavity. This estimate agrees very well with the rise time of the waveforms in Fig.6. Once the lasing is reached, the roundtrip gain coefficient for the multimode instability at p=1.2 becomes of 0.38 per cavity roundtrip (see also [25]). That is, the amplitude of an unstable non lasing mode increases by a factor of exp{0.38N} after N roundtrips. This process has the characteristic time 24 scale of 1/0.38= 2.6 cavity roundtrip. Therefore reaching the steady regime of multimode self-pulsations just in additional 2-3 roundtrips after the onset of lasing emission (by t~500 ps in Fig.6) appears very reasonable. In Fig. 6(b) we plot the system attractor, which traces the evolution of the medium polarization P versus carrier density N. Both P and N values are taken in the vicinity of the laser output facet and are normalized at transparency carrier density. The P-N attractor has a characteristic butterfly shape typical for a chaotic behavior in a Lorenz-type system [32]. We observe similar chaotic attractors with increasing pump power. ] W m [ r e w o p t u p t u O 20 10 0 0 (a) (c) ) W m ( r e w o p t t u p u O 18 16 14 12 10 8 6 4 2 0 520 560 600 100 200 300 400 500 600 Time [ps] (b) n o i t a z i r a o p d e z i l l a m r o N All 7 round trips Last 2 round trips 4 5 Normalized carrier density 1,0 0,5 0,0 -0,5 -1,0 3 18 16 14 12 10 8 6 4 2 ) W m ( r e w o p t t u p u O 0 88 176 264 352 440 528 616 704 792 880 Time (ps) (d) 0 7920 8008 8096 8184 8272 8360 8448 8536 8624 8712 8800 Time (ps) 25 ) W m ( r e w o p t t u p u O 10 1 0,1 0,01 1E-3 1E-4 1E-5 1E-6 0 88 176 264 352 440 528 616 704 792 880 Time (ps) (e) FIG. 6. Results of numerical simulations with TW model for QCL with the cavity length of 4 mm (cavity roundtrip time is 88 ps): (a) Plot of the output power as a function of time. The inset spans over last two roundtrips in the cavity. (b) Chaotic P-N attractor. The behavior during all seven cavity roundtrips is depicted with the blue curve. The red curve signifies the last two roundtrips, when the system reaches the steady regime of quasi-periodic (chaotic) self-pulsations. Panels (c) and (d) show another realization obtained with the same model parameters but with the simulation domain extended over 100 cavity roundtrips, depicting, respectively, the first ten and the last ten roundtrips in the cavity. The steady regime of quasi-periodic self-pulsations as seen after 90 roundtrips in (d) is actually reached after ~6 cavity roundtrips (~530 ps) as seen in (c). The panel (e) shows the details of the fast build-up of the lasing regime on a logarithmic scale. The QCL is pumped at p=1.2 times above the lasing threshold. Other parameters are shown in Table I. With increasing pump rate, the optical field waveform develops into a quasi-periodic square wave. An example of the temporal behavior of the optical field amplitude and medium polarization for a wave traveling in the positive z axis direction as well as the carrier density evolution is depicted in Fig. 7. These waveforms are obtained for the pump rate of p=2.2. The optical field behavior [Fig. 7(a)] appears to be quite similar to the one predicted in Ref [23] for a long-cavity unidirectional ring laser at high pump rates. However in our case, the average field amplitude is zero due to quenching of the initially CW lasing mode and emergence of the two symmetric sidebands in the optical spectrum. Like the optical field, the medium polarization shows the square-wave behavior. However its waveform pattern is not identical to that one of the optical field [Fig. 7(b)]. That is not surprising if one takes into account the scattering of counter-propagating waves on the induced gratings of carrier coherence and population. Most importantly, the medium polarization does not follow the optical field adiabatically. Instead, the medium polarization itself defines the optical field dynamics. One can see in Figs. 7(a) and 7(b) that the pattern of the square wave almost repeats itself after each roundtrip in the 26 cavity, indicating that a wave packet with complex envelope travels back and forth in the cavity. Its envelope just slightly changes at each roundtrip, yielding the quasi-periodic chaotic behavior. A different pattern is seen in the waveforms of the optical power and carrier population [Figs. 7(c) and 7(d) respectively]. These variables are either quadratic with respect to the optical field or have a contribution from the product of the optical field amplitude and the medium polarization [see Eqs.(7)- (11)]. Each waveform in Figs. 7(c) and 7(d) exhibits spikes that are superimposed on a steady level. The spikes are caused by intermittent behavior of the optical field and polarization, while the steady-state component attests for the symmetry of the field and polarization square wave patterns. The period of spikes is roughly a half of that for polarization waveform, yielding the Lorenz-type attractor in Fig. 6(b). All these features seen in the time-domain waveforms are directly imprinted into the optical and radio frequency (RF) spectra. In Fig. 8 we show evolution of these spectra with the pump rate and compare these with the Rabi oscillation frequency. Tper= 88 ps = Tcav 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 . u . a e d u t i l p m a d e F l i 3 2 1 0 -1 -2 n o i t l a z i r a o p d e z i l a m r o N -0.4 220 240 260 280 300 320 340 Time (ps) 360 380 400 420 440 (a) (b) -3 220 240 260 280 300 320 340 Time (ps) 360 380 400 420 440 27 0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00 220 ) W ( r e w o p t u p t u O (c) y t i s n e d r e i r r a c d e z i l a m r o N 5,5 5,0 4,5 4,0 3,5 3,0 240 260 280 300 320 220 240 260 280 300 320 Time (ps) (d) Time (ps) FIG. 7. Results of numerical simulations with TW model for QCL with the cavity length of 4 mm: The optical field waveform for the wave propagating in the positive z axis direction (a) and normalized medium polarization associated with this wave (b), output power (c) and normalized carrier density (d). All values are taken nearby the output facet of the laser cavity. The QCL model parameters are the same as in Fig. 6 but the pump rate is of p=2.2. Due to the standing wave pattern of the optical field in the cavity and induced carrier coherence and carrier population gratings, there are several Rabi oscillation modes in the cavity. In order to extract their frequencies, we have made an additional small-signal analysis of Eqs.(7)-(11). This time, we keep fixed the amplitude of the optical standing wave in the cavity. We find that Rabi oscillations are possible at the main frequency Rabi as well as at the two other frequencies (1+2-1/2)1/2ΩRabi and (1-2-1/2)1/2ΩRabi , where:  Rabi  p  0 T T 1 2 _ ( p ) eff (18) Out of these three eigen modes, only the main one (18) is associated with the gain medium variables in the lower 4×4 matrix block of Eq. (16), the one which is responsible for RNGH instability. In Fig. 8, we also compare the extracted spectra with the evolution of the peak gain frequency Ωmax for multimode RNGH instability (see Fig. 2) as well as with the possible frequency of multimode instability due SHB effect, which is predicted in [16] to be well below Rabi and Ωmax :   [16] SHB 1 T 1 1 p  3 TT 1 2 . 28 (19) All these frequencies (ΩRabi, Ωmax, and [16] SHB ) are obtained using a small-signal approach. Because the optical power is 2E while the carrier dynamics is governed by terms EP , in Figs. 8(a) and 8(b) we plot these frequencies scaled by a factor of 2. Note that the peak gain frequency Ωmax is quite close to the Rabi oscillation frequency (compare the red and blue curves in Fig. 8). The evolution of the RF power spectrum in Fig. 8(a) almost follows along the curves for 2 Rabi and 2Ωmax frequencies, exhibiting the spectral broadening and the frequency shift of the modulation band that increases with the pump rate. Note however that it is unrealistically to expect the exact matching between the numerical simulations for large-amplitude self-pulsations and the outcomes of the small-signal analysis. At the same time the spectral broadening and the frequency shift in Fig. 8(a) are clearly much larger than for a possible multimode instability at the frequency [16] SHB due to SHB effect [16] (green curve). Therefore we attribute the spectral behavior in Fig. 8(a) to RNGH-like instability. Since the output power waveform as in Fig. 7(c) can be regarded as a series of ultrafast spikes superimposed on a steady intensity level, the modulation band seen in the RF power spectrum at the frequency 2ΩRabi is primarily due to the spiking behavior of the output power. In Fig. 8(b) we plot the frequency-domain representation of the carrier density dynamics at different values of p. The spectra show broadening and modulation of the carrier density at frequencies up to twice the Rabi oscillation frequency. This behavior is similar to the evolution of RF spectra in Fig. 8(a). However the modulation band nearby 2ΩRabi is less pronounced and the spectra exhibit almost all frequency components down to Ω=0. This behavior can be attributed to a slow response of the carrier density and carrier population grating to rapid variations EP in Eqs.(10)-(11). The cutoff frequencies are of 1/ 2 T  1 122 GHz and 1/ 2 gT  172 GHz for the average density and population grating, respectively. 29 The medium polarization waveforms in the frequency domain representation and the optical spectra are displayed in Figs. 8(c) and 8(d), respectively. As expected, the two sets of spectra show very good resemblance. In both sets, the initially lasing mode at Ω=0 (the carrier wave) is quenched and the spectra reveal two symmetric sidebands. Since the RF modulation spectra of the output power exhibit modulation bands at the frequency 2ΩRabi [see Fig. 8(a)] one would expect to observe the modulation sidebands at frequencies ±ΩRabi in the optical spectra as well. However these considerations do not take into account large phase variations due to intermittent behavior of the optical field, as seen in Fig. 7 (a). Obviously, the phase of the optical field does not contribute to the RF power spectrum. At the same time, the large phase hops of ±π impact directly the overall optical spectrum, warping its envelope as compare to the envelope of the RF power spectrum. From the waveform in Fig. 7 (a) one can see that the average repetition frequency of such phase hops is much lower than the spectral band of each spike in Fig. 7(c). As a result, the modulation sidebands in the optical spectrum are shifted to lower frequency. Therefore one of the possible origin for experimentally reported clamping of the optical spectrum bandwidth (see Sec. III.B) is that the optical field waveform develops into a quasi-periodic square wave, showing the intermittent behavior. 30 RF power spectrum 2 x Peak gain frequency max 2 x Rabi frequency 2 x SHB frequency from Gordon et all 0 20 40 60 80 100 Normalized Fourier frequency Spectra of Polarization 4 3 2 1 120 3 2 4,0 3,5 3,0 2,5 2,0 1,5 1,0 (a) l d o h s e r h t e v o b a s s e c x e p m u P (c) 3,5 3,0 2,5 2,0 1,5 l d o h s e r h t e v o b a s s e c x e p m u P 1,0 -50 -40 -30 -20 -10 0 10 20 1 30 40 50 Normalized Fourier frequency (b) 4,0 3,5 3,0 2,5 2,0 1,5 1,0 0 l d o h s e r h t e v o b a s s e c x e p m u P Carrier density 2 x Peak gain frequency max 2 x Rabi frequency 20 40 60 80 100 Normalized Fourier frequency Optical spectra (d) 3,5 3,0 2,5 2,0 1,5 l d o h s e r h t e v o b a s s e c x e p m u P 4 3 2 1 120 3 2 1,0 -50 -40 -30 -20 -10 0 10 20 Normalized Fourier frequency 1 30 40 50 FIG. 8. Results of numerical simulations with TW model for QCL with the cavity length of 4 mm: Evolution of QCL (a) RF power spectrum, (b) carrier density, (c) polarization and (d) optical spectrum when pumped above the lasing threshold. Red and blue curves indicate expected location of modulation sidebands originating from ΩRabi and Ωmax. The green curve SHB . The frequency is normalized to longitudinal mode spacing of 11.4 GHz. The model parameters are shows that one for the same as in Figs. 6 and 7. [16] 31 D. RNGH instability in a QCL with short cavity We now move on to the case of short-cavity QCLs and consider an example of QCL with the cavity length L=100 µm, for which the linear stability analysis predicts the second threshold of pth2= 2.35 (see Fig.3, solid blue curve). The results of TW model simulations for the pump rate of p=1.5, which is below the second threshold, are shown in Figs.9 (c) and (d). After several roundtrips in the cavity, the laser reveals an onset of damped relaxation oscillations and a transit to CW lasing regime, as can be seen from both output power waveform [Fig. 9(a)] and spiral-shape P-N attractor [Fig. 9(b)]. The model simulations are thus in agreement with the predictions of the linear stability analysis in Fig. 3. The QCL reveals a different behavior when pumped above the second threshold. At p=2.5, the first emission burst of high peak power is followed by steady regime of regular self-pulsations [Fig. 9(c)]. The FWHM pulsewidth of the first emission burst is of 0.5ps while for regular self-pulsations it is of 0.6 ps width. Note that these self-pulsations are much faster as compare to the relaxation oscillations at below second threshold in Fig. 9(a). The system attractor plotted in the P-N plane indicates that the laser transits to regular self-pulsations just after a few roundtrips in the cavity [Fig. 9(d)] once it is at lasing. Indeed in Fig.9, the gain coefficient for multimode instability is of 0.5 per roundtrip, yielding the characteristic time scale of just two cavity roundtrips. The last portion of the time domain used in the numerical simulation is depicted in Figs. 9(e) and (f). Note that the period of regular self-pulsations is close to the cavity roundtrip time of 2.2 ps while the optical field amplitude (and the medium polarization) changes sign at each half period (This behavior attests that the considered regime is not a usual mode-locking). This can be seen from the 8-like shape of P-N attractor in Fig. 9(f). Figure 10 shows the optical and RF power spectra when the pump rate p is in the range of 2.2 to 3.9 times above the lasing threshold. In case of short-cavity QCL operating at just above the second threshold, only two main frequency components are excited in the optical spectrum [Fig. 10(a)]. With 32 increasing pump rate, the spectrum remains symmetric with respect to the initially lasing mode and the two main spectral components are located at the two nearest-neighbor modes. Respectively, the RF power spectrum shows the main harmonic at about twice of the cold cavity mode spacing [Fig. 10(b)]. As discussed in Sec.III A, the multimode RNGH instability occurs when the frequency Ωmax of the maximum gain for instability is on resonance with the first adjacent cavity mode [Fig. 10(a), blue curve]. However, with increasing pump rate, the frequency of self-pulsations does not follow the increasing frequency Ωmax or Rabi oscillation frequency (red curve). We attribute this behavior to the cavity roundtrip self-repetition condition (see Sec. II.C). Instead of the continuous frequency rise, higher order modes start to appear in the optical spectrum. Note that the frequency of self-pulsations is slightly lower than the cold cavity mode spacing. The group velocity reduction can be attributed to propagation phenomenon of a high-energy pulse in resonant medium, the process which can be regarded as continual absorption of energy from the pulse leading edge and re-emission of energy into the pulse trailing edge [27]. A quite similar behavior is reported in [23] for a short-cavity unidirectional ring laser. However in case of the ring laser, the optical field does not change the sign at each half-period and, as a consequence, the main lasing mode at Ω=0 is not quenched. One can trace a few other similarities between multimode instability in QCLs discussed here and the RNGH instability in unidirectional ring lasers from Ref. [23]. In particular, one very important question has been challenged but has not been answered in [23]. More specifically, the mechanism, which is responsible for appearance of regular self- pulsations in case of short-cavity laser instead of chaotic pulsations in the long-cavity case, have not been elucidated. This qualitative change in dynamic behavior becomes even more wondering if one consider the affinity between the 8-shape attractor in case of regular self-pulsations in Fig. 9(f) and the butterfly-shape attractor in case of chaotic self-pulsations in Fig. 6(b). 33 (a) 30 20 10 ) W m ( r e w o p t u p t u O (c) ) W m ( r e w o p t u p t u O 140 120 100 80 60 40 20 0 -20 0 (b) 0 5 10 15 20 25 30 Normalized carrier density (a.u.) 40 50 20 10 Normalized carrier density 30 0 0 20 40 60 Time (ps) 80 100 . ) . u a ( n o i t a z i r a o p d e z i l a m r o N l (d) n o i t a z i r a o p d e z i l l a m r o N 6 5 4 3 2 1 0 -1 15 10 5 0 -5 -10 20 40 Time (ps) 60 80 0 34 (e) ) W m ( r e w o p t u p t u O 80 60 40 20 0 (f) n o i t a z i r a o p d e z i l l a m r o N 10 5 0 -5 -10 100 102 104 106 108 110 14 Time (ps) 18 16 Normalized carrier density 20 22 24 FIG. 9. Results of numerical simulations with TW model for QCL with the cavity length of 100 μm: Waveform (a) and P- N attractor (b) in case of p=1.5 (below 2nd threshold), the system is stable. Waveform (c) and P-N attractor (d) for p=2.5, zoom- in at the end of simulation domain which shows regular self-pulsations with sub-picosecond pulse width (e) and P-N attractor (f). Below we propose one possible explanation on the effect of the cavity length. To get an insight into its cause we study the P-N attractors in further details. The P and N variables used here are introduced following the approach of Ref. [22]. In particular, the variable P measures the order parameter in the system (the carrier density in the coherent state) and has the same units as the carrier density N. In Figs. 6 and 9, both variables P and N are normalized on the transparency carrier density. In Fig. 9 the order parameter P is high. It reaches about a half of the value for N that precedes the emission pulse when P=0. This feature is observed as in the first emission burst [Fig. 9(d)] as well as in the regime of regular RNGH self-pulsations [Fig. 9(f)]. This behavior indicates an "off-diagonal long range order" in the system, as predicted in the pioneering paper by Graham and Haken [8]. Note that a similar behavior is also predicted for Dicke superradiance (SR) [33,22]. The similarity is even far more striking. Thus as shown in [8], the master equation for RNGH instability behavior can be put in the form that bears a closed formal analogy to the description of condensation phenomena, such as 35 superconductivity. But in [22] (and also in [34]), the SR emission is also shown to be governed by the master equation in the form of Ginzburg-Landau equation. Continuing these parallels we note that in the pioneering work of Risken and Nummedal [7], the approximate analytic solution for the regime of RNGH self-pulsations was obtained in the form of a hyperbolic secant pulse. This was achieved by considering the case, in which the optical filed is entirely defined by the medium polarization, so as PE  . The same relationship between E and P as well as the same hyperbolic secant pulse shape applies to the case of Dicke superradiance [22,35]. There is however one important difference between the two regimes. The RNGH instability is usually analyzed in a CW operating laser when the active medium is under continuous wave pumping. The SR pulse emission occurs from a strongly pumped active medium, when the initial optical field is close to zero. In practice this is achieved with pumping by short and intense optical or electrical pulses [33,35,36,37,38,39]. Taking all these considerations into account, we attribute the first emission burst in Figs. 9(c) and (d) to SR. It was shown that SR in short samples is different from the cooperative emission in long samples [22,34,40]. The borderline between the two cases is set by the coherence length of the SR pulse. In case of long samples, the situation is such that sample domains of the size of the coherence length emit independently of each other. The output pulse is the result of incoherent superposition of SR emission from different sample domains. As a result, the overall pulse width broadens and its amplitude decreases. In a similar way, we may conjecture that the regular RNGH self-pulsations occur when the sample is shorter than the coherence length. Otherwise, in the case of a long sample, different sample domains are not mutually coherent with respect to instable cavity modes and compete with each other at the initial stage. As a result, an erratic pattern of the field is established in the cavity at a later stage. As seen in Fig. 7, the optical field pattern almost repeats itself on subsequent roundtrips in the cavity, however the 36 coherence length remains smaller than the cavity length. It is roughly the average time between the phase hops in Fig. 7(b). A more detailed investigation about the cavity length effect on the coherence length of RNGH self- pulsations is left for another study. (a) Optical power spectra l d o h s e r h t g n s a i l e v o b a s s e c x e p m u p 4,4 4,2 4,0 3,8 3,6 3,4 3,2 3,0 2,8 2,6 2,4 2,2 -2,5 Rabi frequency Peak gain frequency -2,0 -1,5 -1,0 -0,5 0,0 0,5 1,0 1,5 2,0 Normalized frequency (b) l d o h s e r h t e v o b a s s e c x e p m u p 4,4 4,2 4,0 3,8 3,6 3,4 3,2 3,0 2,8 2,6 2,4 2,2 4,4 4,2 4,0 3,8 3,6 3,4 3,2 3,0 2,8 2,6 2,4 2,2 2,5 RF power spectra 5 6 0 1 3 2 Normalized frequency 4 FIG. 10. Results of numerical simulations with TW model for QCL with the cavity length of 100 μm: Optical (a) and RF power (b) spectra. E. The role of carrier diffusion We have shown in Sec.II.C that the carrier coherence grating and carrier population grating are both responsible for lowering the multimode instability threshold. The diffusion coefficient D, which affects their effective relaxation times Tg, T2_g and T2_eff, is sensitive to the temperature. Unfortunately the literature data about the temperature effect on RNGH self-pulsations are quite controversial (see Sec. III.B), which is likely because the temperature also affects many other QCL parameters. On the other hand, the relaxation rates enhancement due to the carrier diffusion is proportional to the square of the 37 p photon wave enumber k [s see the discu ussion to Eqs s.(7)-(11)]. T Therefore w we elucidate t the role of th he carrier d diffusion by considering various emi ission wavel lengths. First we c consider the case of QC CLs emitting g in the MI IR spectral range and d discuss the effect of e electron diffu fusion in the plane of act tive QWs in these device es. In Fig. 1 1 we compa are the increm ments for m multimode R RNGH instab bility in QC CLs samples emitting at 4 µm (gree en curve) an nd 10 µm (re ed curve) w wavelengths . Without ca arrier diffusi ion, the seco ond threshold d is unrealis stically low, at below p= =1.005 in l long-cavity samples (b lack dash-d dotted curve e). The rela axation due e to diffusio on rises the e second t threshold. As s expected, t this effect is more prono ounced for sh horter wavel length QCLs s, where rela axation is f faster (comp pare red and d green curv ves). Thus fo or devices w with the cav vity length o of 4 mm, th he second t threshold pth2 2 is of 1.04 a at the wavele ength of 10 µ µm, while pt th2=1.32 in Q QCL emittin g at 4 µm. T The effect o of the diffusi ion on the se econd thresh hold is seen a as in the long g cavity dev vices (solid c curves) as in the short c cavity QCLs s (vertical da ashed lines). In all consid dered cases, the carrier d diffusion doe es not rise th he second t threshold to p prohibitively y high levels s. FIG. 11. Ma c carrier diffusion 4 4 µm (green cu d dashed lines de aximum RNGH n (black dash- urve) or 10 µm enote the secon H instability in dotted curve) a m (red curve). T nd threshold for ncrement vs p as well as with The solid curve r short-cavity d pump excess a h the effect of t es are obtained devices (L=100 above threshold the carrier diff d for long-cavit 0 µm). Other p d for QCLs w fusion when th ty devices (L=4 parameters are without relaxat e emission wa 4 mm), while t indicated in Ta tion due to velength is the vertical able I. The implic cation of the carrier diffu usion effect i is totally dif fferent in the e case QW la aser diodes o operating i n the VIS o or NIR spect tral range. E Even though the ambipo 38 lar diffusion n in the QW Ws of these d devices is much weaker than the diffusion of electrons in QCLs (see Table I), the relaxation due to diffusion is more perceptible because of the shorter wavelength and smaller spatial period of the induced gratings. It is commonly acknowledged that QW laser diodes with monolithic FP cavities do not show RNGH instabilities in the range of pump currents that can be reached in practice. In Fig. 12, we study the increment for multimode RNGH instability in a FP-cavity GaN semiconductor laser diode operating at 420 nm and in a GaAs LD at 850 nm wavelength. In all cases examined in Fig. 12, the two LDs show quite similar behavior. In Figs. 12 (a) and (b) we consider, respectively, the eigen solutions of the 4×4 and 5×5 matrix blocks in Eq.(16) and plot the largest real part of the Lyapunov exponents. The first matrix is related to the instability caused by the carrier population and carrier coherence gratings, which is the focus of this paper. The second matrix would reproduce the results of the original RNGH theory if both induced gratings are removed ( , g TT _2 g 0 ) while the carrier relaxation and carrier dephasing processes preserve their initial times scale (T1 and T2_eff). Recall that the RNGH theory [7,8] was originally established for a unidirectional ring laser, where the gratings cannot be formed. It predicts the second threshold pth2 at more than 9 times above the lasing threshold. The second threshold of the original RNGH theory would be recovered by our model if, in particular, T _2 / g T _2 eff 0 (see section III.F below). However, neither with carrier diffusion ( T _2 / g T _2 eff 5.0~ see Table 1) nor without it ( T g T _2  _2 eff ), the VIS-NIR LDs cannot reach this condition. As a result, the second threshold emerging from the 5×5 matrix block in Eq.(16) is very high in the VIS-NIR LDs (Fig. 12(b)) due to the presence of induced gratings. The carrier coherence and carrier population gratings have a different implication on the instability emerging from the 4×4 matrix block [Fig.12(a)]. Thus in a FP cavity lasers as we consider here the induced coherence and population gratings lower significantly the second threshold. However without the ambipolar diffusion of carriers, the model predictions for the second threshold in VIS-NIR LDs 39 w would be unr realistically low, of pth2 ≈1 (dash-do otted curves in Fig. 12). The relaxati ion due to th he carrier d diffusion ren nders our mo odel to be m more realisti c (solid curv ves in Fig. 12(a)) becau use the char racteristic t time constan nts T1=1 ns and T2=0.1 ps are now w reduced to o much shor rter effective e relaxation times of T Tg=0.1-0.2 p ps for the car rrier populat ion grating, T2_g~0.05ps s for the coh herence gratin ng and T2_ef ff ~0.09ps f for the effec ctive dephasi ing rate. Th his lifetime r reduction str rongly suppr resses all ef ffects induce ed by the s standing wav ve pattern of f the optical l field in the e cavity. As a consequen nce, the sec ond thresho ld values p predicted fro om the linea ar stability a analysis are o of several h hundred time es above the e lasing thre eshold, as i ndicated in Fig. 12(a). R Recall that th his analysis is based on the truncate ed set of cou upled-mode e equations ( (7)-(11) and the model p predictions f for the secon nd threshold d cannot be quantitative ly accurate at such a h high pump r ate. Neverth heless these predictions are in a reas sonable agre eement with the outcom mes of our n numerical sim mulations ba ased on the T TW rate equ uation model l, which we d discuss next (a) -3x1010 0 100 (b) FIG. 12. (a) L Lyapunov expo G GaAs LDs (blu c calculated from o other parameter Maximum RN onent from the ue curves), cal m (a) 4x4 and ( rs are shown in NGH instability e 5×5 matrix bl lculated with (b) 5x5 matrix n Table I. y increment fro lock vs pump and without c x block. Inset in om the 4×4 ma excess above t carrier diffusio n (b) shows zo atrix block in the lasing thre on (solid curve oom at p from Eq (16) and (b eshold for GaN es and dash-do 1 to 100. The c b) largest real N LDs (black c oted curves re cavity length i part of the curves) and spectively) s L=4 mm, In Fig. 13 we show the e output wav veform and P-N attracto or simulated numerically y with TW m model for a a 100 µm lo ong single-s ection GaN LD pumpe ed above the e second thr reshold, at p p=400. At s uch high 40 . 100 6x1010 0 1x108 3x1010 0 0 -1x108 1 0 0 10 ] z H [ t n e m e r c n i H G N R  = 850 nm  = 420 nm  = 850 nm  = 420 nm m with diff. m with diff. m w.o. diff. m w.o. diff. 102 101 Pump ex 103 3 xcess above e threshold 104 105 p pump rate, L LD exhibits b behavior that t is similar t o QCLs bias sed at above e the RNGH instability th hreshold. T The output w waveform at each cavity facet reveal s the SR em mission burst followed by y regular RN NGH self- p pulsations [c compare to F Fig. 9(c)]. N Note that dur ring the SR a and approach hes a half of f the initial c carrier densit ty stored in emission bu urst, the ord er parameter r is large the system. At the pump p rates below w second t threshold, LD D shows an n ordinary be ehavior of a a Class-B la aser, which i is characteri ized by exci itation of d damped relax xation oscill lations follow wed by a tran nsition to CW W lasing reg gime [as in F Fig. 9(a)]. (a) (b) FIG. 13. Res a and P-N attract sults of numeri tor (b). The red ical simulation d (blue) trace c ns with TW mo corresponds to odel for GaN L the wave prop LD with the ca pagating in the avity length of forward (back f 100 μm: Wa kward) direction aveform (a) n. We thus c onclude tha t VIS-NIR r range QW L LDs have ve ery high sec cond thresho old because of rapid r relaxation of f the carrier coherence a and populatio on gratings. Whereas th he RNGH in nstability thre eshold in s single-sectio n QW LD s is not re eachable un nder realistic c experimen ntal settings s, there are e several c consideration ns that indire ectly suppor rt our conclu usion. Reaching the SR em mission in s short-length mesa-etche ed structure s comprisin ng GaAs Q QWs was a attempted in n [41] unde er short-puls se optical p pumping. H However no evidence o of SR emiss sion was o observed. W We may attrib bute this to t the fact that t threshold c condition for r RNGH ins stability has not been r reached in th hese experim ments as the p pump rate w was too low. According to Fig. 12, lowering of f RNGH inst tability thres shold (and r reaching SR R emission) i in single- s section LD c can be achiev ved by reduc cing the cont tribution of d diffusion to the relaxatio on rates of th he carrier c coherence an nd populatio on gratings. This bring us to the i insight that 41 low-dimens sional semic conductor heterostructures such as quantum dots (QDs) or quantum dashes (QDash) can be used as active gain material in order to avoid prohibitively high second threshold. Surprisingly, in Ref. [42], starting from totally different considerations, a similar conclusion was made about semiconductor heterostructures suitable for SR emission. Thus according to [42], SR emission is not possible with the active gain medium utilizing bulk semiconductor material or QW heterostructures. One needs to use QDs or to introduce an additional quantization degree in a QW by applying a strong magnetic field. The SR emission from magneto-excitons in InGaAs/GaAs QW was confirmed in Ref.[35] under short-pulse optical pumping. There is another way to reduce the second threshold in a semiconductor laser. It consists in incorporating a saturable absorber [17] (see also Introduction and Section II.C). Technically this is achieved by implementing several separately contacted sections in the monolithic cavity of LD. The cavity sections which are positively biased provide the optical gain while a negatively biased section behaves as a saturable electroabsorber. Under moderate negative bias applied to the absorber section, the laser exhibits passive mode-locking or Q-switching operation. However with further increasing negative bias, yielding shorter recovery time and larger absorption coefficient, a different emission regime occurs. At threshold of Q-switched lasing operation and under pulsed current pumping, the laser reveals features of SR-like emission. These features have been experimentally observed in multi-section GaAs, AlGaInAs and GaN QW lasers [36-39]. Unfortunately all experimental studies reported in the literature do not distinguish the first emission burst from the subsequent self-pulsations, which can be modulated with a Q-switching pulse envelope. Nevertheless a clear Rabi splitting was observed in the optical spectrum of multi-section LDs when these were expected to produce SR emission [36]. Although there were no reports on experimental or theoretical studies on RNGH instability in a multi-section laser diode, we believe that this subject will receive further attention in future. 42 F. The role of carrier population and carrier coherence gratings In Fig. 14 we study the effect of coherence grating on the gain spectrum (increment) for multimode instability by examining the highest gain frequency max in Eq.(16) as a function of the pump rate p. These numerical studies complement and confirm the outcomes of our analytical studies from Ref.[25]. As a reference, we show the frequency max obtained by numerically solving the eigen problem (16) gT with QCL parameters from Table 1 and coherence grating relaxation time 2_  (solid blue curve) as 0 well as the calculated Rabi frequency (18) (dashed blue curve) under the same operation conditions. These two reference curves are thus obtained in the presence of mode coupling via scattering on the coherence grating in addition to the coupling via carrier population grating. The effect of coherence grating can be excluded from Eq.(16) by considering the limit gT  . We 0 2 _ achieve this via reducing 2 _ gT by a factor of 1011 while maintaining the average decoherence rate ( T 2 _ eff  ) and the mode coupling via SHB-induced grating of the carrier distribution ( 0 gT  ).This 0 represents very well the limit of coherence grating. gT  , when there is no mode coupling via scattering on the 0 2 _ We find that the system (16) still reveals instability and this instability is caused by the SHB effect only. For the pump rates up to p~10, the spectral shape of the instability increment Re(Λ) is similar to the one depicted in Fig.2. (Compare this shape with the warped instability gain curve at much higher pump rate of p~50-60 in Fig. 1.) Thus the instability gain spectra do not indicate a change in the mechanism of multimode instability. In Fig. 14 we add the superscript "(SHB)" in order to distinguish 43 this case and plot the frequency ) ( SHB calculated numerically from Eq.(16) and max ) SHB ( Rabi from Eq.(18) for the case when 2_gT is reduced by a factor of 1011 (solid and dashed red curves, respectively). In Ref. [16], the expression (19) was obtained for the highest increment frequency for the multimode instability caused by the SHB effect, that is in the case we have just discussed above. In Fig. 14 we denote this frequency as [16] SHB (green curve) and plot it as a function of the pump rate using the parameters of QCL from Table 1. The following conclusions can be made from comparison between different curves in Fig. 14: (i) gT In the presence of coherence grating ( 2_  ), the highest instability gain is at the offset 0 frequency max which is very close to the Rabi frequency Rabi . Like the last one, 2 max exhibits a linear growth with the pump rate p (solid blue curve in Fig.14). This behavior is a signature of a multimode RNGH-like instability (see the Introduction). (ii) Without coherence grating ( 2 _ gT  ), both the frequency 0 max and the Rabi frequency reduce to ) ( SHB and max ) SHB ( Rabi , respectively (solid and dashed red curves). However   ) ( SHB max 2   does not follow anymore the linear growth of   ) ( SHB Rabi 2  and slightly deviates from the Rabi  oscillation frequency. Thus even at the pump rate as high as p=5 times above the lasing threshold, ( SHB max ) is only by a factor of 1.1 lower than ( SHB Rabi ) . Therefore we cannot attribute it to a markedly different behavior than the one observed in the presence of additional mode scattering on the coherence grating. Inclusion of the coherence grating just slightly increases the highest instability frequency in the 4×4 submatrix in Eq.(16). (iii) The frequency  16 SHB [Eq.(19)] obtained in Ref [16] for the case without coherence grating, that is for case we discussed in (ii), is significantly lower than our frequency ( SHB max ) and the Rabi 44 flopping frequency ( SHB Rabi ) . As a consequence of such low frequency it was attributed to a multimode instability of a different kind and labeled as a multimode instability caused by the SHB effect. In fact the original system of the coupled mode equations in Ref [16] contains a set of errors. Its adiabatic approximation for slowly varying medium, polarization does not recover the well-known Class-B laser model (See a discussion to our Eqs. (7)-(11) and the Appendix A). As a consequence of these errors, the frequency our model. [16] SHB << ( SHB Rabi ) , which is not the case in In ref [25] we analyze this problem analytically and arrive to the same conclusions. In particular we obtain a closed form expression for the second threshold that indicates that coherence grating just slightly reduces the second threshold. The main source for low-threshold RNGH instability in QCLs is the SHB effect and a set of the relaxation and diffusion time constants that result in Tg being comparable to T1. 2x1025 max Rabi 2 2 Rabi [16] 2, T2_g=0.13 ps 2, T2_g=0.13 ps (SHB), T2_g=1.3E-24 s (SHB), T2_g=1.3E-24 s 2, Eq.(19) max SHB 2  1x1025 0 1 2 3 4 5 Pump excess above lasing threshold, p FIG. 14. We plot the squares of the following frequencies as a function of the pump rate, using parameters of QCL from max  , solid blue curve) and corresponding Rabi oscillations (dashed red Table 1 (Tg=0.927 ps, T2_eff=0.14 ps and T2_g=0.13 ps unless stated otherwise): the highest instability gain frequency calculated numerically from the eigen problem of Eq.(16) ( 2 _ gT (solid red curve) and the Rabi frequency frequency (18) (dashed blue curve); the frequency 0 ( ) ( SHB max ) 45 SHB Rabi curve) obtained without coherence grating (T2_g is reduced by a factor of 1011); and finally the frequency from Eq.(19) (green curve).. [16] SHB calculated The coherence grating has a much stronger impact on the eigen solutions of the upper 5×5 block in Eq.(16). In Fig. 15 we plot the spectra of the Lyapunov exponents with the largest real parts in the 4×4 (red curves) and 5×5 (blue curves) blocks of Eq.(16) for the two pump rates p=9 [Fig.15(a)] and p=10 [Fig.15(b)]. Once again we compare cases when the coherence grating is present ( gT 2 _  0 , solid curves) and when it is absent ( gT  2 _ 0 , dashed curves). We observe that without coherence grating (when gT  2 _ 0 ) the increment of instability in the 4×4 matrix block (red curves) increases and its spectrum shifts to lower frequencies. It shows similar behavior at the two considered pump rates (p=9 and p=10). The behavior of the instability increment in the 5×5 block has much in common with the one in 4×4 block but the increment values are shifted down on the vertical axis in Fig. 15 toward negative Lyaponov exponents. Only without coherence grating ( gT  2 _ 0 ,while T2_eff is unchanged), the increment becomes positive and RNGH instability occurs at p=10 (see the behavior of the dashed blue curve in the inset of Fig.15(b)). Note that the RNGH instability does not occur at p=9 because T2,eff/T1≠0 [7]. Once the coherence grating is present ( 2_ gT  ), the RNGH instability threshold in 5×5 matrix 0 block increases by several times, to around p=60 in particular case (See Fig. 1). In this way it remains only one pleasurable explanation for the low second threshold observed in experiments with QCLs. Namely we conclude that it is related with the instability arising from the 4×4 matrix block in Eq.(16) due to a combined effect of the carrier population and carrier coherence gratings. 46 (a) t n e n o p x e v o n u p a y L t s e g r a l e h t f o t r a p l a e R 4x1011 3x1011 2x1011 1x1011 0 -1x1011 -2x1011 2.0x109 0.0 -2.0x109 8.0x1012 p=9 9.0x1012 1.0x1013 1, 4x4 matrix, T2_g=1.3E-24s 2, 5x5 matrix, T2_g=1.3E-24s 1, 4x4 matrix, T2_g=0.13 ps 1, 5x5 matrix, T2_g=0.13 ps (b) t n e n o p x e v o n u p a y L t s e g r a l e h t f o t r a p l a e R 4x1011 3x1011 2x1011 1x1011 0 -1x1011 -2x1011 1x109 0 -1x109 8x1012 p=10 9x1012 1x1013 1, 4x4 matrix, T2_g=1.3E-24s 2, 5x5 matrix, T2_g=1.3E-24s 1, 4x4 matrix, T2_g=0.13 ps 1, 5x5 matrix, T2_g=0.13 ps 0.0 4.0x1012 8.0x1012 1.2x1013 1.6x1013 0.0 4.0x1012 8.0x1012 1.2x1013 1.6x1013   FIG. 15. Spectra of instability increments (largest real parts of the Lyapunov exponents) of the two matrix blocks of 4×4 (red curves) and 5×5 (blue curves) sizes for QCL with parameters from Table 1. The pump rate excess above threshold is p=9 (a) and p=10 (b) The solid curves stand for the case considered in this paper when we include the coherence grating effects (T2,g=0.13ps), while the dashed curves represent the case when the coherence grating relaxation time T2_g is reduced by a factor of 1011, suppressing all effects caused by the coherence grating. The insets show zooms to the highest RNGH instability increments in the 5×5 matrix block when gT  (no coherence grating effects). 2 _ 0 IV. CONCLUSION In conclusion we have clarified conditions for multimode RNGH instability in a Fabry-Perot cavity laser and proposed a new explanation for low second threshold in QCLs, which is based on a combined effect of the carrier coherence and carrier population gratings. We have studied the impact of the cavity length on the waveform of RNGH self-pulsations and shown that short-cavity QCLs can produce a regular train of ultrashort pulses. There is even a possibility to obtain supperradiant-like emission. Our findings open a practical way of achieving ultra-short pulse production regimes in the mid-infrared spectral range. Applying the same analysis to conventional laser diodes we find that carrier diffusion in the active bulk semiconductor material or quantum wells prevents the LDs from RNGH self-pulsations. 47 ACKNOWLEDGEMENTS The authors would like to acknowledge the support from the Swiss National Science Foundation (SNF), Ministry of Education, Science and Technological Development (Republic of Serbia), ev. no. III 45010, COST ACTIONs BM1205 and MP1204 as well as the European Union's Horizon 2020 research and innovation programme under the grant agreement No 686731 (SUPERTWIN). DBo is grateful to Vladimir Kocharovsky and Ekaterina Kocharovskaya for valuable discussions. APPENDIX A. Adiabatic approximation test In this appendix we consider adiabatic-following approximation that applies to bidirectional ring lasers operating on two counter-propagating waves or single-mode FP cavity lasers. In these lasers, the standing wave pattern of the field in the cavity leads to the spatial hole burning in the distribution of population inversion. At the same time, our considerations presented below do not apply for a single- frequency unidirectional ring laser from Refs. [7,8] because there is no standing wave pattern in the cavity of such laser and associated with it spatial hole burning effect. The adiabatic-following approximation for the medium polarization is valid when the polarization dynamics is slow and follows instantaneously the optical field in the cavity [9]. For solid state lasers, including semiconductor LDs and QCLs, the dephasing time T2 is much smaller than the relaxation T1. Therefore under adiabatic-following approximation, the behavior of these systems fall into Class-B laser dynamics. These dynamical systems are very well studied, including the cases with mode coupling due to back-scattering and induced population grating. One example can be found in a review [24]. 48 In order to obtain the adiabatic-following approximation for our coupled-mode system (7)-(10), we set the time derivatives in Eqs (8) and (9) to zero, yielding the following instantaneous amplitudes of polarization grating harmonics:    i  2  E   0  E   2   T 2    i  2  E     2 _g T (A1) (A2) where for the purpose of this test we have excluded the diffusion terms, assuming that T 2 _ eff  T 2 _g  T 2 , and gT T . Substituting these in Eqs. (7), (9) and (10), we find that 1 n dE  c dt   dE  dz  2 NT 2 _ eff 2 c n       0 E   0  E    2  1 2 l E 0  d (    0  2    pump T 1 e 2 ikz dt    0   2 ikz e   2 )  2 e  2 ikz T 1   e   2 2 ikz  1     a E E *  (  E E *   )  2 Re( E E e  *    2 ikz )    (A3) (A4) where we have introduced the saturation parameter a T T  2 1 2 / 2  . This system of equations is in perfect agreement with the well-known Class-B laser model (e.g. see Eqs. (1.10) and (1.11) in Ref. [24]). In contrast to our coupled-mode expansion, the adiabatic approximation that follows from the equations (8)-(10) of Ref. [16] disagrees with the well-established coupled-mode model for a Class-B laser. REFERENCES [1] J. Faist, Quantum cascade lasers, (Oxford University Press, 2013). [2] A. Hugi, G. Villares, S. Blaser, H. C. Liu, and J. Faist. "Mid-infrared frequency comb based on a quantum cascade laser." Nature 492, no. 7428, 229-233, (2012). 49 [3] H. A. Haus, IEEE J. Sel. Top. Quantum Electron. 6, 1173 (2000). [4] R. Paiella et al., "High-speed operation of gain-switched midinfrared quantum cascade lasers," Appl. Phys Lett. vol 75, p 2536, (1999). [5] C. Wang et al., "Mode-locked pulses from mid-infrared Quantum Cascade Lasers," Opt. Express vol 17, p 12929, (2009). [6] Ch. Y. Wang, L. Diehl, A. Gordon, C. Jirauschek, F. X. Kärtner, A. Belyanin, D. Bour, S. Corzine, G. Höfler, M. Troccoli, J. Faist, and F. Capasso, "Coherent instabilities in a semiconductor laser with fast gain recovery", Phys Rev A 75, 031802(R) (2007). [7] H. Risken and K. Nummedal, "Self-Pulsing in Lasers," J. Appl. Phys. vol 39, p 4663, (1968). [8] R. Graham and H. Haken, "Quantum Theory of Light Propagation in a Fluctuating Laser-Active Medium", Z. Phys. vol 213, 420, (1968). [9] M. D. Crisp, "Adiabatic-following approximation", Phys. Rev. A, vol. 8, no. 4, pp. 2128–2135, (1973). [10] L. W. Hillman, J. Krasinski, R. B. Boyd, and C. R. Stroud, Jr., Phys. Rev. Lett. 52, 1605 (1984). [11] E. M. Pessina, G. Bonfrate, F. Fontana, and L. A. Lugiato, "Experimental observation of the Risken-Nummedal-Graham-Haken multimode laser instability", Phys. Rev. A 56, 4086 (1997). [12] T. Voigt, M.O. Lenz, F. Mitschke, E. Roldan, G.J. de Valcarcel, "Experimental investigation of Risken–Nummedal–Graham–Haken laser instability in fiber ring lasers", Appl. Phys. B 79, 175 (2004). [13] J.L. Font, R. Vilaseca, F. Prati, E. Roldan, "Coexistence of single-mode and multi-longitudinal mode emission in the ring laser model", Opt. Commun. 261, 336 (2006). [14] E. Roldan, G. J. de Valcarcel, J. F. Urchueguıa, J. M. Guerra, "Observability of the Risken– Nummedal–Graham–Haken instability in Nd:YAG lasers", J. Opt. Soc. Am. B 20 (5), 816 (2003). [15] E. M. Pessina, F Prati, J Redondo, E. Roldan and G. J. de Valcarcel, "Multimode instability in ring fibre lasers", Phys. Rev. A 60, 2517 (1999). [16] A. Gordon, C. Y. Wang, L. Diehl, F. X. Kärtner, A. Belyanin, D. Bour, S. Corzine, G. Höfler, H. C. Liu, H. Schneider, T. Maier, M. Troccoli, J. Faist, and F. Capasso, "Multimode regimes in quantum cascade lasers: From coherent instabilities to spatial hole burning", Phys. Rev. A 77 (5), 053804 (2008). [17] H. Knapp, H. Risken and H.D. Vollmer, Instability of laser cw oscillations in presence of a passive medium, Appl. Phys. 15, 265 (1978). [18] M. Bugajski, K. Pierściński, D. Pierścińska, A. Szerling, K. Kosiel, "Multimode instabilities in mid-infrared quantum cascade lasers", Photonics Letters of Poland, Vol. 5, No. 3, pp. 85-87, (2013). 50 [19] Vl. V. Kocharovsky, A. A. Belyanin, E. R. Kocharovskaya, and V. V. Kocharovsky in Advanced Lasers edited by O. Shulika, I. Sukhoivanov (Springer Series in Optical Sciences, 2015), Chap. 4, p.49 [20] R H Pantell, H E Puthoff, Fundametals of Quantum Electronics, (Willey & Sons Inc, 1969). [21] S. Appelt, A. B. Baranga, C. J. Erickson, M.V. Romalis, A. R.Young, and W. Happer, "Theory of spin-exchange optical pumping of 3He and 129Xe", Phys. Rev. A 58, 1412 (1998). [22] D. L. Boiko, P. P. Vasil'ev, "Superradiance dynamics in semiconductor laser diode structures", Opt. Express vol 20, p 9501, (2012). [23] L. A. Lugiato et al., "Multimode instabilities in a homogeneously broadened ring laser", Phys Rev. A , vol. 32, pp. 1563-1575, (1985). [24] N.V. Kravtsov, E. G. Lariontsev, A. N. Shelaev , "Oscillation regimes of ring solid-state lasers and possibilities for their stabilization", Laser Physics 3, 21 (1993). [25] N. Vukovic, J. Radovanovic, V. Milanovic and D. L. Boiko, "Analytical expression for the second threshold in quantum cascade lasers," arXiv:1606.06914 [cond-mat.mes-hall] [26] E. Roldan, G.J. de Valcarcel, F. Prati, F. Mitschke, T. Voigt, Multilongitudinal mode emission in ring cavity class B lasers, in "Trends in Spatiotemporal Dynamics in Lasers. Instabilities, Polarization Dynamics, and Spatial Structures", edited by O. Gomez-Calderon and J.M. Guerra, (Research Signpost, Trivandrum, pp. online: http://arxiv.org/abs/physics/0412071v1. 81-7736-235-6, Available ISBN: India, 2005) 1-80, [27] S. L. McCall and E. L. Hahn, "Self-Induced Transparency," Phys. Rev., vol. 183, no. 2, pp. 457– 485, (1969). [28] X. Zeng and D. L. Boiko, "1/f noise in external-cavity InGaN diode laser at 420 nm wavelength for atomic spectroscopy," Optics Letters, 39 (6), 1685 (2014). [29] A. K. Wojcik, P. Malara, R. Blanchard, T. S. Mansuripur, F. Capasso, and A. Belyanin, "Generation of picosecond pulses and frequency combs in actively mode locked external ring cavity quantum cascade lasers," Appl. Phys. Lett. 103, 231102 (2013). [30] Y. Wang and A. Belyanin, "Active mode-locking of mid-infrared quantum cascade lasers with short gain recovery time," Opt. Express 23, 4173 (2015). [31] :A. E. Siegman, Lasers (University Science Books, Sausalito, CA, 1986), Chapter 13, Eq (10) [32] E.N. Lorenz, Deterministic non periodic flow. J. Atmosph. Sci. 20, 130–141(1963). [33] R. H. Dicke, "Coherence in spontaneous radiation processes," Phys. Rev. 93, 99–110 (1954). 51 [34] D. L. Boiko P.P. Valsil'ev, "Longitudinal polariton condensation and superradiant emission in semiconductor edge-emitting laser structures," arXiv: 1112.1298 [cond-mat.-el] [35] G. T. Noe, J.-H. Kim, J. Lee, Y. Wang, A. K. Wójcik, S. A. McGill, D. H. Reitze, A. A. Belyanin & J. Kono, Nature Physics 8, 219 (2012). [36] P.P.Vasil'ev Rep. Prog. Phys. 72, 076501 (2009). [37] M. Xia, R. V. Penty, I. H. White, and P. P. Vasil'ev, Optics Express 20 , 8755 (2012). [38] V. F. Olle, P. P. Vasil'ev, A. Wonfor, R. V. Penty, and I. H. White, Opt. Express 20, 7035 (2012). [39] D. L. Boiko, X. Zeng, T. Stadelmann, S. Grossmann, A. Hoogerwerf, T. Weig, U. T. Schwarz, L. Sulmoni, J.-M. Lamy, N. Grandjean, "Superfluorescent emission in electrically pumped semiconductor laser," arXiv:1302.0263 [physics.optics] [40] F. T. Arecchi and E. Courtens, "Cooperative phenomena in resonant electromagnetic propagation," Phys. Rev. A 2, 1730 (1970). [41] L.Kappei, "Dynamics and coherent effects of high density plasmas in semiconductor nanostructures", Ph.D. Thesis , EPFL (2005). [42] A. A. Belyanin, V. V. Kocharovskii, and Vl. V. Kocharovskii, "Novel Schemes and Prospects of Superradiant Lasing in Heterostructures," Laser Physics 2, 952 (1992). 52
1409.6521	2	1409	2015-10-12T12:27:16	Influence of Hyperfine Interaction on the Entanglement of Photons Generated by Biexciton Recombination	[ "cond-mat.mes-hall", "quant-ph" ]	The quantum state of the emitted light from the cascade recombination of a biexciton in a quantum dot is theoretically investigated including exciton fine structure splitting (FSS) and electron-nuclear spin hyperfine interactions. In an ideal situation, the emitted photons are entangled in polarization making the biexciton recombination process a candidate source of entangled photons necessary for the growing field of quantum communication and computation. The coherence of the exciton states in real quantum dots is affected by a finite FSS and the hyperfine interactions via the effective magnetic field known as the Overhauser field. We investigate the influence of both sources of decoherence and find that although the FSS combined with a stochastic exciton lifetime is responsible for the main loss of entanglement, the two effects cannot be minimized independently of each other. Furthermore, we examine the possibility of reducing the decoherence from the Overhauser field by partially polarizing the nuclear spins and applying an external magnetic field. We find that an increase in entanglement depends on the degree as well as the direction of the nuclear spin polarization.	cond-mat.mes-hall	cond-mat	Influence of Hyperfine Interaction on the Entanglement of Photons Generated by Biexciton Recombination Erik Welander,∗ Julia Hildmann,† and Guido Burkard Department of Physics, University of Konstanz, D-78457 Konstanz, Germany The quantum state of the emitted light from the cascade recombination of a biexciton in a quantum dot is theoretically investigated including exciton fine structure splitting (FSS) and electron-nuclear spin hyperfine interactions. In an ideal situation, the emitted photons are entangled in polarization making the biexciton recombination process a candidate source of entangled photons necessary for the growing field of quantum communication and computation. The coherence of the exciton states in real quantum dots is affected by a finite FSS and the hyperfine interactions via the effective magnetic field known as the Overhauser field. We investigate the influence of both sources of decoherence and find that although the FSS combined with a stochastic exciton lifetime is responsible for the main loss of entanglement, the two effects cannot be minimized independently of each other. Furthermore, we examine the possibility of reducing the decoherence from the Overhauser field by partially polarizing the nuclear spins and applying an external magnetic field. We find that an increase in entanglement depends on the degree as well as the direction of the nuclear spin polarization. PACS numbers: 78.67.Hc, 03.67.Bg, 73.21.La, 71.70.Jp I. INTRODUCTION A reliable source of entangled photons is a require- ment for many protocols used in the rapidly develop- ing field of quantum communication1. An established method for creating polarization entangled photons is by parametric down-conversion2,3. However, this technique suffers from being both inefficient and stochastic, causing problems since many quantum communication protocols require an on-demand source. An alternative source is thus desired, and here we consider the biexciton cascade recombination4,5. In a quantum dot (QD), the biexci- ton, which is composed of two conduction band electrons and two valence band holes, can under ideal conditions recombine under the emission of two photons entangled in polarization4,6. The biexciton recombines via one of two possible intermediate exciton states, each consisting of one conduction band electron and one valence band hole. In most quantum dots, the two optically active exciton states are energetically separated by a quantity known as the fine structure splitting (FSS) arising from higher or- der electron-hole exchange interactions7 -- 9. A finite FSS can affect the coherence of the emitted two-photon state in two ways. If the FSS is larger than the linewidth of the emitted light, the photons become distinguishable via a simple frequency measurement which reveals the "which- way" information and destroys the entanglement6. How- ever, even if the splitting is smaller than the linewidth but still finite, the initially coherent exciton state acquires a random phase before recombining, due to the stochastic life time. Methods of reducing and eliminating the FSS include applying magnetic10 and electric9,11 -- 16 fields as well as strain17 -- 20. In addition to the dephasing from a finite FSS, the in- termediate exciton state is affected by the spins of the 105-106 nuclei present in a III-V group semiconductor QD. The spins of the electron21 -- 24 and hole25 -- 27 consti- tuting the exciton couple to nuclear spins via the hyper- fine interaction and are subject to an effective nuclear magnetic field, known as the Overhauser field. Because of the large number of nuclear spins, the Overhauser field can be considered to be stochastic and is another source of decoherence, including an additional random phase of the intermediate exciton state. Experimentally, various techniques for creating entan- gled light using semiconductor microstructures have been demonstrated5,10,28 -- 30. One successful approach relies on the application of an external magnetic field perpendic- ular to the growth direction of the QD10,31, which tunes the energy levels of the optically active excitons by hy- bridization to optically inactive states. This requires the in-plane g-factors of the electrons and holes to have op- posite signs. InAs dots with AlGaAs barrier material having this property were reported31 and by applying an in-plane magnetic field, the FSS were tuned to zero. Nevertheless, a complete theoretical explanation of the partial loss of entanglement is still missing. Understand- ing the dynamics of the intermediate exciton states is essential to investigate the entanglement of the emitted light. In fact, any dephasing and loss of coherence of the exciton state will be reflected in the final photon state. In this paper, we investigate the interplay between the dephasing due to a finite FSS together with a stochas- tic recombination time and the decoherence caused by the hyperfine interaction. Our results concern a dot for which the FSS is tuned to zero by an in-plane magnetic field, which requires the in-plane electron and hole g- factors to have opposite signs. One way of reducing the fluctuations of the Overhauser field is by dynamic nu- clear spin polarization32,33, causing the nuclear spins to have a preferred direction. The polarized nuclear ensem- ble produces a finite effective magnetic field, which may modify the exciton energies and eigenstates. Since the elimination of the FSS has been demonstrated using an external in-plane magnetic field, a tempting idea could be to use the effective magnetic field produced by the polarized nuclear ensemble, to reduce both the FSS and the Overhauser field fluctuations. However, this turns out not to be possible, since an effective in-plane mag- netic field for both holes and electrons is required. For heavy holes, the in-plane component of the nuclear mag- netic field vanishes. Furthermore, we show that the effect from the two sources of decoherence cannot be minimized independently of each other. We consider the effect of finite nuclear spin polarization and find that the entanglement of the emitted light can be improved by nuclear spin polarization. The efficiency of the entanglement improvement depends on the direc- tion along which the nuclear spins are polarized, which is explained by the fact that the hyperfine coupling tensor is not isotropic. We find that the maximum enhancement of the entanglement is achieved when nuclear spins are polarized along the direction for which the coupling ten- sor has its largest components, in our case in the growth direction of the quantum dot. A nuclear polarization in the growth direction gives an additional contribution to the FSS and therefore in- creases the dephasing. To cancel the effective nuclear field in the growth direction and minimize the FSS at the same time, we propose applying an external mag- netic field along a specific direction having an in-plane and a perpendicular component. Combining a finite nu- clear spin polarization along the growth direction of the QD with an external magnetic field, we find a significant improvement of the two-photon state entanglement. II. THEORETICAL MODEL We consider a QD of a cubic semiconductor contain- ing one exciton consisting of one electron and one heavy hole. The spins of the electron and the hole couple to the spins of the atomic nuclei in the QD. This effect is to a good approximation described by the contact hyperfine Hamiltonian HHF = N Xn=1 I(n) ·(cid:16)A(n) e Se + A(n) h Sh(cid:17) , (1) where the summation runs over all nuclear spins in the dot, I(n) is the spin operator of the n-th nuclear spin, (n) Se(h) is the electron(hole) spin operator, and A e(h) are the hyperfine coupling tensors between the n-th nuclear spin and corresponding electron(hole) spin component. In this work, we are interested in heavy holes, for which tensor is finite. This allows us to define an effective magnetic field by moving the electron and hole spin operators outside of the summation and obtain only the z − z component of the A(n) h HHF = µB BHF · feSe + µBBz HFfhSh z , (2) 2 σ− ↓⇑i σ+ σ+ ↑⇓i σ− ↓⇑i↑⇓i ω 0ie Figure 1. Energy diagram of the exciton states in the quan- tum dot. The topmost level is the biexciton state ↓⇑i ↑⇓i consisting of two excitons. The two intermediate levels are the exciton states ↓⇑i and ↑⇓i energetically separated by the FSS ω, which depends on the built-in FSS δ1 and an applied magnetic field. The lowest level is the semiconduc- tor ground state 0ie containing no excitons. The biexciton recombines to one of the intermediate exciton states under the emission of a photon with polarization depending on the exciton state. The remaining exciton then recombines under the emission of another photon of orthogonal polarization. In the ideal situation, this creates a pair of photons entangled only in polarization. The existence of a FSS can degrade the entanglement, even if ω < Γ where Γ is the linewidth of the emitted light because of the stochastic lifetime of the exciton which causes dephasing and leads to a statistical mixture in- stead of a pure state. Furthermore, because of the interaction with stochastic magnetic field originating from the nuclear spins in the quantum dot, the exciton energy levels are not sharply defined, meaning that ω also fluctuates which causes even more dephasing. where fe is the electron-nuclear spin coupling tensor. The heavy hole - nuclear spin coupling is of Ising type, described by the coupling constant fh between the z- component of hole and nuclear spins. The vector BHF is known as the Overhauser field and acts like an effective magnetic field from the perspective of the exciton. Be- cause of the large number of nuclear spins24 (105 − 106) for a typical quantum dot, the Overhauser field is often modelled as a stochastic magnetic field and will be con- sidered further in Section III B. We consider the case of diagonal hyperfine coupling tensors, which for an elec- tron in a III-V semiconductor QD can be written34 as A(n) x,y,z ∝ gx,y,z are the electron g- factors and F (rn)2 is the electron density at the atomic site n. An important feature is the dependence of the spa- tial direction x, y, z, which indicates that fluctuations of the different spatial components of the Overhauser field influence the energy of the electron differently. F (rn)2, where gx,y,z e e A basis for the Hilbert space of the electron spin He is given by {↑i ,↓i}, where ↑ (↓) corresponds to the spin sz = 1/2 (sz = −1/2) state, and for the heavy hole the Hilbert space Hh is spanned by {⇑i ,⇓i}, with citons are given by 3 E1 = δ1 + δ2 +p(2δ0 + δ1 − δ2)2 + 4(gx E2 = −δ1 − δ2 +p(2δ0 − δ1 + δ2)2 + 4(gx 4 4 e + gx h)2µ2 BB2 x e − gx h)2µ2 (6a) BB2 x . (6b) Similar expressions were found by Bayer et. al.36, where it should be noted that our expressions differ slightly, due to a sign error. Demanding the bright exciton states to be degenerate, i.e. E1 = E2, gives an equation for a critical field Bcr, 2(δ1 + δ2) +q(2δ0 + δ1 − δ2)2 + 4(gx =q(2δ0 − δ1 + δ2)2 + 4(gx e − gx g )2µ2 BB2 cr, e + gx g )2µ2 BB2 cr (7) for which the FSS vanishes. The FSS can be written E1 − E2, which may be expanded in Maclaurin series in Bx to the second order, which gives the approximation ω(Bx) ≈ δ1 + e gx µ2 B(cid:2)4gx e hδ0 −(cid:0)gx 2 + gx h 0 − (δ1 − δ2)2 4δ2 2(cid:1) (δ1 − δ2)(cid:3) B2 x. (8) A similar expression was also presented by Stevenson et al.31. Solving ω(Bcr) = 0 provides an expression for the critical magnetic field, given by Bcr = ±s δ1 [(δ1 − δ2)2 − 4δ2 0] e gx hδ0 −(cid:0)gx 2 + gx h e 2(cid:1) (δ1 − δ2)(cid:3) µ2 B(cid:2)4gx . (9) For the general case involving arbitrary magnetic fields and complex δ1, δ2, an analytical diagonalization of the Hamiltonian in Eq. (5) is not known. However, the en- ergy splitting between the bright and dark excitons δ0 is larger than all other relevant energies, including the magnetic coupling elements. Therefore, we can apply the Schrieffer-Wolff37,38 transformation, which provides us with an effective Hamiltonian for the bright exciton subspace: where , H0 = H = H0 + H2, 1 2(cid:18)δ0 − hz− δ∗ 1 δ1 δ0 + hz−(cid:19) , ⇑ (⇓) corresponding to the hole spin states jz = 3/2 (jz = −3/2). The Hilbert space of the exciton is given by the product space HX = He ⊗ Hh and is spanned by the basis vectors {↓⇑i ,↑⇓i ,↑⇑i ,↓⇓i}. The states ↑⇓i and ↓⇑i are known as bright since they can recombine under the emission of a single photon whereas ↑⇑i and ↓⇓i are known as dark. The idealized recombination chain of the biexciton is given by 0iph ⊗ ↓⇑i↑⇓i σ+i ⊗ ↑⇓i + σ−i ⊗ ↓⇑i ↓ √2 ↓ σ+i σ−i + σ−i σ+i √2 ⊗ 0ie where σ±i are photon states of circularly polarized light, and 0iph(e) is the photon (crystal) vacuum. In reality the intermediate state undergoes time-evolution before the exciton has recombined which may lead to degradation of the entanglement of the emitted light. The final state of the intermediate exciton state can be written using a density matrix ρX =(cid:18) p γ∗ 1 − p(cid:19) , γ (3) where p and 1 − p are the populations of the states σ+i⊗↑⇓i and σ−i⊗↓⇑i, and γ is the off-diagonal ma- trix element required to describe a quantum mechanical superposition of the basis states. Here, only there bright excitons are taken into consideration. The concurrence of the emitted light is then given by35 C = 4pp(1 − p)γ, (4) where we note that the off-diagonal elements are essential for the entanglement. Under the influence of a magnetic field B = (Bx, By, Bz)T the exciton system is described by the Hamiltonian7 H = 1 2  δ0 − hz− δ∗ 1 h∗ e h∗ h δ1 δ0 + hz− hh h∗ e he h∗ h δ∗ 2 −δ0 + hz+ hh he δ2 −δ0 − hz+   (5) in the heavy exciton basis {↓⇑i ,↑⇓i ,↑⇑i ,↓⇓i}, where δ0 is the splitting between bright and dark excitons, is the FSS for bright (dark) excitons, hz± = δ1(2) e(h)(cid:17) and µBBz(gz e ± gz gα e(h) are effective g-factors for electrons(holes) along the α-axis. For the case B = Bx x and δ1, δ2 ∈ R, the eigen- value problem H Xi = E Xi is analytically solvable and the two eigenenergies corresponding to the two bright ex- h), he(h) = µB(cid:16)Bxgx e(h) + iBygy and H2 = 1 4  he2 δ0−be δ0h∗ e hh δ2 0−b2 e + hh2 δ0+bh + δ0h∗ e hh δ2 0−b2 h δ0heh∗ h δ2 0−b2 e he2 δ0+be + δ0heh∗ h δ2 0 −b2 h + hh2 δ0−bh   , e(h). The FSS ω can now be found with be(h) = µBBzgz by solving the eigenvalue problem H ni = En ni (10) (11) (12) (13) where 1 + Ω2 2, ω =qΩ2 he2be 1 Ω1 = hz− − δ2 0 − b2 2 δ0h∗ 1 ehh + δ2 0 − b2 4 δ1 + e e Ω2 =(cid:12)(cid:12)(cid:12)(cid:12) + h hh2bh 1 δ2 0 − b2 2 δ0h∗ ehh 1 δ2 0 − b2 4 h(cid:12)(cid:12)(cid:12)(cid:12) (14) (15) (16) , . for n = 1, 2 and taking the difference ω = E1 − E2. Using the explicit form of H given by Eqs. (11) and (12) we find Demanding that ω(Bcr) = 0 again defines critical mag- netic fields, not necessarily along x, for which the FSS is eliminated. Inspecting Eq. (14), we realize that ω = 0 only if Ω1 = 0 and Ω2 = 0. Ω1 can be tuned to zero by adjusting the total magnetic field along z. Ω2 depends on in in-plane components of the magnetic field, Bx and By and the bright exciton coupling δ1, which may be complex. The phase of δ1 is related to the geometry of the quantum dot whereas the phase of h∗ ehh depends on the anisotropy of the g-tensors. For a quantum dot with an isotropic in-plane g-tensor, h∗ ehh is real and in this situation, no magnetic field can completely eliminate the FSS caused by a complex δ1. However, δ1 ∈ R combined with a isotropic g-tensor, Eq. (16) reveals that h∗ ehn < 0 is a criterion for the existence of a magnetic field such that Ω2 = 0. In turn, this requires that the in-plane g- factors of the electron and hole have opposite signs and this implies that not all quantum dots can be tuned to support degenerate bright excitons, also noted in exper- imental work31,36 where InAs dots surrounded by differ- ent barrier materials were studied. The corresponding in- plane g-factors were extracted, and here the values for the case of Al0.33Ga0.67As31 as the barrier material are given in Table I. Experimental39 -- 41 and theoretical42 studies show that the g-factor of InAs and InGaAs QDs can be tuned over large range of values and here we use gz e = 2, gz h = −5 to give a total exciton g-factor gz h = −3, measured in experiments39. There are two main sources of loss of entanglement: (1) the fine-structure splitting combined with the stochas- tic exciton life time and (2) the stochastic Overhauser field that affect the intermediate exciton state. Both mechanisms lead to the acquisition of an unknown phase which causes a reduction of the entanglement. To inves- tigate further, we consider the effect of the two above- mentioned mechanisms on the density operator of the intermediate exciton state. We choose the diagonal basis 1i ,2i which are eigenvectors of H, and an initial density operator in matrix form as e + gz ρ(t = 0) =(cid:18)ρ11 ρ12 ρ21 ρ22(cid:19) , (17) where ρ11 + ρ22 = 1 and ρ∗ 12 = ρ21. We can now deter- mine the time-evolution for the density operator via the Heisenberg equation of motion, 4 ) s n 1 = t ( 1 C ) s n 1 = τ ( 2 C 1 0.8 0.6 0.4 0.2 0 1 0.8 0.6 0.4 0.2 0 (a) (b) 0 0.5 1 1.5 2 2.5 Magnetic Field Bx [T] Figure 2. Comparison between the concurrence of a stochastic FSS with a definite recombination time C1(t) and the con- currence in the presence of a known FSS with a stochastic recombination time C2(τ ). (a) The concurrence of the en- tangled light C1(t), when a stochastic Overhauser field with σx = σy = σz = 5 mT causes an unknown relative phase between the exciton states when the exciton undergoes time- evolution of t = 1 ns. (b) The concurrence of the entangled light C2(τ ) when the known FSS causes an unknown rela- tive phase between the exciton states. The exciton undergoes time-evolution for a stochastic recombination time, which on average is given by τ = 1 ns. We see that there is a trade-off when attempting to minimizing both sources of decoherence since at Bx = Bcr ≈ 1 T, C2(τ ) has a maximum, while C1(t) has a minimum. This indicates that both sources of decoher- ence have to be taken into consideration simultaneously. which has the solution ρ(t > 0) =(cid:18) ρ11 ρ21e−iωt ρ12eiωt ρ22 (cid:19) , (19) with the FSS ω given by Eqs. (14) -- (16). If the FSS is stochastic as one would expect from an Overhauser field we may find its contribution by statistical averaging hρi =Z ∞ −∞ fΩ(ω)(cid:18) ρ11 ρ21e−iωt ρ12eiωt ρ22 (cid:19) dω, (20) where fΩ(ω) is the probability density function of the FSS. For the quantum dot hosting the exciton we assume a stochastic Overhauser field with Gaussian distribution. The FSS, however, does not have a Gaussian distribution because of the nonlinear way the exciton eigenenergies depend on an applied magnetic field, given by Eqs. (14) -- (16). Therefore, the statistical averaging is performed by considering a Gaussian distribution for the Overhauser field with the probability density function i ρ = [ρ, H], (18) fB(B) = 1 σxσyσz(2π)3/2 e−B2 x/2σ2 x−B2 y/2σ2 y −B2 z /2σ2 z , (21) where σx, σy, σz are the standard deviations of the Over- hauser field along x, y, z. We numerically evaluate hρi =Z ∞ −∞ fB(B)(cid:18) ρ11 ρ12eiω(B)t ρ21e−iω(B)t ρ22 (cid:19) dω, (22) e(h)Bα −→ gα using the parameter values given in Table I, from which we can extract an entanglement measure, here the con- currence C1(t) by using Eq. (4). An external magnetic field, B0 may also be present, and both the Overhauser field and the external magnetic field are included by re- e(h)B0 placing gα e(h)Bα in Eqs. (5)-(16), where α ∈ {x, y, z} and f x (h) = 0. To investigate the effect of the stochastic exciton life time we consider a Poissonian recombination process which corresponds to an exponential life time t with prob- ability density function ft(t) = e−t/τ /τ where τ is the average life time. Calculating the statistical average of the density matrix gives α + f α (h) = f y e−t/τ 0 hρi =Z ∞ =(cid:18) ρ11 ρ21 1−iωτ τ (cid:18) ρ11 ρ22 (cid:19) , 1+iωτ ρ12 ρ21e−iωt ρ12eiωt ρ22 (cid:19) dt (23) which also have decaying concurrence43 C2(τ ) ∝ 1 + i ω τ −1 for ω 6= 0. This suggests choosing ω ≪ 1/τ to maximize the concurrence. The two different concur- rences are shown in Fig. 2. We can see that there is a tar- get conflict when applying a magnetic field along x. For a critical magnetic field strength Bx = Bcr the fine struc- ture splitting is eliminated and C2(τ ) has a maximum, but the concurrence C1(t) when considering a stochas- tic magnetic field from the nuclear spins has a minimum. The reason is that the FSS is most sensitive to changes in the magnetic field at this point. To obtain a more com- plete picture we need to take both sources of decoherence into account simultaneously, which we achieve by aver- aging the concurrence in Eq. (23) using the probability distribution for the stochastic magnetic field fB(B) given by Eq. (21), which is done numerically by evaluating Z ∞ −∞ C(τ ) =(cid:12)(cid:12)(cid:12)(cid:12) fB(B) 1 + iω(B)τ . dω(cid:12)(cid:12)(cid:12)(cid:12) III. RESULTS (24) To obtain quantitative results, we choose a set of pa- rameter for the quantum dot given in Table I. A. Dominant Source of Decoherence In order to improve the concurrence we first establish which source of decoherence causes more loss of concur- rence, the FSS or the Overhauser field. From Fig. 2 5 gx,y h gx,y e 1.21 −0.13 2 −5 50 μeV 10 μeV 1 ns gz e gz δ0 δ1 τ h Table I. Table of the parameters for an InAs quantum dot surrounded by GaAs, used in the numerical calculations, re- ported from experiments31. The condition that gx h < 0 is necessary to allow the elimination of the FSS. δ1 is in general complex, but here chosen real. It should be noted, that these parameters are not representative for all quantum dots, but selected in order to allow the FSS to be tuned to zero. e gy this is not obvious, because at Bx = Bcr the FSS is min- imized but the dephasing from the Overhauser field is maximized. Taking both into account and allowing in addition a magnetic field to be applied along z as well we find the concurrence as function of the applied mag- netic field depicted in Fig. 3. From our calculations we find a maximum value for the concurrence Cmax ≈ 0.54. This is consistent with experimentally reported values of typically around 0.530,44,45. However, it should be noted, that a broad range of values for the concurrence have been reported, spanning between 0.1646 and 0.829. We see that the two global maxima are located at (Bx = ±Bcr, By = 0, Bz = 0) which indicates that the FSS is a stronger source of decoherence than the Over- hauser field. Still, the concurrence does not reach unity but is rather close to the minimum observed in Fig. 2a. From these observations we conclude that in order to maximize the concurrence, we should keep Bx = ±Bcr to eliminate the FSS and now focus on reducing the uncer- tainty of the Overhauser field. One way of achieving this is to polarize the nuclear spins, which has been exper- imentally realized47 -- 53, and is investigated in the next In addition to the two global maxima, there section. ) ) τ τ ( ( C C e e c c n n e e r r r r u u c c n n o o C C 0.5 0.4 0.3 0.2 0.1 0 -2.5 Bx [T] 0 -1 2.5 1 0 Bz [T] Figure 3. Concurrence when the combined effect of dephas- ing from a stochastic magnetic field due to the nuclear spins and a stochastic exciton lifetime is considered. The maxima occur when the FSS is eliminated and reach values limited by the dephasing from the stochastic Overhauser field. This indicates that the dominant source of decoherence is the FSS combined with the stochastic lifetime of the exciton. An im- provement of the maximum concurrence can be achieved by reducing the Overhauser field fluctuations. There are also four local maxima located roughly at Bx = 1.5 T, Bz = 0.25 T, demonstrating the importance of treating both sources of decoherence simultaneously. ] T [ t x e x B 8 6 4 2 0 -2 -4 -6 -8 1 0.8 0.6 0.4 0.2 0 0 1 ϕ/π 2 x Figure 4. Concurrence as function of nuclear polarization an- gle ϕ and applied magnetic field B ext along x. We assume a fixed degree of η = 90% polarization along (nx, 0, nz) = (sin ϕ, 0, cos ϕ). The total magnetic field is given by B = (B ext x + ηBmax sin ϕ, 0, 0), which implies an applied magnetic field B ext z = −ηBmax cos ϕ along z. For every angle of po- larization, there are two values of the external magnetic field maximizing the concurrence, which are located at Bx = ±Bcr, which minimizes the FSS. are four local maxima located close to Bx = 1.5 T, Bz = 0.25 T. Although the concurrence is smaller at these points than at the global maxima, they indicate the significance of including the effects of both sources of decoherence simultaneously. B. Effect of Nuclear Spin Polarization It is clear that, within our model, when the FSS is elim- inated, the remaining reduction of the entanglement orig- inates from the Overhauser field. To investigate how the fluctuations of the Overhauser field vary as function of the nuclear spin polarization we consider a simple model for the Overhauser field along one spatial direction B = M Xn=1 Anβn, (25) where M is the number of nuclear spins, βn are binary stochastic variables taking the values ±1 with probability where pβ(±1) = e±S eS + e−S , S = µBgNB0 kBTN , (26) (27) gN is the nuclear g-factor, B0 is an external magnetic field and TN is the nuclear spin temperature. The polarization η is given by η = hβi = eS − e−S eS + e−S = tanh S, and the variance is consequently In Appendix A it is shown that σ2 =(cid:10)β2(cid:11) − hβi2 = sech2 S = 1 − η2. B ∼ N(M η, C(1 − η2)), when M → ∞, 6 (28) (29) (30) where N(µ, σ2) is a Gaussian distribution with mean µ and standard deviation σ, M is the number of nuclear spins, and C depends on M and An. Typically, C will have to be determined experimentally or by numerical simulations and we do not attempt to calculate it here, but the general form Eq. (30) does not depend on the specific QD. Since the fluctuations of the Overhauser field decrease with increasing polarization we now assume that the nuclear spins are polarized to degree η along n = (nx, ny, nz)T , where n2 z = 1. The assumption that the nuclear spin can be polarized along an arbitrary direction relies on experimental demonstrations54. This gives an effective magnetic field BHF = Bmaxηn, with variances x + n2 y + n2 z)](cid:1) . z z (σ2 x, σ2 y, σ2 x , Bext y , Bext x], Cy[1 − η2n2 z ) =(cid:0)Cx[1 − η2n2 y], Cz[1 − η2n2 (31) Together with the applied magnetic field Bext = )T the total effective magnetic field de- (Bext pends on 7 variables: Bext , nx, ny, nz, and η. In order to narrow the search for optimal parameters, we make the following observations: first, x and y are equivalent and we set ny = By = 0. Second, Fig. 3 shows that the concurrence C(τ ) has its maximum for Bz = 0 and we thus set Bext z = −Bmaxηnz. Finally we let tan ϕ = nx/nz and the total effective magnetic field is given by y , Bext x , Bext B = (ηBmax sin ϕ + Bext x )x, (32) and depends on the three free parameters η, ϕ, and Bext x . For η = 0.9 the result is shown in Fig. 4 and we find that for every ϕ there are two applied magnetic fields along Bx locally maximizing the concurrence. As expected from the discussion in the previous section, these occur when Bx = ±Bcr. We may thus set Bx = Bcr and study con- currence as a function of the polarization angle ϕ which is shown in Fig. 5, where we observe that the concur- rence is maximized by minimizing fluctuations along z. Finally we can investigate the concurrence as a function of polarization, shown in the inset of Fig. 5. We find that an increased nuclear spin polarization along z leads to an increased concurrence. We also see that a nuclear spin polarization perpendicular to z has almost no effect on the concurrence. This can be explained by the fact that the g-factors for the x- and y-directions are much smaller than the one along z. e c n e r r u c n o C 1 0.9 0.8 0.7 0.6 0.5 η = 0.5 η = 0.7 η = 0.9 e c n e r r u c n o C 0.9 0.8 0.7 0.6 0.5 x z 0 0.2 0.4 0.6 0.8 1 Polarization degree η 0 0.5 1 1.5 2 Polarization angle ϕ/π 5. The the concurrence under condition Figure (Bx, By, Bz) = (Bcr, 0, 0) for different degrees η and angles ϕ of the nuclear spin polarization. The nuclear spins are polarized along (sin ϕ, 0, cos ϕ) and we can observe a strong dependence on the angle. For a polarization along x (ϕ = ±π/2) the increase of concurrence is almost absent in comparison to a polarization along z (ϕ = nπ, n ∈ Z). Inset: Concurrence as function of the nuclear spin polarization along x (purple) and z (cyan). A nuclear polarization along z leads to a significant improvement in concurrence whereas this effect is all but absent in the case of polarization along x. Although polarization along any direction n would lead to a reduction of the fluctuations of the Overhauser field along n, because the in-plane g-factors are smaller than along the growth direction, in-plane fluctuations have less effect on the FSS. IV. SUMMARY 7 We have theoretically investigated the entanglement between two photons emitted from a cascade recombina- tion of a biexciton in a quantum dot. The entanglement was examined using the concurrence as a quantitative measure. We considered the two main sources of loss of concurrence, the FSS combined with a stochastic inter- mediate exciton lifetime and the stochastic Overhauser field. We found that the two sources of decoherence can- not be minimized independently of each other, and that the FSS is the dominant source of decoherence and must be minimized in order to maximize concurrence. Fur- thermore, we showed that reducing the uncertainty of the Overhauser field by nuclear spin polarization together with an applied magnetic field along a certain direction can improve the concurrence of the emitted light. The increase in entanglement depends strongly on the degree as well as the direction of nuclear spin polarization rela- tive to the growth axis of the QD. This effect is caused by the difference between in-plane g-factors and the g-factor along the growth direction. ACKNOWLEDGEMENTS We acknowledge funding from the Konstanz Center of Applied Photonics (CAP), DFG within SFB 767, and BMBF under the program Q.com-H. ∗ Current affiliation: Institut fur Festkorpertheorie, ISSN 0028-0836. Westfalische Wilhelms-Universitat, Munster, Germany † Current affiliation: Department of Physics, McGill Uni- versity, Montreal, Canada 1 N. Gisin, G. Ribordy, W. Tittel, and H. Zbinden, Rev. Mod. Phys. 74, 145 (2002). 2 P. G. Kwiat, K. Mattle, H. Weinfurter, A. Zeilinger, A. V. Sergienko, and Y. Shih, Phys. Rev. Lett. 75, 4337 (1995). 3 P. G. Kwiat, E. Waks, A. G. White, I. Appelbaum, and P. H. Eberhard, Phys. Rev. A 60, R773 (1999). 4 O. Benson, C. Santori, M. Pelton, and Y. Yamamoto, Phys. Rev. Lett. 84, 2513 (2000). 5 R. M. Stevenson, C. L. Salter, J. Nilsson, A. J. Bennett, M. B. Ward, I. Farrer, D. A. Ritchie, and A. J. Shields, Phys. Rev. Lett. 108, 040503 (2012). 6 C. Santori, D. Fattal, M. Pelton, G. S. Solomon, and Y. Ya- mamoto, Phys. Rev. B 66, 045308 (2002). 7 M. Bayer, G. Ortner, O. Stern, A. Kuther, A. A. Gor- bunov, A. Forchel, P. Hawrylak, S. Fafard, K. Hinzer, T. L. Reinecke, et al., Phys. Rev. B 65, 195315 (2002). 8 E. Kadantsev and P. Hawrylak, Phys. Rev. B 81, 045311 (2010). 9 E. Welander and G. Burkard, Phys. Rev. B 86, 165312 (2012). 10 R. M. Stevenson, R. J. Young, P. Atkinson, K. Cooper, D. A. Ritchie, and A. J. Shields, Nature 439, 179 (2006), 11 B. D. Gerardot, S. Seidl, P. A. Dalgarno, R. J. Warbur- ton, D. Granados, J. M. Garcia, K. Kowalik, O. Krebs, K. Karrai, A. Badolato, et al., Applied Physics Letters 90, 041101 (2007). 12 A. J. Bennett, M. A. Pooley, R. M. Stevenson, M. B. Ward, R. B. Patel, A. B. de la Giroday, N. Skold, I. Farrer, C. A. Nicoll, D. A. Ritchie, et al., Nat Phys 6, 947 (2010), ISSN 1745-2473. 13 A. J. Bennett, M. A. Pooley, R. M. Stevenson, I. Farrer, D. A. Ritchie, and A. J. Shields, Phys. Rev. B 84, 195401 (2011). 14 M. Ghali, K. Ohtani, Y. Ohno, and H. Ohno, Nat Commun 3, 661 (2012). 15 M. M. Vogel, S. M. Ulrich, R. Hafenbrak, P. Michler, L. Wang, A. Rastelli, and O. G. Schmidt, Applied Physics Letters 91, 051904 (2007). 16 A. Hogele, S. Seidl, M. Kroner, K. Karrai, R. J. Warbur- ton, B. D. Gerardot, and P. M. Petroff, Phys. Rev. Lett. 93, 217401 (2004). 17 J. D. Plumhof, V. Kr´apek, F. Ding, K. D. Jons, R. Hafen- brak, P. Klenovsk´y, A. Herklotz, K. Dorr, P. Michler, A. Rastelli, et al., Phys. Rev. B 83, 121302 (2011). 18 R. Trotta, E. Zallo, E. Magerl, O. G. Schmidt, and A. Rastelli, Phys. Rev. B 88, 155312 (2013). 8 45 R. Hafenbrak, S. M. Ulrich, P. Michler, L. Wang, A. Rastelli,and O. G. Schmidt, New J. Phys. 9, 315 (2007). 46 A. Muller, W. Fang, J. Lawall, and G. S Solomon, Phys. Rev. Lett. 103, 217402 (2009). 47 C. W. Lai, P. Maletinsky, A. Badolato, and A. Imamoglu, Phys. Rev. Lett. 96, 167403 (2006). 48 A. S. Bracker, E. A. Stinaff, D. Gammon, M. E. Ware, J. G. Tischler, A. Shabaev, A. L. Efros, D. Park, D. Gershoni, V. L. Korenev, et al., Phys. Rev. Lett. 94, 047402 (2005). 49 E. A. Chekhovich, M. N. Makhonin, K. V. Kavokin, A. B. Krysa, M. S. Skolnick, and A. I. Tartakovskii, Phys. Rev. Lett. 104, 066804 (2010). 50 M. Munsch, G. Wust, A. V. Kuhlmann, F. Xue, A. Lud- wig, D. Reuter, A. D. Wieck, M. Poggio, and R. J. War- burton, Nat Nano 9, 671 (2014), ISSN 1748-3387, letter. 51 R. Oulton, A. Greilich, S. Y. Verbin, R. V. Cherbunin, T. Auer, D. R. Yakovlev, M. Bayer, I. A. Merkulov, V. Stavarache, D. Reuter, et al., Phys. Rev. Lett. 98, 107401 (2007). 52 R. V. Cherbunin, S. Y. Verbin, T. Auer, D. R. Yakovlev, D. Reuter, A. D. Wieck, I. Y. Gerlovin, I. V. Ignatiev, D. V. Vishnevsky, and M. Bayer, Phys. Rev. B 80, 035326 (2009). 53 P.-F. Braun, B. Urbaszek, T. Amand, X. Marie, O. Krebs, B. Eble, A. Lemaitre, and P. Voisin, Phys. Rev. B 74, 245306 (2006). 54 M. N. Makhonin, K. V. Kavokin, P. Senellart, A. Lemaıtre, A. J. Ramsay, M. S. Skolnick, and A. I. Tartakovskii, Nat Mater 10, 844 (2011), ISSN 1476-1122. 55 K. Athreya and S. Lahiri, Measure Theory and Probability Theory, Springer Texts in Statistics (Springer, 2006), ISBN 9780387329031. 19 S. Seidl, M. Kroner, A. Hogele, K. Karrai, R. J. Warbur- ton, A. Badolato, and P. M. Petroff, Applied Physics Let- ters 88, 2031 (2006). 20 L. Sapienza, R. N. E. Malein, C. E. Kuklewicz, P. E. Kre- mer, K. Srinivasan, A. Griffiths, E. Clarke, M. Gong, R. J. Warburton, and B. D. Gerardot, Phys. Rev. B 88, 155330 (2013). 21 I. A. Merkulov, A. L. Efros, and M. Rosen, Phys. Rev. B 65, 205309 (2002). 22 A. V. Khaetskii, D. Loss, and L. Glazman, Phys. Rev. Lett. 88, 186802 (2002). 23 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), ISSN 0028- 0836. 24 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). 25 J. Fischer, W. A. Coish, D. V. Bulaev, and D. Loss, Phys. Rev. B 78, 155329 (2008). 26 C. Testelin, F. Bernardot, B. Eble, and M. Chamarro, Phys. Rev. B 79, 195440 (2009). 27 B. Eble, C. Testelin, P. Desfonds, F. Bernardot, A. Baloc- chi, T. Amand, A. Miard, A. Lemaıtre, X. Marie, and M. Chamarro, Phys. Rev. Lett. 102, 146601 (2009). 28 A. Dousse, J. Suffczynski, A. Beveratos, O. Krebs, A. Lemaitre, I. Sagnes, J. Bloch, P. Voisin, and P. Senel- lart, Nature 466, 217 (2010), ISSN 0028-0836. 29 M. A. M. Versteegh, M. E. Reimer, K. D. Jons, D. Dalacu, P. J. Poole, A. Gulinatti, A. Giudice, and V. Zwiller, Nat Comm 5, 5298 (2014), letter. 30 T. Huber, A. Predojevi´c, M. Khoshnegar, D. Dalacu, P. J. Poole, H. Majedi, and G. Weihs, Nano Lett. 14, 7107 (2014), letter. 31 R. M. Stevenson, R. J. Young, P. See, D. G. Gevaux, K. Cooper, P. Atkinson, I. Farrer, D. A. Ritchie, and A. J. Shields, Phys. Rev. B 73, 033306 (2006). 32 W. A. Coish and D. Loss, Phys. Rev. B 70, 195340 (2004). 33 H. Ribeiro and G. Burkard, Phys. Rev. Lett. 102, 216802 (2009). 34 W. A. Coish and J. Baugh, physica status solidi (b) 246, 2203 (2009), ISSN 1521-3951. 35 W. K. Wootters, Quant. Inf. and Comp. 1, 27 (2001). 36 M. Bayer, O. Stern, A. Kuther, and A. Forchel, Phys. Rev. B 61, 7273 (2000). 37 J. R. Schrieffer and P. A. Wolff, Phys. Rev. 149, 491 (1966). 38 R. Winkler, in Two- Dimensional Electron and Hole Systems (Springer-Verlag Berlin, Heidelberg, 2003). Spin-Orbit Coupling Effects 39 T. Nakaoka, S. Tarucha, and Y. Arakawa, Phys. Rev. B 76, 041301 (2007). 40 S. Takahashi, R. S. Deacon, A. Oiwa, K. Shibata, K. Hi- rakawa, and S. Tarucha, Phys. Rev. B 87, 161302 (2013). 41 V. V. Belykh, A. Greilich, D. R. Yakovlev, M. Yacob, J. P. Reithmaier, M. Benyoucef,and M. Bayer, Phys. Rev. B 92, 165307 (2015). 42 R. Zielke, F. Maier, and D. Loss, Phys. Rev. B 89, 115438 (2014). 43 G. Pfanner, M. Seliger, and U. Hohenester, Phys. Rev. B 78, 195410 (2008). 44 R. J. Young, R. M. Stevenson P. Atkinson, K. Cooper, D. A. Ritchie, and A. J. Shields, New J. Phys. 8, 29 (2006). Appendix A: The Probability Distribution of the The characteristic function of Y is given by Overhauser field in a QD 9 In this section we show that the probability distribu- tion of a weighted sum of N identical stochastic vari- ables approaches a Gaussian distribution when N → ∞. This is closely related to the well-known Central Limit Theorem55 in probability theory. Here we make small extension by considering a sum of identically distributed variables with different coefficients. The aim is to find the probability density function for the sum ϕ Y (t) = N Yn=1 ϕ X (ant) ≈ N Yn=1(cid:18)1 − a2 nσ2t2 2 (cid:19) , (A8) where ϕ X is the characteristic function of any of the Xn which we expand Taylor series to the second order in the second step. Now we consider Y = N Xn=1 anXn, (A1) and ϕ Y (t) ≈ exp" N Xn=1 ln(cid:18)1 − a2 nσ2t2 2 (cid:19)# where Xn are identically distributed stochastic variables with expectation value E[X] and variance σ2, and an are finite coefficients which, in general, also depend on N . For a QD we may choose the coefficients to match the electron position probability density function: N Xn=1 ≈ − ln(cid:18)1 − S1σ2t2 2N − an = Ψ (−l/2 + nl/N )2 N , (A2) where which implies that (A9) (A10) (A11) (A12) a2k n σ2kt2k 2k ∞ Xk=1 N a2 nσ2t2 2 (cid:19) = − Xn=1 Skσ2kt2k Xk=2 2kN 2k−1 , ∞ Sk =Z l 0 Ψ(x)4k dx. an = lim N→∞ Ψ (−l/2 + nl/N )2 N (A3) lim N→∞ N Xn=1 N Xn=1 =Z l/2 −l/2 Ψ (x) 2 dx = 1, where l is the size of the quantum dot. We introduce new variables with vanishing expectation value, Xn = Xn − E[Xn], (A4) and form the new sum Y = N Xn=1 an Xn, which is related to Y by We make the restriction that Ψ(x)2 is bounded on [0, l] which means that there is some constant C such that Ψ(x)2 ≤ C. This also ensures the existence of all Sk and for N approaching infinity we may keep only the first term in Eq. (A11) and we find ϕ Y (t) ≈ exp(cid:18)− S1σ2t2 2N (cid:19) , (A13) (A5) and from this we obtain the probability density function of Y as f Y (y) = √N σ√2πS1 exp(cid:18)− N y2 2S1σ2(cid:19) , (A14) Y = Y + Σ(N )E[X], (A6) where Σ(N ) = an. N Xn=1 which is a Gaussian distribution with variance S1σ2/N . S1 depends on the the coefficients an but the general form is always a Gaussian distribution regardless of what wave function is considered. (A7)
1809.01155	2	1809	2018-10-17T09:39:03	Parity Anomaly in the non-linear response of Nodal-Line Semimetals	[ "cond-mat.mes-hall", "hep-th" ]	Nodal-line semimetals are topological semimetals characterized by one-dimensional band-touching loops protected by the combined symmetry of inversion $\mathcal{P}$ and time-reversal $\mathcal{T}$ in absence of spin-orbit coupling. These nodal loops can be understood as a one-parameter family of Dirac points exhibiting the parity anomaly associated to $\mathcal{P}\mathcal{T}$ symmetry. We find that the parity anomaly also appears in the non-linear optical response of these systems in an analogous way to the linear response transport. We analyze the presence of a tilting term in the Hamiltonian as an element that does not spoil $\mathcal{P}\mathcal{T}$ symmetry: while it is $\mathcal{P}*\mathcal{T}$-symmetric, it breaks separately both $\mathcal{P}$ and $\mathcal{T}$ symmetries, allowing for the potential experimental observability of the linear and non-linear Hall conductivities in appropriate nodal-line semimetals.	cond-mat.mes-hall	cond-mat	Parity Anomaly in the non-linear response of Nodal-Line Semimetals Alberto Mart´ın-Ruiz1, 2, ∗ and Alberto Cortijo1, † 1Materials Science Factory, Instituto de Ciencia de Materiales de Madrid, CSIC, Cantoblanco; 28049 Madrid, Spain. 2Centro de Ciencias de la Complejidad, Universidad Nacional Aut´onoma de M´exico, 04510 Ciudad de M´exico, M´exico Nodal-line semimetals are topological semimetals characterized by one-dimensional band-touching loops protected by the combined symmetry of inversion P and time-reversal T in absence of spin- orbit coupling. These nodal loops can be understood as a one-parameter family of Dirac points exhibiting the parity anomaly associated to P ∗ T symmetry. We find that the parity anomaly also appears in the non-linear optical response of these systems in an analogous way to the linear response transport. We analyze the presence of a tilting term in the Hamiltonian as an element that does not spoil P ∗ T symmetry: while it is P ∗ T -symmetric, it breaks separately both P and T symmetries, allowing for the potential experimental observability of the linear and non-linear Hall conductivities in appropriate nodal-line semimetals. We also propose effective actions for both the linear and non-linear electromagnetic responses of tilted nodal-line semimetals. We find that the linear Hall-like response of tilted Nodal-line semimetals is an axion response due to the parity anomaly, extending the class of systems that display such electromagnetic response. I. INTRODUCTION lines in momentum space, Nodal line semimetals (NLSM) are three-dimensional semimetallic systems where the valence and conduction level cross each other at bands closest to the Fermi a one-dimensional in con- trast to a discrete set of points, as it happens in Weyl semimetals[1, 2]. These semimetals are endowed with a topological Z2 invariant that provides topological stabil- ity of this band structure. In practical grounds, this Z2 topological charge is the Berry phase π acquired by any curve in momentum space that is threaded by the nodal loop (if the curve is not threaded by the nodal loop, the Berry phase is simply zero). In absence of spin-orbit cou- pling (SOC), it is known that the combined symmetry of parity and time reversal symmetry P∗T is the symmetry that allows this topological protection[3]. In presence of SOC, the situation is richer and several combination of symmetries can stabilize the nodal loops[3, 4]. Until now, the only experimental signature of the Z2 topological invariant appears in the detection of the Berry phase in magnetotransport measurements[5 -- 16]. Then it is relevant to explore other physical phenomena where this invariant can be observed. In this regard, it has been proposed that the Z2 topological charge could also leave its imprint in other transport coefficients[17 -- Indeed, the Z2 topological invariant can be un- 19]. derstood as a three dimensional variant of the parity anomaly in two spatial dimensions[18, 19]. In the Quan- tum Field Theory context, the parity anomaly appears in two dimensional massless Dirac systems as the fail- ure of keeping the P ∗T symmetry after regularizing the theory in a gauge invariant way[20, 21]. In the effec- tive electromagnetic action, it manisfests as a radiatively induced Chern-Simons term that gives rise to a finite, ∗ [email protected] † [email protected] quantized anomalous Hall conductivity. Then, by notic- ing that both nodal loops in three dimensional P ∗ T symmetric NLSM and the nodal points in massless P ∗T symmetric Dirac fermions in two dimensions have codi- mension dc = 1, it is expected that they will share the same topological invariant[3, 19]. There is another equiv- alent, but more operational definition of anomaly: Given a theory displaying some symmetry, add to it some term that explicitly breaks this symmetry and compute sensi- ble quantities beyond the tree level (including quantum corrections). When sending back this term to zero, if these responses do not go to the responses obtained in ab- sence of the symmetry breaking parameter, we say that there is a quantum anomaly. In this operational defi- nition the enphasis is put on the adding-removing the symmetry breaking term and the need for regulators is hidden in the word sensible: finite and gauge invariant. This is the definition of anomaly that we will use in the present work. It has been stated in Refs.[18, 19] that NLSM exhibit a non zero Hall conductivity for each value of the polar angle φ that defines the one-dimensional loop in momen- tum space, so when considering all the values of φ, the total Hall current vanishes: (cid:90) e2 2π2 sign(m)k0E × JH = dφeφ = 0, (1) where the vector eφ = − sin φ x+cos φ y is the unitary po- lar vector in cylindrical coordinates in momentum space. Even at finite P ∗ T symmetry breaking mass term, the angular integration of (1) gives zero. The reason behind the cancellation of the Hall conductivity in Eq.(1) is be- cause each point at the nodal line has another point in the nodal loop related by inversion symmetry, so, when sum- ming over all the possible points labeled by φ, each pair of points related by inversion symmetry will contribute oppositely to the Hall conductivity, explaining the result (1)[22]. The scenario is reminiscent to what happens in graphene, where there are two nodal points in the Bril- louin Zone related by inversion symmetry. Adding the 8 1 0 2 t c O 7 1 ] l l a h - s e m . t a m - d n o c [ 2 v 5 5 1 1 0 . 9 0 8 1 : v i X r a same symmetry breaking parameter in both nodes breaks inversion but not time reversal symmetry, so the system displays a valley Hall effect (or spin Hall effect when con- sidering spin) instead of an anomalous Hall effect[23]. If a symmetry breaking mass with opposite sign is added to each node, then a quantum anomalous Hall effect is obtained[24]. Nevertheless, for any fixed angle φ, the corresponding point in the nodal loop displays the parity anomaly, and using the standard bulk-boundary corre- spondence associated to the Chern-Simons term in two spatial dimensions, we can use the parity anomaly for each value of φ to account for the drumhead surface states appearing in NLSM[25, 26]. The purpose of the present work is twofold: first, we study the non-linear Hall con- ductivity in NLSM to see how the parity anomaly also extends to non-linear effects, along the lines of previous studies[27, 28]. Second, we analyze the way to add terms that, respecting the P ∗T symmetry, might alter the bal- ance between inversion-related points in the nodal loop, and obtain non-vanishing results for the linear Hall con- ductivity (1) and for the non-linear Hall responses. In Section II we describe the effective k · p two-band model for NLSM including a tilting term, describing also its properties under P and T symmetries. III we review the fate of the linear Hall conductivity for NLSM in absence of the tilting term, and how this pre- cise term allows for a non-vanishing Hall conductivity, implying that in principle the term coming from the par- ity anomaly could be observed. In Sec.IV we study the Berry curvature dipole moment which is the element that governs the Hall-like non-linear optical response in ab- sence of interband transitions. In Sec.V we study the non-linear response associated to interband transitions and their relation with the parity anomaly. We end by commenting the obtained results in Sec. VI. In Sec. II. MODEL AND PRELIMINARIES In the present work we will use the following simpli- fied P∗T symmetry breaking model[25] (throughout this work we will use = 1): H(k) = σ0vt · k + ρ)σ1 + mσ2 + v3k3σ3, 0 − k2 (k2 1 Λ (2) 1 + k2 ρ = k2 where k2 2. The Pauli matrices σ represent an effective orbital basis, not necessarily the spin degree of freedom[4]. The parameter Λ is a momentum scale that comes from any particular lattice realization of the k · p model (2). Alternatively, one can work with a four- band model that is linear in all the momenta (this situa- tion corresponds to systems with SOC), but the relevant bands will be these that are crossed by the Fermi level, leaving with an anisotropic two-band model and a param- eter Λ depending on the SOC coupling constant (among other lattice parameters), after projecting out the high energy sector. Since the physics described in this work does not depend on this parameter, in the rest of the 2 FIG. 1. (color online) Cross sections of the Fermi surface for k3 = 0, for fixed chemical potential and tilt (µ < vtk0) for the case of vt = vt e1 (α = 0). We observe that the separation between the conduction and valence band closes as m → 0, as expected. Panel (a) shows the generic case for m (cid:54)= 0. At m = 0 (panel (b)) the two bands touch each other at points over the loop, defining the integration limits of the integral of the polar angle in momentum space. paper, we will make non dimensional variables absorbing this extra model-dependent parameter. The presence of the mass term mσ2 breaks P ∗ T symmetry. When m and vt are zero, the model consists in two bands that touch each other at a one-dimensional ring in momen- tum space (located in the k3 = 0 plane). In this case (m = vt ≡ vt = 0) the system is not only invariant un- der the combined P ∗ T symmetry, it is also symmetric under both P and T symmetries separately. However, the topological stability of the nodal line comes from the invariance of this combined symmetry, so any term in the Hamiltonian H(k) invariant under P∗T will not alter the topological protection of H(k). This is precisely the case of the term σ0vt · k. This shift in energies has been used in graphene to study the half-integer contribution of each Dirac point to the quantum Hall effect[29], and tilted nodal loops have been predicted to occur in alkaline-earth stannides, germanides, and silicides[30] and in other ma- terials displaying non-symmorfic symmetries[31, 32]. In the orbital basis employed in the model (2), the inver- sion and time reversal symmetries P and T are imple- mented by the operators P = σ3 and and T = Kσ3 respectively (K stands for complex conjugation and for both symmetries the momentum k has to reverse its di- rection), so P ∗T symmetry is implemented by P T = K, that is, the reality condition for the Hamiltonian H(k): H∗(k) = H(k). From this it is clear that the term σ0vt·k breaks time reversal and inversion symmetries separately (due to the flip of the sign of k) but leaves invariant P∗T symmetry. As mentioned in the introduction, the topo- logical invariant associated to the loop remains intact despite the presence of σ0vt · k. III. LINEAR HALL CURRENT Let us analyze how the presence of the term σ0vt · k modifies the Hall conductivity in Eq.(1). This analysis ϕ-ϕ+k1k2a)k1k2ϕ0b)3 FIG. 2. Evolution of the conductivity σ13 (in units of the quantum of conductance σ0 = e2/h) a) as a function of the dimensionless mass gap m ≡ mΛ/k2 0 = 0.3 and vt ≡ vtΛ/k0 = 0.6 and b) as a function of the tilt vt for fixed m = 0.01 and µ = 0.1. 0 for µ ≡ µΛ/k2 is not useful only to prepare us for the non-linear opti- cal/transport responses, but it is interesting by its own, since the orientation of the loop in momentum space will play a crucial role for the detection of the Hall effect in NLSM, as we will see. The part of the electric current density (as a response to an electric field E) associated to the parity anomaly can be computed, for a two-band model, with the follow- ing expression: JH = d3kf s 0 (k)E × Ωs k, (3) (cid:90) (cid:88) s=± e2 8π3 k and f s where Ωs 0 (k) are the Berry curvature and the equilibrium distribution function associated to the band s, respectively. We have computed the current (3) for zero and non-zero mass term m. The Berry curvature for the Hamiltonian (2) has been computed in Ref.[19]. Let us focus first on the m = 0 limit that is amenable of analytical tractability. In this limit, the Berry curvature of the band s can be written as k = sπ sign(m)δ(k3)δ(kρ − k0)eφ, Ωs (4) being independent of the tilt parameter vt. Due to the presence of the tilting vector vt, it is expected that for some range of chemical potentials, both conduction and valence bands will contribute to (3)[33]. In Fig. 1(a) we plot a cross section of the Fermi surface for k3 = 0 and a given nonzero value of the symmetry breaking parameter m. When m = 0, as shown Fig. 1(b), the electron and hole Fermi surfaces touch each other at two points over the nodal loop, defining the integration limits of the integral of the polar angle in momentum space. For k3 = 0 and kρ = k0, the zero-temperature equi- librium distribution functions for electrons and holes only depend on the polar angle in momentum space: 0 = Θ[s(µ − vtk0 cos(φ − α))], where α is the angle be- f s tween the vector vt and the horizontal axis. From these expressions, it is easy to see that the integration over the polar angle will not run over all the angles defining the nodal loop, but they will depend on the ratio µ/vtk0 and the angle α, as shown in Fig.1. Performing the angular integration, the Hall current takes the form (the details are presented in the Appendix A): (cid:115) JH = e2 2π2 sign(m)k0 1 − µ2 t k2 v2 0 E × (vt × t), (5) where t is the unit vector perpendicular to the plane containing the nodal loop in momentum space, and vt = vt/vt. We see that, while the chemical potential does not exceed vtk0 (physically, the conduction band does not extend over the whole nodal loop), we obtain a non- vanishing Hall current. While the Berry curvature still corresponds to the P ∗ T situation, the effect of the time reversal symmetry breaking tilting term (still preserving P ∗ T ) is to reduce the integration region over the Hall conductivity is constructed. The current (5) is obtained by using the m → 0 limit of k. We have also computed the m (cid:54)= the Berry curvature Ωs 0 situation numerically to observe the evolution of the current JH for small yet finite values of m. The relation of the current JH , the electric field, and the vectors vt and t is still valid as written in Eq.(5). In Fig. 2(a) (Fig. 2(b)) we plot the component σ13 as a function of m at fixed µ and tilt vt (as a function of tilt vt at fixed µ and m). It is interesting to observe that there is a value of the mass m above it the conductivity goes to zero, even for a non-zero tilt. We can understand this behavior by con- sidering the Fermi distributions (A2a) and (A2b) at non zero values of m. As it can be seen in Fig. 1, the in- tegration contour in (3) is defined by the values of the Fermi distribution functions for valence and conduction bands over the nodal loop. For the angle φ, we have the condition −1 ≤ cos φs ≤ 1 with s = 1 (s = −1) for elec- trons (holes). From the dispersion relations (18) over the nodal loop, this condition translates into the inequalities (again, for s = ±1): − (vtk0 − sµ) ≤ m ≤ (vtk0 + sµ). (6) 0.20.40.60.8m0.51.01.52.02.53.03.5σ13/σ0a)0.51.01.52.02.5vt1234σ13/σ0b)(cid:90) We can assume that vtk0 > µ (otherwise the conduction band covers all the angles φ and the valence band does not contribute, obtaining a vanishing result as it happens in (1)). It is easy to see that, if we want both bands to contribute to the integration in (3), both inequalities (6) must simultaneously hold, restricting the possible values of m to the interval [0, vtk0 + µ), explaining the result plotted in Fig. 2(a). We finish this section discussing which type of electro- magnetic action gives rise to the current (5). Since we are working in the infinite space case (no boundaries) the Hall current (5) can be written as the functional deriva- tive of an effective action term Γ(1) H /δAa (we write C∗ = e2 for simplicity) that takes the gauge invariant form: H , JH,a = δΓ(1) 1 − µ2 t k2 v2 0 2π2 sign(m)k0 (cid:113) Γ(1) H = d3xdtC∗µνρσqµAν∂ρAσ, (7) where we have defined the vector qµ = (0, k0 vt × t), hav- ing dimensions of inverse length. The main difference with the proposals available in the literature for untilted NLSM is that there is an explicit vector qµ that does not vanish at the end of the calculation[18, 19]. We identify this action as the axion electromagnetic action in three spatial dimensions, and conclude that, tilted NLSM display an axion term in their electromag- netic response due to the parity anomaly, in contrast to Weyl semimetals, where the the axion term appears due to the chiral anomaly[34, 35]. Also, the situation is dif- ferent to the case of axion insulators: Here we are dealing with a gapless system displaying the parity anomaly[36]. IV. NON-LINEAR HALL CURRENT In systems lacking inversion symmetry, there is a non-linear analog to the previously discussed anoma- lous Hall current[27, 28, 37, 38]. In three dimensional Weyl and Dirac semimetals, the generated photocurrents have attracted attention due to their connection with the chiral anomaly[38 -- 41]. Under the effect of an os- cillating electric field, E = eiωtE + e−iωtE∗, two non- linear currents can be generated: a photogalvanic cur- a = J (0) rent J 0 and a second harmonic current, a = e2iωtJ (2ω) J 2ω J (0) a + J (0)∗ a + e−2iωtJ (2ω)∗ c , J (2ω) a = χabc(ω)EbEce2iωt. a = χabc(ω)EbE∗ , where (8) a a Both currents are proportional to the same non-linear conductivity tensor χs abc(ω) that reads, in the collisionless limit (ωτ (cid:29) 1) as (cid:90) d3k χs abc(ω) = iadc e3 ω 8π3 f s 0 ∂bΩs d, from which it is customary to define the Berry curvature dipole moment (BCDM): (cid:90) d3k Dab = 8π3 f s 0 ∂aΩs b. (9) (10) 4 In what follows we will devote ourselves to the analysis of Dab, instead of the full expression (9). In the case of zero tilting, we will analyze the BCDM keeping finite the breaking parameter m and performing the limit m → 0 at the end of the calculations. A. Zero tilting For finite m term, the Berry curvature is (s = ±1): (11) Ωs k = s eφ, mv3kρ Λε3 k 0 − k2 (cid:113) k = −ε+ 3k2 3 + (k2 m2 + v2 ρ)2/Λ2 is the disper- where εk = sion relation for the conduction band (for the valence k = −εk) at zero tilting, and kρ is again band, ε− the radial momentum in cylindrical coordinates. Since Ω3 = 0, we directly have Da3 = 0. Also, by inspection, it is easy to see that D3a vanishes as well (the integrand is odd in k3). By symmetry, we also have that D11 = −D22. In parallel to the discussion of the Hall current done in Ref.[19], it is convenient to define first the one-parameter family of BCDM parametrized by the polar angle, Dab = ab in terms of two functions depend- (cid:82) dφDφ ab, and write Dφ ing only on the mass term and the chemical potential: 11 = −Dφ Dφ 22 = sin φ cos φQ(m, µ), 12 = X(m, µ) − cos2 φQ(m, µ), Dφ 21 = −X(m, µ) + sin2 φQ(m, µ). Dφ (12a) (12b) (12c) Without loss of generality, in absence of the tilting term, we will assume that the chemical potential µ crosses the conduction band (meaning that we will use the band s = 1), so the functions Q and X are defined as follows (the origin of these two functions can be seen in Appendix B) X(m, µ) = Q(m, µ) = − 6mv3 8π3Λ3 (cid:90) mv3 8π3Λ dkρdk3f0(εk) kρ ε3 k , (13) dkρdk3f0(εk) 0 − k2 ρ(k2 k3 ρ) ε5 k . (14) The integration domain is determined by the Fermi-Dirac distribution function f0(εk), as usual. It is illuminating to consider the situation when the In chemical potential lies slightly above the bandgap. this case, only momenta close to the nodal loop will con- tribute to the integrals. In this limit, the dispersion can 3 + (2k0qρ/Λ)2, with qρ = kρ − k0. Under this approximation, the func- tion X(m, µ) can be computed analytically (µ ≥ m): X(m, µ) → X0(m, µ) = be approximated by εk (cid:39) (cid:112)m2 + v2 (cid:18) 1 − m . (15) (cid:19) 3k2 1 8π2 sign(m) µ (cid:90) 5 FIG. 3. Plot of the functions 8π2X0(m, µ) (continuous line) and 8π2X(m, µ) (dashed line) for m = 0.1 (left). Plot of the functions 8π2Q(m, µ) (dashed line) and 2× 8π2X0(m, µ) (continuous line) as a function of the dimensionless chemical potential µ for m = 0.1 (middle) and m = 0.01 (right). This expression is the same that appears in the linear Hall current (1) before taking the m → 0 limit. In the zero mass limit, we obtain a finite value, which is a fin- gerprint of the parity anomaly. We have also evaluated numerically the full expression (13) to check the validity of (15). In Fig.3(left) we have plotted both X(m, µ) and X0(m, µ). For small values of the chemical potential and the mass term, both expressions are in agreement. Also, we can understand the behavior of D12 as a function of m plotted Fig. 3(right) along the lines discussed for the linear case at the end of Sec. III. Contrary to X(m, µ), the function Q(m, µ) is not peaked around the nodal loop so we have to evaluate it numerically. In Fig. 3(middle, right) we have plotted Q(m, µ) for two values of the parameter m, and compared with X0(m, µ). Crucially, we observe that when the mass parameter becomes smaller, the function Q(m, µ) ap- proaches to 2X0(m, µ). This implies that the function Q(m, µ) also becomes non-zero in the zero-mass limit due to the parity anomaly. We can write the φ-dependent BCDM components in a compact form by using the unit vector normal to the nodal loop t = (0, 0, 1) and the unitary vector perpen- dicular to t, r = (cos φ, sin φ, 0): Dφ ab = abctcX(m, µ) + bcdrarctdQ(m, µ). (16) Now we see that the situation is similar but somewhat different to the linear case. As we mentioned, the expres- sion in Eq.(1) is zero even for finite m after integrating over φ. In Eq.(16) we can integrate over φ at finite m obtaining a non-zero result: Dab(m, µ) = Dφ abdφ = π [2X(m, µ) − Q(m, µ)] abctc. (17) This is in agreement with the symmetry analysis per- formed in Ref.[28]. Only the antisymmetric part of the matrix Dab survives to the integration, so in three spatial dimensions, this antisymmetric pseudotensor transforms as a polar vector, in our case the normal to the nodal loop t playing such role. This also suggest the possibility that, for m (cid:54)= 0, the NLSM might become ferroelectric. (cid:90) Let us consider now the m = 0 case. We have seen that in this limit, the function Q(m, µ) approaches to 2X0(m, µ), taking the function X0(m, µ) the exact value X0(0, µ) = 1 8π2 . This means that each component of Dφ ab is non-zero, but after integrating over φ we find a vanishing value for Dab(0, µ) after using in (17) the fact that Q(0, µ) = 2X(0, µ). B. Non-zero tilting Let us compute the BCDM in presence of the tilting term. As we have seen in Sec. III, this term does not alter the expression of the Berry curvature Ωs k but strongly modifies the integration contours for electrons and holes through the modification of the dispersion relations, (cid:114) k = vt · k + s εs v2 3k2 3 + m2 + 1 Λ2 (k2 0 − k2 ρ)2. (18) Previously, in absence of tilting, it was possible to write down simple expressions for the BCDM in terms of the functions X(m, µ) and Q(m, µ), since in that case only the conduction band contributed to the integral (µ > m). We were then able to take the m = 0 limit for these expressions and find an analytical expression that allowed us to compare with the parity anomaly in the linear Hall current. A finite tilting makes both bands to contribute to the BCDM and the calculations at finite mass become more involved, precluding any analytical treatment. For this reason, we will analytically compute the BCDM at finite tilting directly at m = 0. We leave the full calculations of Dab to the appendix C. We quote here the final results. For D12 we obtain: sign(m) D12 = 2π2 vtk0 1 − µ2 v2 t k2 0 cos 2α. (19) We stress here that we have calculated the total D12, integrated over the angle φ. We see that the presence of a P breaking tilting term produces a non-zero result for the BCDM. In Fig. 4(a,b) we compare the numerical (cid:18) µ (cid:19)(cid:115) 012345μ0.00.20.40.60.81.08π2X(m,μ)8π2X0(m,μ)0.00.51.01.52.02.53.0μ0.00.51.01.52.08π2Q(m,μ)2×8π2X0(m,μ)0.00.51.01.52.02.53.0μ0.00.51.01.52.08π2Q(m,μ)2×8π2X0(m,μ)6 FIG. 4. Comparison of D12 computed numerically (continuous lines) at m (cid:54)= 0 and the analytical result (dashed lines), m = 0. (a): D12 as a function of vt for m = 0.01 and µ = 0.3. (b): D12 as a function of the dimensionless chemical potential µ for vt = 0.8 and m = 0.01. (c): D12 as a function of m for the same values of µ and vt as before. results at finite m with the analytical calculation with m = 0. In Fig. 4(c) we show the behavior of D12 as a function of m. In the limit of zero mass, the numerical value matches the analytical result. As it happened with the linear Hall conductivity, at sufficiently large values of m (for given values of µ > m and vt), D12 goes to zero. This can be understood along the same lines as in in Sec. III: For some critical value of m, the conduction band covers the whole nodal loop, leading to a zero result due to integration over φ. We also obtain that D21 = D12. In Appendix C we also provide details of the calculation of D11(= −D22). This component takes the form (cid:19)(cid:115) (cid:18) µ vtk0 D11 = − sign(m) 2π2 1 − µ2 v2 t k2 0 sin 2α. (20) Collecting all the components of the BCDM, we can write it in a way similar to a quadrupole moment, in terms of the vectors t and vt: (cid:18) µ (cid:19)(cid:115) Dab = sign(m) · (cid:2)2SaSb − S2(δab − δa3δb3)(cid:3) , 1 − µ2 v2 t k2 0 vtk0 4π2 · (21) where the vector S is defined as S = vt + t × vt. Inserting the BCDM (21) into Eq. (8) we find (again, in the collisionless regime) e3 J (0) = sign(m) ·(cid:2)E × E∗ − (S · E)(S × E∗) − (t · E)(t × E∗)(cid:3) , (22) 1 − µ2 v2 t k2 0 2π2iω vtk0 · such that the real-valued non-linear Hall current is ob- tained as J 0 = J (0) +J (0)∗, where J (0)∗ is the complex conjugate of (22). We can propose an effective electromagnetic action term that leads to the non-linear current J 0 in the same way as in the linear case for the current in Eq.(5). In the low energy (long wavelength) limit the corresponding lo- H reads (Fµν = ∂µAν − ∂νAµ is the cal effective action Γ(2) field strength): H = − Γ(2) µνραβAµIm(cid:0)FανF ∗ d3xdtλH (cid:1) . (cid:90) (23) βρ (cid:18) µ (cid:19)(cid:115) In the low frequency regime, we might compare the term λH µνραβ with the expression of the real-valued cur- rent J (0) + J (0)∗, and obtain, after defining the vectors Sµ = (0, S), tν = (0, t), and using the Lorentz metric ηρσ: (cid:18) µ (cid:19)(cid:115) λH µνραβ = e3 ω sign(m) · (cid:0)2µρσSνSσ − S2µνρ + S2µρσtσtν 1 − µ2 v2 t k2 0 vtk0 2π2 (cid:1) . η0αη0β · (24) This term is not topological in the same sense that it is the Chern-Simons term for the anomalous Hall conduc- tivity, although it is directly related to the Berry curva- ture. Also, we have to remember that the BCDM Dab is a property of the Fermi surface and it is dissipative in ori- gin (here we considered only the collisionless regime), in stark contrast to the linear Hall case that, although not acquiring quantized values, it is a non-dissipative prop- erty coming from both the filled bands and the Fermi surface[42, 43]. The remarkable result is that this non- linear response for tilted NLSM, although non topologi- cal, acquires contributions to the parity anomaly because it ultimately depends on the Berry curvature. V. INTERBAND EFFECTS In previous sections we have studied the appearance of the parity anomaly in non-linear responses for fre- quencies ω smaller than 2µ, that is, neglecting interband transitions. However, it is known that the Berry curva- ture modifies the interband processes that can contribute to the non-linear optical responses[44, 45]. Moreover, it has been stated recently that it is possible to obtain a quantized circular photogalvanic response in shifted Weyl semimetals[46]. In this section we study interband effects in the non-linear response of NLSM to see if they are sen- sible to the parity anomaly. 0.30.40.50.60.70.80.9vt0.0000.0050.0100.0150.0200.0250.030D120.20.40.60.8μ0.0050.0100.0150.0200.0250.030D120.20.40.60.8m0.0010.0020.0030.0040.0050.006D127 FIG. 5. (Left) Plot of g(ω, m) as a function of m for three different values of ω. (Right) Plot of the function g(ω, m) as a function of ω for three different values of m. A. Injection current The injection current is defined as the part of the pho- togalvanic effect that grows linearly with time: = βab(ω)[E(ω) × E∗(ω)]b. dJa dt (25) For a two-band model, as the one used in the present work, the tensor βab(ω) is defined as (cid:90) βab(ω) = iπe3 8π3 d3k ∂Ek,21 ∂ka Ωbδ(ω − Ek,21), (26) where Ek,21 stands for the difference between the con- duction and valence dispersion relations, Ek,21 = ε2,k − ε1,k = 2εk. As in previous sections, in the pertinent quantities the polar angle φ factorizes in the integrals so we will write the tensor βab as a function of this angle. Also, we will follow the strategy of computing βφ ab for non-zero values of m, and analyze the limit m → 0. The angle depen- dence can be casted in terms of the components of the vectors r and t, so the tensor βφ ab(ω) will take the follow- ing angular structure: βφ ab(ω) = ie3g(ω, m)rabcdrctd, (27) It is possible to obtain an analytical expression of the function g(ω, m) (see the details in Appendix D). In terms of the dimensionless variables ω = Λω/k2 0 and m = Λm/k2 g(ω, m) = − 4 m ω3 (cid:110)(cid:112)(ω/2)2 − m2 − 1 −(cid:2)(ω/2)2 − m2(cid:3) (cid:16)(cid:112)(ω/2)2 − m2 − 1 (cid:17)(cid:111) 0, this function reads: × arctan (28) . In Fig. 5 we have plotted the function g(ω, m). In the left panel we observe the evolution of g(ω, m) for several values of ω varying m. The most relevant information we extract is that this function vanishes when the sym- metry breaking parameter approaches to zero, even for a given angle φ. This is in stark contrast to the case of the BCDM, where, for a given φ, it remained non zero at m = 0. This implies that the injection current does not receive a finite contribution from the parity anomaly. In the right panel of Fig. 5 we have plotted g(ω, m) for sev- eral values of m, varying ω. While the most salient fea- √ ture of the function g(ω, m) is that it vanishes at m = 0, it also vanishes for frequencies ω ≥ 2 m2 + 1. Although the injection current does not receive con- tributions from the parity anomaly, it is finite at non- zero symmetry breaking parameter m, even after inte- grating over the angle φ. It is instructive to see the be- havior of (25) when circularly polarized light is applied to the NLSM. For right circular polarization, we have E × E∗ = −iE2 n, where n is the unit vector normal to the polarization plane. Inserting this in (27) and inte- grating over φ we have, in vector form: = πe3g(ω, m)E2 t × n, dJ dt (29) that is, provided that m is different from zero, an in- jection current appears perpendicular to the normal to the polarization plane and perpendicular to the vector normal to the plane containing the nodal loop. From an experimental perspective, the current in Eq.(29) appears to grow indefinitely with time. This does not occur in real settings, and after some characteristic time τ the interband processes should equilibrate with other dissipative effects, leading to a steady current Jτ = τ πe3g(ω, m)E2 t × n. (30) B. Shift current Another non-linear optical response that comes from inter-band transitions and it is susceptible of being mod- ified by the Berry curvature is the shift current, defined = σabb(ω)Eb(ω)Eb(ω). The tensor σabb can through J shif t be computed through the formula f ss(cid:48) 0 I abb ss(cid:48) δ(ω − Ess(cid:48)), σabb(ω) = e32π2 (cid:88) (cid:90) (31) a d3k s(cid:54)=s(cid:48) 012345m0.000.050.100.150.200.25g(ω,m)ω=10ω=8ω=505101520ω0.0000.0050.0100.0150.0200.025g(ω,m)m=0.15m=0.1m=0.05where, as before, Ess(cid:48) is the difference between the con- duction and valence bands in a two-band model, f ss(cid:48) 0 = 0−f s(cid:48) f s 0 is the difference between the equilibrium distribu- tion functions corresponding to the bands s and s(cid:48), and I abb ss(cid:48) = Im(ra ss(cid:48);a) is constructed from the interband matrix elements of the position operator, related to the off diagonal elements of the non-abelian Berry connection as ra ss(cid:48). The semicolon denotes the covariant derivative: rb ss(cid:48). In absence of tilt, we can write the model (2) as H = i σidi(k), the quantity I abb 12 = − m I abb 4εk ss(cid:48) − Aa (cid:19) ss(cid:48) = i(cid:104)s∂ka s(cid:48)(cid:105) ≡ Aa 12 can be written as abdj − ∂adj ∂2 ss(cid:48) − i(Aa ss(cid:48);a = ∂arb ∂bεk εk ijldl∂bdi (cid:80) ss)rb . (32) s(cid:48)srb (cid:88) ijl (cid:18) Noticing that d2 = m and d3 = v3k3, after some algebra, we obtain I abb 12 = mv3 4ε4 k ∂bεk(∂ad1δb1 − ∂bd1δa1), (33) 0 − k2 with the shorthand notation ∂a = ∂ka . Λ (k2 Now, using d1 = 1 ρ), we can evaluate the derivatives and perform the momentum integration. As before, we will consider first the case with fixed φ and later discuss the integral over this angle. By symmetry considerations, it is easy to see that the non-vanishing components of σφ abb(ω) are σφ 311(ω) = σφ 322(ω) = 2e3 ω 2e3 ω cos2 φg(ω, m), (34a) sin2 φg(ω, m), (34b) with g(ω, m) defined in (28). As in the case of the injection current, the shift current does not survive to the limit m → 0, implying that the parity anomaly does not leave its imprint in the shift current, either, even at fixed φ. If we integrate over φ, for finite values of the parameter m, the total shift current J shif t a is J shif t = π2 e3 ω g(ω, m)(E 2 1 + E 2 2 )t, (35) that is, while the injection current (29) circulates along the direction dictated by the intersection between the polarization plane and the plane containing the nodal loop, the shift current (35) is perpendicular to the nodal loop. Also, we note that there are other contributions to the shift current that come from kinematic reasons, anisotropies in the velocities, or anisotropies in the distri- bution function fk. These contributions are not related to the Berry curvature, so they are not expected to be modified by the parity anomaly, so we will no consider here. To conclude this section, we comment on the effects that the tilting term might cause in the injection and 8 shift currents through the quantities (26) and (32). As it is clear from previous sections, the tilt term does not modify the Berry connection and therefore the Berry cur- vature, but modify the dispersion relations for electrons and holes in the same manner. Then since the dispersion relations enter in these expressions through their differ- ences, we conclude that the only possible way to modify these responses is through the changes in the distribution functions. However, we have seen that the interband ef- fects do not show terms coming from the parity anomaly, so they will vanish in the limit m = 0 even if we introduce the changes induced by the tilting term in the distribu- tion functions, so we will not analyze these effects here. VI. CONCLUSIONS In the present work we have shown how the parity anomaly appears in the non-linear response in P∗T sym- metric NLSM. As it happens for the linear Hall conduc- tivity in these systems, we find that, if the nodal loop is not tilted, the parity anomaly appears in these non-linear Hall responses for finite values of the polar angle char- acterizing the nodal loop, but vanishes after integrating over this polar angle. This occurs because each point at the nodal loop possesses a inversion symmetric partner belonging to the nodal loop as well, so both contributions cancel out, leading to a net zero response. We have found that tilting the nodal loop preserves the P ∗ T symme- try, but breaks this balance between inversion symmetric points on the nodal loop, giving rise to a finite linear and non-linear Hall responses. While the parity anomaly shows up in the non-linear Hall responses for the range of frequencies where interband transitions can be neglected, it does not appear in these non-linear responses coming from interband transitions, as the injection and shift cur- rents. Here we have considered NLSM in absence of SOC, where the P ∗ T symmetry topologically stabilizes the nodal loop. In presence of SOC other combinations of symmetries might stabilize the nodal loop. Then, a per- tinent question is if some phenomenon similar to appear- ance of linear and non-linear Hall responses due to the parity anomaly will appear there. In this case, the mini- mal number of bands involved is the description of NLSM with SOC is four. However, the bandstructure will con- sist in two high-energy gapped bands, and two low energy bands forming the nodal loop, with an effective model for these two bands similar to (2), although the momentum dependence of d1(k) might be linear. In this case, all the results obtained in the present work are still valid, and what remains is to add any possible term coming from the gapped, high-energy sector that otherwise does not depend on the chemical potential. These symmetries are also compatible with the presence of tilting of the nodal loops[31, 32]. (cid:90) (cid:88) s=± e2 8π3 VII. ACKNOWLEDGEMENTS We thank Laszlo Oroszlany for comments on the manuscript. A. M. was supported by the CONACyT postdoctoral Grant No. 234774. A.C. acknowledges financial support through the MINECO/AEI/FEDER, UE Grant No. FIS2015-73454-JIN. and the Comunidad de Madrid MAD2D-CM Program (S2013/MIT-3007). Appendix A: Details of the computation of the Hall current (5) For finite tilt oriented forming an angle α with the kx 3), both valence and conduc- axis, vt = vt(cos α, sin α, vt tion bands contribute to the Hall conductivity in (3): σH ab = d3kf s 0 (k)abcΩs c. (A1) In the massless limit, we will use the expression (4) for Ωs k, and at zero temperature, the Dirac deltas in the ex- pression for the Berry curvature simplify the distribution functions for electrons and holes in (A1) as 0 = Θ(µ − vtk0 cos(φ − α)), f + 0 = 1 − Θ(µ − vtk0 cos(φ − α)). f− (A2b) (A2a) The valence band (A2b) contributes with two terms. The first one gives zero after integration over φ. The second term goes with an extra minus sign with respect to (A2a). This sign cancels out the extra minus sign from the Berry curvature Ωs k for the valence band, implying that in this case the contributions of electrons and holes add up, in- stead of cancelling each other: σH ab = e2k0 4π2 sign(m)abc dφΘ(µ − vtk0 cos(φ − α))eφ c . (A3) From this expression it becomes clear that the integration domain of the polar variable φ is not the whole nodal loop but those values of φ that satisfy cos(φ − α) ≤ µ/vtk0, and the integration limits in (A3) are φ± ≡ α ± φ0 = α ± arccos (cid:18) µ (cid:19) (A4) . vtk0 These angles are clearly depicted in Fig. 1a for the case α = 0 (tilting along the horizontal axis). If we choose E = E3 z, it is easy to perform the integral in (A3) with the previous integration limits, obtaining (cid:90) 9 If we denote t = (0, 0, 1) as the normal vector to the plane containing the nodal loop in momentum space, we have E × (vt × t) = (− cos αE3,− sin αE3, 0) so we can compactly write (cid:115) JH = e2 2π2 sign(m)k0 1 − µ2 v2 t k2 0 E × (vt × t). which is the expression (5). Appendix B: Computation of the functions X(m, µ), X0(m, µ), and Q(m, µ) . The functions X(m, µ) and Q(m, µ) originate from the derivatives of the Berry curvature at finite m ( the ex- pression (11)). Q(m, µ) comes from (cid:18) 1 (cid:19) (cid:0)k2 = ∂Ω1 ∂k1 = − mv3 Λ − 6mv3 Λ3 k2 ∂ ∂k1 ε3 k k2 ρ sin φ cos φ ε5 k (cid:1) . (B1) 0 − k2 ρ Separating the angular part, the integration of this ex- pression over kρdkρ and dk3 gives Q(m, µ). On the other hand, ∂Ω2 ∂k1 = mv3 Λ 0 − k2 ρ) − 6k2 ρ cos2 φ(k2 Λ2ε5 k 1 ε3 k . (B2) (cid:33) The first term of the right hand side is the origin of the function X(m, µ), and again we see in the second term the function Q(m, µ). The approximate function X0(m, µ) comes from ap- proximating the integral in X(m, µ) for values of µ slightly above the gap. Then we can approximate εk (cid:39) 3 + (2k0qρ/Λ)2, with qρ = kρ − k0: 3k2 (cid:112)m2 + v2 X0(m, µ) = mv3 8π3Λ f0 (m2 + v2 k0 + qρ 3 + ( 2k0 3k2 Λ qρ))3/2 dqρdk3. (B3) (cid:32) (cid:90) The distribution function f0 limits the integration to the range m < εk < µ. With the following change of vari- ables, qρ = Λ r sin θ,we have, after 2k0 integrating over θ: r cos θ and k3 = 1 v3 (cid:90) √ µ2−m2 (cid:18) 0 m 8π2 1 8π2 sign(m) (m2 + r2)3/2 dr r (cid:19) 1 − m µ , (B4) (cid:115) (cid:115) 13 = − e2 σH 4π2 k0 sign(m) 1 − µ2 v2 t k2 0 cos α, (A5a) X0(m, µ) = = 23 = − e2 σH 4π2 k0 sign(m) 1 − µ2 v2 t k2 0 sin α. (A5b) which is the result (15). D12 = − sign(m) 8π2 f s 0 (εs k)δ(k3)(kρ cos2 φδ(cid:48)(kρ − k0) + D12 = sign(m) 2π2 vtk0 1 − µ2 v2 t k2 0 cos 2α. (C9) (C2) The D21 component reads (cid:18) µ (cid:90) (cid:88) s=± Appendix C: Computation of the components of the BCDM at finite tilting We start with the detailed computation of D12. In cylindrical coordinates, we have D12 = 1 8π3 f s 0 (εk) ∂Ωs 2 ∂k1 kρdkρdk3dφ. (C1) Using the expression (4) for Ωs k and the expression of derivatives in cylindrical coordinates, ∂k1 = cos φ∂kρ − sin φk−1 ρ ∂φ we have (cid:90) (cid:88) s=± (cid:90) (cid:88) s s + sin2 φδ(kρ − k0))dkρdk3dφ, where we have expressed δ(cid:48) = ∂kρ δ. To simplify further, we use the distributional definition of the derivative of the Dirac delta: (C3) kρf0δ(cid:48)(kρ − k0) = −δ(kρ − k0)(f0 + kρf(cid:48) 0), s 8π2 D12 = sign(m) to obtain 0 kρ cos2 φ) · (f s ≡ D(1) dkρdk3dφδ(k3)δ(kρ − k0) · (cid:90) (cid:88) s 0 cos 2φ − f s(cid:48) 12 + D(2) 12 . The first piece D(1) 12 is similar to the integral defining the linear Hall conductivity (see Appendix A). We then use the expressions (A2a) and (A2b) for the conduction and valence band distribution functions that define the angular integration limits (A4). As in the linear case, the opposite sign of the Berry curvature for electrons and holes cancels out with the opposite sign present in their distribution functions, so the contributions for electrons and holes add up leading, after integrating over φ: (C4) (cid:19)(cid:115) 1 2π2 sign(m) (cid:18) µ 0 (εs(k)) =(cid:80) vtk0 D(1) 12 = tity(cid:80) ∂εs(k) s s∂kρ f s For the calculation of D(2) s s∂kρ εs(k)∂εf s = vt cos(φ − α) − s 12 , we need to compute the quan- 0 (εs(k)), with 0 − k2 ρ) 0 − k2 Λ2 (k2 ρ)2 (C6) and ∂εf0(ε) = −δ(µ − ε). We insert the previous expres- sions for each s in D(2) 12 , and, due to the extra minus sign from the Berry curvature in (4), we obtain 2kρ(k2 3 + 1 (cid:113) 3k2 v2 ∂kρ Λ2 , 12 = − sign(m)k2 D(2) 0 2π2Λ · δ(µ − vtk0 cos(φ − α)). dφ dkρ sign(k0 − kρ)δ(kρ − k0) · (cid:90) (cid:90) 1 − µ2 v2 t k2 0 cos 2α. (C5) 10 We perform the integral over kρ by substituting the Delta function by the Heat kernel regularization (Erf is the er- ror function): (cid:90) (cid:90) dkρ sign(k0 − kρ)δ(kρ − k0) = dkρ sign(k0 − kρ)e − (kρ−k0 )2 a2 = = lim a→0 = lim a→0 (cid:20) √ 1 a 1 2 π Erf (cid:18) k0 (cid:19) a = 0, (C8) implying that D(2) result 12 = 0. This leaves us with the final (cid:21) − 1 (cid:19)(cid:115) f s 0 (εk) ∂Ωs 1 ∂k2 kρdkρdk3dφ. (C10) D21 = 1 8π3 D21 = −(cid:88) (cid:90) Following the same steps done in the main text, this com- ponent takes the intermediate form: s sign(m) 8π2 s dkρdk3dφδ(k3)δ(kρ − k0) · · (f s 0 cos 2φ + f s(cid:48) 0 kρ sin2 φ). (C11) The first term in the parenthesis is the same as D(1) 12 , while the second term is the same appearing in D(2) 12 ex- cept for the replacement cos2 φ → sin2 φ. D(2) 12 vanishes independently of the angular dependence so does D(2) 21 , and we conclude that D21 = D12. Finally, let us evaluate the D11 component: (cid:90) (cid:88) D11 = 1 8π3 s=± = − sign(m) ∂Ωs 1 ∂k1 (cid:90) f s 0 (cid:88) ± 8π2 [sin φδ(kρ − k0)]. · ∂ ∂k1 kρdkρdk3dφ = dkρdk3dφf s 0 δ(k3)kρ · Applying the definition of ∂k1 in cylindrical coordinates and the definition of the derivative of the Delta function, we obtain dkρdk3dφ sin 2φδ(k3) · D11 = sign(m) (cid:90) (cid:88) 8π2 s=± · [2f s 0 ]δ(kρ − k0). 0 + kρf s(cid:48) (cid:90) (cid:88) As before, the integral involving f s(cid:48) (cid:18) µ (cid:19)(cid:115) sign(m) D11 = s=± 4π2 = − sign(m) 2π2 vtk0 1 − µ2 v2 t k2 0 (C12) (C13) 0 vanishes, so dφ sin 2φdk3δ(k3)f s 0 δ(kρ − k0)dkρ = sin 2α, (C14) (C7) after integrating over φ with the integration limits (A4). It is immediate to see that D22 = −D11. Appendix D: Computation of the function g(ω, m) Let us evaluate the function g(ω, m) for an arbitrary frequency. We start with the definition g(ω, m) = − 2mv3 Λ3 · Θ(kρ) kρdkρdk3 · 0 − k2 ρ(k2 k2 ρ) ε4 k δ (ω − 2εk) . First, we express the Dirac delta function in terms of the roots of the argument. To this end, we have to employ the formula δ(f (x)) = δ(x − xi) f(cid:48)(xi) , (D2) ∂εk ∂kρ and case we have ω = 2εk, wherefrom we obtain the roots ρ±(cid:48)± = ±(cid:48) k∗ 0 ∓ Λ k2 (ω/2)2 − m2 − v2 3k2 3. (D3) (cid:114) (cid:113) 11 Since our integral is only for positive values of kρ (it is a radius in polar coordinates), then we are left with only two roots, which corresponds to k∗ ρ+(cid:48)± > 0. Therefore we can write δ(kρ − k∗ ρ+) δ(ω − 2εk) = 2ε(cid:48) kkρ=k∗ δ(kρ − k∗ ρ−) 2ε(cid:48) kkρ=k∗ ρ± = k∗ Θ(k∗ Θ(k∗ ρ+) + ρ−). ρ− ρ+ (D1) (D4) The constraint imposed by the step function defines the limits of integration over the variable k3. We can also directly verify that (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)kρ=k∗ ρ± ρ±(cid:1) ρ±(cid:0)k2 0 − k∗ 2 Λ2εkk∗ ρ± = − 2k∗ εkkρ=k∗ ρ± = ω/2. = − 2k∗ ρ±(cid:0)k2 ρ±(cid:1) 0 − k∗ 2 Λ2(ω/2) (D5) (cid:90) (cid:88) i for an arbitrary (continuously differentiable) function f (x) with its roots xi giving f (xi) = 0. In the present Inserting these results into the integral expression and integrating with respect to kρ: (cid:1)(cid:34) 0 − k2 ρ (cid:90) k3 (cid:0)k2 (cid:90) (cid:2)k∗ 2 ρ ε4 k 1 ω3 δ(kρ − k∗ ρ+) 2ε(cid:48) kkρ=k∗ ρ−Θ(k∗ ρ+ ρ+) + Θ(k∗ ρ−)(cid:3) dk3. ρ+Θ(k∗ ρ+) − k∗ 2 (cid:35) δ(kρ − k∗ ρ−) 2ε(cid:48) kkρ=k∗ ρ− Θ(k∗ ρ−) dkρdk3 = (D6) ρ+ = 0 yields k3 ∈ [−k∗ 3, +k∗ 3], g(ω, m) = − 2mv3 Λ3 = − 4mv3 Λ (cid:112)(ω/2)2 − m2 − (k2 Now we determine the integration region for k3. On the one hand, the condition k∗ where k∗ 0/Λ)2. On the other hand, we can see that the condition k∗ 3 = 1 v3 ρ− = 0 has no real solutions for k3. Therefore, we are only left with the first integral: g(ω, m) = − 4mv3 Λω3 (cid:90) +k∗ 3 (cid:20) −k∗ 3 (cid:113) This integral can be performed in a simple fashion. The result is (cid:21) 0 − Λ k2 (ω/2)2 − m2 − v2 3k2 3 dk3. (D7) g(ω, m) = (cid:40) = − 4mv3 Λω3 2k2 0k∗ 3 − Λ v3 v3k∗ (cid:34) 3 (ω/2)2 − m2 − v2 (cid:113) (cid:110)(cid:112)(ω/2)2 − m2 − 1 −(cid:2)(ω/2)2 − m2(cid:3) arctan 3 + [(ω/2)2 − m2] arctan (cid:32) (cid:112)(ω/2)2 − m2 − v2 (cid:16)(cid:112)(ω/2)2 − m2 − 1 (cid:17)(cid:111) 3k∗2 v3k∗ 3 3k∗ 2 3 (cid:33)(cid:35)(cid:41) .(D8) (D9) After some algebraic simplifications and defining the dimensionless quantities m and ω, we obtain g(ω, m) = − 4 m ω3 [1] A. A. Burkov, M. D. Hook, and L. Balents, Phys. Rev. [2] N. P. Armitage, E. J. Mele, and A. Vishwanath, Rev. B 84, 235126 (2011). Mod. Phys. 90, 015001 (2018). 12 [3] C. Fang, Y. Chen, H.-Y. Kee, and L. Fu, Phys. Rev. B 92, 081201 (2015). [4] Q. D. Gibson, L. M. Schoop, L. Muechler, L. S. Xie, M. Hirschberger, N. P. Ong, R. Car, and R. J. Cava, Phys. Rev. B 91, 205128 (2015). [24] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). [25] Y. H. Chan, C. K. Chiu, M. Y. Chou, and A. P. Schnyder, Phys. Rev. B 93, 205132 (2016). [26] R. A. Molina and J. Gonz´alez, Phys. Rev. Lett. 120, 146601 (2018). [5] J.-W. Rhim and Y. B. Kim, Phys. Rev. B 92, 045126 [27] E. Deyo, L. E. Golub, E. L. Ivchenko, and B. Spivak, (2015). [6] J. Hu, Z. Tang, J. Liu, X. Liu, Y. Zhu, D. Graf, K. Myhro, S. Tran, C. N. Lau, J. Wei, et al., Phys. Rev. Lett. 117, 016602 (2016). [7] M. N. Ali, L. M. Schoop, C. Garg, J. M. Lippmann, E. Lara, B. Lotsch, and S. S. P. Parkin, Science Advances 2, e1601742 (2016). [8] J. Liu and L. Balents, Phys. Rev. B 95, 075426 (2017). [9] E. Emmanouilidou, B. Shen, X. Deng, T.-R. Chang, A. Shi, G. Kotliar, S.-Y. Xu, and N. Ni, Phys. Rev. B 95, 245113 (2017). [10] J. Zhang, M. Gao, J. Zhang, X. Wang, X. Zhang, M. Zhang, W. Niu, R. Zhang, and Y. Xu, Frontiers of Physics 13, 137201 (2017). [11] S. Pezzini, M. R. van Delft, L. M. Schoop, B. V. Lotsch, A. Carrington, M. I. Katsnelson, N. E. Hussey, and S. Wiedmann, Nature Physics (2017). [12] S. A. Ekahana, S.-C. Wu, J. Jiang, K. Okawa, D. Prab- hakaran, C.-C. Hwang, S.-K. Mo, T. Sasagawa, C. Felser, B. Yan, et al., New Journal of Physics 19, 065007 (2017). [13] M. Matusiak, J. R. Cooper, and D. Kaczorowski, Nature ArXiv e-prints (2009), 0904.1917. [28] I. Sodemann and L. Fu, Phys. Rev. Lett. 115, 216806 (2015). [29] H. Watanabe, Y. Hatsugai, and H. Aoki, Phys. Rev. B 82, 241403 (2010). [30] H. Huang, J. Liu, D. Vanderbilt, and W. Duan, Phys. Rev. B 93, 201114 (2016). [31] H. Huang, K.-H. Jin, and F. Liu, Phys. Rev. B 96, 115106 (2017). [32] S. A. Ekahana, S.-C. Wu, J. Jiang, K. Okawa, D. Prab- hakaran, C.-C. Hwang, S.-K. Mo, T. Sasagawa, C. Felser, B. Yan, et al., New Journal of Physics 19, 065007 (2017). [33] S. Ahn, E. J. Mele, and H. Min, Phys. Rev. Lett. 119, 147402 (2017). [34] A. A. Zyuzin and A. A. Burkov, Phys. Rev. B 86, 115133 (2012). [35] Y. Chen, S. Wu, and A. A. Burkov, Phys. Rev. B 88, 125105 (2013). [36] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, 195424 (2008). [37] J. E. Moore and J. Orenstein, Phys. Rev. Lett. 105, Communications 8, 15219 (2017). 026805 (2010). [14] H. Yang, R. Moessner, and L.-K. Lim, Phys. Rev. B 97, [38] Y. Zhang, Y. Sun, and B. Yan, Phys. Rev. B 97, 041101 165118 (2018). (2018). [15] C. Li, C. M. Wang, B. Wan, X. Wan, H.-Z. Lu, and X. C. Xie, Phys. Rev. Lett. 120, 146602 (2018). [39] A. Cortijo, Phys. Rev. B 94, 235123 (2016). [40] T. Morimoto, S. Zhong, J. Orenstein, and J. E. Moore, [16] L. Oroszl´any, B. D´ora, J. Cserti, and A. Cortijo, Phys. Phys. Rev. B 94, 245121 (2016). Rev. B 97, 205107 (2018). [41] H. Rostami and M. Polini, Phys. Rev. B 97, 195151 [17] S. T. Ramamurthy and T. L. Hughes, Phys. Rev. B 95, (2018). 075138 (2017). [18] A. A. Burkov, Phys. Rev. B 97, 165104 (2018). [19] W. B. Rui, Y. X. Zhao, and A. P. Schnyder, Phys. Rev. [42] F. D. M. Haldane, Phys. Rev. Lett. 93, 206602 (2004). [43] A. M. Essin, J. E. Moore, and D. Vanderbilt, Phys. Rev. Lett. 102, 146805 (2009). B 97, 161113 (2018). [44] V. I. Belinicher and B. I. Sturman, Soviet Physics Us- [20] N. A. Redlich, Phys. Rev. Lett. 52, 18 (1984). [21] N. A. Redlich, Phys. Rev. D 29, 2366 (1984). [22] H. B. Nielsen and M. Ninomiya, Nuc. Phys. B 185, 20 pekhi 23, 199 (1980). [45] J. E. Sipe and A. I. Shkrebtii, Phys. Rev. B 61, 5337 (2000). (1981). [46] F. de Juan, A. G. Grushin, T. Morimoto, and J. E. [23] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 Moore, Nat. Comm. 8, 15995 (2017). (2005).
1511.04962	1	1511	2015-11-16T14:12:34	Ultrafast photodetection in the quantum wells of single AlGaAs/GaAs-based nanowires	[ "cond-mat.mes-hall" ]	We investigate the ultrafast optoelectronic properties of single Al0.3Ga0.7As/GaAs-core-shell-nanowires. The nanowires contain GaAs-based quantum wells. For a resonant excitation of the quantum wells, we find a picosecond photocurrent which is consistent with an ultrafast lateral expansion of the photogenerated charge carriers. This Dember-effect does not occur for an excitation of the GaAs-based core of the nanowires. Instead, the core exhibits an ultrafast displacement current and a photo-thermoelectric current at the metal Schottky contacts. Our results uncover the optoelectronic dynamics in semiconductor core-shell nanowires comprising quantum wells, and they demonstrate the possibility to use the low-dimensional quantum well states therein for ultrafast photoswitches and photodetectors.	cond-mat.mes-hall	cond-mat	Ultrafast photodetection in the quantum wells of single AlGaAs/GaAs- based nanowires N. Erhard1,2, S. Zenger1, S. Morkötter1, D. Rudolph1, M. Weiss2,3, H. J. Krenner2,3, H. Karl3, G. Abstreiter1,2,4, J. J. Finley1,2, G. Koblmüller1, and A. W. Holleitner1,2 1 Walter Schottky Institut and Physics Department, Technical University of Munich, D-85748 Garching, Germany 2 Nanosystems Initiative Munich (NIM), Schellingstr. 4, D-80799 München, Germany Institute of Physics, Universität Augsburg, Universitätsstr. 1, D-86135 Augsburg, Germany Institute for Advanced Study, Technische Universität München, D-85748 Garching, Germany 3 4 Abstract: We investigate the ultrafast optoelectronic properties of single Al0.3Ga0.7As/GaAs-core-shell-nanowires. The nanowires contain GaAs-based quantum wells. For a resonant excitation of the quantum wells, we find a picosecond photocurrent which is consistent with an ultrafast lateral expansion of the photogenerated charge carriers. This Dember-effect does not occur for an excitation of the GaAs-based core of the nanowires. Instead, the core exhibits an ultrafast displacement current and a photo-thermoelectric current at the metal Schottky contacts. Our results uncover the optoelectronic dynamics in semiconductor core- shell nanowires comprising quantum wells, and they demonstrate the possibility to use the low-dimensional quantum well states therein for ultrafast photoswitches and photodetectors. Keywords: GaAs/AlGaAs-core-shell nanowires, quantum well, ultrafast photodetector, ultrafast photoswitch 1 Semiconducting nanowires have attracted increasing attention as building blocks for optoelectronic devices such as solar cells,(1)(2)(3) photodetectors,(4)(5)(6)(7) LEDs,(8)(9)(10) single photon emitters,(11)(12)(13) and lasers.(14)(15) Nanowires further allow to integrate quantum confined electron states into semiconductor circuits. Such states can have two-dimensional (2D)(16)(17), one-dimensional (1D),(16)(17)(18) and zero- dimensional (0D) character(11)(12)(13)(18)(19)(20)(22), which are isolated from the free surface and host the active device region. Among the various different material systems, GaAs-AlGaAs nanowires are a particularly interesting model system for radial 2D quantum wells (QWs), since the embedded QWs are nearly lattice- matched to the matrix material with negligible strain and piezoelectric effects.(23) So far, GaAs-AlGaAs based QWs in nanowires have been studied with respect to their optical, electronic and structural properties using e.g. photoluminescence spectroscopy(3) and inelastic light scattering.(2) However, investigation of the carrier dynamics and electrical transport properties are largely hampered for NWs consisting of both a bulk-like (3D) GaAs core and a (2D) QW. For all-electrical experiments, charge carriers have to be introduced by doping. Since multiple channels cannot be doped selectively, disentangling the transport characteristic of multiple channels is highly non-trivial.(24) Here, we apply a picosecond photocurrent spectroscopy to single GaAs-AlGaAs based QW-nanowires.(25) Since the core and the QWs are made out of GaAs in such nanowires, an optical excitation generates electrons and holes in both. We reveal that photocurrent spectroscopy allows to resolve the 3D joint density of states of the core states. In turn, the photocurrent response of the QWs shows up as a distinct, extra photocurrent response. Therefore, the QW states can be optoelectronically addressed and exploited for photodetection. Using this technique, we demonstrate that the QW states in GaAs-AlGaAs core-shell nanowires can be exploited in ultrafast photoswitches and photodetectors. In particular, we find that the quantum well states exhibit a picosecond optoelectronic response at cryogenic temperatures, which we interpret in terms of a lateral photo-Dember effect.(26)(27) This effect originates from the different diffusivity of electrons and holes in III-V semiconductor components. The diffusivities are enhanced at low dimensions.(28) Therefore, the QW states are particularly suited for such a picosecond photocurrent response. We characterize the optoelectronic properties of the QW nanowires using energy-, space-, and time-resolved photocurrent measurements. Our results may pave the way to use quantum well states in core-shell nanowires for applications in high-speed optoelectronic circuits. The 10 µm long nanowires are grown on a Si (111) substrate by molecular beam epitaxy (Figure 1a and supporting information).(17) Figure 1b schematically shows the QW-nanowire structure. The nominally 2 undoped nanowires consist of a GaAs-core (green) with a diameter of 60-80 nm which is surrounded by a 30 nm thick Al0.3Ga0.7As-shell (light yellow), followed by a 2 nm thick GaAs-QW layer (green) and a 30 nm thick Al0.3Ga0.7As-shell (light yellow). A capping layer of 10 nm GaAs (green) prevents the oxidation of the outer AlGaAs-shell. The nanowires are mechanically transferred onto a preprocessed sapphire substrate and contacted by evaporated Ti/Au contacts with a thickness of 10 nm/90 nm without annealing (Figure 1c). The contacts form lateral, coplanar strip lines with a total length of ~58 mm, a width of 5 µm and a separation of 6 µm. The strip lines are needed to perform the time-resolved photocurrent experiments,(29)(30)(31) which are discussed below. Due to the nanowire composition, two different metal/nanowire contacts are formed: (i) At the nanowire base (white triangle in Figure 1c), the GaAs-core and the QWs are in direct contact to the Ti/Au. (ii) At the nanowire tip (black triangle), only the GaAs cap is in direct contact to the Ti/Au, while the GaAs-based QWs and core are further separated from the metals by the AlGaAs shell(s). The I-V- characteristics of the nanowire-circuits have an s-shape at room temperature (Figure 2a), which is consistent with having Schottky-contacts at both the nanowire base and the tip.(32) The overall dark current IDark is small (~0.5 nA at 5 V), which can be explained by the fact that the nanowires are nominally undoped in combination with the presence of Schottky contacts. The dark current drops with decreasing temperature such that we do not measure any current signal above noise level for temperatures below 100 K. In a next step, we detect two-dimensional photocurrent maps of the contacted NWs (Figure 2b and supporting information). In particular, we measure the time-averaged photocurrent IPhoto at a laser energy of 1.7 eV, zero bias, and a bath temperature Tbath of 8 K (Figure 2c); i.e. a temperature and bias at which no dark current is detectable. The chosen laser energy is smaller than the band gap of the AlGaAs-shells (~1.9 eV). Hereby, charge carriers are predominantly excited in the GaAs-core and the GaAs-QWs of the nanowires. We observe a positive time-averaged photocurrent close to the nanowire base (white triangle in Figure 2c). This finding is consistent with the expectation that at this position, the core and the QWs are in contact to the metals. At the nanowire tip (black triangle), the signal of the time-averaged photocurrent is not detectable by time-averaged photocurrent measurements. The details of the photocurrent generation mechanisms at both the base and the tip of the nanowire will be discussed below when we present the time- resolved photocurrent data. We note that the spatial dependence of the photocurrent in Figure 2c corresponds to the laser spot shape with a diameter of about 5 µm. From the photocurrent direction 3 (photogenerated electrons moving to the left), we can deduce that the Schottky contact at the nanowire base is slightly n-type (supporting information). To explore the dimensionality of the electronic states within the GaAs-core and the GaAs-QWs, we measure the time-averaged photocurrent IPhoto as a function of the laser excitation energy (Figure 3a). For each energy, a spatially resolved photocurrent map is detected similar to the one in Figure 2c. The maps are fitted by two-dimensional Gaussians, and the corresponding maximum photocurrent is plotted as a function of the laser energy in Figure 3a. The depicted errors are fit errors. For energies below 1.5 eV, we do not observe any photocurrent signal which is consistent with the fact that GaAs has a band gap of ~1.51 eV at 10 K.(33) In the range between 1.5 eV and 1.7 eV, the photocurrent increases with excitation energy, exhibiting a square root dependence. This finding is consistent with a 3D joint density of states and with the fact that the GaAs-core diameter of 60-80 nm exceeds the Bohr radius of ~12 nm in GaAs. In other words, the GaAs-core exhibits 3D electronic states. At an excitation energy of ~1.68 eV, we observe a step-like increase of the photocurrent (red area in Figure 3a). We interpret it to correspond to the optical excitation of the QW-states. When we subtract the fitted square root dependence representing the GaAs-core from the overall photocurrent of the nanowire, we can extract an extra photocurrent within the quantum well IQW (inset of Figure 3a). Within the experimental error, IQW exhibits an E-1/2 behavior which is characteristic for a 1D density of states. For a two-dimensional QW, one would expect a constant density of states. In fact, the GaAs-QW forms a hexagonal tube within the nanowire. Such a QW-tube exhibits electron states which are usually a superposition of 2D states (electrons confined to the six facets of the hexagonal QW-tube) and 1D states (electrons confined to the six edge-channels of the hexagonal QW tube). For the ground state, the electrons are confined to the 1D edge-channels of the QW-tube which principally matches the observed behavior of IQW.(16)(17) However, for thin QWs as present in our sample, thickness fluctuations and the associated localization of excitons in “natural” quantum dots are expected to play the dominant role.(18)(34) For excitation energies just below the lowest QW transition energy, the absorption in the QW is enhanced due to the formation of quantum dot excitons. This effect is more pronounced in 2D structures like the GaAs- QWs and does not occur for the 3D GaAs core. Therefore, the energy dependence of IQW can also be explained by an enhanced absorption due to exciton effects in the QW. As shown in experiments performed on NWs from the same growth run presented in Ref.(34), carrier transport is well preserved in the QW continuum. Activation of carriers from interface fluctuations to the 2D continuum is highly efficient due to the 4 shallow confinement of these QDs even at low temperatures. In a photocurrent experiment, absorption in such localized states occurs and gives rise to the observed deviation of the ideal 2D density of states. Thus we can conclude, that the measured photocurrent indeed comprises photogenerated, but dissociated electron-hole pairs in the QW. These charge carriers are in fact mobile, such that a current can be detected in the overall macroscopic circuit. Figure 3b shows the photoluminescence of another QW-nanowire from the same wafer. Indeed, the onset of the photoluminescence at 1.51 eV corresponds to the band gap of the GaAs-core. A set of emission peaks (red) occurs at ~1.7 eV which matches the lowest transition energy for the extra photocurrent from the QW (red area in Figure 3a). The sharp photoluminescence emission lines are consistent with the expectation of localized excitons. Localization may originate from a combination of local thickness and alloy fluctuations in the QW and crystal structure,(18)(34)(35) i.e. with twinning defects between wurtzite and zinc-blende segments. (36) We now turn to the ultrafast optoelectronic dynamics occurring in the GaAs-core of the nanowires. We use an on-chip THz-time domain photocurrent spectroscopy, where an optical femtosecond pump laser excites the electronic states within the nanowire. This laser is the same as in the presented time-integrated photocurrent measurements in Figures 2c and 3a. Since the contacts form strip lines, the photocurrents give also rise to electromagnetic transients in the metal strip lines with a bandwidth of up to 2 THz. The transients run along the strip lines, and they are detected on-chip by a time-delayed optical femtosecond probe pulse in combination with an Auston-switch.(25) We use ion-implanted amorphous silicon for this ultrafast photodetector with a time-resolution of about 1 ps. (37)(29)(30)(31) The current ISampling across the Auston-switch is sampled as a function of the time-delay t between the two laser pulses, and it is directly proportional to the ultrafast photocurrents in the nanowire.(29)(30)(31) Hereby, the photocurrents in the nanowire can be measured with a picosecond time-resolution. Figures 4a and 4b show ISampling vs t for resonant excitation of the GaAs-core (Fig. 4a with Elaser = 1.53 eV at Tbath = 8 K and Fig. 4b with Elaser = 1.42 eV at Tbath = 296 K). In each panel, the bottom trace corresponds to a laser excitation at the nanowire base (white triangle). The top trace relates to the nanowire tip (black triangle). The traces in-between correspond to the positions marked by black dots in Figure 2c. Each curve of ISampling is fitted by a red line which has two independent fit-components: (i) A Lorentzian with a full width of half maximum (FWHM) of ~1.5 ps and (ii) 5 an exponentially modified Gaussian with a decay time of ~5 ps. First, we discuss the Lorentzian response. It can be explained by an ultrafast displacement current with the following current density (29)(30) 𝒋𝐃𝐢𝐬𝐩𝐥𝐚𝐜𝐞𝐦𝐞𝐧𝐭 = 𝜀0𝜀 ∂𝑬 ∂𝑡 , (1) with 𝜀 (𝜀0) the relative (vacuum) permittivity. In this ultrafast process, photogenerated charge carriers redistribute within the semiconductor nanowire to locally decrease electric fields E.(29)(38)(39) The data in Figure 4 are measured at zero bias. Since the sign of the fast displacement current is positive along the whole nanowire, we can deduce that an intrinsic electric field exists along the nanowire. This finding is consistent with a dominant Schottky barrier at the nanowire base as discussed in the context of Figure 2c. For nominally undoped nanowires, Schottky fields stretch far into the nanowires. The amplitude of the ultrafast displacement is given by the strength of the electric field in combination with the optical absorption at each position. The absorption is maximum in-between the two metal contacts, which explains the spatial amplitude variation of the ultrafast displacement current in Figures 4a and 4b. According to the fits (red lines), the second ultrafast photocurrent contribution can be described by an exponentially modified Gaussian. Intriguingly, the sign of this contribution does not change equally when the exciting laser is scanned across the nanowire (Figures 4a and 4b). This contribution is consistent with a photo-thermoelectric current generated at the nanowire base(29)(31)(40) 𝐼Thermo = (𝑆Nanowire − 𝑆Stripline)Δ𝑇/𝑅, (2) with Δ𝑇 the local temperature increase, R the total resistance of the electrical circuit and 𝑆Nanowire (𝑆Stripline) the Seebeck coefficient of the nanowire (Ti/Au contact). At the nanowire base, a direct interface exists to the electron bath in the metal contact. Consistent with this interpretation, all presented data (time-resolved and time-integrated) are independent of the applied bias voltage in an investigated range of -5 V to +5 V. In particular, a photo-thermoelectric current 𝐼Thermo is not expected to have a bias dependence as the Seebeck-coefficients are independent of the bias voltage. Intriguingly, the ultrafast photocurrent dynamics change at low temperatures, when the electronic states in the QWs are excited. Figure 4c depicts the time-resolved ISampling for Elaser = 1.59 eV at TBath = 8 K, and Figure 4d for Elaser = 1.70 eV and at TBath = 296 K. At room-temperature, we observe qualitatively the same time-resolved photocurrent behavior as for the resonant excitation in the GaAs-core. In turn, the room- 6 temperature photocurrent in the GaAs-QWs is consistent with an ultrafast displacement current and a photo- thermoelectric current generated at the nanowire base. At 8 K (Figure 4c), however, the photocurrent in the QW becomes negative for a laser excitation at the nanowire base (white triangle) and positive for an excitation at the nanowire tip (black triangle) (independent of the applied bias). The spatial sign change of the photocurrent is only observed at low temperatures. The origin of the optoelectronic dynamics in the QWs is revealed by power-dependent photocurrent measurements discussed in the next section. We perform a power-dependence of the pump laser pulse at TBath = 8 K and numerically compute the absolute area ASampling below the time-resolved photocurrent ISampling as in Figure 4. This amplitude is plotted in Figure 5a. When only the GaAs-core is excited (Elaser = 1.53 eV), ASampling exhibits an initial fast increase with a saturation for the pump laser power exceeding 100 µW. This saturation can be understood by an ultrafast displacement current which screens the electric field of a Schottky barrier.(41) The higher the laser power, the more charge carriers are generated to screen the electric field. In turn, the displacement current increases as a function of the laser power. As soon as sufficient charge carriers are photogenerated to totally screen the electric field in the NW, the displacement current saturates. We note that we detect an equivalent saturation behavior for all measurements at room temperature. Intriguingly, for an excitation of the QWs at low temperatures, we do not observe a saturation as a function of laser power. Instead, the amplitude of the photocurrent increases linearly with laser power in the examined range. The spatial sign change and the linear increase of the photocurrent in the QWs can be understood by a lateral photo-Dember effect.(27) Figures 5b and 5c schematically depict the photocurrent generation mechanism of such a photo- Dember effect. The exciting laser exhibits a Gaussian-like intensity distribution (red areas). The laser is partially shadowed by the Ti/Au contacts which are on top of the nanowire. Proportional to the incident laser intensity profile, charge carriers are photogenerated in the nanowire QW. Typically, the diffusivity of the electrons greatly exceeds the diffusivity of the holes in III-V semiconductor quantum wells.(28) Thereby, the electron cloud expands faster than the hole cloud (Figures 5b and 5c), which results in asymmetric electron and hole clouds shortly after the laser excitation. In turn, a so-called Dember current density jDember is generated towards the contacts with opposite sign at the nanowire base and tip (arrows in Figures 5b and c). Hereby, one can explain the sign change of the ultrafast photocurrent for the nanowire being excited at its base or tip. In addition, the photo-Dember effect also explains that at the nanowire base (white triangle), jDember has the opposite direction and sign than the displacement current. The latter has the direction of the 7 n-type Schottky field. In this interpretation, the photo-Dember effect has a maximum amplitude at the contacts, where the asymmetry of the electron and hole clouds is maximum due to the shadowing effect, although the optical absorption would be maximum in the center between the contacts. Consistently, at the center, we detect a superposition of ultrafast currents with cancelling directions (Figure 4c). We further note that a photo-Dember effect should give a linear power dependence when detected in a THz-time domain spectroscopy,(30) as we observe in Figure 5a. Generally, the well-known Einstein relation is modified in degenerate electron systems.(42) In 1D and 2D systems, the diffusivity-mobility-ratio is larger than in bulk material and it increases with charge carrier density.(28) For illumination with a 500 µW and 160 fs laser pulse, the nanowire QW can be considered to be highly degenerate leading to a high diffusivity-mobility-ratio. Furthermore, the diffusivity in GaAs-QWs increases with photon energy,(43) and the diffusion length increases with decreasing temperature and increasing laser power.(44) Hence, diffusion effects such as the lateral photo-Dember effect are enhanced for QW excitation at low temperatures. In agreement with this interpretation, we observe the lateral photo- Dember effect in the QW only at 8 K but not at room-temperature. At room-temperature, the photogenerated electrons and holes are very likely emitted through a thermionic emission from the QW into the GaAs-core within ~1 ps (cf. supporting information). In turn, directly after the photoexcitation, all photocurrents generated in the QW are suppressed because of a thermionic transmission from the QW to the core. At 8 K, however, the thermionic emission rate is strongly reduced, and the photogenerated charge carrier dynamics predominantly occur within the QWs. We note that the nanowires are designed such that the tunneling rates through the 30 nm thick AlGaAs barrier are too low to significantly influence the charge carrier distribution in the QW (cf. supporting information). The thermionic emission explains why we only observe the lateral photo-Dember effect at 8 K. In summary, we study the ultrafast photocurrents in AlGaAs/GaAs-core-shell nanowires containing 2 nm thick QWs. The photogenerated charge carriers enable a lateral photo-Dember effect at 8 K for laser excitation of the QWs. At room temperature, the data are consistent with photogenerated charge carriers being emitted within 1 ps to the GaAs-core. For a resonant excitation of the 3D core of the nanowire, other optoelectronic dynamics emerge. These are an ultrafast displacement current and a photo-thermoelectric current. Both currents originate at the interface of the nanowires to the metal contacts of the overall circuit. 8 The first current is driven by the electric fields at a Schottky barrier. The second one comprises the diffusion and drift of hot electrons to and from the interface. Our results reveal the optoelectronic dynamics in AlGaAs/GaAs-core-shell nanowires for ultrafast photoswitches and photodetectors. AUTHOR INFORMATION Corresponding Authors E-Mail: [email protected] The authors declare no competing financial interest. ACKNOWLEDGEMENTS We gratefully acknowledge financial support of the ERC-grant NanoREAL, the Deutsche Forschungsgemeinschaft via the SFB 631, the Emmy Noether Program (H.J.K. and M.W., KR3790/2-1), the Center for NanoScience (CeNS), and the International Graduate School of Science and Engineering (IGSSE) of the Technische Universität München. REFERENCES (1) Kempa, T. J.; Tian, B.; Kim, D. R.; Hu, J.; Zheng, X.; Lieber, C. M. Single and Tandem Axial p-i-n Nanowire Photovoltaic Devices. Nano Lett., 2008, 3456–3460. (2) Mariani, G.; Wong, P.-S.; Katzenmeyer, A. M.; Léonard, F.; Shapiro, J.; Huffaker, D. L. Patterned Radial GaAs Nanopillar Solar Cells. Nano Lett., 2011, 2490–2494. (3) Krogstrup, P.; Jørgensen, H. I.; Heiss, M.; Demichel, O.; Holm, J. V.; Aagesen, M.; Nygard, J.; Fontcuberta i Morral, A. Single-nanowire solar cells beyond the Shockley-Queisser limit. Nature Photonics, 2013, 306–310. (4) Hayden, O.; Agarwal, R.; Lieber, C. M. Nanoscale avalanche photodiodes for highly sensitive and spatially resolved photon detection. Nature Materials, 2006, 352–356. 9 (5) Thunich, S.; Prechtel, L.; Spirkoska, D.; Abstreiter, G.; Fontcuberta i Morral, A.; Holleitner, A. W. Photocurrent and photoconductance properties of a GaAs nanowire. Appl. Phys. Lett., 2009, 083111. (6) Gallo, E. M.; Chen, G.; Currie, M.; McGuckin, T.; Prete, P.; Lovergine, N.; Nabet, B.; Spanier, J. E. Appl. Phys. Lett. 2011, 98 (24), 241113, DOI:10.1063/1.3600061. (7) Bulgarini, G.; Reimer, M. E.; Hocevar, M.; Bakkers, E. P. A. M.; Kouwenhoven, L. P.; Zwiller, V. Avalanche amplification of a single exciton in a semiconductor nanowire. Nature Photonics, 2012, 455–458. (8) Haraguchi, K.; Katsuyama, T.; Hiruma, K.; Ogawa, K. Appl. Phys. Lett.GaAs p-n junction formed in quantum wire crystals. 1992, 60 (6), 745, DOI:10.1063/1.106556. (9) Tomioka, K.; Motohisa, J.; Hara, S.; Hiruma, K.; Fukui, T. Nano Lett. 2010, 10 (5), 1639–1644, DOI:10.1021/nl9041774. (10) Guo, W.; Zhang, M.; Banerjee, A.; Bhattacharya, P.G. Nano Lett. 2010, 10 (9), 3355–3359, DOI:10.1021/nl1011027x. (11) Claudon, J.; Bleuse, J.; Malik, N. S.; Bazin, M.; Jaffrennou, P.; Gregersen, N.; Sauvan, C.; Lalanne, P.; Gerard, J.-M. Nature Photonics 2010, 4 (3), 174–177. (12) Babinec, T. M.; HausmannBirgit J. M.; Khan, M.; Zhang, Y.; Maze, J. R.; Hemmer, P. R.; Loncar, M. Nature Nanotechnoloy [Online] 2010, 5 (3), 195–199. http://dx.doi.org/10.1038/nnano.2010.6. (13) Reimer, M. E.; Bulgarini, G.; Akopian, N.; Hocevar, M.; Bavinck, M. B.; Verheijen, M. A.; Bakkers, E. P. A. M.; Kouwenhoven, L. P.; Zwiller, V. Nat Commun 2012, 3, 737. (14) Mayer, B.; Rudolph, D.; Schnell, J.; Morkötter, S.; Winnerl, J.; Treu, J.; Müller, K.; Bracher, G.; Abstreiter, G.; Koblmüller, G.; Finley, J. J. Nature Comms 2013, 4, 2931, DOI:10.1038/ncomms3931. (15) Saxena, D.; Mokkapati, S.; Parkinson, P.; Jiang, N.; Gao, Q.; Tan, H. H.; Jagadish, C. Nature Photon 2013, 7 (12), 963–968, DOI:10.1038/nphoton.2013.303 10 (16) Fickenscher, M.; Shi, T.; Jackson, H. E.; Smith, L. M.; Yarrison-Rice, J. M.; Zheng, C.; Miller, P.; Etheridge, J.; Wong, B. M.; Gao, Q.; Deshpande, S.; Tan, H. H.; Jagadish, C.; Optical, Structural, and Numerical Investigations of GaAs/AlGaAs Core–Multishell Nanowire Quantum Well Tubes, Nano Lett., 2013, 13, 1016–1022. (17) Funk, S.; Royo, M.; Zardo, I.; Rudolph, D.; Morkötter, S.; Mayer, B.; Becker, J.; Bechtold, A.; Matich, S.; Döblinger, M.; Bichler, M.; Koblmüller, G.; Finley, J. J.; Bertoni, A.; Goldoni, G.; Abstreiter, G.; High Mobility One- and Two-Dimensional Electron Systems in Nanowire-Based Quantum Heterostructures, Nano Lett., 2013, 13, 6189–6196. (18) Shi, T.; Jackson, H. E.; Smith, L. M.; Jiang, N.; Gao, Q.; Tan, H. H.; Jagadish, C.; Zheng, C.; Etheridge, J. Emergence of Localized States in Narrow GaAs/AlGaAs Nanowire Quantum Well Tubes. Nano letters, 2015. (19) Loitsch, B.; Rudolph, D.; Morkötter, S.; Döblinger, M.; Grimaldi, G.; Hanschke, L.; Matich, S.; Parzinger, E.; Wurstbauer, U.; Abstreiter, G.; Finley, J.; Koblmüller, G., Advanced Materials 2015, 27, 2195. (20) Holmes, M. J.; Choi, K.; Kako, S.; Arita, M.; Arakawa, Y., Nano Lett. 2014, 14, 982-986 (21) Heiss, M.; Fontana, Y.; Gustafsson, A.; Wüst, G.; Magen, C.; O’Regan, D. D.; Luo, J. W.; Ketterer, B.; Conesa-Boj, S.; Kuhlmann, A. V.; Houel, J.; Russo-Averchi, E.; Morante, J. R.; Cantoni, M.; Marzari, N.; Arbiol, J.; Zunger, A.; Warburton, R. J.; Fontcuberta i Morral, A. Nat Mater [Online] 2013, 12 (5), 439–444. http://dx.doi.org/10.1038/nmat3557. (22) Weiss, M.; Kinzel, J. B.; Schülein, Florian J R; Heigl, M.; Rudolph, D.; Morkötter, S.; Döblinger, M.; Bichler, M.; Abstreiter, G.; Finley, J. J.; Koblmüller, G.; Wixforth, A.; Krenner, H. J. Dynamic acoustic control of individual optically active quantum dot-like emission centers in heterostructure nanowires. Nano Lett., 2014, 2256–2264. (23) Pistol, M.-E.; Pryor, C.; Band structure of core-shell semiconductor nanowires, Phys. Rev. B, 2008, 78. 11 (24) Chen, G. McGuckin T.; Hawley, C. J.; Gallo, E. M.; Prete, P.; Miccoli, I.; Lovergine, N.; Spanier, J. E.; Subsurface Imaging of coupled carrier transport in GaAs/AlGaAs core-shell nanowires, Nano Lett., 2015, 15 ,75-79. (25) Auston, D. H. Impulse Response of Photoconductors in Transmission Lines. IEEE Journal of Quantum Electronics, 1983, 639–648. (26) Dember, H. Über die photoelektronische Kraft in Kupferoxydul-Kristallen. Physikalische Zeitschrift, 1931, 554. (27) Klatt, G.; Hilser, F.; Qiao, W.; Beck, M.; Gebs, R.; Bartels, A.; Huska, K.; Lemmer, U.; Bastian, G.; Johnston, M. B.; Fischer, M.; Faist, J.; Dekorsy, T.; Terahertz emission from lateral photo-Dember currents, Opt Express, 2010, 18, 4939–4947. (28) Choudhury, S.; Singh, L. J.; Ghatak, K. P.; The Einstein relation in quantum wells and quantum well wires: simplified theory and a suggestion for experimental determination, Nanotechnology, 2004, 15, 180–188. (29) Prechtel, L.; Padilla, M.; Erhard, N.; Karl, H.; Abstreiter, G.; Fontcuberta i Morral, Anna; Holleitner, A. W.; Time-Resolved Photoinduced Thermoelectric and Transport Currents in GaAs Nanowires, Nano Lett., 2012, 12, 2337–2341. (30) Erhard, N.; Seifert, P.; Prechtel, L.; Hertenberger, S.; Karl, H.; Abstreiter, G.; Koblmüller, G.; Holleitner, A. W.; Ultrafast photocurrents and THz generation in single InAs-nanowires, Ann. Phys., 2013, 525, 180–188. (31) Erhard, N.; Holleitner, A. W.; Chapter 19 of Semiconductor Nanowires: Materials, Synthesis, Characterization and Applications. Woodhead Publishing Group: Sawston, 2014. arXiv:1409.0300 (2014). (32) Zhang, Z. Y.; Jin, C. H.; Liang, X. L.; Chen, Q.; Peng, L.-M.; Current-voltage characteristics and parameter retrieval of semiconducting nanowires, Appl. Phys. Lett., 2006, 88, 073102. (33) Rudolph, D.; Funk, S.; Döblinger, M.; Morkötter, S.; Hertenberger, S.; Schweickert, L.; Becker, J.; Matich, S.; Bichler, M.; Spirkoska, D.; Zardo, I.; Finley, J. J.; Abstreiter, G.; Koblmüller, G.; 12 Spontaneous Alloy Composition Ordering in GaAs-AlGaAs Core–Shell Nanowires, Nano Lett., 2013, 13, 1522–1527. (34) Weiss, M.; Schülein, F. J. R.; Kinzel, J.B.; Heigl, M; Rudolph, D.; Bichler, M.; Abstreiter, G.; Finley J.J.; Wixforth, A.; Koblmüller, G.; Krenner, H.J.; Radio frequency occupancy state control of a single nanowire quantum dot. J. Phys. D 2014, 47, 394011. (35) Hess, H. F.; Betzig, E.; Harris, T. D.; Pfeiffer, L. N.; West, K. W.; Near-field spectroscopy of the quantum constituents of a luminescent system, Science (New York, N.Y.), 1994, 264, 1740–1745. (36) Rudolph, D.; Schweickert, L.; Morkötter, S.; Hanschke, L.; Hertenberger, S.; Bichler, M.; Koblmüller, G.; Abstreiter, G.; Finley, J. J.; Probing the trapping and thermal activation dynamics of excitons at single twin defects in GaAs–AlGaAs core–shell nanowires, New J. Phys., 2013, 15, 113032. (37) Kastl, C.; Karnetzky, C.; Karl, H.; Holleitner, A. W. Nat Commun [Online] 2015, 6. http://dx.doi.org/10.1038/ncomms7617 (38) Zhou, X.; Numerical physics of subpicosecond electrical pulse generation by nonuniform gap illumination, IEEE J. Quantum Electron., 1996, 32, 1672–1679. (39) Dunn, G. M.; Walker, A. B.; Vickers, A. J.; Wicks, V. R.; Transient response of photodetectors, Journal of Applied Physics, 1996, 79, 7329–7338. (40) Fu, D.; Zou, J.; Wang, K.; Zhang, R.; Yu, D.; Wu, J.; Electrothermal Dynamics of Semiconductor Nanowires under Local Carrier Modulation, Nano Lett., 2011, 11, 3809–3815. (41) Prechtel, L.; Song, L.; Schuh, D.; Ajayan, P.; Wegscheider, W.; Holleitner, A. W.; Time-resolved ultrafast photocurrents and terahertz generation in freely suspended graphene, Nat Comms, 2012, 3, 646. (42) Landsberg, P. T.; Einstein and statistical thermodynamics. III. The diffusion-mobility relation in semiconductors, Eur. J. Phys., 1981, 2, 213–219. (43) Yoon, H.; Wake, D.; Wolfe, J.; Morkoç, H.; In-plane transport of photoexcited carriers in GaAs quantum wells, Phys. Rev. B, 1992, 46, 13461–13470. 13 (44) Lesiak, A.; Chwalisz, B.; Wysmolek, A.; Potemski, M.; Stepniewski, R., Thierry-Mieg, V.; Carrier Diffusion in GaAs/AlAs Type II Quantum Well, Acta Physica Polonica, 2005, 108, 755–760. 14 Figure Captions: Figure 1 (a) Scanning electron microscope (SEM) image and (b) schematic drawing of the QW-nanowires grown on a Si (111) substrate. The GaAs-core, the 2 nm QW and the GaAs capping layer are depicted in green. The AlGaAs inner and outer shell are shown in light yellow and the Si substrate in dark blue. (c) Schematic profile of the contacted QW-nanowire lying on a sapphire substrate. The hatched areas are the metal contacts to the nanowire. The black and white triangles mark the two different contact types at the top and the base of the nanowires, as discussed in the text. Figure 2 (a) Dark current voltage characteristic of a QW-nanowire contacted by two Ti/Au contacts (T ~ 296 K). (b) Optical microscope image of an investigated QW-nanowire. (c) Time integrated photocurrent IPhoto map of the nanowire. The dashed lines mark the position of the strip line contacts. Black dots indicate the nanowire orientation and the excitation positions for time-resolved photocurrent measurements (T ~ 8 K, USD = 0 V, PLaser = 500 µW, ELaser = 1.7 eV). Figure 3 (a) Time-integrated photocurrent IPhoto as a function of excitation energy. The black line is a square root fit through the data points. The red area marks the photocurrent generated in the quantum wells, which is depicted in the inset as photocurrent IQW. The red line in the inset has the shape of a 1D density of state function (T ~ 8 K, USD = 0 V, PLaser = 500 µW). (b) Photoluminescence signal of a second QW-nanowire grown on the same wafer (T ~ 4.2 K, ELaser = 1.88 eV). Figure 4 Time-resolved photocurrent Isampling at ~ 8 K for a laser excitation energy of (a) 1.53 eV and (b) 1.70 eV and at ~ 296 K for (c) 1.42 eV and (d) 1.59 eV. The different traces correspond to the laser excitation positions along the nanowire (black dots in Figure 2) starting at the position marked by an open black triangle (bottom trace). The data are made offset for a better visibility (USD = 0 V, PLaser = 500 µW, PProbe,1.53eV = 116 mW, PProbe,1.70eV = 70 mW, PProbe,1.42eV = 96 mW, PProbe,1.59eV = 96 mW). 15 Figure 5 (a) Absolute integrated amplitude ASampling of the time-resolved photocurrent Isampling as a function of laser power for an excitation at the position marked by a white triangle. The red line is a linear fit from the data obtained for excitation at 1.70 eV, the black line is a saturation fit from the data obtained for excitation at 1.53 eV. (b) and (c): Schematic drawing visualizing the lateral photo-Dember effect with the Dember current density jDember for the position marked by black and white triangles, respectively. 16 Figures: Figure 1 17 2µmAl2O3abcTi/AuTi/AuAlGaAsGaAs Figure 2 18 Figure 3 19 Figure 4 20 ISampling(pA)8412016ISampling(pA)0-1010Time (ps)0105Time (ps)0105ac ISampling(pA)120ISampling(pA)5010Time (ps)0105Time (ps)010515bd8K296K296K8K Figure 5 21 c 00│ASampling│(pAps)0.40.2200400Power (µW)1.70eV1.53eV0.68KjDemberabAlGaAsGaAs-QW+++++-----jDember+++Ti/Au+++++-----Ultrafast photodetection in the quantum wells of single AlGaAs/GaAs- based nanowires N. Erhard, S. Zenger, S. Morkötter, D. Rudolph, M. Weiss, H. Krenner, H. Karl, G. Abstreiter, J. Finley, G. Koblmüller, and A. W. Holleitner - Supplementary Material - S1: Nanowire growth parameters. The GaAs-QW-nanowires are grown in a solid source MBE chamber on a SiO2/Si(111) substrate patterned by nanoimprint lithography (1). The GaAs-cores of the nanowires are grown by catalyst-free vapor-solid (VLS) growth employing a substrate temperature of 610°C, a beam equivalent arsenic pressure BEP(As4) of 1.9 × 10-6 mbar , and a Ga flux of 0.25 Å/s for a duration of 150 min. After decreasing the substrate temperature to 490 °C and increasing the BEP to 3.5 × 10-5 mbar, the GaAs-cores are overgrown subsequently in situ by Al0.3Ga0.7As, GaAs, Al0.3Ga0.7As and a final GaAs capping layer with nominal widths of 30 nm, 2 nm, 30 nm and 10 nm, respectively. The nominal Al content x[Al] as well as the nominal thicknesses di are obtained from the measured flux ratios and calibrations as described in ref (2). S2: Energy dispersive X-ray (EDX) spectroscopy on a GaAs-QW-nanowire. We perform EDX measurements on the QW-nanowires to confirm the nanowire geometry presented in the main article. Therefore, the nanowires are drop-casted onto C-coated Cu grids and analyzed using a 200kV-JEOL2011 equipped with an EDAX Apollo XL detector. Supplementary Figure S1 a (b) shows a transmission electron microscope (TEM) image of the nanowire tip (base) of a second QW-nanowire grown on the same wafer. The EDX spectra in Supplementary Figure S1 c and d are taken within the area marked with a red circle in Figure S1a and b, respectively. The integration time is set to 100 s. For the nanowire tip, we deduce an Al content of (10 ± 2)% which supports the assumption that there exists an AlGaAs-shell on top of the nanowire. The nanowire base contains (3 ± 2)% Al which is due to the fact that the GaAs-cores are grown first into the pre-defined holes in the SiO2/Si(111) substrate. S3 Time-integrated photocurrent measurement. All presented optoelectronic measurements are performed under vacuum conditions at ~10-6 mbar. An objective of a microscope focuses the exciting laser light of a mode-locked titanium:sapphire laser onto the nanowire. The laser spot has a circular shape with a diameter of ~5 µm. The laser is chopped at a frequency of about 2.3 kHz resulting in a photocurrent 𝐼Photo = 𝐼on − 𝐼off detected with a lock-in amplifier utilizing the reference signal provided by the chopper. 𝐼on (𝐼off) is the current through the nanowire device when the laser is switched on (off). (3) Supplementary Figure S1: Transmission electron microscope (TEM) image of the GaAs-QW-nanowire tip (a) and base (b). (c) EDX spectra of the nanowire tip at the position black triangle in (a) and (d) of the nanowire base at the position white triangle in (b). S4: Band structure of a contacted GaAs-QW-nanowire. Supplementary Figure S2 sketches the assumed band structure along a GaAs-QW-nanowire contacted by two gold contacts consistent with the detected photocurrent. The sketch is consistent with all presented measurements. The GaAs-core (green) has a band gap of 1.51 eV at ~8 K (2) and forms an n-type Schottky barrier to the gold contact. The AlGaAs-shells (light yellow) have a higher band gap of about 1.92 eV (4) to 1.96 eV (5). For simplicity, we assume that the outer AlGaAs-shell together with the GaAs capping layer forms an n-type Schottky contact to the gold which is consistent with the s-shape of the dark current through the nanowire (Figure 2a). However, we note that as we excite only the GaAs in the nanowire, the photocurrent response is not significantly influenced by the contact type of the AlGaAs-shells. We solve the static Schrödinger equation for the 2 nm GaAs-QW using the effective masses for electrons and holes from ref.(6), and the Al0.3Ga0.7As parameters from ref.(7). Hereby, we obtain the transition energy for excitation from the first heavy (light) hole state to the electron ground state to be ~1.74 eV (~1.77 eV) at 0 K. The value is larger than the 1.68 eV observed in PL- and photocurrent measurements (cf. main article). However, for the calculation, we use bulk values for Al0.3Ga0.7As which may explain the deviation. The transition energy of 1.68 eV at 8 K is estimated to be ~1.59 eV (~780 nm) at 296 K using the Varshni- equation E(T)=E(0)-(αT2/(β+T)) (8). In turn, we use a laser excitation energy of 1.59 eV to resonantly excite the QW at room temperature which is also in accordance with the PL measurements in the main article (Figure 3). Supplementary Figure S2: (a) Schematic drawing of the contacted QW-nanowire lying on a sapphire substrate. The black and white triangles mark the two different contact types as discussed in the paper. GaAs is depicted in green and the AlGaAs in light yellow. The two dashed rectangles highlight the contact regions as sketched in Figure 5 of the main manuscript. (b) Sketch of the band structure of the contacted GaAs-QW-nanowire along the dashed blue line in (a). The bend, dotted arrow on the side panel highlights the corresponding direction within the nanowire. EC (EV) depicts the conduction (valence) band bottom (top) edge. The dashed line EF depicts the Fermi level. The dotted lines in the QW mark the first light hole, the first heavy hole and the electron ground state from bottom to top in the QW. Al2O3AuGaAsAlGaAsECEVEFab S5: Thermionic emission and tunneling from the QW into the core through the AlGaAs-barrier. The emission rate from thermionic tunneling rate 𝜏𝑒 𝑘𝐵𝑇 −1 can be estimated as:(9) 1/2 1 𝜏𝑒 = ( 2 ) 2𝜋𝑚𝑖𝑑𝑊 𝑒𝑥𝑝 (− Δ𝐸 𝑘𝐵𝑇 ) , with 𝑘𝐵 the Boltzmann constant, 𝑇 the temperature, 𝑚𝑖 the effective mass, 𝑑𝑊 the width of the QW, and Δ𝐸 the effective barrier height. The tunneling rate 𝜏𝑡 −1 from the first sublevel in the QW is given by ref.(10): 1 𝜏𝑡 = 𝜋ℏ 2𝑚𝑖𝑑𝑊 2 𝑒𝑥𝑝 (− 2𝑑𝐵√2𝑚𝑏𝑖Δ𝐸 ℏ ) , with ℏ the reduced Planck constant, 𝑑𝐵 the thickness of the barrier and 𝑚𝑏𝑖 the effective mass in the barrier. The calculated transmission times are summarized in the following table. At 300 K thermionic transmission is strong and the photogenerated holes are transmitted to the GaAs-core within the first ps. For 8 K, thermionic emission is strongly suppressed. In addition, the tunneling through the 30 nm thick AlGaAs- shell is weak. In turn, the photogenerated charge carriers stay in the QW within the investigated timescale of 100 ps. 𝜏𝑒 at 300 K [s] 𝜏𝑒 at 8 K [s] 4 × 10-13 5 × 10-14 9 × 10-13 1 × 1038 5 × 1019 4 × 1038 𝜏𝑡 [s] 3 × 10-4 3 × 10-8 2 × 10-3 electron light hole heavy hole REFERENCES (1) Hertenberger, S.; Funk, S.; Vizbaras, K.; Yadav, A.; Rudolph, D.; Becker, J.; Bolte, S.; Döblinger, M.; Bichler, M.; Scarpa, G.; Lugli, P.; Zardo, I.; Finley, J. J.; Amann, M.-C.; Abstreiter, G.; Koblmüller, G.; High compositional homogeneity in In-rich InGaAs nanowire arrays on nanoimprinted SiO2/Si (111), Appl. Phys. Lett., 2012, 101, 043116. (2) (3) (4) Rudolph, D.; Funk, S.; Döblinger, M.; Morkötter, S.; Hertenberger, S.; Schweickert, L.; Becker, J.; Matich, S.; Bichler, M.; Spirkoska, D.; Zardo, I.; Finley, J. J.; Abstreiter, G.; Koblmüller, G.; Spontaneous Alloy Composition Ordering in GaAs-AlGaAs Core–Shell Nanowires, Nano Lett., 2013, 13, 1522–1527. Erhard, N.; Holleitner, A. W.; Chapter 19 of Semiconductor Nanowires: Materials, Synthesis, Characterization and Applications. Woodhead Publishing Group: Sawston, 2014. arXiv:1409.0300 (2014). Lambert, B.; Caulet, J.; Regreny, A.; Baudet, M.; Deveaud, B.; Chomette, A.; Optical determination of the Al x Ga 1-x As energy gap variation versus the Al concentration in MBE-grown samples, Semicond. Sci. Technol., 1987, 2, 491–493. (5) Bosio, C.; Staehli, J.; Guzzi, M.; Burri, G.; Logan, R.; Direct-energy-gap dependence on Al concentration in Al_{x}Ga_{1-x}As, Phys. Rev. B, 1988, 38, 3263–3268. (6) Heiss, M.; Conesa-Boj, S.; Ren, J.; Tseng, H.-H.; Gali, A.; Rudolph, A.; Uccelli, E.; Peiró, F.; Morante, J. R.; Schuh, D.; Reiger, E.; Kaxiras, E.; Arbiol, J.; Fontcuberta i Morral, Anna; Direct correlation of crystal structure and optical properties in wurtzite/zinc-blende GaAs nanowire heterostructures, Phys. Rev. B, 2011, 83. (7) Adachi, S.; Properties of aluminium gallium arsenide. IEE, INSPEC: London, 1993. (8) (9) Varshni, Y. P.; Temperature dependence of the energy gap in semiconductors, Physica, 1967, 34, 149–154. Schneider, H.; Klitzing, K.; Thermionic emission and Gaussian transport of holes in a GaAs/Al_{x}Ga_{1-x}As multiple-quantum-well structure, Phys. Rev. B, 1988, 38, 6160–6165. (10) Fox, A. M.; Miller, D.; Livescu, G.; Cunningham, J. E.; Jan, W. Y.; Quantum well carrier sweep out: relation to electroabsorption and exciton saturation, IEEE J. Quantum Electron., 1991, 27, 2281– 2295.
1210.1971	2	1210	2013-03-10T19:10:01	Full-counting statistics for molecular junctions: Fluctuation theorem and singularities	[ "cond-mat.mes-hall" ]	We study the full-counting statistics of charges transmitted through a single-level quantum dot weakly coupled to a local Einstein phonon which causes fluctuations in the dot energy. An analytic expression for the cumulant generating function, accurate up to second order in the electron-phonon coupling and valid for finite voltages and temperatures, is obtained in the extended wide-band limit. The result accounts for nonequilibrium phonon distributions induced by the source-drain bias voltage, and concomitantly satisfies the fluctuation theorem. Extending the counting field to the complex plane, we investigate the locations of possible singularities of the cumulant generating function, and exploit them to identify regimes in which the electron transfer is affected differently by the coupling to the phonons. Within a large-deviation analysis, we find a kink in the probability distribution, analogous to a first-order phase transition in thermodynamics, which would be a unique hallmark of the electron-phonon correlations. This kink reflects the fact that although inelastic scattering by the phonons once the voltage exceeds their frequency can scatter electrons opposite to the bias, this will never generate current flowing against the bias at zero temperature, in accordance with the fluctuation theorem.	cond-mat.mes-hall	cond-mat	Full-counting statistics for molecular junctions: the fluctuation theorem and singularities Y. Utsumi,1, 2, ∗ O. Entin-Wohlman,3, 4, 2 A. Ueda,5 and A. Aharony3, 4, 2 1Department of Physics Engineering, Faculty of Engineering, Mie University, Tsu, Mie, 514-8507, Japan 2Kavli Institute for Theoretical Physics China, CAS, Beijing 100190, China 3 Physics Department, Ben Gurion University, Beer Sheva 84105, Israel 4 Raymond and Beverly Sackler School of Physics and Astronomy, Tel Aviv University, Tel Aviv 69978, Israel 5Faculty of Pure and Applied Sciences, Division of Applied Physics, University of Tsukuba, Tsukuba, Ibaraki, 305-8573, Japan (Dated: October 31, 2018) We study the full-counting statistics of charges transmitted through a single-level quantum dot weakly coupled to a local Einstein phonon which causes fluctuations in the dot energy. An analytic expression for the cumulant generating-function, accurate up to second order in the electron-phonon coupling and valid for finite voltages and temperatures, is obtained in the extended wide-band limit. The result accounts for nonequilibrium phonon distributions induced by the source-drain bias voltage, and concomitantly satisfies the fluctuation theorem. Extending the counting field to the complex plane, we investigate the locations of possible singularities of the cumulant generating- function, and exploit them to identify regimes in which the electron transfer is affected differently by the coupling to the phonons. Within a large-deviation analysis, we find a kink in the probability distribution, analogous to a first-order phase transition in thermodynamics, which would be a unique hallmark of the electron-phonon correlations. This kink reflects the fact that although inelastic scattering by the phonons once the voltage exceeds their frequency can scatter electrons opposite to the bias, this will never generate current flowing against the bias at zero temperature, in accordance with the fluctuation theorem. PACS numbers: 05.30.-d, 72.70.+m, 71.38.-k, 73.63.Kv, 05.70.Ln I. INTRODUCTION It has been revealed in electric transport measurements that the electron-phonon interaction induces unique fea- tures in the nonequilibrium current through molecular junctions1,2 and atomic wires.3 -- 5 In particular, when the phonon energy is small compared to the resonance width on the junction, the inelastic phonon scattering increases/decreases the current for small/large transmis- sion probabilities as the source-drain bias voltage V ex- ceeds the local phonon frequency ω0.1,3 -- 5 Quite a number of theoretical microscopic models, e.g. Ref. 6,7, including density-functional theories,8 have been devoted to such junctions (see Ref. 9 and references therein). They re- vealed that a simplified model,10 of a single-level quan- tum dot coupled to a local Einstein phonon mode causing fluctuations of the dot energy level,11 seems to suffice to capture this ubiquitous feature. Recently, the current noise of an atomic wire has been measured,5 and it was observed that the electron-phonon interaction can en- hance or reduce the noise, depending on the value of the transmission probability. Based on the theory of Avriller and Levy Yeyati,12 the negative correction is understood as resulting from the anti-bunching of two electrons: An electron cannot be inelastically scattered by a phonon when the final state is already occupied by another elec- tron. Avriller and Levy Yeyati considerations follow from the theory of full-counting statistics (FCS),13,14 which is most convenient for analyzing nonequilibrium electric transport. Indeed, considerable effort has been invested in recent years in exploiting FCS to study various aspects of nonequilibrium quantum transport [e.g. Refs. 12 -- 28 and references therein]. FCS refers to the probability dis- tribution Pτ (q) of the charge q to be transmitted through a quantum conductor during a certain measurement time τ at out-of-equilibrium conditions (we set e = = 1). The effect of coupling to a vibrational mode on the elec- tric transport has also been investigated in this context, both for a weak12,21 -- 25 and a strong26 -- 28 electron-phonon coupling. In general, it is a rather formidable task to calculate the FCS of interacting electrons. For this reason, most of the investigators have taken advantage of the Keldysh field-theory technique. There, the characteristic function (CF), Zτ (λ) =Xq Pτ (q)eiqλ , (1) or the scaled cumulant generating-function (CGF)29 per- taining to the steady state, F (λ) = lim τ →∞ 1 τ lnZτ (λ) , (2) can be formally written as a 'partition function' or a 'free energy', respectively, defined along the Keldysh time- contour. The λ appearing in these formulae is termed the counting field or the counting parameter. However, even if one calculates the CGF successfully, one still needs to find a way to characterize the electronic correlations in it. One promising approach would be to utilize the location distribution of the zeros of the CF, or equivalently that of the singularities of the CGF, by allowing the counting field λ in Eq. (1) to attain complex values,30 -- 33 similarly to the Yang-Lee theory of phase transitions.34 This idea is based on the recent observation that, upon transform- ing λ into u, u = eiλ, (3) the singularities of the CGF of noninteracting electrons transported between two terminals are all on the neg- ative real axis of the u−plane.30,35,36 It suggests that singularities off the negative real axis would character- ize electronic correlations. From this aspect, molecular junctions are rather advantageous since second-order per- turbation theory in the electron-phonon coupling would capture most relevant features of the electron-phonon correlations in them, allowing for obtaining the location distribution of the singularities. Another recent ingredient is the fluctuation theorem (FT).37 -- 49 The FT is a consequence of micro-reversibility and can be understood as a microscopic extension of the second law of thermodynamics. Despite its simple ap- pearance, a detailed-balance like relation [see Eq. (6) below], the FT reproduces the linear-response results, i.e. it ensures the fluctuation-dissipation theorem and Onsager's reciprocal relations close to equilibrium,38 -- 43 while conveying invaluable information at nonequilib- rium conditions. For molecular junctions, the FT has been addressed using the master-equation approach of FCS for incoherent electron transport.26,27 The FT is considered to be a basic symmetry, such as gauge in- variance, which the CF should fulfill. In the present paper we investigate the FCS of elec- trons coupled to phonons under out-of-equilibrium con- ditions. The quantity to be calculated and analyzed is the cumulant generating-function.12,21 -- 25 Employing the Luttinger-Ward functional,50 -- 52, we obtain its full an- alytic expression, or equivalently all cumulants, in the wide-band limit, treating systematically the nonequilib- rium phonon distribution via a self-consistent condition accurate up to second order in the electron-phonon cou- pling. There are other attempts in the literature23,25 to account for the nonequilibrium phonon effect avoiding the Luttinger-Ward functional; these produce contradict- ing results; our first two cumulants agree with those of Ref. 23, which has not gone beyond the second cumulant. (Our results disagree with those of Ref. 25.) The structure of the paper is as follows. We be- II with brief general explanations of the gin in Sec. FT, the large-deviation analysis and singularities of the CGF. We then introduce in Sec. III the model Hamil- tonian and present analytical results for the CGF and detailed explanations of the calculations and the approx- imations involved. A self-consistent calculation based on the nonequilibrium Luttinger-Ward functional,50 -- 52 is relegated to Appendices A and B. In Sec. IV we analyze the singularities of the CGF and demonstrate the prob- 2 ability distribution within the large-deviation analysis. Section V summarizes our results. Technical details of the calculations are given in Appendices C and D. II. THE FLUCTUATION THEOREM AND SINGULARITIES The definition of the probability distribution Pτ (q) in quantum systems requires special care.13,14 Full-counting statistics theory begins with by introducing the CF [see Eq. (1) and Eq. (14) below for the definition employed in the Keldysh technique] and then defining the quasi- probability distribution by the inverse Fourier transform of the CF, Pτ (q) = 1 2π Z π −π dλe−iλqZτ (λ) . (4) More details are given in Sec. III. The n-th cumulant, in steady state, is given by the n-th derivative of the CGF, Eq. (2), hhI nii = lim τ →∞ hhqnii τ = . (5) ∂nF (λ) ∂(iλ)n (cid:12)(cid:12)(cid:12)λ=0 For example, the average current is the first cumulant, and the current noise is the second one. The FT in the context of quantum electric trans- port relates the probability of the current to flow oppo- sitely to the bias voltage because of thermal agitations, Pτ (−q),38 -- 48 to the distribution Pτ (q), Pτ (−q) = Pτ (q)e−βqV , (6) where β is the inverse temperature. The FT can be equiv- alently written in terms of the CGF F , Eq. (2), as F (λ) = F (−λ + iβV ) . (7) The relation (7) restricts the possible locations of the singularities of the CGF in the λ−plane. As an ex- ample, we depict in Fig. 1 (a) the branch cuts corre- sponding to the continuous singularities of the CGF per- taining to two-terminal transport of noninteracting elec- trons, Eq. (25) below. This CGF is 2π−periodic along the real axis of λ, which guarantees integer values of charge.13 The branch cuts, depicted by thick lines, are at Re λ = (2n − 1)π, where n is an integer. The FT en- sures that the branch cuts are symmetrically distributed around λ = iβV /2. (In Fig. 1 (a), the upper left thick line is identical to the lower right thick line, etc.) The 2π−periodicity is removed by the conformal transforma- tion Eq. (3). Then the branch cuts are on the negative real axis of the u-plane [Fig. 1 (b)].30,31,35,36 The steady-state probability distribution, beyond the central-limit theorem, is derived within the theory of large deviations.29 At steady state, realized in the τ → ∞ limit, we scale q = Iτ and Zτ ≈ eτ F . Then the integral of the inverse Fourier transform Eq. (4) can be estimated by the saddle-point approximation53 and the result is writ- ten with the rate function29 I as Pτ ≈ e−τ I. Since Pτ is real and positive, the saddle point is expected to re- side on the imaginary axis of complex λ−plane. Then the rate function is written in the form of a Legendre-Fenchel transform, I(I) = − lim τ →∞ 1 τ ln Pτ (Iτ ) = max λ {iλI − F (λ)} . (8) Here λ is a purely imaginary number. In most cases, the CGF is real, i.e. the imaginary part of the expo- nent of the Fourier integral (4), lnZτ (λ) − iλq, is zero on the imaginary axis of λ−plane. Then the imaginary axis is expected to be the steepest contour of the integral (4).53 There are few exceptions where singularities of the CGF are on the imaginary axis17,18 as we will also find below. The relation between the CGF and the rate func- tion is analogous to that of thermodynamic potentials. It suggests that singularities on the imaginary axis of the λ−plane would also cause characteristic features in the rate function. (a) (b) iβV/2 −π π λ u − u + e−βV/2 u FIG. 1: The thick lines represent branch cuts corresponding to the continuous singularities of the CGF of noninteracting electrons, Eq. (25), in the λ−plane (a) and in the u−plane (b). The FT ensures that the branch cuts are symmetrically distributed around λ = iβV /2 in the λ−plane (a). In the u−plane (b) the FT relates the branch cuts located inside and outside the dotted circle. III. MODEL HAMILTONIAN AND CUMULANT-GENERATING FUNCTION Our explicit calculations are carried out for a simple model [Fig. 2], a single-level quantum dot coupled to a local Einstein phonon, which induces fluctuations in that level energy.9 -- 11,54 The model Hamiltonian is H = Hmol + Hph + Hlead + Htun,± , (9) where the ± subscript is due to the presence of counting fields (see below). The molecular bridge or the atomic wire is described by the Hamiltonian Hmol, 0c0 , Hmol = [ǫ0 + γ(b + b†)]c† in which c0 (c† 0) destroys (creates) an electron on the lo- calized level representing the molecule, of energy ǫ0, b and (10) 3 b† are the destruction and creation operators of the vibra- tions to which the electron is coupled while residing on the dot, and γ is the strength of the electron-phonon cou- pling. The vibrational modes obey the harmonic Hamil- tonian Hph = ω0b†b , (11) where ω0 is the frequency of the Einstein phonon. The leads are represented by free electron gases, of creation and destruction operators c† rk and crk, and eigen energies ǫrk, Hlead = Xr=L,RXk ǫrkc† rkcrk . (12) Finally there is the tunneling Hamiltonian, coupling the leads to the molecule. This part of the Hamiltonian is augmented by the counting fields. Those appear as phase factors on the operators crk and c† rk, Htun,± = Xr=L,RXk Jre±iλr c† rkc0 + H.c. , (13) where JL and JR are the tunneling amplitudes between the left and the right lead, and the dot. µ L ω0 J L J R µ R ε 0 FIG. 2: Schematic description of the single-level quantum dot coupled to an Einstein vibrational mode. In terms of the tunneling Hamiltonian Eq. (13), the characteristic function is14 Zτ (λ) =DTK exp(cid:16)−iZK dtHtun,±(t)I(cid:17)E , (14) where K denotes the Keldysh contour, which runs from t = −τ /2 to t = τ /2 on the upper branch and returns to t = −τ /2 on the lower one (see Fig. 3) and TK is the time- ordering operator along that contour. The subscript I indicates time dependence in the interaction picture. The ± notation indicates the branch of the Keldysh contour on which the tunneling Hamiltonian is effective, + for the upper branch in Fig. 3 and − for the lower one. In the long-time limit, the scaled CGF depends only on the the difference of the two counting fields λ = λL − λR . (15) As λL+λR counts the number of electrons flowing into the dot, the fact that the CGF depends solely on λ implies current conservation. K + K − t = τ/2 t = −τ/2 FIG. 3: The Keldysh contour. Since in our case the electron-phonon coupling γ is weak, we carry out the calculation up to second-order accuracy, O(γ2). There is a subtle point in this expan- sion. A naive second-order perturbation theory is not capable of producing the correct nonequilibrium phonon distribution.6,8,21 -- 23,55,56 We therefore have to perform a re-summation of infinite diagrams by adopting the linked cluster expansion, see e.g. Ref. 19, or a more advanced method, the nonequilibrium Luttinger-Ward functional, Φ.50 -- 52,57 The first approach had been adopted in Refs. 24 and 25, while the second had been employed implic- itly by Gogolin and Komnik, 16 and hence by later stud- ies12,21 -- 23 based on Ref. 16. Although the second scheme is physically transparent since it relies on the self energy of the electron Green function, the price to pay is that a self-consistent calculation is required in order to ensure various conservation laws. Here we follow the second approach based explicitly on the Luttinger-Ward func- tional, which enables us to perform the self-consistent 4 calculations of the CGF in a transparent manner. De- tails, are given in Appendices A and B. In terms of the Luttinger-Ward potential, the cumu- lant generating-function is F (λ) = F0(λ) − ¯Φ(2)(λ) , (16) where F0 is the CGF pertaining to noninteracting elec- trons, and given in terms of the Keldysh Green function gλ [see Eq. (19)] of the electronic part of the Hamiltonian Eq. (9), F0(λ) = 1 2π Z dω ln detg−1 detg−1 λ (ω) 0 (ω) , (17) with the trace being performed over the 2 × 2 Keldysh space. The CGF satisfies the normalization condition F0(0) = 0. The effect of the electron-phonon interaction is included in the scaled Luttinger-Ward potential ¯Φ(2), which consists of linked diagrams up to O(γ2) accuracy, lim τ →∞ 1 τ Φ(gλ) = ¯Φ(λ) = ¯Φ(2)(λ) + O(γ4) . (18) A. The electronic part The inverse of the Keldysh Green function gλ reads gλ(ω)−1 =(cid:20) ω − ǫ0 + iPr=L,R Γr[f − −iPr=L,R Γrf − r (ω) − f + r (ω)e−iλr r (ω)]/2 iPr=L,R Γrf + −ω + ǫ0 + iPr=L,R Γr[f − r (ω)eiλr r (ω) − f + r (ω)]/2(cid:21) , (19) where ΓL and ΓR are the partial widths of the localized level induced by the coupling with the leads, Γr = 2πνrJr2 , r = L, R . (20) Here νL and νR are the densities of states in the left and right leads. Each of the leads is specified by its chemical potential µr, such that µL − µR = V , and their electron/hole distribution is accordingly given by f ± r (ω) = 1 e±β(ω−µr) + 1 . (21) Our calculation is confined to the extended wide- band limit,21 -- 23 in which either the level broadening Γ = Pr=L,R Γr is larger than the all other energy scales except for the dot level ǫ0 or the dot level ǫ0 is larger than all other energy scales, i. e. or V , kBT, ω0 ≪ Γ , V , kBT, ω0, Γ ≪ ǫ0 . (22) (23) In both cases, we can neglect the energy dependence of the dot density of states. The normalized dot density of states is then replaced by its value at the Fermi energy [see Eq. (C3)] ρ0 = Γ2 0 + Γ2 . 4ǫ2 The calculation detailed in Appendix C yields 13 F0(λ) = (arccoshXλ)2 2πβ βV 2 8π − , where (24) (25) Xλ =(T − 1) cosh βV 2 − T cosh βV + 2iλ 2 . (26) Here T is the transmission of the localized level, T = αρ0 , (27) written in terms of the normalized density of states Eq. (24) and the transmission probability at resonance α = 4ΓLΓR Γ2 . (28) The counting field dependence is only through Xλ. This result obeys the FT since when λ → −λ + iβV , Xλ is unchanged. This CGF possesses continuous lines of singularities for Xλ ∈ [1,∞). As discussed in Sec. II, in the u−plane these become branch cuts (−∞, u−] and [u+, 0), where the branch points are at u± = e−βV /2(x±√x2 − 1), with x = cosh(βV /2)(1 − 1/T ) − 1/T [see Figs. 1 (a) and (b)]. The zeroth-order CGF (25) takes a particular sim- ple form for a symmetric bridge ΓL = ΓR at resonance ǫ0 = 0, for which the transmission is perfect T = 1 [see Eqs. (27), (28), and (24)] F0(λ) = iλ(V + iλ/β)/(2π) , (29) which describes Gaussian thermal fluctuations. On the other hand, when the bridge is extremely askew, ΓR ≪ ΓL or ǫ0 ≫ Γ and thus T ≪ 1 one may expand the CGF to obtain for details) g = 2γ2 πΓ2 . 5 (31) The relevant diagrams are shown in Fig. 4: the second- order diagrams [(a) and (b)] and two of higher-order dia- grams [(c) and (d)]. Diagram (a) represents the Hartree term, which is ignored below since it is independent of the phonon distribution [see Eq. (B3)]. Diagram (b) rep- resents the Fock term, which depends on the phonon dis- tribution function at equilibrium [see Eq. (B8)]. The ac- tual nonequilibrium phonon distribution function can be obtained only by summing up to an infinite order (in the electron-phonon coupling) of diagrams. The simplest way is to collect all ring-diagrams, such as (b), (c) and (d) in Fig. 4, is to exploit the random-phase approximation (RPA) [see Eq. (B10)] which yields F0(λ) ≈ T V 2π (cid:16) eiλ − 1 1 − e−βV + e−iλ − 1 eβV − 1(cid:17) , which is the bi-directional Poisson form. B. The phonon-induced part (30) ¯ΦRPA(gλ) = 1 4π Z dω ln det Dλ(ω)−1. (32) This approximation accounts for the relaxation of the phonon mode by the particle-hole excitations in the elec- trodes. The RPA is expected to provide results accurate up to second order in γ, The nonequilibrium Luttinger-Ward functional Φ in Eq. (18) results from the coupling of the charge carriers to the vibrational modes. We expand it diagrammati- cally in powers of the small parameter (see Appendix B ¯ΦRPA(gλ) = ¯Φ(2)(λ) + O(g2) . (33) The dressed phonon Green function Dλ in Eq. (32) is given by [see Appendix D] Dλ(ω)−1 =(cid:20) (ω2 − ω2 0)/2ω0 − Π++ Π−+ λ (ω) λ (ω) −(ω2 − ω2 Π+− λ (ω) 0)/2ω0 − Π−− λ (ω)(cid:21) . (34) In Eq. (34) there appears the Keldysh particle-hole prop- agator Π, whose lesser/greater components are given by iΠ±∓ λ (ω) = γ2 2π Z dω′g±∓ λ (ω′ + ω/2)g∓± λ (ω′ − ω/2) . (35) It is convenient to express these propagators in the forms iΠ±∓ λ (ω) = Xr,r′=L,R i Π±∓ rr′ λ(ω)e±i(λr −λr′ ) , (36) where Π±∓ in the r and r′ leads. Explicitly, rr′ λ describe the particle-hole excitations created where sr = ±1 for r = L/R and αrr′ = 4ΓrΓr′/Γ2. Here zi (i = 1,··· , 4) are z1 = eβω/2Zλ+, z2 = eβω/2Zλ−, z3 = e−βω/2Zλ+, and z4 = e−βω/2Zλ−, where Zλ± =Xλ ±qX 2 λ − 1 . (38) The lesser components are obtained from the symmetry relations Π+− Π+− λ (ω) = Π−+ λ (ω) = Π−+ λ (−ω) , λ (−ω) . (39) For λ = 0, Eq. (37) reduces to the well-known form i Π−+ rr′ λ(ω) = gαrr′ρ2 0 β eβ(ω+sr′ V )/2Xi zi − zj 1 × (zi + e−β(sr′ V −ω)/2) ln ziYj6=i , (37) Having determined the dressed phonon Green function Eq. (34), we now use it to obtain ¯ΦRPA, Eq. (32). By us- ing Eqs. (D10) and (D16), the determinant of the dressed (zi + e−β(srV +ω)/2) i Π−+ rr′ 0(ω) = gαrr′ρ0 µr − µr′ − ω eβ(µr−µr′ −ω) − 1 . (40) phonon Green function is found to be − det D−1 λ (ω) =(cid:18) ω2 − ω2 2ω0 − ReΠR(ω)(cid:19)2 0 + Aλ(ω) , (41) where the retarded component ΠR is expressed by the lesser and greater components upon exploiting the Kramers-Kronig relation, Eq. (D17). The counting-field dependent phonon life time broadening, √Aλ is a crucial ingredient in our analysis. It is given by λ (ω)Π+− Aλ(ω) = Π−+ [ Π−+ λ (ω) λ (ω) + Π+− 1 4 − λ (ω) + φS λ (ω)]2 , (42) where Π±∓ rr′λ(ω) and λ (ω) =Prr′ Π±∓ λ (ω) =2gρ0[1 − 2ρ0]Xi Ys,s′=± iφS × ln zi βzi Yj6=i 1 zj − zi (eβ(sV +s′ω)/2 + zi) Hence a , (43) for the condition (22) and zero for the condition (23). Collecting the results, we find that the derivative of Eq (32) is ∂ ¯ΦRPA(gλ) ∂λ =Z dω 4π ∂λAλ(ω) 2ω0 − ReΠR(ω)(cid:17)2 (cid:16) ω2−ω2 0 . + Aλ(ω) (44) The ω−integrations is carried out to leading order in g. The integrand has four poles, which within that accuracy are given by ±[ω0 + ReΠR(ω0)] ± ipAλ(ω0) , where we have used Aλ(ω) = Aλ(−ω). The real part (45) also shifts the of the retarded component in Eq. argument of A, which can be neglected since it is already proportional to the small parameter g, Eq. (31). (45) 6 (46) (47) The locations of poles of the integrand in Eq. (44) are complex functions of λ. Consider them first in the limit λ → 0. Since A0(ω0) = (2gρ2 0ω0)2 is a positive real number, then at small enough λ, the real part of the squire root satisfies RepAλ(ω0) > 0 and thus the integration yields up to O(g). It follows that ∂ ¯ΦRPA(gλ) ∂λ ≈ ∂λpAλ(ω0) , ¯Φ(2)(λ) =pAλ(ω0) − 2gρ2 0ω0 , which satisfies the normalization condition ¯Φ(2)(0) = 0. Away from the origin λ = 0, RepAλ(ω0) may be nega- tive. When this is the case we obtain ¯Φ(2)(λ) = −pAλ(ω0) − 2gρ2 appears branch cut 0ω0 . at λ satisfying (48) RepAλ(ω0) = 0. In order to carry out a large-deviation analysis it suf- fices to consider the imaginary axis of the λ−plane, since there the rate function (8) is real. On the imaginary axis, Aλ(ω0) is a real function. When Aλ(ω0) > 0 the imaginary part of ¯Φ(2) is zero and thus the imaginary axis can serve as the steepest contour (the steepest as- cent path)53 of the integral Eq. (4). When Aλ(ω0) ≤ 0, there appears a branch cut on the imaginary axis. One might have thought that the branch cut would be detri- mental to the saddle-point approximation of the integral in Eq. (4). Here we point out that a complex integration along a path encircling the branch cut oscillates rapidly in the τ → ∞ limit and thus would be averaged out. There- fore we will neglect the contribution from the branch cut. (47), and (48), supplemented by Eqs. (37) and (43), are the main results of this paper. Closed expressions, but confined to the first two cumu- lants, have been derived in Ref. 23. In contrast, we obtain analytic expressions for the entire CGF, which en- able us to examine its singularities and to fully analyze the rate function itself, as will be detailed in the Sec. IV. To conclude this section we verify that our results obey the FT. Exploiting the extended detailed-balance rela- tion, Equations (42), (a) (c) (b) (d) FIG. 4: Second-order diagrams, the Hartree (a) and Fock (b) terms. Diagram (c) is one of the fourth-order ones, and diagram (d) is one of sixth-order diagrams (a ring diagram). The full lines denote electron propagators, and the dashed ones phonon propagators. Π−+ rr′ −λ+iβV (ω) = Π−+ rr′ λ(ω) = Π+− rr′ λ(ω)eβ(ω−µr+µr′ ) , and the relations Π−+ −λ+iβV (ω) = Π+− Π−+ −λ+iβV (ω) = Π+− −λ+iβV (ω) = φS φS λ (ω)eβω , λ (ω) , λ (ω) , we find A−λ+iβV (ω) = Aλ(ω) . Therefore Eqs. (47) and (48) satisfy the FT, ¯Φ(2)(−λ + iβV ) = ¯Φ(2)(λ) . (49) (50) (51) (52) IV. RESULTS AND DISCUSSION In the following we confine ourselves to a symmetric junction, α = 1 (ΓL = ΓR), and thus the normalized density of states on the dot at the Fermi level dominates the transmission probability, T = ρ0 [see Eq. (27)]. The numerical results are all obtained for the electron-phonon coupling constant, Eq. (31), g = 0.1, unless otherwise specified. A. Average current and noise Figure 5 (a) shows the source-drain bias voltage depen- dence of the current hhIii at a finite temperature βω0 = 10 (solid lines) and at zero temperature (dashed lines) for a perfect, T = 1, and a relatively weak, T = 0.5, transmis- sion probabilities. At perfect transmission the current is suppressed above the threshold V > ω0, because elec- trons are inelastically backscattered by phonons. When the transmission is weak, T = 0.5, the current is slightly enhanced above the threshold. These results are consis- tent with previous ones.21,22 A finite temperature tends to smear the kink structure of the T = 1 curve; it af- fects far less the average current at weak transmission, T = 0.5, where the solid and dashed lines almost overlap. Figure 5 (b) depicts the current noise hhI 2ii. At per- fect transmission the noise is absent below the thresh- old V < ω0 at zero temperature. Thermal fluctua- tions which arise at finite temperatures induce additional noise below the threshold. Although the current is sup- pressed above the threshold V = ω0 by the inelastic phonon scattering, the noise is significantly enhanced. This indicates that inelastic phonon scattering broadens the probability distribution of the current. In the case of a weak transmission, T = 0.5, the temperature effect is less dramatic -- the noise is simply enhanced. B. Singularities and the rate function At zero temperature, it is possible to obtain a simpler form for the scaled CGF, which is useful for finding its singularities. For positive voltages V ≥ 0, the explicit form of the electronic part of the CGF, Eq. (25), reads 7 (a) T =1 T =0.5 T =1 (b) 0 ω / K R I 1 0 -1 3 0 ω / K R 2 I 2 1 0 T =0.5 -2 -1 0 V/ω 0 1 2 FIG. 5: The source-drain bias voltage dependence of the current (a) and the current noise (b) for T = 1 and 0.5. Solid lines -- βω0 = 10; dashed lines -- zero temperature. The vertical axes are normalized by ω0/RK, where RK = 2π is the resistance quantum. for V ≥ ω0. Below, we investigate the analytic properties of the phonon-induced part of the CGF. 1. Elastic phonon scattering As at zero temperature phonons cannot be excited when 0 < V < ω0, and electron transport at such volt- ages is hence affected only by elastic phonon scatter- ing. Figure 6 (a) shows schematically the square-root branch cut of ¯Φ(2), Eqs. (47) and (48). We find that the u−plane is separated into two by the the brach cut. This branch cut intersects the real axis at u0 = 1 − 1/T [u0 is indicated by empty dots in Fig. 6 (a)] and u1 = (T − 1)(V + ω0)/[V (T −1)+T ω0]. Around these points, Aλ can be expanded as Aλ ≈ g2 4V 2(T − 1)2 (u − u0)2 , (56) F0(λ) = V 2π ln u , u = 1 + T (u − 1) , (53) and where u is defined in Eq. (3). The function Aλ which determines the phonon part is given by Aλ(ω0) =4 g2T 2[(1 − 1/u)(T − 1)V + T ω0]2 , (54) for 0 < V < ω0 and Aλ(ω0) =g2T 2{[V (1 − T + u2(2T − 1)) + (T − 1 + u(2 − u + 2T (u − 1))) ω0]2 − uT 2(V − ω0)[2u ω0 + u(V + (2u − 1) ω0)]}/u4 , (55) Aλ ≈ g2 4T 4(T ω0 + (T − 1)V )4 V 2(T − 1)2 (u − u1)2 . (57) Upon sweeping the transmission T from 0 to 1, u0 in- creases from −∞ to 0, while u1 increases but at TC = V /(V + ω0) jumps from +∞ to −∞. Accordingly, we may define two regimes, or phases, 32 I (T < TC) and II (T > TC) as indicated in Fig. 6 (a). This classification roughly captures the behavior of the average current and the current noise. Figures 6 (b) and (c) depict the correc- tions induced by the electron-phonon interaction in the average current and in the current noise, respectively, hhI niiph = ∂n ¯Φ(2) ∂(iλ)n(cid:12)(cid:12)(cid:12)λ=0 . (58) We find that the electron-phonon interaction always in- creases the average current under the conditions adopted here [Fig. 6 (b)]. On the other hand, the noise can be ei- ther enhanced or suppressed, depending on which regime the transmission is in [Fig. 6 (c)]. In regime I, one of the intersection points, u1, is on the positive real axis, outside the unit circle u = 1 [the left panel of Fig. 6 (a)]. Therefore, in the λ−plane, there is a nonanalytic point on the positive iλ−axis, which induces a weak non-convexity of the CGF as shown in Fig. 7 (a). [The non-convex region is indicated by an arrow there. For comparison, we also plot the g = 0 case (the dotted line) for which the CGF is convex.] Figure 7 (b) exhibits the Legendre transform of the CGF, I = iλ∗I − F (λ∗), where λ∗ satisfies I = ∂F (λ∗)/∂(iλ∗). The Legendre transform is multi-valued around I/hhIii0 ∼ 3 (the thick line in the figure) because of the non-convexity of the CGF. In contrast, the Legendre-Fenchel transform, Eq. (8), chooses the minimum value among them and pro- vides the physical rate function I. [For the relation be- tween the Legendre transform and the Legendre-Fenchel transform, see Ref. 29.] Then, similar to the way a first- order phase transition manifests itself in thermodynam- ics, a kink appears in the rate function. We note that the location of the kink does not coincide with that of the peak; the peak of the rate function is at I = hhIii where λ = 0 and I = 0. The physical consequence is that the elastic phonon scattering broadens the distri- bution by enhancing the probability of currents larger than the average value. It is important to note that the kink is a consequence of the non-convexity of the CGF on the positive real axis of the u−plane, a feature which is absent in the noninteracting-electrons case. In regime II the two intersection points are on the neg- ative real axis of the u−plane [the right panel of Fig. 6 (a)]. The CGF and the rate function pertaining to this case are plotted in Figs. 7 (c) and (d). The CGF is convex and the corresponding rate function is concave. In this regime the peak position is shifted from that for g = 0 [dashed line]. The elastic scattering by the phonons can either broaden or shrink the width of the rate function depending on the transmission probability as we deduce from Fig. 6 (c). 2. Inelastic phonon scattering Above threshold V ≥ ω0, inelastic phonon scattering becomes possible. The analytic properties of the scaled CGF in this regime depend on the bias voltage. The branch cuts in the u−plane are schematically shown in Fig. 8 (a). Three branch points, u1, u2, and u3 can be obtained by searching for the roots of Aλ(u) = 0 [the 8 (a) u 0 u 1 u 1 u 0 (I) T < T C (II) CT < T 0.1 (b) 0.05 I 0 ω / K R h p I 0 0 0.05 0 ω / K R h p 2 I 0 -0.05 0 (c) I II II T C 0.5 T T C 0.5 T 1 1 (a) Schematic picture of the branch cut of ¯Φ(2) in the FIG. 6: complex u−plane below threshold, 0 < V < ω0. The dashed circles are the unit ones. The corrections induced by elas- tic electron-phonon scattering to the average current and the current noise are portrayed in panel (b) and (c), respectively. The bias voltage is V /ω0 = 0.5. See text for the significance of TC. filled dots in Fig. 8 (a)]. In addition there is another point, u0 = 1−1/T [the empty dots in Fig. 8 (a)], around which Aλ can be expanded Aλ ≈ g2 2V (T − 1)(ω0 − V ) T (u − u0)3 . (59) The branch point u1 is always on the negative real axis such that u1 ≤ u0. From the positions of the other two branch points, u2 and u3, we identify three regimes, see Fig. 8 (a). In regime I the two branch points are on the positive real axis outside the unit circle. As T increases u2 and u3 approach one another and meet at T = T−. Then in regime II, the two branch points are located symmetrically off the real axis. Upon further increasing T , u2 and u3 move in the complex plane and at T = T+ they meet on the real axis again. In regime III, the two branch points are on the positive real axis inside the unit circle. In Figs. 9 (a) and (b) we plot the CGF and the rate function pertaining to regime I. The overall tendencies are similar to those found in regime I below threshold, 10 (a) 0 I / ) λ ( (cid:127) 8 6 4 2 0 -3 -2 0 0 I −2 0 iλ 3 (b) / ) Ι ( − ∼ −4 0 1 2 3 4 5 /I I 0 (c) 4 2 ) K R V / ( / ) λ ( (cid:127) 0 -5 -2 0 (d) 0 iλ ) K R V / ( / ) Ι ( − -0.5 -1 0 0.5 /I (V/R )K 5 1 FIG. 7: The scaled CGF (a) and its Legendre transform (b) in regime I (T = 0.1). Solid lines are for g = 0.1 and dashed lines are for g = 0. The bias voltage is below thresh- old, V /ω0 = 0.5. The CGF is plotted as a function of the counting field on the imaginary axis. The axes are normal- ized by hhIii0 = T V /RK. The scaled CGF (c) and the rate function (d) in regime II (T = 0.5). 9 We note that our identification of the three regimes roughly captures the behavior of the corrections to the current and the noise induced by the electron-phonon in- teraction [Figs. 8 (b) and (c)], which oscillate as a func- tion of the transmission probability.5,12,24 (a) u 1 u0 u2 u3 u1 u2 u3 u0 u0 u1 u3 u2 (I) T < T − (II) T − < T < T + (III) +T < T (b) (c) 0.5 0 -0.5 0 ω / K R h p I -1 0 1.5 1 0.5 0 0 ω / K R h p 2 I -0.5 0 I II III T + T − 0.5 T I II 0.5 T 1 III 1 (a) Schematic picture of the branch cut of ¯Φ(2) in the FIG. 8: complex u−plane above threshold V > ω0. The dashed circles are the unit ones. (b) The corrections induced by the electron- phonon coupling to the current and (c) to the current noise. The bias voltage is V /ω0 = 1.5. The boundaries between I and II and between II and III are at T− ≈ 0.551 and T+ ≈ 0.908, respectively. Regime II roughly corresponds to the negative- noise correction region. C. Fluctuation theorem 0 < V < ω0. In the shaded area of Fig. 9 (a) the CGF is nonanalytic and non-convex. This gives rise to a stronger kink structure in the rate function [Fig. 9 (b)]. Figures 9 (c) and (d) present the CGF and the rate function in regime II. The overall tendency is again similar to those of regime II below threshold. Because of the non- analyticities off the real axis, the statistics would not be reduced to that of noninteracting electrons. The analysis of the CGF and the rate function in regime III [see Fig. 8] is rather subtle. At zero temper- ature, there develops a non-convex region in the CGF, and the origin λ = 0 might enter it. When this happens, second-order perturbation fails since the rate function does not satisfy the relation I(I = hhIii) = 0 [this relation is related to the normalization condition F (λ = 0) = 0]. We therefore study regime III at finite temperatures, tak- ing as an example βω0 = 10, and show that the FT is (a) 10 0 I / ) λ ( (cid:127) 8 6 4 2 0 -3 -2 0 0 I −2 / ) Ι ( − −4 0 iλ 3 (b) 0 1 2 3 4 5 /I I 0 4 (c) ) K R V / ( 2 / ) λ ( (cid:127) 0 -2 -5 0 (d) 0 iλ ) K R V / ( -1 / ) Ι ( − -2 0 0.5 /I (V/R )K 5 1 FIG. 9: The CGF (a) and the rate function (b) in I (T = 0.1). The solid lines are for g = 0.1 and the dashed ones are for g = 0. The bias voltage is above threshold, V /ω0 = 1.5. Panels (c) and (d) show the CGF and the rate function, respectively, pertaining to II (T = 0.6). Axes notations are the same as in Fig. 7. crucial for obtaining a physically-reasonable result. Note that when the symmetry (7) holds, the FT (6) is also preserved within the large-deviation analysis,58 I(I) = min = min ξ {ξI − F (iξ + iβV )} ξ∗ {−ξ∗I − βIV − F (−iξ∗)} =I(−I) − βIV . 10 (60) Figure 10 exhibits the CGF and the rate function at perfect transmission. For comparison, we plot the corre- sponding curves for noninteracting electrons, Eq. (29). As we have already noted when discussing the current noise, Fig. 5 (b), the width of the rate function is en- hanced by inelastic phonon scattering [Fig. 10 (b)]. The CGF obeys the FT, Eq. (7), and the curves are symmet- ric around the dot-dashed vertical line at iλ = −βV /2 [Fig. 10 (a)]. The peak of the probability distribution is shifted in the negative direction and the probability to find large current fluctuations is suppressed as com- pared with the noninteracting case [Fig. 10 (b)]. In the shaded area of Fig. 10 (a), the CGF is non-analytic and non-convex. Correspondingly, the rate function has a non-differentiable point at I = 0, see Fig. 10 (b). As a result, although the probability to observe currents smaller than the average value is enhanced by the inelas- tic phonon scattering, the probability to find negative currents I < 0 is strongly suppressed. This is consis- tent with the FT (6), which states that although ther- mal agitations generate current flowing in the opposite direction to the source-drain bias, that probability is ex- ponentially suppressed at low temperatures. Note that previous studies report on a finite current flowing oppo- sitely to the bias at zero temperature [see e.g. Eq. (13) in Ref. 12] in disagreement with the FT, although it may be quantitatively small. This can be easily seen by cal- culating the probability distribution of the transmitted charge q using the CGF as given by Eq. (13) of Ref. 12 and the inverse Fourier transform Eq. (4), Pτ (q) ≈ = 1 −π 2π Z π e−¯q1 ¯q ¯q0−q (¯q0 − q)! 1 dλe−iλq+i¯q0 λ+¯q1(e−iλ−1) θ(¯q0 − q) , (61) where the parameters ¯q0 and ¯q1 are defined in Ref. 12. In the limit of zero temperature β → ∞ this probability distribution remains finite at q < ¯q0 including negative the q regime, which violates the FT (6) . D. Discussion Recently, Kumar et al 5 have explored experimentally the possibility to identify different regimes, as implied by the sign of the correction to the noise induced by the coupling with the phonons,12,24 as depicted in Fig. 8 (c). Above the threshold, our classification predicts three regimes, similarly to Ref. 5. However, the critical points quoted there, T Kumar = 1/2 ± 1/(2√2) ≈ 0.85, 0.15, are ± (a) 4 2 0 -2 0 I / ) λ ( (cid:127) -4 -1.5 0 (b) 0 I / ) Ι ( − -2 -4 -1 -0.5 /iλ βV 0 0.5 0 1 /I I 0 2 FIG. 10: (a) The CGF and (b) the rate function for T = 1, V /ω0 = 1.5, and βω0 = 10. The solid (dotted) lines show results with (g = 0.1) and without (g = 0) electron-phonon interaction, respectively. In the shaded area of panel (a) the CGF for interacting case is non-analytic and non-convex, re- sulting in a non-differentiable point of the rate function at I = 0 (b). Axes are normalized by hhIii0 = V /RK. different from ours. Figure 11 summarizes the regimes found in Sec. IV B. Above the threshold, our critical points T± depend on the bias voltage, with T− = 1/2 and T+ = 1 in the V → ω0 limit. Hence, there is no one-to-one correspondence between the classification of Ref. 5 and ours although we expect the regime II above the threshold roughly correspond to the negative phonon- induced noise regime. Since oscillations in higher cumu- lants are ubiquitous12,24,59 -- 61 and are dominated by sin- gularities close to λ = 0 as detailed in Ref. 59,61, it seems to be legitimate to utilize the location distribution of the singularities itself for the classification.31 -- 33 11 It is worthwhile to expound upon this point. Previous studies5,12,23,24 have considered the changes of slope of the nth cumulant at threshold, ∆hhI nii′ ph = ∂V hhI niiphV =ω0+0 − ∂V hhI niiphV =ω0−0 , (62) as guidelines for the classifications of the various regimes. If we adopt this scheme, we find [see Eqs. (47), (54) and (55)] ∆( ¯Φ(2))′ = ∂V ¯Φ(2)(cid:12)(cid:12)(cid:12)V =ω0+0 − ∂V ¯Φ(2)(cid:12)(cid:12)(cid:12)V =ω0−0 which upon expanding becomes , (63) (5 − 8T )iλ (17 − 60T + 48T 2)(iλ)2 (77 − 392T + 642T 2 − 336T 3)(iλ)3 ∆( ¯Φ(2))′ = T 2 2 + gT 2 4 + gT 2 12 + ··· . (64) We see that the slope difference of the n = 1 cumu- lant changes its sign at T = 0.635, that of n = 2 at T = 0.434, 0.816, etc. In the n = 2 case, the result is compatible with that of Ref. 23. For n = 3, we ob- tain three zeros (at T = 0.449, 0.576, 0.885) similarly to Refs. 12,24, though the positions are different. In gen- eral, the coefficient of λn in the expansion is an n + 2th- degree polynomial function of T , yielding oscillations as a function of T , which is one example of the universal oscillations.59,62 From the higher cumulants, we obtain more zeros, but it is unclear what useful information can be extracted of them. In contrast, the topology of the singularity distribution is distinct and, in our opinion, provides a better way to classify regimes where electron transport is affected differently by the coupling with the phonons. However, as a direct connection between the singularity distribution and the phonon-induced noise seems to be absent at the moment, it is not surprising that we obtain just poor quantitative agreements be- tween our critical points and the positions of the zeros for the phonon-induced noise in Figs. 6 (c) and 8 (c). Another comment related to previous studies has to do with the high-bias limit. Urban et. al.25 have suggested that in that limit, i.e., for ω0 ≪ V ≪ Γ, the correction induced by the electron-phonon coupling to the cumu- lants scales as hhI n+1iiph/hhI niiph ∼ V /ω0. We can use our result Eq. (55) to obtain in this regime the expansion 1 T 0.5 0 0 II I T C III T + II T − I 1 V/ω 0 2 Aλ(ω0) ≈ 4g2T 4ω0 + 2g2T 4V 2(4T − 3)iλ + ··· which implies a stronger scaling, , (65) FIG. 11: Phase diagram summarizing the regimes discussed in Sec. IV B. hhI n+1iiph hhI niiph ≈(cid:18)n − 1 2(cid:19)(cid:18) 3 ω0(cid:19)2 2 − 2T(cid:19)(cid:18) V . (66) Our result thus extends that of Ref. 23, where hhI 2iiph ∝ V 4 was reported. We note that our phonon-induced part of the CGF, Eq. (32), is already different form that of Urban et. al., Eq. (11) in Ref. 25. A comment on the validity of second-order perturba- tion theory is called for. Flindt et. al.61 have analyzed the FCS of sequential transport through a quantum dot containing two levels.63 They have assumed identical in- coming rates but different outgoing ones for the two levels and analyzed the approximate CGF, derived by expand- ing in the ratio of the two outgoing rates, taken to be a small parameter. They have noticed a peculiar behavior: The agreement between the cumulants obtained by dif- ferentiating the approximate CGF (with respect to the counting field) and those derived by differentiating the CGF before expanding it was good for the low-order cu- mulants, but failed completely for the higher-order ones. In our case, the CGF is obtained as an expansion in the electron-phonon coupling and thus it is plausible that the unphysical results which we encountered around regime III for V > ω0 at zero temperature, may have a similar origin to the apparent discrepancy reported in Ref. 61. This fault may be resolved by accounting for all orders in the electron-phonon coupling. However, getting an- alytical results seems to be technically complicated and almost inevitably requires numerical methods.57 Another example of the FCS for interacting electrons is found in transport through a quantum dot in the Kondo regime. Recently, Sakano et. al. have calculated the CGF for the SU (N )−impurity Anderson model, 20 us- ing the renormalized perturbation theory. Their result is exact up to cubic order in the source-drain bias volt- age O(V 3) for a particle-hole symmetric case, and is quadratic in u. This is interpreted as the sum of the CGF for single-particle transfer and that for two-particle transfer. In our case, the phonon part of the CGF at zero temperature, to leading order in V , can be obtained from Eqs. (47) and (54) in the form ¯Φ(2) ≈ 2gV T n+1 (T − 1)n (un − 1) 12 V. SUMMARY We have investigated the full-counting statistics of cur- rents mediated by elastic and inelastic electron-phonon scattering. In the extended wide-band limit, we ob- tained analytic expressions for the cumulant generating- function, accurate up to second order in the electron- phonon coupling. Our results are applicable for finite temperatures and bias voltages and satisfy the fluctua- tion theorem. Using those we analyzed the locations of singularities of the CGF. The singularities are symmet- rically distributed in the λ−plane as to obey the fluctu- ation theorem. The singularities in u− plane, which appear because of the electron-phonon interaction, classify specific regimes in which the dependence of the electron transfer statistics on the bare transmission is distinct. For small transmis- sion probabilities we find singularities of the CGF on the positive real axis satisfying u > 1. Around the singular- ities, the CGF is non-convex, which results in a kink of the rate function. Such a kink, derived within the large- deviation analysis resembles a first-order phase transi- tion in thermodynamics. It signifies the tendency of the phonon scattering to enhance the probability to find cur- rents larger than the average value. When the bias voltage is larger than the phonon fre- quency, V > ω0, we find singularities in 0 < u < 1 around perfect transmission. This results in a kink at I = 0 and a strong reduction for I < 0 in the rate function. This behavior can be understood in the following way: In this regime, phonons scatter electrons inelastically opposite to the bias. This broadens the rate function but the probability for current flowing opposite to the bias volt- age is suppressed exponentially at low temperatures in accordance with the fluctuation theorem. The kink structures in the rate function character- izes the electron-phonon interactions. Although mea- surements of the rate function of molecular junctions would be technically demanding, the FCS can be in prin- ciple monitored experimentally,64 as is proven by existing measurements of higher cumulants59,65 and of the FCS itself, 46,66 for metallic and semiconducting nanostruc- tures. This gives hope that our predictions could be put to test. ∞ Xn=1 u − u0 2gT (T − 1)(u − 1) = V , (67) Acknowledgments where u0 = 1 − 1/T . One may interpret the first right- hand side of Eq. (67) as a sum of independent n-electron transfers. However, the second line indicates that the CGF is non-analytic on the negative real axis of the u−plane at u0 ≤ 0. Therefore, within the first order expansion in V , the electron transfer statistics may be re- duced to that of noninteracting electrons.30 This example suggests that there exist certain subtleties in interpreting the CGF for interacting electrons. We thank Christian Flindt, Dimitri Golubev, Akira Oguri and V´aclav Spicka for valuable discussions. We particularly thank Tom´as Novotn´y for his helpful dis- cussion in comparing our results with previous theories. This work was supported by the Bination Science Foun- dation (BSF) of the US and Israel, by the Israel Science Foundation (ISF), by the Okasan-Katoh Foundation, by the Grant-in-Aid for Young Scientists (B) (Grants No. 23740294 and No. 24710111), by the Young Researcher Overseas Visits Program for Vitalizing Brain Circula- tion (R2214) from the JSPS, and by the MEXT kakenhi "Quantum Cybernetics". This paper was written while the authors were members of a research group on molec- ular electronics at the Institute for Advanced Studies, Jerusalem. Appendix A: The Luttinger-Ward potential The way to construct a partition function based on the self energy is to exploit the Luttinger-Ward functional approach,51 or the self-consistent Φ-derivable approximation.50 This method can be straightforwardly extended to a nonequilibrium situation.52 The saddle- point approximation for the CGF can also be constructed by this approach.48 The underlying idea is the observa- tion that by introducing the Luttinger-Ward functional Φ, which includes all skeleton diagrams, the total gener- ating functional can be formally written as F (λ) = lim τ →∞ 1 τ(cid:16)Tr[ln G−1] + Tr[Σ(G) G] − Φ(G)(cid:17) , (A1) where the trace and the product mean integrations over time along the Keldysh contour. The Green function G, also defined on the Keldysh contour, is G−1(1, 2) = g−1 λ (1, 2) − Σ(1, 2; G) , (A2) where the arguments 1, 2, ··· stand for t1, t2, ··· . Here gλ is the Green function of the noninteracting electrons gλ(1, 2) = −i hTKc0(1)Ic† 0(2)Ii , (A3) given explicitly in Eqs. (C1) and (C2). The self energy is a functional of G as well as of the phonon Green function d, d(1, 2) = −i hTK(b(1)I + b†(1)I)(b(2)I + b†(2)I)i , (A4) given in Eqs. (D1) and (D2). The functional derivative of the nonequilibrium Luttinger-Ward functional Φ gives the self energy, 13 It should be emphasized that in this formulation the self energy has to be determined self-consistently in order to satisfy conservation laws.52 To second order in the electron-phonon coupling γ, Eq. (A6) becomes dF dλ ≈ lim τ →∞ 1 τ + lim τ →∞ 1 τ ∂g−1 λ ∂λ (cid:19) Tr(cid:18)gλ Tr(cid:18)gλΣ(gλ)gλ ∂g−1 λ ∂λ (cid:19) , (A8) where we have replaced Σ(G) by Σ(gλ), since it is already O(γ2). Upon using the identity ∂gλ ∂λ = −gλ ∂g−1 λ ∂λ gλ , (A9) and exploiting the self-consistent condition Eq. (A5) [in its O(γ2) form], Eq. (A8) becomes Tr[ln g−1 λ ] − lim τ →∞ 1 τ dΦ(gλ) dλ , (A10) dF dλ ≈ lim τ →∞ 1 τ d dλ and consequently F (λ) = lim τ →∞ 1 τ(cid:16)Tr[ln g−1 λ ](cid:17) − ¯Φ(2)(λ) . (A11) Comparing this expression with the original one, Eq. (A1), one notes that the term depending explicitly on the self energy has disappeared. The lowest-order scaled Luttinger-ward potential ¯Φ(2)(λ) is obtained by expand- ing ¯Φ(gλ) (which may depend on gλ and not on G) up to O(γ2) [see Eq. (18)]. For brevity, the calculation above is presented in the time domain. One may also switch to the frequency rep- resentation in which Eq. (A6) reads dF (λ) dλ = − 1 2π Z dωTr(cid:18)G(ω)τ3 ∂gλ(ω)−1 ∂λ τ3(cid:19) . (A12) The Pauli matrix τ3, τ3 =(cid:20) 1 0 0 −1(cid:21) . (A13) Σ(1, 2; G) = δΦ δG(2, 1) . (A5) appears once we project the time from the Keldysh con- tour on the real time axis, Both functions G and Σ depend implicitly on the count- ing field only through gλ. By differentiating Eq. (A1) with respect to the count- ing field λ and using Eq. (A5) one obtains dF (λ) dλ = − lim τ →∞ 1 τ Tr(cid:18)G ∂g−1 λ ∂λ (cid:19) , where we have used the relation dΦ dλ = Tr(cid:18) δΦ δG ∂G ∂λ(cid:19) . (A6) (A7) ZK dt =Z τ /2 −τ /2 dt+ −Z τ /2 −τ /2 dt− , (A14) where t± ∈ K± (see Fig. 3). This leads to Eqs. (16) and (17) in the main text. Note that Eq. (A6) or (A12) cor- responds to the "generalized current expression" given by Eq. (12) of Ref. 16. The latter is the starting point for several studies of FCS of molecular junctions,12,21 -- 23 e.g. Eq. (1) in Ref. 21, Eq. (8) in Ref. 22, and Eq. (3) in Ref. 12. We emphasize that the simple form of the generalized current expression16 is correct only when the approximate self energy of the electron Green function (A5) is determined self-consistently. This point becomes clearer when one set the counting field in Eq. (12) of Ref. 16 to zero, ¯λ = 0. That equation is then reduced to the expression for the current, analyzed by Hersh- field, et. al. in Refs. 68, where it was demonstrated that second-order perturbation for the self energy can violate current conservation. Although this problem was discussed in the context of the on-site Coulomb interac- tion, we suspect that it will arise for the on-site electron- phonon interaction as well. The safe approach is to ex- ploit the"generalized current expression", Eq. (12) of Ref. 16, with the self energy determined self-consistently. Appendix B: Diagrammatic expansion Given the results of Appendix A, it remains to calcu- late ¯Φ(gλ) up to O(γ2). This is accomplished by expand- ing it perturbatively in γ. The zeroth-order term is just a constant, Inserting Eqs. (C2) and (D2) into Eq. (B2) yields ¯ΦH(gλ) = g πω0 (cid:18)Z dω × Xr Γr 1 Ωλ(ω) Γ Z dω ¯ρ(ω) (ω − ǫ0)2 + Γ2/4(cid:19) Γ(ω − ǫ0) Ωλ(ω) ! , r (ω) − 1 2f + 14 (B3) where the small parameter g is given in Eq. (31). Note the appearance of the distribution Eq. (21) in the form f + r (ω) − 1/2, resulting from the definition of the step function as Θ(0) = 1/2 in the continuous notation.67 By using Eq. (C7), one can check that the FT is fulfilled, ¯ΦH(g−λ+iβV ) = ¯ΦH(gλ). In the main text, we neglect the Hartree term since it does not depend on the phonon distribution. The Fock term [Fig. 4 (b)] reads ¯Φ(0)(gλ) = − lim τ →∞ 1 2τ Tr ln[d] , (B1) ¯ΦF(gλ) =i independent of the counting field. The diagrams con- stituting the second order are depicted in Fig. 4. The Hartree term [Fig. 4 (a)] is = − γ2 2 lim τ →∞ 1 2 lim τ →∞ 1 τ ZK τ ZK 1 d1d2 gλ(1, 2)d(1, 2)gλ(2, 1) d1d2 d(1, 2)Πλ(2, 1) , (B4) ¯ΦH(gλ) = −i γ2 2 lim τ →∞ = −i γ2 2 Xs,s′=+,− s s′dss′ 1 τ ZK (0)Z dω1 2π d1d2 gλ(1, 1)d(1, 2)gλ(2, 2) where we have introduced the particle-hole propagator, Π, gss λ (ω1)Z dω2 2π gs′s′ λ (ω2) . (B2) Πλ(1, 2) = −iγ2gλ(1, 2)gλ(2, 1) . Adopting the form (B4), we find (B5) ¯ΦF(gλ) = − 1 2Z dω 2π(cid:20)d++(ω)Π++ λ (ω) + d−−(ω)Π−− λ (ω) − d−+(ω)Π+− λ (ω) − d+−(ω)Π−+ (B6) λ (ω)(cid:21). Here a constant should be added to keep the normalization condition ¯ΦF(g0) = 0. Inserting Eqs. (D10) and (D16), we obtain ¯ΦF(gλ) = − + λ (ω) + Π+− 0 (ω) 1 2Z dω 2π(cid:20)d−+(ω)(cid:18) Π+− d−+(ω) + d+−(ω) 2 φS λ (ω) + − Π+− 2 d++(ω) − d−−(ω) 2 λ (ω)(cid:19)+d+−(ω)(cid:18) Π−+ λ (ω) + Π−+ 0 2 (ω) − Π−+ λ (ω)(cid:19) (B7) φA λ (ω)(cid:21). Using relations (39) and the corresponding ones for the components of the phonon Green function, e.g., d+−(ω) = d−+(−ω), yields ¯ΦF(gλ) =Xs=± sn+(sω0)(cid:18) i Π−+ λ (ω0) − PZ dω coth(cid:18) βω0 2 (cid:19) φS λ (sω0)(cid:19)+ ω0 φA λ (ω) ω2 − ω2 λ (sω0) + iΠ−+ − iΠ−+ , (B8) (sω0) i 2 2π 2 0 0 where P means the Cauchy principle value and n± is the Bose distribution, n±(ω) = ± 1 e±βω − 1 . (B9) The Fock term depends on the equilibrium phonon dis- tribution n+, which suggests that a re-summation of an infinite series of diagrams is needed in order to account for the nonequilibrium phonon distribution. We carry out this summation within the random-phase approxi- mation, by summing over all ring diagrams [see diagrams (c) and (d) in Fig. 4]. The RPA is also known to be rel- evant for the AC conductance.69 In this way, we obtain the functional ¯ΦRPA(gλ) = = 1 2 1 2 lim τ →∞ lim τ →∞ 1 τ 1 τ Tr ln [1 − d Πλ ] + ¯Φ(0)(gλ) Tr ln D−1 λ , (B10) where we have also included the zeroth-order term, Eq. (B1) and the full phonon propagator, Dλ, D−1 λ = d−1 − Πλ . (32) (B11) space. Equation (B10) yields Eq. This RPA can be also formulated using the Keldysh path-integral approach within the saddle-point approx- imation48 and further accounting for the Gaussian- fluctuation correction around it. in Fourier A comment on the FT and current conservation in the diagrammatic expansion is called for. The FT, including current conservation as represented by Eq. (15), has been proved using perturbation expansion in the interaction.40 Although the proof has been constructed for the Coulomb interaction, it may be extended to the electron-phonon interaction case as well. as g±∓ r (ω) = ±4i Γr Γ2 ¯ρ(ω)f ± r (ω) . 15 (C4) The dependence on the counting field is contained in the function Ωλ, Ωλ(ω) = − × [f + L (ω)f − detgλ(ω)−1 detg0(ω)−1 = 1 + T (ω) R (ω)(eiλ − 1) + f + R (ω)f − L (ω)(e−iλ − 1)] , (C5) where the transmission of the localized level is frequency dependent, T (ω) = α¯ρ(ω) . (C6) From Eqs. (17) and (C5) we can see the FT is satisfied since Ω−λ+iβV (ω) = Ωλ(ω) . (C7) Within the extended wide-band limit approximation, the frequency dependent normalized density of state (C3) can be replaced by its value at the Fermi energy (24). Then the transmission becomes energy independent, as shown in Eq. (27), and consequently the computation of the integral determining the zeroth-order CGF [see Eq. (17)] is straightforward. The key observation is that the variable transformation z = exp[β(ω′ − (µL + µR)/2)] , (C8) transforms Eq. (C5) into a simpler form, (z − Zλ+)(z − Zλ−) Ωλ(ω′) = (1 + z e−β(µL−µR)/2)(1 + z eβ(µL−µR)/2) , Appendix C: The electronic part (C9) The electronic Keldysh Green functions are obtained by inverting the matrix Eq. (19) gλ(ω) =(cid:20) g++ λ (ω) g+− g−+ λ (ω) g−− λ (ω) λ (ω)(cid:21) , (C1) to obtain gss λ (ω) = 1 Ωλ(ω)  1 s(ω − ǫ0) + iΓ/2 + is Xr=L,R gss r (ω)  , gss r (ω) Ωλ(ω) gss (C2) eisλr , λ (ω) = Xr=L,R where s = −/+ for s = +/−. The lesser and greater Green functions, in the absence of the counting field, gss r , are expressed in terms of the density of states on the lo- calized level normalized by Γ, the width of the resonance on the localized level, ¯ρ(ω) = Γ2/4 (ω − ǫ0)2 + Γ2/4 , (C3) − iπ coth [δ(ω − ω0) − δ(ω + ω0)] , d∓±(ω) = − 2πi[δ(ω − ω0) − δ(ω + ω0)] n∓(ω) . (D2) where Zλ± and Xλ are given in Eq. (38) and Eq. (26), respectively. Then we obtain ∂F0(λ) ∂(iλ) = −1 2πβ 2∂iλXλ Zλ+ − Zλ− ln Zλ− Zλ+ , (C10) and consequently the CGF (25) by integrating over λ. Appendix D: The phonon-induced part We first derive the dressed phonon Green function Dλ Eq. (34). The free Keldysh phonon Green function is given by d(ω) =(cid:20) d++(ω) d+−(ω) d−+(ω) d−−(ω)(cid:21) , (D1) whose four components are d±±(ω) =Re (ω + i0+)2 − ω2 0 2ω0 βω 2 Here 0+ is a positive infinitesimal, and n± is the Bose distribution, Eq. (B9). Using the matrix form of the particle-hole Keldysh Green Function (B5), Πλ(ω) =(cid:20) Π++ Π−+ λ (ω) Π+− λ (ω) Π−− λ (ω) λ (ω) (cid:21) , (D3) Eq. (34) is obtained as the matrix form of Eq. (B11), Dλ(ω)−1 = d(ω)−1 − τ3Πλ(ω)τ3 , where τ3 is the third Pauli matrix, Eq. (A13). (D4) Analytic expressions for the four components of the particle-hole propagator, Eq. (B5) or equivalently Eq. (35), are obtained in the extended wide-band limit.21 -- 23 Since the lesser and greater components are related to one another [see Eqs. (39)] it suffices to compute the greater component. Π−+ rr′ λ, i Π−+ rr′ λ(ω) = r γ2 2π Z dω′ g−+ 0Z dω′ f − = gαrr′ρ2 (ω′ + ω/2)g+− (ω′ − ω/2) Ωλ(ω′ + ω/2)Ωλ(ω′ − ω/2) r′ r (ω+)f + r′ (ω−) Ωλ(ω+)Ωλ(ω−) , (D5) where the small parameter g is given in Eq. (31), and ω± = ω′ ± ω/2 . (D6) After a lengthy but straightforward calculation, exploit- ing the variable transformation Eq. (C8), we obtain Eq. (37). The calculation of Π++ and Π−− is facilitated by considering the combinations Π++ ± Π−−. The only λ−dependence of the casual and anti-casual electronic Green functions is contained in their denominator, Ωλ [see Eqs. (C2)]. Therefore we may write iΠ±± λ (ω) = γ2 2π Z dω′ g±± (ω+)g±± (ω−) 0 Ωλ(ω+)Ωλ(ω−) 0 Upon using the relations g++ 0 = g+− 0 = g−+ g−− 0 + gR = g−+ 0 − gR = g+− 0 + gA , 0 − gA , . (D7) (D8) where gR,A are the retarded and advanced Green func- tions, gR,A(ω) = 1 ω − ǫ0 ± iΓ/2 , (D9) λ (ω) + Π−− λ (ω) = Π+− λ (ω) + Π−+ λ (ω) + φS λ (ω) , (D10) we find Π++ with iφS λ (ω) = where the relation γ2 1 Ωλ(ω+)Ωλ(ω−) − 1(cid:17) 2π Z dω′(cid:16) × [gR(ω+)gR(ω−) + gA(ω+)gA(ω−)] , (D11) Z dω′gR/A(ω+)gR/A(ω−) = 0 (D12) 16 has been used. Since Ωλ = 1 for ω ≫ max(V , 1/β) the integral in Eq. (D11) is bounded, and therefore in the extended wide-band limit21 -- 23 the terms in the square brackets there can be replaced by 2Re gR(0)2 ≈ (cid:26) 8ρ0[1 − 2ρ0]/Γ2 2/ǫ2 0 (V , kBT, ω0 ≪ Γ) (V , kBT, ω0, Γ ≪ ǫ0) (D13) , yielding Eq. (43); iφS λ (ω) ≈ (cid:26) 2gρ0[1 − 2ρ0]S 0 where (V , kBT, ω0 ≪ Γ) (V , kBT, ω0, Γ ≪ ǫ0) , (D14) S =Z dω′(cid:16) =Xi Ys,s′=± 1 Ωλ(ω+)Ωλ(ω−) − 1(cid:17) (eβ(sV +s′ω)/2 + zi) ln zi βzi Yj6=i 1 . zj − zi (D15) Turning now to the combination Π++ − Π−−, we use for the Eqs. Keldysh component of the Green function, to obtain (D8) and the relation gK = g−+ 0 + g+− 0 Π++ λ (ω) − Π−− λ (ω) = 2ReΠR(ω) + φA λ (ω) . (D16) The retarded component is derived from the relation ΠR(ω) = γ2 2πiZ dω′ gK(ω+)gA(ω−) + gR(ω+)gK(ω−) 2 , which is rewritten by exploiting the Kramers-Kronig re- lation as ΠR(ω) = i 2π Z dω′ Π−+ (ω′) − Π+− ω − ω′ + i0+ 0 0 (ω′) . (D17) In Eq. (D16) we obtain γ2 1 iφA λ (ω) = 2π Z dω′(cid:16) Ωλ(ω+)Ωλ(ω−) − 1(cid:17) × Im(cid:16)gK(ω+)gA(ω−) + gR(ω+)gK(ω−)(cid:17) . Using Eq. (C7), we can verify the symmetry (D18) −λ+iβV (ω) = φA φA λ (ω) . (D19) One can now convince oneself that in the extended wide- band limit, in which Re[gR(ω)] ≈ −ǫ0/[ǫ2 0 + Γ2/4] and gK(ω) = − Γr Γ 4i Γ Xr ρ0tanh[β(ω − µr)/2] (D20) is O(1/Γ), φA λ may be safely neglected, since φA λ ∝ (cid:26) 1/Γ2 (V , kBT, ω0 ≪ Γ) 1/(Γǫ0) (V , kBT, ω0, Γ ≪ ǫ0) . (D21) ∗ Electronic address: [email protected] 1 R. H. M. Smit, Y. Noat, C. Untiedt, N. D. Lang, M. C. van Hemert, and J. M. van Ruitenbeek, Nature 419, 906 (2002). 2 M. Kiguchi, O. Tal, S. Wohlthat, F. Pauly, M. Krieger, D. Djukic, J. C. Cuevas, and J. M. van Ruitenbeek, Phys. Rev. Lett. 101, 046801 (2008). 3 O. Tal, M. Krieger, B. Leerink, J. M. van Ruitenbeek, Phys. Rev. Lett. 100, 196804 (2008). 4 N. Agrait, C. Untiedt, G. Rubio-Bollinger, and S. Vieira, Phys. Rev. Lett. 88, 216803 (2002). 5 M. Kumar, R. Avriller, A. Levy Yeyati, and J. M. van Ruitenbeek, Phys. Rev. Lett. 108, 146602 (2012). 6 J. K. Viljas, J. C. Cuevas, F. Pauly, and M. Hafner, Phys. Rev. B 72, 245415 (2005). 7 L. de la Vega, A. Martin-Rodero, N. Agrait, and A. Levy Yeyati, Phys. Rev. B 73, 075428 (2006). 8 T. Frederiksen, M. Brandbyge, N. Lorente, and A.-P. Jauho, Phys. Rev. Lett. 93, 256601 (2004). 9 M. Galperin, M. A. Ratner, and A. Nitzan, J. Phys. Cond. Matt. 19, 103201 (2007). 10 R. Egger and A. O. Gogolin, Phys. Rev. B 77, 113405 (2008); O. Entin-Wohlman, Y. Imry, and A. Aharony, Phys. Rev. B 80, 035417 (2009). 11 T. Holstein, Ann. Phys. 8, 325 (1959). 12 R. Avriller and A. Levy Yeyati, Phys. Rev. B 80, 041309(R) (2009). 13 L. S. Levitov and G. B. Lesovik, JETP Lett. 58, 230 (1993); L. S. Levitov, H. Lee, and G. B. Lesovik, J. Math. Phys. 37, 4845 (1996). 14 Quantum Noise in Mesoscopic Physics, Vol. 97 of NATO Science Series II: Mathematics, Physics and Chemistry edited by Yu. V. Nazarov (Kluwer Academic Publishers, Dordrecht/Boston/London, 2003). 15 A. Komnik and A. O. Gogolin, Phys. Rev. Lett. 94, 216601 (2005); 16 A. O. Gogolin and A. Komnik, Phys. Rev. B 73, 195301 (2006). 17 D. A. Bagrets, Y. Utsumi, D. S. Golubev, and G. Schon, Fortschr. Phys. 4, 917 (2006). 18 D. A. Bagrets, Phys. Rev. Lett. 93, 236803 (2004). 19 Y. Utsumi, D. S. Golubev, and G. Schon, Phys. Rev. Lett. 96, 086803 (2006). 20 R. Sakano, Y. Nishikawa, A. Oguri, A. C. Hewson, and S. Tarucha, Phys. Rev. Lett. 108, 266401 (2012); R. Sakano, A. Oguri, T. Kato, and S. Tarucha, Phys. Rev. B 83, 241301 (2011). 21 F. Haupt, T. Novotn´y, and W. Belzig, Phys. Rev. Lett. 103, 136601 (2009). 22 F. Haupt, T. Novotn´y, and W. Belzig, Phys. Rev. B 82, 165441 (2010). 23 T. Novotn´y, F. Haupt, and W. Belzig, Phys. Rev. B 84, 113107 (2011). 24 T. L. Schmidt and A. Komnik, Phys. Rev. B 80, 041307(R) (2009). 25 D. F. Urban, R. Avriller, and A. Levy Yeyati, Phys. Rev. B 82, 121414(R) (2010). 26 L. Simine and D. Segal, Phys. Chem. Chem. Phys., 14, 13820 (2012). 27 G. Schaller, T. Krause, T. Brandes, and M. Esposito, arXiv:1206.3960. 17 28 S. Maier, T. L. Schmidt, and A. Komnik, Phys. Rev. B 83, 085401 (2011). 29 H. Touchette, Phys. Rep. 478, 1 (2009). 30 A. G. Abanov and D. A. Ivanov, Phys. Rev. Lett. 100, 086602 (2008); A. G. Abanov and D. A. Ivanov, Phys. Rev. B 79, 205315 (2009). 31 D. Kambly, C. Flindt, and M. Buttiker, Phys. Rev. B 83, 075432 (2011). 32 D. A. Ivanov, A. G. Abanov, Europhys. Lett. 92, 37008 (2010). 33 C. Flindt and J. P. Garrahan, Phys. Rev. Lett. 110, 050601 (2013). 34 C. N. Yang and T. D. Lee, Phys. Rev. 87, 404 (1952); T. D. Lee and C. N. Yang, Phys. Rev. 87, 410 (1952). 35 M. Vanevi´c, Yu. V. Nazarov, and W. Belzig, Phys. Rev. Lett. 99, 076601 (2007). 36 F. Hassler, M. V. Suslov, G. M. Graf, M. V. Lebedev, G. B. Lesovik, and G. Blatter, Phys. Rev. B 78, 165330 (2008). 37 D. J. Evans, E. G. D. Cohen, and G. P. Morriss, Phys. Rev. Lett. 71, 2401 (1993); ibid. 71, 3616 (1993); G. Gallavotti and E. G. D. Cohen, Phys. Rev. Lett. 74, 2694 (1995); G. Gallavotti, Phys. Rev. Lett. 77, 4334 (1996). 38 J. Tobiska and Yu. V. Nazarov, Phys. Rev. B 72, 235328 (2005). 39 H. Forster and M Buttiker, Phys. Rev. Lett. 101, 136805 (2008). 40 K. Saito and Y. Utsumi, Phys. Rev. B. 78, 115429 (2008); arXiv:0709.4128. 41 D. Andrieux, P. Gaspard, T. Monnai, and S. Tasaki, New J. Phys. 11 043014 (2009). 42 M. Esposito, U. Harbola, and S. Mukamel, Rev. Mod. Phys. 81, 1665 (2009). 43 M. Campisi, P. Hanggi, and M. Talkner, Rev. Mod. Phys. 83, 771 (2011) 44 A. Altland, A. De Martino, R. Egger, and B. Narozhny, Phys. Rev. Lett. 105, 170601 (2010); Phys. Rev. B 82, 115323 (2010). 45 R. Lopez, J. S. Lim and D. Sanchez, Phys. Rev. Lett. 108, 246603 (2012); R. Sanchez, R. Lopez, D. Sanchez, and M. Buttiker, Phys. Rev. Lett. 104, 076801 (2010); J. S. Lim, D. Sanchez, and R. Lopez, Phys. Rev. B 81, 155323 (2010); D. Sanchez, Phys. Rev. B 79, 045305 (2009). 46 Y. Utsumi, D. S. Golubev, M. Marthaler, K. Saito, T. Fu- jisawa, and G. Schon, Phys. Rev. B 81, 125331 (2010); B. Kung, C. Rossler, M. Beck, M. Marthaler, D. S. Golubev, Y. Utsumi, T. Ihn, and K. Ensslin, Phys. Rev. X 2, 011001 (2012). 47 S. Nakamura, Y. Yamauchi, M. Hashisaka, K. Chida, K. Kobayashi, T. Ono, R. Leturcq, K. Ensslin, K. Saito, Y. Utsumi, and A. C. Gossard, Phys. Rev. Lett. 104, 080602 (2010); Phys. Rev. B 83, 155431 (2011). 48 Y. Utsumi and K. Saito, Phys. Rev. B. 79, 235311 (2009). 49 O.-P. Saira, Y. Yoon, T. Tanttu, M. Mottonen, D. V. Averin, and J. P. Pekola, Phys. Rev. Lett. 109, 180601 (2012). 50 G. Baym and L. P. Kadanoff, Phys. Rev. 124, 287 (1961); G. Baym, Phys. Rev. 127, 1391 (1962). 51 J. M. Luttinger and J. C. Ward, Phys. Rev. 118, 5 (1960); J. M. Luttinger, Phys. Rev. 119, 4 (1960). 52 Yu. B. Ivanov, J. Knoll, H. Van Hees, and D. N. Voskre- sensky, Nucl. Phys. A 657, 413 (1999). 53 C. M. Bender and S. A. Orszag, Advanced Mathemati- cal Methods for Scientists and Engineers, (Springer, New York, 1999). 54 A. Mitra, I. Aleiner, and A. J. Millis, Phys. Rev. B 69, 245302 (2004). 55 O. Entin-Wohlman, Y. Imry, and A. Aharony, Phys. Rev. B 81, 113408 (2010). 56 A similar problem has been encountered for a local moment coupled with a nonequilibrium electron gas, see e.g. A. Rosch, J. Paaske, J. Kroha, and P. Wolfle, Phys. Rev. Lett. 90, 076804 (2003); O. Parcollet and C. Hooley, Phys. Rev. B 66, 085315 (2002). 57 T.-H. Park and M. Galperin, Phys. Rev. B 84, 205450 (2011). 58 J. L. Lebowitz and H. Spohn, J. Stat. Phys. 95, 333 (1999). 59 C. Flindt, C. Fricke, F. Hohls, T. Novotn´y, K. Neto`cn´y, T. Brandes, and R. J. Haug, Proc. Natl. Acad. Sci. USA 106, 10116 (2009). 60 D. S. Golubev, M. Marthaler, Y. Utsumi, and Gerd Schon, Phys. Rev. B 81, 184516 (2010). 61 C. Flindt, T. Novotn´y, A. Braggio, and A-P. Jauho, Phys. Rev. B 82, 155407 (2010). 62 M. V. Berry, Proc. R. Soc. A 461, 1735 (2005). 63 W. Belzig, Phys. Rev. B 71, 161301(R) (2005). 64 Y. Utsumi, D. S. Golubev, M. Marthaler, G. Schon, and 18 K. Kobayashi, Phys. Rev. B 86, 075420 (2012). 65 B. Reulet, J. Senzier, and D. E. Prober, Phys. Rev. Lett. 91, 196601 (2003); Yu. Bomze, G. Gershon, D. Shovkun, L. S. Levitov, and M. Reznikov, Phys. Rev. Lett. 95, 176601 (2005). 66 S. Gustavsson, R. Leturcq, B. Simovic, R. Schleser, T. Ihn, P. Studerus, K. Ensslin, D. C. Driscoll, and A. C. Gossard, Phys. Rev. Lett. 96, 076605 (2006); T. Fujisawa, T. Hayashi, R. Tomita, and Y. Hirayama, Science 312, 1634 (2006). 67 For a precise treatment of the step function, see Sec. 2.8 in A. Kamenev, Field Theory of Nonequilibrium Systems (Cambridge University Press, Cambridge, 2011). 68 S. Hershfield, J. H. Davies, and J. W. Wilkins, Phys. Rev. Lett. 67, 3720 (1991); S. Hershfield, J. H. Davies, and J. W. Wilkins, Phys. Rev. B 46, 7046 (1992). 69 A. Ueda, O. Entin-Wohlman, and A. Aharony, Phys. Rev. B 83, 155438 (2011).
1009.2202	1	1009	2010-09-11T20:20:51	Nonlinear optical response of hole-trion systems in quantum dots in tilted magnetic fields	[ "cond-mat.mes-hall" ]	We discuss, from a theoretical point of view, the four wave mixing spectroscopy on an ensemble of p-doped quantum dots in a magnetic field slightly tilted from the in-plane configuration. We describe the system evolution in the density matrix formalism. In the limit of coherent ultrafast optical driving, we obtain analytical formulas for the single system dynamics and for the response of an inhomogeneously broadened ensemble. The results are compared to the previously studied time-resolved Kerr rotation spectroscopy on the same system. We show that the Kerr rotation and four wave mixing spectra yield complementary information on the spin dynamics (precession and damping).	cond-mat.mes-hall	cond-mat	physica status solidi, 9 November 2018 Nonlinear optical response of hole -- trion systems in quantum dots in tilted magnetic fields Paweł Machnikowski*,1, Tilmann Kuhn2 1 Institute of Physics, Wrocław University of Technology, 50-370 Wrocław, Poland 2 Institute of Solid State Theory, University of Munster, 48149 Munster, Germany Received XXXX, revised XXXX, accepted XXXX Published online XXXX Key words: nonlinear optical spectroscopy, spin, decoherence, magnetooptical Kerr effect ∗ Corresponding author: e-mail [email protected], Phone: +48-71-3204546, Fax: +48-71-3283696 We discuss, from a theoretical point of view, the four wave mixing spectroscopy on an ensemble of p-doped quantum dots in a magnetic field slightly tilted from the in-plane configuration. We describe the system evolution in the density matrix formalism. In the limit of coherent ultrafast optical driving, we obtain analytical formulas for the single system dynamics and for the response of an inhomogeneously broadened ensemble. The results are compared to the previously studied time- resolved Kerr rotation spectroscopy on the same system. We show that the Kerr rotation and four wave mixing spectra yield complementary information on the spin dy- namics (precession and damping). 1 Introduction. The properties of confined spins in semiconductor nanostructures are interesting because of their expected important role in quantum computing and spintronic devices. In particular, the hole spin attracts much attention because of its enhanced coherence time. In sys- tems that efficiently couple to optical fields, like quantum wells and self-assembled quantum dots, an interesting pos- sibility is to study the spin dynamics using optical spec- troscopy tools. A method which is particularly suited for this purpose is the time-resolved Kerr (or Faraday) rota- tion (TRKR, TRFR) [1,2,3], where the evolution of the occupations of the hole and trion Zeeman states is traced by investigating the rotation of the polarization plane of the reflected or transmitted probe pulse. In a recent work [4], we proposed a general descrip- tion for the dynamics of a confined hole-trion system in a tilted magnetic field. We showed that a TRKR experiment provides rich information about the rates of spin precession and decoherence (both longitudinal and transverse). In par- ticular, the optical response at slightly tilted magnetic fields contains contributions related both to the hole spin relax- ation (longitudinal decoherence) with respect to the hole Copyright line will be provided by the publisher spin quantization axis, as well as the dephasing of spin co- herences (transverse decoherence) with respect to this axis. Under favorable experimental conditions (a sufficient sep- aration of time scales), the corresponding two decoherence rates can be deduced from a single run of an experiment, although in a realistic system the latter may be convoluted with (or even dominated by) the inhomogeneous dephas- ing. Although the TRKR or TRFR method may be the most obvious choice for the investigation of the spin dynamics the spin precession of the trion and hole in a tilted field will lead to transitions between optically active and in- active states which should be manifested in any form of the optical response. In particular, signatures of the spin- related dynamics should be visible in the four wave mixing (FWM) nonlinear spectroscopy. As the FWM spectroscopy is one of the most widely used methods in the investigation of semiconductors and their nanostructures [5,6] it may be interesting to study how the spin precession and decoher- ence affect the FWM response and what information on the spin-related kinetics can be extracted from this kind of experiments. Copyright line will be provided by the publisher 0 1 0 2 p e S 1 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 0 2 2 . 9 0 0 1 : v i X r a 2 P. Machnikowski, T. Kuhn: Nonlinear optical response of hole -- trion systems In this contribution, we study the FWM response of the same system of confined holes as discussed in our previous work [4]. We show that the FWM signal is also affected by the spin dynamics in a way that allows one to extract the Larmor frequencies of electrons and holes. The avail- able information on the decoherence rates is less specific than that contained in the TRKR response and is available only as long as the hole spin dephasing times are not much longer than the life time of the optical coherence. On the other hand, the spins undergo partial refocusing in the two- pulse optical echo experiment which leads to the appear- ance of components in the optical response that depend on the inherent spin dephasing rates but are insensitive to the inhomogeneous distribution of the Larmor frequencies. In this way, the FWM method may be a valuable experimen- tal tool to study the spin dynamics, complementary to the TRKR or TRFR techniques. The paper is organized as follows. In Sec. 2, we define the model under study. In Sec. 3 we derive the FWM re- sponse of the system. Next, in Sec. 4, we discuss the result and compare the spin-related information contained in the FWM signal to that available from the TRKR response. 2 Model. We study a system similar to that discussed in our previous work [4], consisting of an ensemble of quantum dots or trapping centers in quantum wells with one confined hole in each of the dots. The system is excited with a pump pulse at the time t = −τ and a probe pulse at t = 0. Unlike in our previous work, we are now interested in the FWM response from the system. Experimentally, the relevant third-order response is isolated by choosing the appropriate excitation and detection directions (see Ref. [4] for a discussion). In the modeling, we assume the pump and probe pulses to have phases φ1 and φ2, respectively, and calculate the terms in the response of an inhomoge- neously broadened system that carry the phase 2φ2 − φ1. Out of many possible configurations of the polarizations of the excitation and detection, we choose the circularly co- polarized one, with both pulses having the σ+ polarizations and the detection being performed at the same polarization. The system is placed in a magnetic field tilted by the angle ϑ to the normal, which defines the trion spin quantization. The hole spin is quantized along the axis at an angle ϕ from the normal to the sample, defined by the components of the highly anisotropic hole Land´e tensor. We will assume that the system geometry is very close to the Voigt configura- tion, with the magnetic field only slightly tilted from the system plane, so that cos ϕ ≈ 0 and sin ϕ ≈ 1. The system dynamics is a composition of many pro- cesses. For the optical hole-trion transition, we assume the radiative recombination rate γ1 and the additional pure de- phasing rate γ0. The hole and the trion spin precess in the magnetic field with the Larmor frequencies ωh and ωt, respectively. The longitudinal relaxation rate for the hole spin is T −1 1 = κ+ + κ−, where κ−, κ+ are the rates for the spin flip transitions to the lower and to the upper Copyright line will be provided by the publisher 2 = T −1 Zeeman level, related by the detailed balance condition κ−/κ+ = exp(¯hωh/kBT ), where kB is the Boltzmann constant and T is the temperature. The hole spin transverse relaxation rate is T −1 1 /2 + κ0, where κ0 is the ad- ditional pure dephasing rate. We assume that the electron spin dephasing is slow compared to the trion lifetime so that the electron spin coherence is dominated by the latter. The simulations presented in this contribution are per- formed for T = 4 K and for the magnetic field of 7 T tilted at an angle π/2 − ϑ = 4◦ from the system plane. For the in-plane and perpendicular components of the hole g-tensor equal to 0.04 and 0.6, respectively, this yields the hole Larmor frequency ωh = 0.036 ps−1 and the hole spin quantization axis oriented at ϕ = 44◦ from the normal di- rection. The trion Larmor frequency is 0.16 ps−1 (corre- sponding to the electron g-factor of 0.26). We will assume the decoherence times 1/γ1 = 1.2 ns and T1 = 2 ns and no additional pure dephasing effects (γ0 = κ0 = 0). 3 The FWM response. Our theoretical analysis is based on the method developed in Ref. [4]: the system evo- lution is studied in the density matrix formalism, with the optical and spin-related dephasing included via a Lindblad dissipator in the evolution equation. In the limit of coher- ent ultrafast optical driving, this approach yields analytical formulas for the single system dynamics, which allows one to perform averaging over an inhomogeneous distribution of various parameters in the ensemble. For a σ+ probe, the only element of the density ma- trix just after the probe pulse that carries the e2iφ2 phase dependence is ρ31(0+) = ρ13(0−)e2iφ2 sin2 α2 2 , where φ2 is the phase of the probe pulse, α2 is its area, and 0− denotes the time instant just before the arrival of the pulse. The evolution of the system state is then calcu- lated using the Master equation in the Lindblad form, like in Ref. [4]. The σ+ interband coherence at a time t > 0 is ρ31(t) = X ±,± d±,±eλ±,±tρ13(0+)e−iEt/¯h, where the amplitudes and the exponents are given by d1 = d2 = d3 = − d4 = − cos ϕ + 1 4 cos ϕ − 1 4 cos ϕ + 1 4 cos ϕ − 1 4 , , , , λ1 = λ2 = λ3 = λ4 = 2iωt − (2iωh − β) 4 2iωt + (2iωh − β) 4 , , −2iωt − (2iωh − β) 4 −2iωt + (2iωh − β) 4 , . Here E is the interband transition energy (at zero field) and the dephasing constants are Γ = 4γ0 + 2γ1 + κ0 + κ+ + κ− pss header will be provided by the publisher 3 and β = (κ− − κ+)(cid:2)(cos ϕ − 1) sin2 ϕ + 1(cid:3) . Now, we need to find ρ13(0−). For a σ+ -- polarized pump, just after the pump pulse (at t = −τ +), the only non-zero element linear in the pump amplitude is ρ31. By solving the Master equation one finds at t = 0− ρ13(0−) = X i cieλ∗ i τ eiEτ /¯h, (1) where ci = ρ13(−τ +)di. ρ31(t) = Thus, at the time t one finds sin α1 sin2 α2 1 2 2 e−iE(t−τ )/¯h X ij ei(2φ2−φ1) didjeλit+λ∗ j τ . When averaging over the distribution of interband energies E in a typical QD ensemble, the exponent produces a nar- row echo peak around t = τ. Since the spin-related evo- lution (frequencies λi) is slow on this time scale, we can put t = τ under the summation. Then the magnitude of the time-integrated response is proportional to FWM ∼ (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) X i 2 dieλiτ(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 1 (1 + cos ωtτ )e−Γ τ /2h(cos ϕ − 1)2e−βτ /2 8 +(cos ϕ + 1)2eβτ /2 + 2(1 − cos2 ϕ) cos ωhτi. (2) h, σ2 In order to take into account also the inhomogeneity of the Land´e tensors at different trapping centers this result should be averaged over a certain distribution of the Lar- mor frequencies ωh and ωt, which we will assume to be Gaussian and characterized by the variances σ2 t . We assume here that the distribution of Land´e tensors is un- correlated to the spectral positions of the trion transitions. In Fig. 1(a), we show the calculated time-integrated FWM signal in the absence of inhomogeneous distribution of the Larmor frequencies. As follows from Eq. (2), oscil- lations at various frequencies are present in the optical sig- nal, corresponding to combinations of the hole and trion frequencies. A more transparent picture is obtained after performing a Fourier-transform of this signal, as shown in Fig. 1(b). Here one can see the zero-frequency line, as well as lines at the frequencies ±ωt, ±ωh, and ±ωt ± ωh. In a real system, the optical response is affected by the inhomogeneous distribution of the relevant parameters. In the present case, the Larmor frequencies are usually not identical for each hole-trion system due to variations of the g-factors in the nanostructure. In Fig. 2 we show the results of a simulation performed for the same parame- ters as in the previous case but with the additional effect (a) ) . u . b r a ( l a n g i s M W F (b) ) . u . b r a ( ] M W F [ T F e R 0 1 t [ns] 2 -0.2 -0.1 0 w [ps-1] 0.1 0.2 Figure 1 The FWM signal from a system without inho- mogeneous broadening of the Larmor frequencies. (a) The time-integrated FWM signal as a function of the delay time; (b) The real part of the Fourier transform of the time- resolved signal. (a) ) . u . b r a ( l a n g i s M W F (b) ) . u . b r a ( ] M W F [ T F e R 0 1 t [ns] 2 -0.2 -0.1 0 w [ps-1] 0.1 0.2 Figure 2 The FWM signal from a system with inhomo- geneously broadened Larmor frequencies. (a) The time- integrated FWM signal as a function of the delay time; (b) The real part of the Fourier transform of the time-resolved signal. of inhomogeneous broadening of the Larmor frequency, which is assumed to be equal to 20% of their average val- ues. Due to this inhomogeneity effect, the oscillations in the time integrated response are damped and the signal is dominated by the monotonic decay (except for short delay times). Correspondingly, all the non-zero frequency peaks in the Fourier spectrum are strongly broadened. However, the central peak remains unaffected. 4 Discussion and conclusions. An interesting point in the discussion presented above is the presence of a zero-frequency component which produces a central peak in the Fourier transform, composed of two Lorentzian con- tributions: one with the width (Γ + β)/2 and another one with (Γ −β)/2. Interestingly, as follows from Eq. (1), there are no zero-frequency components in the polarization evo- lution between the pulses. Therefore, the non-oscillatory part of the FWM response can be interpreted as a result of partial refocusing of the Larmor precession by the probe pulse. The presence of this zero-frequency component, which is insensitive to the inhomogeneous distribution of the g- factor, opens a possibility of extracting useful information on the spin-related system parameters. If the trion life- Copyright line will be provided by the publisher 4 P. Machnikowski, T. Kuhn: Nonlinear optical response of hole -- trion systems [4] P. Machnikowski and T. Kuhn, Phys. Rev. B 81, 115306 (2010). [5] P. Borri, W. Langbein, S. Schneider, U. Woggon, R. L. Sellin, D. Ouyang, and D. Bimberg, Phys. Rev. Lett. 87, 157401 (2001). [6] A. Vagov, V. M. Axt, T. Kuhn, W. Langbein, P. Borri, and U. Woggon, Phys. Rev. B 70, 201305(R) (2004). time is sufficiently long compared to the hole spin decoher- ence times (β not too small compared to Γ ) then the two Lorentzian components of the central line, with the widths Γ ± β can be separated by fitting, from which the values of Γ and β can be deduced. The value of β, along with the detailed balance relation, allows one to extract the rates κ±, hence the longitudinal decoherence time. On the other hand, Γ involves both the spin-related rates κ±,0 and the optical lifetime and dephasing rates γ0,1. The latter, how- ever, can often be deduced independently for a given sys- tem, which can allow one to find also the value of κ0 and to calculate the intrinsic transverse spin decoherence time T2. This should be compared with the information avail- able from a TRKR experiment [4]. There, the hole-related response originates from the occupations of the hole Zee- man sublevels after the trion recombination. Therefore, the TRKR experiment allows one to extract the hole spin- related information even if the exciton lifetime and co- herence time are very short, as was indeed the case in the experiments [1,2]. However, as the information on the transverse spin dephasing in that experiment is deduced from the spin precession signal, it is convoluted with the inhomogeneous effect which may even completely domi- nate the decay of coherent precession signal. Therefore, in a TRKR experiment essentially only the inhomogeneous transverse decoherence time T ∗ 2 is available. Thus, we have shown that a FWM experiment per- formed on an ensemble of quantum dots or other trapping centers doped with excess holes can provide useful infor- mation on the properties of hole spin precession and de- coherence. Although this information is only available un- der favorable conditions related to the various decoherence rates in the system (long exciton coherence) it is insensi- tive to inhomogeneous effects, in particular to a variation of g-factors. Therefore, we conclude that the FWM exper- iments may be a useful tool to extract spin-related infor- mation which is complementary to that available from the TRKR study. Acknowledgments.This work was supported in part by the Alexander von Humboldt Foundation within a Re- search Group Linkage Grant. P.M. acknowledges support from the TEAM programme of the Foundation for Polish Science, co-financed from the European Regional Devel- opment Fund. References [1] M. Syperek, D. R. Yakovlev, A. Greilich, J. Misiewicz, M. Bayer, D. Reuter, , and A. D. Wieck, Phys. Rev. Lett. 99, 187401 (2007). [2] M. Kugler, T. Andlauer, T. Korn, A. Wagner, S. Fehringer, R. Schulz, M. Kubov´a, C. Gerl, D. Schuh, W. Wegscheider, P. Vogl, and C. Schuller, Phys. Rev. B 80, 035325 (2009). [3] T. Korn, M. Kugler, M. Griesbeck, R. Schulz, A. Wag- ner, M. Hirmer, C. Gerl, D. Schuh, W. Wegscheider, and C. Schuller, New J. Phys. 12, 043003 (2010). Copyright line will be provided by the publisher
1205.1264	1	1205	2012-05-07T01:21:14	Electron Number Dependence of Spin Triplet-Singlet Relaxation Time	[ "cond-mat.mes-hall" ]	In a GaAs single quantum dot, the relaxation time T_{1} between spin triplet and singlet states has been measured for the last few even number of electrons. The singlet-triplet energy separation E_{ST} is tuned as a control parameter for the comparison of T_{1} between different electron numbers. T_{1} shows a steady decrease from 2-electrons, 4-electrons, to 6-electrons, and we found this implies an enhancement of the spin-orbital coupling strength in a multi-electron quantum dot.	cond-mat.mes-hall	cond-mat	Electron Number Dependence of Spin Triplet-Singlet Relaxation Time Gang Cao, Ming Xiao,∗ HaiOu Li, Tao Tu, and GuoPing Guo† Key Laboratory of Quantum Information, Chinese Academy of Sciences, University of Science and Technology of China, Hefei 230026, People's Republic of China (Dated: August 5, 2018) In a GaAs single quantum dot, the relaxation time T1 between spin triplet and singlet states has been measured for the last few even number of electrons. The singlet-triplet energy separation EST is tuned as a control parameter for the comparison of T1 between different electron numbers. T1 shows a steady decrease from 2-electrons, 4-electrons, to 6-electrons, and we found this implies an enhancement of the spin-orbital coupling strength in a multi-electron quantum dot. The spin singlet-triplet states of an electron pair in a quantum dot have been demonstrated as potential solid- state qubits [1 -- 3]. Principally, any even number of elec- trons would form the singlet-triplet configuration. Ex- perimentally, spin blockade effect for a variety of even number of electrons have been observed [4] and the coher- ent manipulation of multi-electron singlet-triplet-based qubits has been recently studied [5]. A question that naturally arises is whether the multi-electron interaction interferes with the singlet-triplet coherence, such as en- hancing relaxing or dephasing. Here we study the singlet-triplet relaxation time T1 in a single quantum dot for different even number of elec- trons. Since T1 strongly depends on the singlet-triplet en- ergy separation EST , we control EST by tuning the quan- tum dot shape with confinement gates. For a given elec- tron number, T1 is measured with the pump-and probe technique [6] in a tunable range of EST . T1 shows large decrease (roughly 3 times) when the electron number in- creases from 2 to 6. An increase in the spin-orbit coupling strength with larger electron number is found to explain the observed decrease in T1. Fig. 1(a) shows a scanning electron microscopy (SEM) image of the gate-defined single GaAs quantum dot. The left barrier of the dot is closed and the electrons only tunnel through the right barrier. The dc current through the quantum point contact (QPC) is recorded to count the charge number on the dot. A gap between the QPC and the dot is created to maximize the charge counting sensitivity. In this experiment the gap is closed tightly and the QPC dc bias voltage V dc QP C is small to minimize the back-action effect [7, 8]. Fig. 1(b) shows the charge stability diagram measured by the QPC while gate P and RB are biased at dc voltages. We will measure T1 of the spin singlet-triplet states for 2e, 4e, and 6e, respectively. As shown in Fig. 2(a), a sequence of square-wave volt- age pulses is applied on the plunger gate P to explore the energy spectroscopy of the quantum dot by pumping the electrons to excited states [9]. The gray-scale plot shows the QPC response averaged over many duty cycles with a lock-in amplifier, in the 1e ↔ 2e transition region. Dur- ∗Electronic address: [email protected] †Electronic address: [email protected] (a) Q (b) LT RT 300nm LB RB P + 0e 1e 2e FIG. 1: (a) A SEM image showing the geometry of our sam- ple. The dotted circle is the location of the quantum dot. Gate P is used to control the QD electron energy with re- spect to the Fermi level of the electron reservoir. Sometimes voltage pulse ∆VP will be applied on gate P to dynamically explore the quantum dot energy spectrum. (b) Gray-scale plot of the QPC differential current as functions of voltages VP and VRB. Voltages on other gates are: VLB = -1.40 V, VLT = VRT = -1.50 V, VQ = -0.90 V, and V dc QP C = 0.3 mV. ing each duty cycle, the low-level pulse Vl and high-level pulse Vh bring the QD electrons into the spin ground state Si twice, and correspondingly produce two charge transition lines, denoted as Sh and Sl. When the pulse amplitude is large enough, the high-level pulse Vh pumps the electrons into the spin excited state Ti as well. In fact, we see an additional line denoted as Th between Sh and Sl when ∆VP ≡ Vh − Vl ≥ 12.3 mV. Using the energy-voltage conversion factor 0.07 meV/mV, we determined EST as 0.86 meV in this example. In order to detect the relaxation process from Ti to Si, we applied a sequence of three-step pulses [6, 10], as illustrated in Fig. 2(b). At the low voltage level Vl, the energy of both Ti and Si lie above the Fermi level EF of the electron reservoir. This resets the quantum dot by emptying out the Si state. Then at the high voltage level Vh, the energy of both Ti and Si drop below EF . Therefore the electrons are pumped into the spin excited state Ti with a certain probability. Vh sustains for a waiting time TW , during which period Ti relaxes to Si. Finally at the medium voltage level Vm, EF is brought between the energy of Ti and Si. If the electrons have already relaxed into Si, no electron jumping will occur in such an energy configuration. On the other hand, if the relaxation has not completed and the electrons still have a probability of occupying state Ti, one electron will jump out of the dot and another electron will jump (a) )) (b) (a) (b) 2 time Tae Sae Tae Sae time Tae Sae (c) (d) (c) (d) FIG. 2: (a) A sequence of square-waves is applied on gate P . The pulse frequency is typically 600 Hz. The gray-scale plot shows the numerically differentiated QPC current measured by a lock-in amplifier with time constant 300 ms. This graph is taken around the 1e ↔ 2e transition region. VRB = -0.76 V and all other gate voltages are the same as in Fig. 1(b). (b) The mechanism of the pump-and-probe measurement for the spin relaxation time when a sequence of three-step pulses is applied on gate P . (c) The gate-averaged QPC current over a sequence of 4000 pump-and-probe pulses. It begins with the low-level pulse, followed by a high-level pulse (TW = 0.5 ms in this example). Finally from 0 ms to 3 ms, the spin-bump is read in the medium-level pulse step. (d) The spin-bump height as a function of TW . Open dots are the experimental data. Solid curve is the fitting with an exponential decay. in to fill state Si. As a consequence, the QPC current shows a "up" and "down" switching. So the occurrence probability of the QPC current switching at the Vm level, as a function of TW , can tell us the speed of the Ti−Si relaxation process. Repeating the above process many times, the "up" and "down" switchings of the QPC current at the Vm level average out as Gaussian distributions, called a "spin bump". The dependence of the spin bump height on TW contains the information of T1. Fig. 2(c) shows a typi- cal trace of the gate-averaged QPC current recorded by a high-bandwidth oscilloscope. The spin bump occurs when the pulse time is from 0 ms to 3 ms. We increase the waiting time TW and see that the spin bump height rapidly decays. Fig. 2(d) shows the extracted depen- dence of the spin bump height on TW , which can be de- scribed by an exponential decay with a rate 1/T1. The fitting in Fig. 2(d) gives T1 = 0.88 ± 0.01 ms. It is well known that T1 strongly depends on the energy splitting EST , because of the energy dependence of the phonon emission rate and the electron-phonon interac- tion [11 -- 14]. EST keeps increasing when we sweep down VP to squeeze the electron number from 6 to 2, because the dot size shrinks and the exchange interaction energy increases. During this process, a longer T1 is observed for a smaller electron number. But in this situation we FIG. 3: (a) Tuning EST with gate voltage VLB for 2e. (b) For 2e, T1 as a function of EST . (c) T1 for different electron numbers. EST is fixed at 0.75 ± 0.01 meV for the black closed squares, and 0.62 ± 0.01 meV for the red open circles. (d) 1/T1 as a function of EST . The red triangles, green squares, and blue circles are for 2e, 4e, and 6e, respectively. The solid curves are the theoretic fittings. The red dotted curve is the same fitting with the red solid curve, except that the dot radius is enlarged from 16 nm to 20nm. can not tell if the change in T1 comes from the difference in EST or in the electron number. A meaningful compar- ison can only be made at the same value of EST while the electron number is varied. The control experiment on EST can be done with tuning some of the confinement gates, mainly gate LB. When increasing VLB (and slightly compensating VRB), we found that EST continuously decreases, as shown in Fig. 3(a) for the case of 2e. This control on EST is most likely fulfilled by changing the quantum dot shape [12, 15]. A more positive voltage on gate LB pulls the electron wave functions on one side of the dot and makes the dot less circular. This gives rise to smaller EST . In this experiment, we can tune EST from 0.93 to 0.75 meV for 2e by increasing VLB over 460 mV. When EST goes up, T1 monotonically increases, as can be seen in Fig. 3(b) for 2e. For 4e and 6e, we also ob- served a monotonically increase of T1 with EST . How- ever it becomes more and more difficult to collect enough data points for larger electron numbers. When VLB is the same, EST for 4e and 6e is usually about one half of EST for 2e. It is therefore not easy for EST to reach the same magnitude for different electron numbers. An ex- haust tuning of one confinement gate at relatively large dot size turns out to result in double or multiple dots. Nonetheless, we managed to bring up EST for 4e and 6e to reach the lower limit of EST for 2e. In Fig. 3(c) we present T1 for 2e, 4e, and 6e for two fixed values of EST , 0.75 meV and 0.62 meV, respectively. We see that at each fixed value of EST , T1 substantially decreases with increasing electron number. For example, when EST = 0.75 meV, T1 drops from 0.75 ms for 2e to only 0.23 ms for 6e, which is a change of more than 3 times. To quantitatively evaluate the change in T1, we use the simplified model [12, 14]: 1/T1 = M 2 SO λ5 l c4 32πρ¯h3(cid:16) Ξ2a4 j Z π/2 dθsin5θe−(a2sin2θ)/(2λ2 l ) l Z π/2 j )(cid:17) dθAj(θ)2sin5θe−(a2sin2θ)/(2λ2 0 + Pj=l,t e2β 2a4 j c4 λ3 0 Here the first integral comes from the deformation potential electron-phonon coupling and is nonzero only for the longitudinal phonons. The second integral is from the piezoelectric coupling and contains contribu- tions from both the longitudinal and transverse phonons. The piezoelectric constants are as following: Al(θ) = 3√2sin2(θ)cosθ/4, At1 (θ) = √2sin(2θ)/4, and At2 (θ) = √2(3cos2(θ) − 1)sinθ/4. The coupling strength for the deformation potential and piezoelectric interactions is Ξ = 6.7 eV and eβ = 1.4 × 109 eV/m, respectively [11, 12]. The mass density ρ is 5300 kg/m3, and the sound speed for the longitudinal and transverse phonons is cl = 4730 m/s and cl = 3350 m/s, respectively [12]. The dot radius a is estimated as about 16 nm using transport experiments. The characteristic energy, where the phonon wavelength λl,t = ¯hcl,t/EST matches the dot size and the relaxation rate 1/T1 reaches a maximum, is therefore E∗ ST = ¯hcl,t/a = 0.19 or 0.14 meV. This is much smaller than the energy range in our experiment, and we see from Fig. 3(d) that our data points lie to the right side of the peak points in the simulated curves. The spin-orbit strength MSO is an independent fac- tor in our simulation. A value 0.33 µeV gives a good match to our data for 2e, shown as the red curve and 3 red symbols in Fig. 3(d). This is consistent with the reported 0.37 µeV in the literature [12]. MSO also shows a dependence on the electron number. The green (blue) curve in Fig. 3(d) tells us that MSO for 4e (6e) is 0.36 (0.55) µeV, which is about 1.1 (1.7) times of MSO for 2e. These changes result in measurable increase in the relaxation rate 1/T1 due to its quadratic dependence on MSO. The dot radius a may also increase with larger electron number. In Fig. 3(d) the red dotted curve is the predicted 1/T1 for 2e with larger a (20 nm) and same MSO (0.33 µeV) as the red solid curve. The peak po- sition shifts to the left and the magnitude decreases on the right side of the peak. So the enlargement of dot size seems unable to explain the observed increase of 1/T1. It has been pointed out that the Coulomb interac- tion between electrons has large influence on a number of factors in the singlet-triplet relaxation process, and most likely induces stronger spin to charge coupling [13]. So the multi-electron Coulomb interaction may be re- sponsible for the enhancement of the spin-orbit coupling strength in a multi-electron quantum dot. The inclusion of more orbital states could possibly strengthen the spin- orbital coupling too. In conclusion, we studied the relaxation time of spin singlet-triplet states for the last few even electron num- bers. With increasing electron number, even only from 2 to 6, T1 was found to substantially decrease, possi- bly due to the enhanced spin-orbital coupling strength. Therefore the exact electron number may be a matter when we choose a singlet-triplet as the basis of a qubit. This work was supported by the NFRP 2011CBA00200 and 2011CB921200, and NNSF 10934006, 11074243, 11174267, 91121014, and 60921091. [1] J. M. Elzerman, R. Hanson, L. H. Willems van Beveren, B. Witkamp, L. M. K. Vandersypen, and L. P. Kouwen- hoven, Nature 430, 432 (2004). [2] 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). [3] Yasuhiro Tokura, Wilfred G. van der Wiel, Toshiaki Obata, and Seigo Tarucha, Phys. Rev. Lett. 96, 047202 (2006). [4] A. C. Johnson, J. R. Petta, C. M. Marcus, M. P. Hanson and A. C. Gossard, Phys. Rev. B 72, 165308 (2005). [5] A. P. Higginbotham, F. Kuemmeth, C. Barthel, C. M. Marcus, M. P. Hanson, and A. C. Gossard, APS March Meeting, Session L29 (2012). [6] S. Tarucha, K. Ono, T. Fujisawa, W. G. van der Wiel, and L. P. Kouwenhoven, "Interactions, Spins and the Kondo Effect in Quantum-Dot Systems" in "Electron Transport in Quantum Dots", Edited by J. P. Bird, Springer (2003). [7] Ramn Aguado and Leo P. Kouwenhoven, Phys. Rev. Lett. 84, 1986 (2000). [8] H. O. Li, M. Xiao, G. Cao, C. Zhou, R. N. Shang, T. Tu, G. C. Guo, G. P. Guo, and H. W. Jiang, Appl. Phys. Lett. 100, 092112 (2012). [9] J. M. Elzerman, R. Hanson, L. H. Willems van Beveren, L. M. K. Vandersypen, and L. P. Kouwenhoven, Appl. Phys. Lett. 84, 4617 (2004). [10] S. Sasaki, T. Fujisawa, T. Hayashi, and Y. Hirayama, Phys. Rev. Lett. 95, 056803 (2005). [11] V. N. Golovach, A.V. Khaetskii, and D. Loss, Phys. Rev. Lett. 93, 016601 (2004). [12] 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). [13] V. N. Golovach, A.V. Khaetskii, and D. Loss, Phys. Rev. B 77, 045328 (2008). [14] J. P. Hansen, S. A. Sorngard, M. Forre, and E. Rasanen, Phys. Rev. B 85, 035326 (2012). [15] S. Amasha, K. MacLean, Iuliana P. Radu, D. M. Zum- buhl, M. A. Kastner, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 100, 046803 (2008).
1308.5056	1	1308	2013-08-23T06:49:43	Electronic properties of delta-doped phosphorus layers in silicon and germanium	[ "cond-mat.mes-hall" ]	The Thomas-Fermi-Dirac (TFD) approximation and an sp3d5s* tight binding method were used to calculate the electronic properties of a delta-doped phosphorus layer in silicon. This self-consistent model improves on the computational efficiency of "more rigorous" empirical tight binding and ab initio density functional theory models without sacrificing the accuracy of these methods. The computational efficiency of the TFD model provides improved scalability for large multi-atom simulations, such as of nanoelectronic devices that have experimental interest. We also present the first theoretically calculated electronic properties of a delta-doped phosphorus layer in germanium as an application of this TFD model.	cond-mat.mes-hall	cond-mat	Electronic properties of delta-doped phosphorus layers in silicon and germanium J. S. Smith,∗ J. H. Cole,† and S. P. Russo‡ Chemical and Quantum Physics, School of Applied Sciences, RMIT University, Melbourne VIC 3001, Australia (Dated: February 20, 2018) The Thomas-Fermi-Dirac (TFD) approximation and an sp3d5s∗ tight binding method were used to calculate the electronic properties of a δ-doped phosphorus layer in silicon. This self-consistent model improves on the computational efficiency of "more rigorous" empirical tight binding and ab initio density functional theory models without sacrificing the accuracy of these methods. The computational efficiency of the TFD model provides improved scalability for large multi-atom sim- ulations, such as of nanoelectronic devices that have experimental interest. We also present the first theoretically calculated electronic properties of a δ-doped phosphorus layer in germanium as an application of this TFD model. PACS numbers: 71.10.Ca,31.15.aq,73.22.-f,71.15.-m I. INTRODUCTION New in situ phosphorus δ-doping techniques in sin- gle crystal silicon and germanium are achieving unprece- dented carrier concentrations inside highly confined lay- ers (δ-layers) of the host material.1,2 δ-doped P layers in Si (Si:P) are ideally one monolayer (ML) thick and at very low temperatures exhibit electronic properties that are a combination of those of the dopant and the host material. Through advances in scanning tunnelling microscope lithography3 and molecular beam epitaxy4 a new generation of δ-doped structures in Si is now be- ing realised. These structures include: δ-doped Si:P lay- ers,5 quantum dots,6,7 quantum dot arrays,8 quantum wires9,10 and a single electron transistor.11 Using in situ doping techniques one in four atoms in- side a Si(001) ML (1/4 ML) can be substituted with a P donor.1 Highly doped planes of P atoms form layers, in- side which the donor atoms are inherently disordered.1 Because of the high doping densities inside these lay- ers, we expect the donor electrons to behave similarly to an inhomogeneous electron gas. However, the ma- jority of recent theoretical models describe the donor electrons using explicit donor atom configurations rather than an average donor electron density. Experimental verification of either type of theoretical model is cur- rently unavailable as recent measurements of the Si:P δ- layer band structure are inconclusive when compared to the theory.12 Presently, it is only through computational modelling that the electronic properties of Si:P δ-layers can be quantitatively understood. The most recent empirical and ab initio models of Si:P δ-layers are designed to maximise the completeness of the system's mathematical description rather than com- putational efficiency. Here, we show the efficacy of these models to be reproducible with a one-dimensional (1D) model of the donor electron potential using the Thomas- Fermi-Dirac (TFD) approximation13,14 and a parametri- sation of correlation effects.15 We combine this descrip- tion of the donor electrons with an empirical tight bind- ing (TB) method,16,17 which offers the scalability needed for large multi-atom simulations of nanoelectronic de- vices that have experimental interest. One of the earliest models of a δ-doped layer was the application of the Thomas-Fermi (TF) theory13,18,19 to an n-type δ-layer, in general.20 Later, this same TF method was applied to p-type δ-layers in GaAs and Si21 and, combined with TB in a study of n-type δ-doped Si and B layers in GaAs.22 Transport properties in n- type δ-layers in Si have also been studied using the TF method.23 Si:P δ-layers were first modelled using a den- sity functional theory (DFT) method with a planar Wan- nier orbital (PWO) basis,24 and later an antibonding orbital model25 with an sp3s∗ TB method.26 However, the sp3s∗ basis set is too small to reproduce the experi- mentally predicted curvature of the ~kx valley minima.16 A self-consistent three-dimensional (3D) quantum well model27 combined with an sp3d5s∗ TB method16,17 was therefore recently used in both studies of disorder28 and temperature dependence29 in Si:P δ-layers. These stud- ies used the nemo -d package,30 which has also been applied to δ-doped Si:P quantum wires.10,31 Complimen- tary DFT models of Si:P δ-layers and quantum wires have been proposed using the siesta and vasp pack- ages, with localised atomic orbital (LAO) bases32–35 and a planewave basis.36 However, the applicability of these models to realistic device architectures is restricted by the N 3 scaling in calculation time associated with DFT. Historically, the effective mass theory (EMT) has also been used to study n-type δ-doped layers in Si37 and was recently put on a firm theoretical footing for Si:P δ-layers.38 Unfortunately, although computationally ef- ficient, the EMT is only applicable to the lowest lying conduction valleys, which are assumed to be parabolic, and is therefore not suitable for modelling the higher con- duction bands, which are important in the calculation of transport properties. Here, we calculate the electronic properties of a Si:P δ-layer using an sp3d5s∗ TB method and a self-consistent 1D (TB1D) model of the donor potential and, we present the first theoretically calculated electronic properties of a Ge:P δ-layer. This also marks the first time that the TFD approximation has been combined with an sp3d5s∗ basis set. We report on the band structure, valley splitting, Fermi level, electronic density of states (eDOS) and the changes in these properties against ML doping density. The TB1D model is outlined in Section II. The results of the model are discussed for a Si:P δ-layer in Section III and for a Ge:P δ-layer in Section IV. We conclude in Section V. II. METHOD OF CALCULATION A. Hamiltonian The electronic properties of bulk Si and Ge are de- scribed in the tight binding approximation. The Hamil- tonian is defined using an sp3d5s∗ basis set16 with 18 empirical parameters that correctly predict the bulk ef- fective masses of Si and Ge.17 The Hamiltonian is of the form, Hbulk = Xi,u ε(u) i c† i,uci,u + Xi,j6=i,u,v t(uv) ij i,ucj,v + Hb.c. , c† where c† (1) i,u and ci,u are creation and annihilation op- erators. The first term in Hbulk describes the on-site energies (cid:0)ε(u) i (cid:1) of an electron in orbital u on atom i and the second term the coupling (cid:0)t(uv) (cid:1) between electron or- bitals u and v on first nearest-neighbour atoms i and j. The third term ( Hb.c.) describes the boundary conditions which for periodic boundaries are of the form, ij Hb.c. = Xk,l6=k,u,v t(uv) kl c† k,ucl,v , (2) i ij and t(uv) where k and l are first nearest-neighbour atoms via the periodic boundaries of the supercell. The Hamil- tonian terms ε(u) are expressed through TB parameters.39 These parameters are found by an optimi- sation procedure which fits to experimentally measured electronic properties: band energies and effective masses at high symmetry points in the Brillouin zone (BZ).40 We do not carry out such a procedure here, and instead use previously published TB parameters for Si and Ge.17 The sp3d5s∗ basis set comprises 20 hydrogenic orbitals for the description of the bulk Si and Ge band struc- ture. In the δ-layer system, the P donor electrons occupy conduction states that are non-degenerate12 and we are therefore able to ignore electron spin and halve the size of the TB basis set from 20 to 10 orbitals. This results in a Hamiltonian matrix of order 4800 × 4800. To calculate the electronic properties of the structure the Hamilto- nian matrix is diagonalised for all ~k-points in the first irreducible Brillouin zone (IBZ).41 The speed of this di- agonalisation is directly proportional to the order of the 2 TABLE I. Parameters used in the calculations and references in square brackets. m0 is the free electron mass. ǫr Si Ge 11.4 [42] 15.36 [42] ml/m0 0.9163 [43] 1.588 [44] mt/m0 a (A) 0.1905 [43] 0.08152 [44] 5.430 [16] 5.6563 [16] Hamiltonian matrix and the number of ~k-points in the first two-dimensional (2D) BZ or ~k-point grid. We use an effective ~k-point grid of 120×120×1 to converge the Fermi level to within 1 meV. To simulate the δ-layer we use a primitive tetragonal (tP) supercell which has dimensions (a × a × 60a) nm, where a is the lattice constant of the host material. This is equivalent to 120 MLs of bulk "cladding" in the pos- itive and negative z-directions perpendicular to the δ- layer. Periodic boundary conditions are applied in the x and y directions, and also in the z direction after the bulk cladding. Using 120 MLs of bulk cladding the en- ergies of the conduction states of interest are converged to within 1 meV. We do not geometry optimise the bulk Si or Ge as previous DFT investigations have shown that the effect of the P atoms on the positions of the bulk Si atoms is negligible.32,36 However, we expect the incorpo- ration of P atoms into bulk Ge to have a larger effect on the positions of the Ge atoms because the Ge nucleus is much larger than the Si and P nuclei. Although we do not geometry optimise the bulk Ge here, we note that this approximation merits further investigation. B. Donor electron potential The donor electrons are described by a 1D potential that is added to the diagonal terms of the Hamiltonian matrix.22,45 The TFD approximation and an appropriate parametrisation of correlation effects are used to calcu- late the donor potential. In this calculation the doping plane is approximated by a sheet of vanishing thickness with a constant areal doping density, and the validity of the local density approximation is assumed through- out. The TF and Dirac exchange formalism are exact in the high density limit. The TF approximation for δ- doped semiconductor structures has been shown to pro- duce meaningful results in the 1012–1013 cm−2 electron density range20 (one to two orders of magnitude lower than the nominal 1.7×1014 cm−2 considered herein). A full derivation of the TF theory for δ-doped semiconduc- tor structures is presented in the Appendix. For a 2D δ- doped semiconductor structure in the high density limit, the electrostatic potential in the TF approximation is written as VTF(z) = − α2 (αz + z0)4 , (3) with and α = (2 ¯m)3/2e2ν 60π2ǫrǫ03 z0 = (cid:18) 8α3ǫrǫ0 e2nD (cid:19)1/5 , 3 anisotropy of the effective mass term in the calculation of the exchange and correlation (XC) potentials. Finally, the donor potential is written as V (z) = VTF(z) + VX(z) + VC(z) , (12) which in matrix form is (4) (5) where z is distance perpendicular to the δ-layer, ν is the number of equivalent conduction valleys, ǫr is the static dielectric constant, ¯m is the geometric average of the longitudinal (ml) and transverse (mt) effective masses23,37,38,46 ( ¯m = m2/3 ) and nD is the areal donor density. The parameters in Table I are used throughout. t m1/3 l To the electrostatic potential is added an exchange po- tential (VX) derived from the Dirac exchange energy func- tional47 and a correlation potential (VC). The exchange potential is written as VX(z) = − ¯me4 (4πǫrǫ0)2 (cid:18) 9 4π2(cid:19)1/3 1 rs(z) , (6) with and rs(z) = (cid:18) 4π ¯a0 3n(z) 3 (cid:19)−1/3 ¯a0 = 4πǫrǫ02 ¯me2 (7) (8) where rs is the Wigner-Seitz radius, ¯a0 is the effective Bohr radius ( ¯a0 = 1.96 nm for Si, and ¯a0 = 3.80 nm for Ge) and n(z) is the electron density (see Appendix). The correlation energy is approximated using the correlation functional parametrised by Perdew & Wang15 for systems having zero spin polarisation. In the high density limit, this correlation potential is written as VC(z) = − 2A(cid:26)ln(cid:20)1 + 1 2Af(cid:21)(cid:18)1 + 2α1rs(z) 3 − f ′ f (1 + 2Af ) (cid:18) rs(z)(1 + α1rs(z)) 3 with and f = β1r1/2 s + β2rs + β3r3/2 s + β4r2 s (cid:19) (cid:19)(cid:27) , (9) (10) f ′ = β1 2 r−1/2 s + β2 + 3β3 2 r1/2 s + 2β4rs , (11) where A = 0.031091, α1 = 0.21370, β1 = 7.5957, β2 = 3.5876, β3 = 1.6382, and β4 = 0.49294. Following the methods of Refs. 23, 37, and 38 we do not consider the V = Xi Vi c† i ci , (13) where Vi is the discrete form of V (z) for the z coordi- nate. The Hamiltonian describing the δ-layer system is therefore, H = Hbulk + V . (14) Here, the donor potential is treated as an external po- tential and added to only the diagonal or on-site energy terms of the Hamiltonian matrix:22,45 an approximation we believe is validated by the results of the TB1D model as demonstrated in Section III. C. Fermi level pinning Recent measurements of the band structure of a Si:P δ-layer show at least one occupied state in the conduc- tion band (CB).12 Because the δ-layer exhibits metallic characteristics at low temperatures, the Fermi level must be solved for iteratively. In this work, the number of occupied conduction states is increased from zero until the charge neutrality condition is satisfied, i.e. inside a finite area, the number of electrons is equal to the num- ber of donor atoms (each donor atom contributing one donor electron). The energy at which charge neutrality is achieved is defined as the energy of the Fermi level. The total number of electrons at energy E can be writ- ten as N (E) = s f (E) Z(E) , (15) where s is the spin degeneracy, Z(E) is the eDOS eval- uated over the first 2D BZ, and f (E) is the Fermi-Dirac distribution function. For our calculations s is equal to one, therefore, N (E) = Z(E)(cid:18)1 + exp(cid:20) E − EF kBT (cid:21)(cid:19)−1 , (16) where EF is the Fermi level, kB is Boltzmann's constant, and T is temperature. The area of the first 2D BZ is converted to a real space area and the number of donors is calculated from the areal doping density. EF is then solved for iteratively by evaluating Eq. 16 at energies greater than the bulk valence band minimum until charge neutrality is achieved. III. ANALYSIS OF THE MODEL FOR A SI:P δ-LAYER Si is an indirect band gap semiconductor and has a six-fold degenerate CB minimum at ±0.81A in Figure 1, where A = X, Y, Z are points of high symmetry in the face-centred cubic (FCC) BZ.48 Figure 1 shows the band structure of bulk Si plotted in the FCC BZ on the path of high symmetry: L to Γ to X. Figure 2 shows the band structure (CB only) of a Si:P δ-layer plotted in the tP BZ on the path of high symmetry: 0.3M to Γ to 0.6X. The path Γ to Y is equivalent to the path Γ to X by the four-fold symmetry of the Si(001) doping plane and therefore not shown. The path Γ to Z is affected by zone-folding36,49 because of the height of the tP supercell in the z direction and therefore also not shown. The change from an FCC unit cell to a tP supercell projects the conduction valleys in the ±z directions to Γ. These conduction bands are labelled 1Γ and 2Γ in Figure 2. The vertical confinement of electrons by the donor potential shifts these conduction valleys into the bulk band gap region and splits their energy minima by the 1Γ/2Γ valley splitting.32 The change to a tP supercell also folds the bands in the ±x and ±y directions. The ±0.81X and ±0.81Y valley minima of bulk Si are folded to ±0.37X and ±0.37Y respectively. The +X valley is labelled as 1∆ in Figure 2. The projection of the Si conduction valleys is illustrated in Ref. 24 [Fig. 1]. The 1D confinement of the donor electrons in the z di- rection shifts the lowest conduction states of bulk Si into the bulk band gap region. Figure 2 shows three conduc- tion valleys to be partially occupied, or partly below the Fermi level (EF): the 1Γ, 2Γ and 1∆ bands. The Si:P δ-layer is metallic; the CB minimum (E1Γ) is 285 meV below EF. E1Γ is 427 meV below the CB minimum of bulk Si and, in Table II, agrees with the results of all mod- els (excluding DFT1D (LAO)33 and DFT3D34) to within 28 meV. The minimum of the 2Γ band (E2Γ) compares ) V e ( y g r e n E 5 4 3 2 1 0 -1 -2 L Γ X FIG. 1. Band structure of bulk Si calculated using an sp3d5s∗ TB method.16,17 The bands are plotted in the FCC BZ. 0.1 0 -0.1 -0.2 -0.3 -0.4 ) V e ( y g r e n E 4 EF 1∆ E1∆ 2Γ 1Γ E1Γ E2Γ 0.3M Γ 0.6X FIG. 2. Band structure (12 lowest conduction bands only) of a Si:P δ-layer (1/4 ML) [solid lines], the energy minima of the lowest three conduction bands [dotted lines], and the band structure (CB only) of bulk Si [dashed lines].16,17 The bands are plotted in the tP BZ and the energy offset has been chosen to set the bulk Si CB minimum to zero. TABLE II. Energies of conduction valley minima and Fermi levels (meV) at low temperature for a variety of Si:P δ-layer models (1/4 ML). The models are arranged in order of in- creasing magnitude of E1Γ. This work is highlighted in bold. Uncertainty of ±3 meV is labelled by †. Model (basis set) DFT1D (LAO)33 † DFT3D (LAO)36 TB3D (sp3d5s∗)31 DFT1D (PWO)24 † TB1D (sp3d5s∗) EMT1D38 † E1Γ -296 -369 -401 -419 -427 -445 E2Γ -288 -269 -375 -394 -421 -426 E1∆ -165 -68 -249 -250 -287 -257 EF -72 -23 -115 -99 -142 N/A equally well to DFT1D (PWO)24 and is within 5 meV of EMT1D.38 Interestingly, the 1∆ valley minimum (E1∆) of TB1D is 30–38 meV lower than that predicted by the three closest models in Table II. In general, there is good agreement between the literature and TB1D. Contrast- ingly, values for EF differ significantly between the mod- els. This disparity can be explained by the variety in the exact method and ~k-point grid used to solve for the Fermi level between each of the models. To calculate EF, we use a similar procedure to TB3D31 and although TB1D predicts EF to be 27 meV lower than TB3D,31 this discrepancy is explained by the 26 meV difference between the values for E1Γ. The calculated binding en- ergies (EF − E1Γ) of TB3D31 and TB1D agree within 1 meV (see Table III). Similarly, the greatest contributing factor to the dis- crepancies in the values of the conduction valley min- ima between the models in Table II can be explained by the exact methods used to estimate the XC energy. An examination of the conduction valley energy minima of DFT1D (PWO)24 shows a 10±3 meV shift in E1Γ when XC and short-range effects are added to the model, com- pared to a -34 meV shift in the E1Γ for DFT3D31 when XC effects are included (see Ref. 29). These XC energy corrections are relatively small compared to the -130 meV shift in E1Γ that corresponds to including XC effects in TB1D (see Figure 3 for a comparison of the donor po- tentials with and without XC). To approximate the XC energy correction to the semiclassical component of the donor potential, we use the same procedure as EMT1D. In Table II, the values of the conduction valley minima for TB1D agree most strongly with EMT1D. The method used by DFT1D (PWO)24 to estimate the XC energy correction is available and in future could be used to test the validity of the two different approximations to XC. Unfortunately, the details of the exact implementation of XC in TB3D31 are not available. However, the lack of quantitative experimental results for the Si:P δ-layer system means verification of any of the approximations to XC effects is currently impractical. Figure 3 shows the TF donor potential with and with- out XC. The energies of the conduction valley minima are shown as energy levels inside the potential well. XC ef- fects increase the magnitude of the TF potential. We find exchange effects dominate over correlation effects at the nominal 1/4 ML doping density. The exchange poten- tial shifts the energies of the CB minimum by -125 meV, where as the correlation potential shifts the energies by only -5 meV. We note that the system is therefore ade- quately described by the TFD donor potential and that the correlation potential could be ignored for simplicity. The degeneracy of the Si CB minimum results in a valley splitting between the 1Γ and 2Γ bands, named the 1Γ/2Γ valley splitting33 (equal to E2 − E1). Understand- ing this energy splitting is important in the design of "few-electron" quantum electronics in silicon50 and mod- elling transport properties in these systems at low voltage biases.6 Table III shows the values of E2 − E1 calculated by a variety of Si:P δ-layer models. The 1Γ/2Γ valley splitting calculated by TB1D is 6 meV and agrees with that of DFT1D (LAO)33 to within 2±6 meV. There is a difference of 20 meV between the valley splitting pre- dicted by TB3D31 and TB1D. There are also differences of 14±6 and 21±6 meV between the valley splitting pre- dicted by EMT1D38 and DFT1D (PWO)24 respectively and TB1D. These discrepancies can be explained by the confinement of electrons in the z direction.33,38 The val- ley splitting is inversely proportional to the spatial extent of the donors perpendicular to the doping plane.35 There are two characterisitcs of the donor potential that affect the confinement of the donor electrons. These are the rate of decay of the gradient of the potential in the z direction and the depth of the potential well. A com- parison of the donor potentials and valley splittings be- tween the models is consistent with this statement. The donor potential of DFT1D (PWO)24 (see Ref. 24 [Fig. 3]) compares well to Figure 3. The DFT1D (PWO)24 donor potential has a similar depth (∼0.75 eV) to the TB1D potential (0.78 eV) however the rate of decay of ) V e ( l a i t n e t o p n o r t c e l e r o n o D 0 -0.2 -0.4 -0.6 -0.8 5 EF E1∆ E2Γ E1Γ TF + XC TF −6 −4 −2 0 2 4 6 Distance from delta-layer in z (nm) FIG. 3. Donor potential for a Si:P δ-layer (1/4 ML) with [solid line] and without [dotted line] XC effects. The Fermi level and energies of the conduction valley minima are shown as energy levels inside the potential well. TABLE III. Values of valley splitting and binding energy (meV) at low temperature for for a variety of Si:P δ-layer (1/4 ML) models and experiment. The models are arranged in order of increasing EF − E1Γ. This work is highlighted in bold. Uncertainty of ±3 meV is labelled by †. Model (basis set) DFT1D (LAO)33 † Experiment12 (T = 100K) TB1D (sp3d5s∗) TB3D (sp3d5s∗)31 DFT1D (PWO)24 † DFT3D (LAO)36 EMT1D38 E2Γ − E1Γ EF − E1Γ 8 N/A 6 26 25 100 19 224 ∼270 285 286 320 346 470† the gradient of the potential in the z direction is larger for DFT1D (PWO)24 (decaying to zero by ∼4 nm). The larger rate of decay results in stronger confinement of the electrons in the z direction and a larger valley splitting (26±6 meV). The rate of decay of the gradient of the po- tential in the z direction and the depth of the potential are sensitive to ML doping density (see Table IV), donor atom configuration28,33 and basis set size.36 We note that including XC effects in TB1D and DFT1D (PWO)24 and short-range effects in DFT1D (PWO)24 does not affect the magnitude of the valley splitting. And, that when a large basis set is used effects of basis set size become relatively minor.38 Therefore, it is the ML doping density which controls the size of the 1Γ/2Γ valley splitting and it is understanding the proportionality between these prop- erties which is important for engineering valley splitting in Si:P δ-layer systems. Recently published experimental measurements of the Si:P δ-layer band structure performed at T = 100 K confirm the existence of at least one occupied state at Γ.12 This state, named the δ-state, has a binding energy (EF − E1) of ∼190 meV at T = 300 K and ∼270 meV at T = 100 K. A comparison to the theory is difficult as all Si:P δ-layer models explicitly assume lower tem- peratures, in the neighbourhood of T = 4 K. However, it is obvious from Table III that the experimental bind- ing energy of ∼270 meV at T = 100 K compares well to the value of 285 meV calculated by TB1D. At lower experimental temperatures we expect the binding energy of the δ-state to continue to increase as thermal exci- tations are reduced, increasing the confinement of the donor electrons and decreasing the electron screening of the δ-state by intrinsic charge carriers. It is not obvious from the experimental measurements whether there is a valley splitting at Γ. However, the energy resolution of the experimental band structure measurement is low and energy splittings of less than ∼70 meV are not resolvable. Ref. 12 suggests that the 1Γ/2Γ valley splitting is either small and not able to be resolved or large (>150 meV) and significantly underestimated by the theoretical mod- els available in the literature. In Figure 2(b) of Ref. 12 the 1Γ, 2Γ and 1∆ conduction bands calculated in Ref. 33 [Figure 10(d)] are overlaid to show the disparity between theory and experiment. However, Ref. 33 has been shown to overestimate the 1Γ/2Γ valley splitting by ∼50%.36 We expect the experimental valley splitting to increase with stronger confinement at lower temperatures and therefore suggest that a small valley splitting (<70 meV) is more likely. This is consistent with the Si:P δ-layer models in the Table III (excluding DFT3D38). Table IV shows the sensitivity of CB minimum, valley splitting, Fermi level and binding energy to changes in ML doping density. E1Γ increases from -427 meV at the nominal doping density of 1/4 ML (1.696 × 1014 cm−2) to -38 meV at a density of 1/128 ML (5.3 × 1012 cm−2). This shows that at a doping density as low as 1/128 ML, E1Γ can still be resolved from the CB minimum of bulk Si. However, we note that the TF donor potential is most accurate at high doping densities20 (>1013 cm−2) or at ML doping densities greater than 1/64 ML. In the range of doping densities greater than 1/64 ML, E1Γ and EF show a linear response to changes in ML doping den- sity. This trend agrees with DFT1D (PWO),24 as shown in Table IV. The strongest agreement between the Fermi levels of the two models is at a doping density of 1/16 ML TABLE IV. Energies of the CB minimum, valley splitting, Fermi level and binding energy (meV) of a Si:P δ-layer at low temperature for a range of ML doping densities calculated by TB1D and using DFT1D (PWO)24 for comparison. 1/8 ML doping density 1/4 nD (1014 cm−2) E1Γ E2Γ − E1Γ EF EF (DFT1D24) EF − E1Γ 1/16 1/32 1/64 1/128 1.696 0.847 0.424 0.212 0.106 0.053 -38 -427 0 -142 -20 -111 N/A -62 N/A -36 N/A 285 -95 1 -43 -252 -152 -59 0 -29 30 18 162 92 52 6 2 -90 1 -60 6 at which they agree to within 1 meV. A consequence of this behaviour is that the binding energy (EF − E1Γ) also shows a linear response to changes in areal doping density. The TB1D model calculates a binding energy (EF − E1Γ) of 52 meV at a doping density of 1/32 ML which is close to the experimental Si:P single donor level of 45.6 meV.51 This suggests that at a doping density of ∼1/32 ML, the separation of the P atoms is sufficient to recover the physics of a single P donor in Si. An areal doping density of 1/32 ML corresponds to an average donor separation of 2.17 nm. This agrees with the effec- tive Si:P Bohr radius in the effective mass approximation (1.96 nm) calculated using Eq. 8 and the parameters in Table I. It also agrees with Ref. 52 [Figure 1, η = 5.8] which shows the probability density of a single P donor in Si to have decayed significantly by ∼2 nm of bulk Si cladding. This suggests the TB1D model is applicable at ML doping densities as low as ∼2×1013 cm−2. IV. APPLICATION OF THE MODEL TO A GE:P δ-LAYER The similar electronic properties of Ge to Si make it equally attractive for the manufacture of nanoelectronic devices. Carrying on from the success of the δ-doped Si:P fabrication route, δ-doped P in Ge (Ge:P) structures are now being fabricated on the nanometre scale utilising the same in situ doping techniques.53,54 Presently, exper- iments are focused on highly doped Ge:P δ-layers54–56 which could serve as the channel material in the next generation of high-speed nano-transistors.57 Ultra-scaled n-type Ge devices on the Si platform could also be re- alised through the patterning of Ge:P δ-layers into source and drain contacts or quantum wires – analogous to Si:P δ-layers.9,10,58 Recently fabricated Ge:P δ-layers exhibit extremely high 2D carrier concentrations of 1/4 ML – 1/2 ML,56,59 ) V e ( y g r e n E 5 4 3 2 1 0 -1 -2 L Γ X FIG. 4. Band structure of bulk Ge calculated using an sp3d5s∗ TB method.16,17 The bands are plotted in the FCC BZ. 7 ) V e ( y g r e n E 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 EF = −170 meV 2M 1M E1M E2M 1∆ E1∆ 2Γ 1Γ E1Γ E2Γ E1∆ = −192 meV E2Γ = −308 meV E1Γ = −313 meV E2M = −369 meV E1M = −385 meV M Γ X FIG. 5. Band structure (12 lowest conduction bands only) of a Ge:P δ-layer (1/4 ML) [solid lines], the energy minima of the lowest five conduction bands [dotted lines], and the band structure (CB only) of bulk Ge [dashed lines].16,17 The bands are plotted in the tP BZ and the energy offset has been chosen to set the bulk Ge CB minimum to zero. ) V e ( l a i t n e t o p n o r t c e l e r o n o D 0 -0.2 -0.4 -0.6 -0.8 E1∆ E1Γ E1M EF E2Γ E2M Ge:P layer Si:P layer 50 40 30 20 10 ) s t i n u . b r a ( S O D e Ge:P layer Si:P layer Bulk Ge Bulk Si EF,Ge:P EF,Si:P −8 −6 −4 −2 0 2 4 6 8 Distance from delta-layer in z (nm) 0 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 Energy (eV) FIG. 6. Donor potential for a Ge:P δ-layer [solid line] and Si:P δ-layer (1/4 ML) [dotted line]. The Fermi level and energies of the conduction valley minima are shown as energy levels in the potential well for the Ge:P δ-layer [solids lines] and Si:P δ-layer [dotted lines]. FIG. 7. eDOS for the Ge:P δ-layer [heavy solid line] and Si:P δ-layer (1/4 ML) [dashed line], bulk Ge [light solid line] and bulk Si [dotted line]. A Gaussian smearing of 25 meV has been applied for visualisation. which is equivalent to complete substitutional doping of the Ge(001) surface.2 However, presently there is no the- oretical model for the electronic properties of these novel Ge:P nanoelectronic structures. Here, we use our TB1D model to rectify this; comparing the electronic properties of a Ge:P δ-layer to that of a Si:P δ-layer. There is an a priori argument for applying the TB1D model to a Ge:P δ-layer as unlike DFT,60 TB methods are able to repro- duce the band structure of bulk Ge as easily as that of bulk Si (without the need for computationally expensive hybrid density functionals). The bulk band structure of Ge is shown in Figure 4. Ge is an indirect band gap semiconductor and has a four- fold degenerate CB minimum at L in the FCC BZ. The change from an FCC unit cell to a tP supercell projects the conduction valleys at L to M in the tP BZ. In Fig- ure 5, these bands are labelled 1M and 2M . A valley splitting of 16 meV is predicted between the 1M and 2M band minima. The 1D confinement of the donor electrons shifts the lowest lying conduction bands into the bulk band gap region. The 1M and 2M bands are each doubly degenerate as Ge has four equivalent lowest conduction valleys. Similarly to Si, the change from a FCC unit cell to a tP supercell also folds the bands in the ±X and ±Y directions. In Figure 5, the ±0.88X and ±0.88Y valley minima of bulk Ge are folded to ±0.24X and ±0.24Y respectively. The ±Z conduction valleys are folded to Γ. In Figure 5, these bands are labelled 1Γ and 2Γ. A valley splitting of 6 meV is predicted between the 1Γ and 2Γ band minima. Interestingly, the conduction TABLE V. Energies of the CB minimum, valley splitting, Fermi level and binding energy (meV) of a Ge:P δ-layer for a range of ML doping densities. ML doping density nD (1014 cm−2) E1M E2M − E1M EF EF − E1M 1/4 1.563 -385 16 -170 215 1/8 0.781 -211 5 -61 150 1/16 0.391 -117 2 -36 81 1/32 0.195 -66 1 -26 40 1/64 0.098 -37 1 -15 22 valley at Γ in the band structure of bulk Ge is not shifted into the bulk band gap region to the extent of the M and X bands. The projected Z conduction valleys dominate the CB minima of the Ge:P δ-layer band structure at Γ. This is explained by the confinement of the CB minima which is proportional to the curvature of the conduction valleys.38 A comparison of the bands in Figure 4 shows the conduction valley at Γ is narrower than the bands at M and X, and it is therefore not shifted as far into the bulk band gap region. This is shown in Figure 5 where the Γ band of bulk Ge is above the Fermi level. Therefore, at low voltage biases we expect the transport properties of the Ge:P δ-layer to be dominated by the four projected L conduction valleys and two projected Z conduction valleys. The donor potential for the Ge:P δ-layer is shown in Figure 6. At the nominal doping density of 1/4 ML the larger Bohr radius of Ge results in a donor potential which is broadened in the z direction but which exhibits a similar minimum value (0.85 eV) to the donor potential of the Si:P δ-layer (0.78 eV). Therefore, we expect the CB minimum (E1M ) of the Ge:P δ-layer to be shifted further into the bulk band gap region than the CB minimum of the Si:P δ-layer. However, comparing Figures 2 and 5 the opposite is true, the Si:P 1Γ band is shifted further into the bulk band gap region than the Ge:P 1M band. This can be explained by the eDOS near the CB minimum. In Figure 7, there is a larger number of conduction states near the CB minimum which decreases the confinement of any one state. Figure 7 shows the eDOS of the Si:P and Ge:P δ-layers. Excluding energies in the neighbourhood of -0.25 eV, the eDOS of the Ge:P δ-layer is greater than that of the Si:P δ-layer on the domain (-0.4, 0.0) eV. This is explained by the larger number of Ge:P bands or conducting modes below the Fermi level and suggests a higher conductivity for the Ge:P δ-layer than the Si:P δ-layer at low volt- age biases. The larger eDOS for the Si:P δ-layer in the neighbourhood of E = −0.25 eV corresponds to the fill- ing of the Si:P 1∆ band (E1∆,Si:P = −287 meV) before the Ge:P 1∆ band (E1∆,Ge:P = −192 meV). The eDOS of bulk Ge and bulk Si show the extent to which the con- duction valleys are confined or shifted into the bulk band gap region. Table V shows the sensitivity of CB minimum, valley splitting, Fermi level and binding energy to changes in 8 ML doping density. Similarly to the Si:P δ-layer, the CB minimum of the Ge:P δ-layer (E1M ) shows a lin- ear response to changes in areal doping density. This is not true of the Fermi level and therefore binding en- ergy. Their non-linear behaviour can be explained by the larger number of partially filled bands below the Fermi level. Interestingly, at a doping density of 1/4 ML the value of the 2M/1M valley splitting is equal to more than twice the energy splitting at Γ. This suggests the 1M and 2M bands are more strongly confined than the 1Γ and 2Γ bands. This is expected as the CB minimum of bulk Ge at L is 225 meV below the X conduction valley minimum (see Figure 4). The 2M/1M valley splitting of the Ge:P δ-layer (16 meV) is 10 meV greater than the 1Γ/2Γ val- ley splitting of the Si:P δ-layer (6 meV). Therefore, we expect experimental measurements of the Ge:P δ-layer to be more sensitive to the energy splitting of the lowest conduction valleys. The binding energy of the Ge:P δ- state is 70 meV less than that of the Si:P δ-state and we therefore expect this to be resolvable using current ex- perimental techniques. This prediction is consistent with the larger Bohr radius of Ge. V. CONCLUSIONS We find the TB1D model produces results that are comparable to more rigorous TB and DFT calculations of Si:P δ-layers. A self-consistent 1D donor potential is therefore adequate to describe the physics of this highly confined and densely doped 2D system. Self-consistency is built into the semi-classical component of the donor potential. Therefore, we avoid the need to solve itera- tively for the electrostatic field, increasing the computa- tional efficiency of our calculations. The TFD formalism is known to be exact in the high density limit and here is shown to produce meaningful results at ML doping densities as low as ∼2×1013 cm−2. The calculated values for the conduction valley min- ima, valley splittings and Fermi levels agree with the most recent theoretical models and experimental mea- surements, and are internally consistent with the calcu- lated donor potential. We believe this validates both the TB1D model and the addition of the 1D donor potential to only the diagonal terms of the Hamiltonian matrix. XC effects are shown to be a major source of discrepancy between the models discussed herein. At the nominal 1/4 ML doping density, the XC potential for the Si:P δ-layer is found to be dominated by exchange effects (in agreement with Ref. 38). Correlation effects are therefore unimportant in the Si:P δ-layer system. The Ge:P δ-layer is shown to have conduction valley minima at M , Γ and ±0.24X in the tP BZ. A Ge:P δ- state, analogous to the Si:P δ-state12 is predicted at M with a 2M/1M valley splitting of 16 meV. We therefore expect experimental measurements to be more sensitive to the energy splitting of the lowest conduction valleys in a Ge:P δ-layer. The binding energy of this Ge:P δ-state is calculated to be 70 meV smaller than the binding energy of the Si:P δ-state, which suggests that this difference is experimentally resolvable using current experimental techniques. The Ge:P δ-layer exhibits a larger number of partially occupied conducting modes than the Si:P δ- layer. This suggests a higher conductivity for the Ge:P δ-layer at low voltage biases. where, as a consequence of the uncertainty principle, N = 2 h3 V . V is the total occupied volume of phase space, V = 4πν 3 {pF(z)}3V , ACKNOWLEDGMENTS where ν is the number of equivalent conduction valleys. Therefore, 9 The authors' thank A. Budi and D.W. Drumm for use- ful discussions. Appendix: Derivation of the electrostatic potential of the donor electrons in the Thomas-Fermi approximation This derivation follows in part that of March,13 and Ioriatti20 for the TF theory and TF theory of δ-doped semiconductor structures. In what follows, n is electron density and z is a coordinate that runs perpendicular to the doping plane. SI units are used throughout. The TF energy-density functional is written, ETF[n(z)] = T [n(z)] + Uee[n(z)] + UeN[n(z)] , where T [n(z)] is the kinetic energy density functional and, Uee[n(z)] and UeN[n(z)] are the electron-electron and electron-nuclear potential energy density function- als. T [n(z)] = Z t dz , (A.1) where N = 2 h3 4πν 3 {pF(z)}3V , n(z) = 8πν 3h3 {pF(z)}3 {p(z)}5 = (cid:18) 3h3 8πν n(z)(cid:19)5/3 . and Substituting into Eq. A.2, t = t = t = 8πν {p(z)}4 dp 2 ¯mh3 Z pF 8πν 10 ¯mh3 [pF(z)]5 10 ¯mν2/3 (3π23)2/3{n(z)}5/3 0 3 t = ck{n(z)}5/3 , ck = 3 10 ¯mν2/3 (3π23)2/3 . where t is the kinetic energy per unit volume and can be written as t = Z pF 0 n(z) {p(z)}2 2 ¯m Iz(p) dp . Iz(p) dp is the probability of an electron having momen- tum between p(z) and p(z) + dp and can be written (to first order in dp) as Finally, the kinetic energy density functional can there- fore be written as T [n(z)] = ckZ n(z)5/3 dz. The electron-electron potential energy density functional is defined as Iz(p) dp = 3{p(z)}2 {pF(z)}3 dp , Uee[n(z)] = e2 8πǫrǫ0 Z Z n(~r′)n(~r) ~r − ~r′ d~r d~r′ where pF is the maximum momentum of an electron. Therefore, t = Z pF 0 n(z) {p(z)}2 2 ¯m 3{p(z)}2 {pF(z)}3 dp . (A.2) In the local density approximation, and for an infinite plane of charge can be written as Uee[n(z)] = − e2 4ǫrǫ0 Z Z n(z) n(z′) z − z′ dz dz′ . Similarly, the electron-nuclear potential energy density functional is defined as n(z) = n0 = N V UeN [n(z)] = eZ n(~r)VN(~r) d~r and for an infinite plane of charge can be written as UeN [n(z)] = − e2nD 2ǫrǫ0 Z n(z) z dz . the self-consistency of the electrostatic field is achieved by insisting VTF and n are related by the Poisson equation, ∂2 ∂z2 VTF(z) = e2 ǫrǫ0 n(z) − e2nD ǫrǫ0 δ(z) . (A.5) The TF energy-density functional is then Rewriting Eq. A.4, 10 ETF[n(z)] = ckZ n(z)5/3 dz − e2nD 2ǫrǫ0 Z n(z) z dz 4ǫrǫ0 Z Z n(z) n(z′) z − z′ dz dz′ . e2 − (A.3) We seek to minimise this expression for the total energy by varying the electron density subject to a constraint, namely that the total number of electrons remains con- stant. This is expressed through the variational state- ment, δ(E − µN ) = 0 , where µ is the Lagrangian multiplier and the chemical potential, and the total number of electrons is given by N = Z n(z) dz . From this statement it follows that µ = ∂E ∂N . Carrying out the functional derivative of Eq. A.3 with respect to N , µ = 5 3 ck{n(z)}2/3 − e2nD 2ǫrǫ0 z − e2 2ǫrǫ0 Z n(z)z − z′ dz . (A.4) Then, defining the electrostatic part of Eq. A.4 as VTF(z) = − e2nD 2ǫrǫ0 z − e2 2ǫrǫ0 Z n(z)z − z′ dz , ∗ [email protected] † [email protected] ‡ [email protected] 1 H. Wilson, O. Warschkow, N. A. Marks, N. J. S. R. Schofield, T. Reusch, M. Radny, and M. Y. Simmons, Curson, P. Smith, D. R. McKenzie, Physical Review B 74, 195310 (2006). 2 G. Scappucci, O. Warschkow, G. Capellini, W. M. and M. Y. Simmons, Klesse, D. R. McKenzie, Physical Review Letters 109, 076101 (2012). T.-C. Tucker Shen, 3 J. and R. µ = 5 3 ck{n(z)}2/3 + VTF(z) and rearranging for n(z), n(z) = (cid:18) 3 5ck(cid:19)3/2 (cid:0)µ − VTF(z)(cid:1)3/2 . We substitute this expression for n(z) into Eq. A.5, which results in the differential equation, ∂2 ∂z2 VTF(z) = e2 ǫrǫ0 (cid:18) 3 5ck(cid:19)3/2 − e2nD ǫrǫ0 δ(z) . (cid:0)µ − VTF(z)(cid:1)3/2 (A.6) The solution to equation A.6 is of the form, VTF(z) − µ = − α2 (αz + z0)4 where and α = (2 ¯m)3/2e2ν 60π2ǫrǫ03 z0 = (cid:18) 8α3ǫrǫ0 e2nD (cid:19)1/5 . Finally, we set the chemical potential equal to zero, which results in the following expression for the electrostatic potential of the donor electrons, VTF(z) = − α2 (αz + z0)4 . Solid-State Electronics 42, 1061 (1998). 4 J. L. OBrien, S. R. Schofield, M. Y. Simmons, R. G. Clark, A. S. Dzurak, N. J. Curson, B. E. Kane, N. S. McAlpine, M. E. Hawley, and G. W. Brown, Physical Review B 64, 161401 (2001). 5 T.-C. Shen, J. J.-Y. S. Kline, R. Du, Applied Physics Letters 80, 1580 (2002). and Ji, M. A. Zudov, R.- J. R. Tucker, 6 M. Fuechsle, M. Friesen, M. A. Eriksson, Nature nanotechnology 5, 502 (2010). S. Mahapatra, F. A. Zwanenburg, and M. Y. Simmons, 7 A. Fuhrer, M. Fuchsle, T. C. G. Reusch, B. Weber, and 33 D. J. Carter, N. A. Marks, O. Warschkow, and D. R. 11 M. Y. Simmons, Nano Letters 9, 707 (2009). 8 W. Pok, T. C. G. Reusch, G. Scappucci, F. J. and M. Y. Simmons, Ruess, A. R. Hamilton, IEEE Transactions on Nanotechnology 6, 213 (2007). 9 S. R. McKibbin, G. Scappucci, W. Pok, and M. Y. Sim- mons, Nanotechnology 24, 045303 (2013). 10 B. Weber, S. Mahapatra, H. Ryu, S. Lee, A. Fuhrer, T. C. G. Reusch, D. L. Thompson, W. C. T. Lee, G. Klimeck, L. C. L. Hollenberg, and M. Y. Simmons, Science 335, 64 (2012). 11 M. Fuechsle, J. A. Miwa, S. Mahapatra, H. Ryu, S. Lee, O. Warschkow, L. C. L. Hollenberg, G. Klimeck, and M. Y. Simmons, Nature nanotechnology 7, 242 (2012). 12 J. A. Miwa, P. Hofmann, M. Y. Simmons, and J. W. Wells, Physical Review Letters 110, 136801 (2013). 13 N. H. March, in Theory of the inhomogeneous electron gas, edited by S. Lundqvist and N. H. March (Plenum Press, New York, 1983) Chap. 1, pp. 1–7. 14 R. G. Parr and W. Yang, Density-Functional Theory of Atoms and Molecules (Oxford University Press, New York, 1989). 15 J. P. Perdew and Y. Wang, McKenzie, Nanotechnology 22, 065701 (2011). 34 D. W. Drumm, J. S. A. Budi, L. C. L. Hollenberg, Physical Review Letters 110, 126802 (2013). Smith, M. C. Per, and S. P. Russo, 35 A. Budi, D. W. Drumm, M. C. Per, A. Tre- and L. C. L. Hollenberg, 36 D. W. gonning, S. P. Russo, Physical Review B 86, 165123 (2012). Budi, L. C. S. Nanoscale research letters 8, 111 (2013). 37 L. Scolfaro, D. Beliaev, R. Enderlein, P. Russo, A. and Drumm, C. M. Per, L. Hollenberg, and J. Leite, Physical Review B 50, 8699 (1994). 38 D. W. Drumm, L. C. L. Hollenberg, M. Y. Simmons, and M. Friesen, Physical Review B 85, 155419 (2012). 39 J. C. Slater and G. F. Koster, Physical Review 94, 1498 (1954). Bowen, Klimeck, R. 40 G. T. C. Salazar-Lazaro, T. A. Cwik, Superlattices and Microstructures 27, 77 (2000). B. Boykin, and A. Stoica, 41 R. Ramiarez and M. C. Bohm, International Journal of Quantum Chemistry 30, 391 (1986). 42 R. Faulkner, Physical Review 184, 713 (1969). 43 J. Hensel, H. Hasegawa, and M. Nakayama, Physical Review B 45, 13244 (1992). 16 J.-M. Jancu, R. Scholz, F. Beltram, Physical Review B 57, 6493 (1998). 17 T. B. Boykin, G. Klimeck, Physical Review B 69, 115201 (2004). 18 L. H. Thomas, Math. Proc. Cambridge 23, 542 (1927). 19 E. Fermi, Zeitschrift fur Physik 48, 73 (1928). 20 L. Ioriatti, Physical Review B 41, 8340 (1990). 21 L. M. Gaggero-Sager, M. E. Mora-Ramos, and D. A. Contreras-Solorio, Physical Review B 57, 6286 (1998). 22 S. J. Vlaev and L. M. Gaggero-Sager, Physical Review B 58, 1142 (1998). and F. Bassani, Physical Review 138, A225 (1965). and F. Oyafuso, Journal of Physics and Chemistry of Solids 20, 281 (1961). 44 B. Levinger and D. Frankl, 45 M. Graf and P. Vogl, Physical Review B 51, 4940 (1995). 46 W. Rice, and Physical Review B 7, 1508 (1973). Brinkman T. 47 P. A. M. Dirac, Math. Proc. Cambridge 26, 376 (1930). 48 C. J. Bradley and A. P. Cracknell, The mathematical the- ory of symmetry in solids: Representation theory for point groups and space groups, 1st ed. (Oxford University Press, London, 1972). 23 I. Rodr´ıguez-Vargas and L. M. Gaggero-Sager, 49 T. B. Boykin and G. Klimeck, Journal of Applied Physics 99, 033702 (2006). Physical Review B 71, 115215 (2005). 24 G. Qian, Y.-C. Chang, and Physical Review B 71, 045309 (2005). J. R. Tucker, 25 Y.-C. Chang and G. Li, Computer Physics Communications 95, 158 (1996). 50 R. Rahman, J. Verduijn, N. Kharche, G. P. Lansber- and S. Rogge, gen, G. Klimeck, L. C. L. Hollenberg, Physical Review B 83, 195323 (2011). 51 A. K. Ramdas and S. Rodriguez, 26 X. Cartoix`a and Y.-C. Chang, Reports on Progress in Physics 44, 1297 (1981). Physical Review B 72, 125330 (2005). 27 R. Rahman, C. J. Wellard, F. R. Bradbury, M. Prada, and L. C. L. Hollenberg, J. H. Cole, G. Klimeck, Physical Review Letters 99, 036403 (2007). 28 S. Lee, H. Ryu, H. Campbell, L. C. L. Hol- and G. Klimeck, lenberg, M. Y. Physical Review B 84, 205309 (2011). Simmons, 52 C. J. Wellard C. Physical Review B 72, 085202 (2005). and L. L. Hollenberg, 53 G. Scappucci, G. Capellini, B. Johnston, W. M. Simmons, and M. Y. Klesse, Nano Letters 11, 2272 (2011). J. A. Miwa, 54 G. Scappucci, G. Capellini, W. M. Klesse, and M. Y. Simmons, Nanoscale 5, 2600 (2013). 29 H. Ryu, S. Lee, and G. Klimeck, in 55 G. Scappucci, G. Capellini, W. C. T. Lee, and M. Y. 13th International Workshop on Computational Electronics (2009) pp. 1–4. Simmons, Applied Physics Letters 94, 162106 (2009). 56 G. Scappucci, G. Capellini, W. M. Klesse, and M. Y. 30 G. Klimeck, S. S. Ahmed, H. Bae, N. Kharche, R. Rah- man, S. Clark, B. Haley, S. Lee, M. Naumov, H. Ryu, F. Saied, M. Prada, M. Korkusinski, and T. B. Boykin, in IEEE Transactions on Electron Devices (2007) pp. 2079– 2089. 31 H. Ryu, S. Lee, B. Weber, S. Mahapatra, L. C. L. and G. Klimeck, Hollenberg, M. Y. Simmons, Nanoscale (2013), 10.1039/c3nr01796f. 32 D. J. Carter, O. Warschkow, N. A. Marks, and D. R. McKenzie, Physical Review B 79, 033204 (2009). Simmons, Nanotechnology 22, 375203 (2011). 57 Y. Kamata, Materials Today 11, 30 (2008). 58 G. Scappucci, W. M. Klesse, A. R. Hamilton, G. Capellini, D. L. Jaeger, M. R. Bischof, R. F. Reidy, B. P. Gorman, and M. Y. Simmons, Nano Letters 12, 4953 (2012). 59 G. Scappucci, G. Capellini, W. M. Klesse, and M. Y. Simmons, Journal of Crystal Growth 316, 81 (2011). 60 I. Remediakis and E. Kaxiras, Physical Review B 59, 5536 (1999).
1906.08838	1	1906	2019-06-20T20:34:10	Defect-implantation for the all-electrical detection of non-collinear spin-textures	[ "cond-mat.mes-hall" ]	The viability of past, current and future devices for information technology hinges on their sensitivity to the presence of impurities. The latter can lead to resistivity anomalies, the so-called Kondo effect, reshape extrinsically Hall effects or reduce the efficiency of magnetoresistance effects essential in spintronics. Here we demonstrate that atomic defects enable highly efficient all-electrical detection of spin-swirling textures, in particular magnetic skyrmions, which are promising bits candidates in future spintronics devices. Impurities amplify the bare transport signal and can alter significantly the spin-mixing magnetoresistance (XMR) depending on their chemical nature. Both effects are monitored in terms of the defect-enhanced XMR (DXMR) as shown for 3d and 4d transition metal defects implanted at the vicinity of skyrmions generated in PdFe bilayer deposited on Ir(111). The ineluctability of impurities in devices promotes the implementation of DXMR in reading architectures with immediate implications in magnetic storage technologies.	cond-mat.mes-hall	cond-mat	Defect-implantation for the all-electrical detection of non- collinear spin-textures Imara Lima Fernandes1, Mohammed Bouhassoune1, and Samir Lounis1 1Peter Grunberg Institut and Institute for Advanced Simulation, Forschungszentrum Julich and JARA, D-52425 Julich, Germany *[email protected], [email protected] The viability of past, current and future devices for information technology hinges on their sensitivity to the presence of impurities. The latter can lead to resistivity anomalies, the so-called Kondo effect, reshape extrinsically Hall effects or reduce the efficiency of magne- toresistance effects essential in spintronics. Here we demonstrate that atomic defects ena- ble highly efficient all-electrical detection of spin-swirling textures, in particular magnetic skyrmions, which are promising bits candidates in future spintronics devices. Impurities amplify the bare transport signal and can alter significantly the spin-mixing magnetoresis- tance (XMR) depending on their chemical nature. Both effects are monitored in terms of the defect-enhanced XMR (DXMR) as shown for 3d and 4d transition metal defects implanted at the vicinity of skyrmions generated in PdFe bilayer deposited on Ir(111). The ineluctability of impurities in devices promotes the implementation of DXMR in reading architectures with immediate implications in magnetic storage technologies. Defects are inherent to all devices and materials. Being unavoidable, they dramatically re- shape transport properties, often negatively, and thus are key ingredients in settling the competi- 9 1 0 2 n u J 0 2 ] l l a h - s e m . t a m - d n o c [ 1 v 8 3 8 8 0 . 6 0 9 1 : v i X r a 1 tiveness of newly proposed technologies and therefore their survival. In the context of spintronics 1 exploiting the spin rather than the charge to carry information, defects can reduce the efficiency of magnetoresistance effects2, 3 in current perpendicular-to-plane geometries such as the giant magne- toresistance (GMR)4, 5 or tunneling magnetoresistance (TMR)6, 7. Impurities intrinsically give rise to inelastic transport channels allowing the exploration of electron-bosons interactions 8 -- 10 while generating extrinsic contributions to Hall effects 11 -- 16 in current-in-plane geometries. All of this is not surprising since the very fundamental Kondo effect17 results from diluted magnetic impurities leading to an anomalous behaviour of the resistance at low temperature18. Recently a new kind of magnetoresistance effect, the spin-mixing magnetoresistance (XMR), has been discovered 19, 20, which enables the all-electrical detection of non-collinear magnetic states, such as magnetic skyrmions and spin spirals. In contrast to the GMR or TMR effect, XMR requires a single magnetic electrode instead of two and thus its importance in establishing spin- swirling textures with chiral or topological protection properties, in particular skyrmions 21, 22, as future bits for information technology 23. While several investigations were devoted to the impact of inhomogeneities on the motion and stability of skyrmions24 -- 35, their implications in the electri- cal detection are yet to be determined. Similarly to the GMR and TMR effects, it is often expected that defects would reduce the XMR efficiency. In this article, we demonstrate from a full ab initio approach that contrary to the current wisdom, atomic defects are of technological importance in reading non-collinear spin-states since they enable a highly efficient magnetoresistance signal, where the reference transport signal is am- 2 plified by many of the investigated impurities (see Figure 1a). We envision scenarios, where im- purities are manipulated atom-by-atom36, implanted37 or spontaneously generated via intermixing mechanisms. We introduce the defect-enhanced XMR (DXMR) in order to evaluate the potential of defects in magnifying simultaneously the transport and XMR signals. We perform system- atic simulations of atomic resolved transport measurements as probed within scanning tunneling microscopy/spectroscopy (STM/STS) of 3d (V, Cr, Mn, Fe, Co, Ni) and 4d (Nb, Mo, Tc, Ru, Rh) transition metal defects implanted in the Pd surface layer covering the fcc-Fe monolayer de- posited on Ir(111) surface. The latter substrate is known to host few nanometers-wide magnetic skyrmions 38 -- 40 stabilized by the presence of Dzyaloshinskii-Moriya interaction 41, 42. We identified the different mechanisms conspiring to make the universal trends of the various XMRs, as function of the impurities electronic states, distinct from each other. We put forward the DXMR effect as a key-reading tool in man-engineered defective substrates with immediate implications for device applications in the context of non-collinear states. Results Impact of defects on the all-electrical detection. The all-electrical detection of single magnetic skyrmions via the XMR effect19, 20, 43 is made possible via the non-collinearity of the Fe substrate's magnetic moments and the presence of spin-orbit interaction, which lead to a spin-mixing of the electronic states living initially in the majority- and minority-spin channels. Within tunneling trans- port experiments, these mixed states are detectable with a non-magnetic electrode (Figure 1a). In the case of PdFe/Ir(111) substrate, the XMR effect as probed in STM/STS is particularly enhanced 3 Figure 1: All-electrical detection of magnetic skyrmions in the presence of defects. a Illus- tration of a tunneling transport experiment to read magnetic skymions on fcc-PdFe bilayer on the Ir(111) surface with the single-atomic defects. The spin-mixing of the electronic structure within the non-collinear environment leads to changes on the tunneling conductance allowing the mag- netic data information to be sensed in a current perpendicular-to-plane geometry. The various XMRs signals, for the clean system and in the presence of defects, are shown on the the lower panel. Depending on the reference background two types of XMR signals can be measured, b while the clean ferromagnetic substrate leads to the novel defect-induced XMR (DXMR), c the substrate-defect complex background leads to the regular XMRdefect. The defects are indicated by green spheres. 4 bcaXMRcleanXMRdefectDXMRVacPdFefcc-Ir(111)bulkVac!"AuVacPdFefcc-Ir(111)bulkVac!"Au!"#Au!"#AuSuspended contactsSuspended contacts0.038.777.3116XMR (%)for resonant interface and surface states44, which result from various hybridization mechanisms being affected by relativistic effects. We consider a skyrmion passing at the vicinity of a defect as illustrated schematically in Fi- gure 1. Here, it is convenient to define two types of XMR signals contingent on the reference mea- surement with respect to which the efficiency of the MR effect is calculated. The reference point can be either the clean (defect-free) ferromagnetic substrate (see Figure 1b) or the substrate-defect complex (see Figure 1c) in the collinear magnetic ground state. The former leads to the defect- enhanced XMR (DXMR) while the latter is a regular XMR signal but obtained in the presence of the defect and denoted in the following XMRdefect. Such reference points can be conveniently measured in a device. In both cases, we calculate the percent deviation of the conductance on top of a skyrmion from that of a reference point within a collinear magnetic region. We note that the XMR signal collected from an STM experiment on a pinned skyrmion 20 naturally includes the contribution of the pinning defect and should thus correspond to what we name DXMR. Based on the Tersoff-Hammann approach45, the differential conductance dI/dV at a given bias voltage Vbias measured within STM/STS, equipped with a non-magnetic tip, is proportional to the local density of states (LDOS) of the substrate at the energy E = EF + eVbias obtained in vacuum at the tip position. Here, we consider the tip located in second vacuum layer above the Pd substrate as indicated in Figure 1b-c, thus the efficiency of these magnetoresistance effects are calculated from: XMRdefect(E) = LDOSdefect C (E) , (1) NC (E) − LDOSdefect LDOSdefect (E) C 5 which is the equivalent of the conventional defect-free XMR, XMRclean, where all LDOS are ob- tained in the clean region. C and NC correspond respectively to collinear and non-collinear mag- netic regions, as shown in Figure 1b-c. The newly proposed DXMR efficiency is extracted from : DXMR(E) = LDOSdefect NC (E) − LDOSclean LDOSclean (E) C (E) = XMRclean(E) + , (2) C LDOSdefect NC (E) − LDOSclean LDOSclean (E) NC (E) C which obviously measures the enhancement of the XMR efficiency by the defect with respect to the signal obtained in the defect-free region. The latter equation shows that the DXMR signal is related to the "traditional" effect and it is obviously shaped by how the tunneling matrix elements involving the defect are different from those of the defect-free substrate. In fact, by re-expressing it as: DXMR(E) = LDOSdefect NC (E) − LDOSdefect LDOSclean (E) C C (E) + LDOSdefect C (E) − LDOSclean C LDOSclean C (E) (E) , (3) one deduces that besides the detection of the non-collinearity, the DXMR ratio measures the en- hancement of the transport signal emanating from the reference points of the substrate because of the impurity. The larger it is, the better are the conditions for the all-electrical detection. For conciseness, we focus our first analyses on the impact of a V impurity at the closest vicinity of the skyrmion's core on the various XMRs, whose energy-resolved signals are plotted in Figure 2a. The investigated skyrmion has a diameter DSk ≈ 2.2 nm and we recall that in 6 the skyrmion's core, the magnetic moment of the substrate Fe atom is flipped with respect to the ferromagnetic surrounding. Among the XMR signals, the DXMR efficiency is the largest with an impressive ≈ 85%, i.e. an increase of ≈ 230% with respect to the defect-free signal, for an energy injection about eVbias ≈ +0.45 eV (Figure 2a, purple curve). As deduced from Eq. 3, this large efficiency is just an indication that the defect enhances the bare tunneling transport signal. For comparison, the XMRclean (Figure 2a, green curve) and XMRdefect (Figure 2a, orange curve) signals reach a similar efficiency around 26%. In the whole investigated energy window, the latter two signals are very similar in magnitude and have a shape, which in general, seems similar to the DXMR signal. The XMRclean and XMRdefect efficiencies are shaped by the same mechanisms, i.e. the non- collinearity of the substrate magnetic moments and the presence of spin-orbit interaction. In con- trast, the DXMR signal is more sensitive to the hybridization of the electronic states of the defect with the substrate Fe atoms and the way these states decay into vacuum, which settles the tunneling matrix elements. In general, the enhancement factor of the DXMR signal is found similar to the one induced on the collinear substrate by implanted impurities. Origin of enhancement of the bare transport signal and of the DXMR efficiency. The refer- ence background for both the XMRclean and the DXMR signal is the defect-free region away from the skyrmion (collinear magnetic region). Thus, to understand the enhancement of the DXMR effect, we invoke Equation 2 and analyse the vacuum LDOS just above the skyrmion core with and without the implanted V-atom (Figure 2e-f). As illustrated in Figure 2b, the vacuum electronic 7 Figure 2: Efficiency of the XMR signal and the electronic structure. a Comparison of various energy-resolved XMR signals measured on top of the skyrmions core considering the influence of a V impurity located close to skyrmion's core. The dashed line indicates the energy at which the DXMR reaches the maximum efficiency. b Electronic structure in the vacuum on top of the core of the skyrmion core resolved into majority and minority spin-channels for the clean system (blue line) and with a V-impurity (red line). c Superposition of the electronic states in arbitrary units of vacuum, V and Fe at the skyrmion's core. d Impact of chemical nature of the defect on the DXMR signal at an injection energy eVbias ≈ 0.45 eV for 3d (red circles) and 4d (blue squares) impurities located close to the skyrmion center. The gray dashed line indicates the signal of the XMRclean. e-f Illustrative legend of the STM-tip probing the clean (defective) substrate at the location defined by the light blue (red) sphere. The green sphere indicates the impurity. 8 abcDefectCleannon-magnetic tipnon-magnetic tipdefstates of both spin-channels are strongly amplified by the presence of the V-defect affecting the differential conductance by virtue of the Tersoff-Hamann approximation. Here one identifies the main advantage of a substrate hosting impurities. The overall STM/STS signal is increased, which greatly facilitates the detection of buried skyrmions, as it is for PdFe/Ir(111) surface. Thus, defects can act as mediating probes for the electrical detection. The observed vacuum states originate from the decay of the V and Fe substrate's electronic states plotted in Figure 2e-f, as can be noticed from the one-to-one correspondence between the atomic virtual bound states and the features probed in vacuum. The orbital-resolved LDOS, e.g. the pz and dz2 states, leading to the largest tunneling matrix elements are shown in particular for the V defect and for the Fe atom at the skyrmion's core. The latter defines the spin-frame of reference in which the LDOS is plotted and the convention to denote the majority- and minority-spin states. Because of the large exchange splitting of V, the hybridization is weak between the occupied majority-spin states of Fe and the unoccupied ones of V, giving rise to the dz2 and pz virtual bound states located at ≈ + 0.9 eV and ≈ + 1.0 eV, respectively (see upper channel of Figure 2c). Recalling that V replaces a Pd atom from the defect-free substrate, we expect thus a larger intensity of the features observed in vacuum because of the additional impurity-states. Similar conclusions can be drawn for the opposite spin channel. Interestingly, an unoccupied pz virtual bound state at ≈ + 0.53 eV is induced by the Fe minority-spin state in the LDOS of V (see lower channel of Figure 2c), with a larger intensity than the one obtained in the clean Pd-overlayer. 9 Systematic trends. The enhancement of bare transport signal detected within the DXMR signal with respect to the conventional XMR is not limited to the V impurity but occurs for all investigated implanted defects of the 3d and 4d series. The factor of enhancement depends, however, on the chemical nature of the impurities as well as on the injection energy. In Figure 2d, we systematically collect the efficiency of the DXMR as function of the impurities atomic number for an injection energy of eVbias ≈ + 0.45 eV. Overall, the DXMR ratio decreases when increasing the atomic number of the defects. For the 3d elements, V leads to the highest efficiency, which is larger than the 30% efficiency induced by Ni. The latter defect does not alter the defect-free XMR efficiency. Interestingly, defects from the beginning of the 4d series lead to an impressive enhancement of the signal, reaching almost 116% for Nb followed by 94% for Mo which translate to a increase of about 350% and 260%, respectively, with respect to the defect-free XMR efficiency. This implies that the best impurities for the enhancement of the bare tunneling transport signal are early transition elements with a preference for the 4d series. The behavior of the DXMR signal as function of the chemical nature of the defect can be directly related to the filling of the impurities electronic states. By moving from left to right across the 3d (4d) atomic row of the periodic table the unoccupied states of the impurities shift towards the Fermi energy becoming partially or almost fully occupied at the end of the series as can be seen in Figure 3a(b) for the collinear magnetic states. Interestingly, this sequence is accompanied with a transition from an antiferromagnetic coupling to the substrate for V (Nb), Cr (Mo) and Mn (Tc) to a ferromagnetic coupling for Fe (Ru), Co (Rh) and Ni (Pd). The decrease of the impurities LDOS upon the filling of the electronic states, as illustrated by Figure 3e, explains the DXMR trend 10 plotted in Figure 2d as well as the very large ratio induced by the 4d impurities when compared to the 3d ones. The smaller exchange splitting of the 4d elements increases the possibility of having large LDOS around the Fermi energy. Paying closer attention to the electronic features of the vacuum on top of the core of the skyrmion responsible for the discussed effect for the 3d and 4d series (see Figure 3c-d), one iden- tifies that the peaks obtained for the antiferromagnetic V (Nb), Cr (Mo) and Mn (Tc) are more intense than those of the ferromagnetic Fe (Ru), Co (Rh) and Ni (Pd). Here, two concomitant mechanisms are at play. First the hybridization strength is known to decrease from left-to-right and to increase from up-to-down across the transition element series of the peridioc table 33. Second, the pz state induced in the electronic structure of the impurities depends on the magnetic coupling because of the switch of the spin-nature of the impurity states hybridizing with the minority-spin states of Fe. Thus, more impurity states are available when the spin-alignment is rather ferromag- netic. Real-space XMRs contrast and inhomogeneous non-collinearity. Contrasting the DXMR be- havior, the XMRdefect for an injection energy eVbias ≈ + 0.45 eV when plotted as function of the atomic number (see Figure 3f) exhibits a S-like shape with a maximum and a minimum close to the middle of the series with the efficiency increasing up to about 56% with respect to the defect- free signal. By changing the injection energy, the S-shape can be reversed and the location of the extrema can be strongly shifted. Interestingly, while the bare transport signal is enhanced for the early transition elements, the XMR signal is better magnified by impurities with more than 11 Figure 3: Impact of defects on the electronic structure and on the XMRdefect signal. Local density of states (LDOS) of the defect within the collinear configuration for a the 3d-inatom series b the 4d-inatom series. LDOS of the vacuum on top of the core of the skyrmion with c 3d and d 4d impurity close to the skyrmion core. e Chemical trend of the LDOS of the vacuum on top of the core of the skyrmion at +0.45 eV for the 3d-elements (red line) 4d-elements (blue line). The green full line (dashed line) indicates the LDOS of the vacuum within the non-collinear (collinear) magnetic region for the defect free-system substrate at +0.45 eV. f Impact of band-filling on the defect spin-mixing magnetoresistance (XMRdefect) at an injection energy eVbias ≈ 0.45 eV. The gray dashed line indicates the signal of the XMRclean. 12 abcdefhalf-filled d-states. The XMR efficiency can be of use to detect slight changes in the magnetic texture. The XMRclean and XMRdefect depend on the opening angle between neighboring magnetic moments. In the defect-free region, the measurable electrical contrast is expected to exhibit a highly symmetric shape, translating the symmetry of the magnetic texture, with the highest intensity at the center of the skyrmion (see Figure 4a). At the vicinity of a defect, the skyrmion experiences an asymmetric environment impacting its spin-texture and consequently the all-electrical XMR contrast. This is better grasped by visualizing the difference ∆XMRdefect = XMRdefect − XMRclean as done in Figure 4b-c for Cr and Ni impurities. For Cr, the XMRdefect signal strongly decreases at the close vicinity of the impurity since the strong exchange coupling of the impurity and the substrate atoms reduces the surrounding non-collinearity (see Figure 4e), which lowers the XMR ratio. For Ni, however, the change in the XMR signal (Figure 4c) is off-set from the defect with the pattern being less intense than for Cr, which corresponds to the modifications induced in the magnetic texture shown in Figure 4f. The DXMR signal is not that sensitive to the impurity-induced non-collinear modifications and reaches a maximum value on top of all defects. As an example, the Cr-related DXMR signal at a bias energy ≈ 0.45 eV is illustrated in Figure 4d when the impurity is close to the core of the skyrmion. 13 Figure 4: Impact of single atomic defects on XMR-signals of skyrmions. a Defect-free XMR signal compared to d DXMR signal induced by a Cr impurity. Difference between XMRdefect and XMRclean plotted for b Cr and c Ni impurities close to the core of skyrmion. The green circle indicates the position of the impurity. The corresponding skyrmion magnetic textures are shown in e for Cr and f for Ni, where the green arrow represents the impurity magnetic moment while the rest of arrows are those of the Fe substrate atoms. The angle θ is the polar angle defining the magnitude of the z-component of the magnetization for each substrate atom. 14 abcdefθ °−2⁄Discussion In this work, we explore the possibility of using defects to detect non-collinear magnetic textures using all-electrical means. By strongly modifying the tunneling matrix elements, defects increase the intensity of the background transport signal. This would prove useful in case of buried non- collinear spin-textures, which are difficult to access with surface probe techniques. To track the impact of impurities, we defined the DXMR efficiency, which monitors the enhancement of the bare transport signal and of the XMR efficiency induced by the presence of the defects. From the obtained general trends upon filling of the impurities electronic states, our investigations point to the use of early transition elements to enhance the tunneling transport patterns, while elements with more than half-filled d-shells seem to be better suited for a better XMR-detection. Contrary to the usual expectations, defects provide a path to enhance in various ways the all-electrical detection of non-collinear magnetic textures. The DXMR and XMRdefect can be in- corporated in future reading technologies, which enforces the view that controlled engineering of defective-materials is a promising route for device architectures. Methods Computational details. The calculations were performed with the full-potential scalar-relativistic Korringa-Kohn-Rostoker (KKR) Green function method with spin-orbit coupling46, 47. The defect and the single magnetic skyrmions are considered using the embedded technique which allows us to evaluate details of the electronic structure at different positions. The Pd/Fe/Ir(111) slab consists 15 of a fcc-stacked PdFe bilayer deposited on 34 layers of Ir. The embedded cluster consists of 37 Fe atoms and 124 atoms in total, details of the self-consistent calculations can be found at Ref.33. The local density of states for the ferromagnetic-state or skyrmion-state were obtained performing one single interaction with a k-mesh of 200 × 200 and an angular momentum cutoff of lmax = 3 for the orbital expansion of the Green function. Data availability The data that support the findings of this study are available from the corre- sponding authors on request. Acknowledgements This work is supported by the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (ERC-consolidator grant 681405 DYNASORE). We acknowledge the computing time granted by the JARA-HPC Vergabegremium and VSR commission on the supercomputer JURECA at Forschungszentrum Jlich. Author contributions I.L.F. performed the first-principles calculations. All authors discussed the results and helped writing the manuscript. Competing Interests The authors declare that they have no competing financial interests. Correspondence Correspondence and requests for materials should be addressed to I.L.F. (email: i.lima.fernandes@fz- juelich.de) or to S.L. (email: [email protected]). References 16 1. Wolf, S. et al. Spintronics: a spin-based electronics vision for the future. Science 294, 1488 -- 1495 (2001). 2. Evetts, J. et al. Defect-induced spin disorder and magnetoresistance in single-crystal and polycrystal rare-earth manganite thin films. Philosophical Transactions of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences 356, 1593 -- 1615 (1998). 3. Ke, Y., Xia, K. & Guo, H. Oxygen-vacancy-induced diffusive scattering in fe/mgo/fe magnetic tunnel junctions. Phys. Rev. Lett. 105, 236801 (2010). 4. Baibich, M. N. et al. Giant magnetoresistance of (001) Fe/(001) Cr magnetic superlattices. Phys. Rev. Lett. 61, 2472 (1988). 5. Binasch, G., Grunberg, P., Saurenbach, F. & Zinn, W. Enhanced magnetoresistance in layered magnetic structures with antiferromagnetic interlayer exchange. Phys. Rev. B 39, 4828 (1989). 6. Julliere, M. Tunneling between ferromagnetic films. Phys. Lett. A 54, 225 -- 226 (c). URL http://www.sciencedirect.com/science/article/pii/ 0375960175901747. 7. Moodera, J. S., Kinder, L. R., Wong, T. M. & Meservey, R. Large magnetoresistance at room temperature in ferromagnetic thin film tunnel junctions. Phys. Rev. Lett. 74, 3273 (1995). 8. Stipe, B. C., Rezaei, M. A. & Ho, W. Single-molecule vibrational spec- troscopy and microscopy. Science 280, 1732 -- 1735 (1998). URL https:// science.sciencemag.org/content/280/5370/1732. https://science. sciencemag.org/content/280/5370/1732.full.pdf. 17 9. Heinrich, A. J., Gupta, J. A., Lutz, C. P. & Eigler, D. M. Single-atom spin-flip spec- troscopy. Science 306, 466 -- 469 (2004). URL https://science.sciencemag.org/ content/306/5695/466. https://science.sciencemag.org/content/ 306/5695/466.full.pdf. 10. Schweflinghaus, B., dos Santos Dias, M., Costa, A. T. & Lounis, S. Renormalization of electron self-energies via their interaction with spin excitations: A first-principles investiga- tion. Phys. Rev. B 89, 235439 (2014). URL https://link.aps.org/doi/10.1103/ PhysRevB.89.235439. 11. Nagaosa, N., Sinova, J., Onoda, S., MacDonald, A. H. & Ong, N. P. Anomalous hall ef- fect. Rev. Mod. Phys. 82, 1539 -- 1592 (2010). URL https://link.aps.org/doi/10. 1103/RevModPhys.82.1539. 12. Gradhand, M., Fedorov, D. V., Zahn, P. & Mertig, I. Extrinsic spin hall effect from first principles. Phys. Rev. Lett. 104, 186403 (2010). URL https://link.aps.org/doi/ 10.1103/PhysRevLett.104.186403. 13. Lowitzer, S. et al. Extrinsic and intrinsic contributions to the spin hall effect of alloys. Phys. Rev. Lett. 106, 056601 (2011). URL https://link.aps.org/doi/10.1103/ PhysRevLett.106.056601. 14. Fert, A. & Levy, P. M. Spin hall effect induced by resonant scattering on impurities in metals. Phys. Rev. Lett. 106, 157208 (2011). URL https://link.aps.org/doi/10.1103/ PhysRevLett.106.157208. 18 15. Zimmermann, B., Long, N. H., Mavropoulos, P., Blugel, S. & Mokrousov, Y. Influence of complex disorder on skew-scattering Hall effects in L10-ordered FePt alloy. Phys. Rev. B 94, 060406 (2016). URL https://link.aps.org/doi/10.1103/PhysRevB.94. 060406. 16. Bouaziz, J., Lounis, S., Blugel, S. & Ishida, H. Microscopic theory of the residual surface resistivity of rashba electrons. Phys. Rev. B 94, 045433 (2016). URL https://link. aps.org/doi/10.1103/PhysRevB.94.045433. 17. Kondo, J. Resistance minimum in dilute magnetic alloys. Prog. Theor. Exp. Phys. 32, 37 -- 49 (1964). 18. Daybell, M. & Steyert, W. Localized magnetic impurity states in metals: Some experimental relationships. Rev. Mod. Phys. 40, 380 (1968). 19. Crum, D. M. et al. Perpendicular reading of single confined magnetic skyrmions. Nat. Com- mun. 6, 8541 (2015). 20. Hanneken, C. et al. Electrical detection of magnetic skyrmions by tunnelling non-collinear magnetoresistance. Nat. Nanotechnol. 10, 1039 (2015). 21. Bogdanov, A. N. & Yablonskii, D. Thermodynamically stable "vortices" in magnetically ordered crystals. the mixed state of magnets. J. Exp. Theor. Phys. 95, 178 (1989). 22. Rossler, U. K., Bogdanov, A. N. & Pfleiderer, C. Spontaneous skyrmion ground states in magnetic metals. Nature 442, 797 -- 801 (2006). 19 23. Fert, A., Reyren, N. & Cros, V. Magnetic skyrmions: advances in physics and potential applications. Nat. Rev. Mater. 2, 17031 (2017). 24. Fert, A., Cros, V. & Sampaio, J. Skyrmions on the track. Nat. Nanotech. 8, 152 -- 156 (2013). 25. Sampaio, J., Cros, V., Rohart, S., Thiaville, A. & Fert, A. Nucleation, stability and current- induced motion of isolated magnetic skyrmions in nanostructures. Nat. Nanotech. 8, 839 -- 844 (2013). 26. Iwasaki, J., Mochizuki, M. & Nagaosa, N. Universal current-velocity relation of skyrmion motion in chiral magnets. Nat. Commun. 4, 1463 (2013). 27. Iwasaki, J., Mochizuki, M. & Nagaosa, N. Current-induced skyrmion dynamics in constricted geometries. Nat. Nanotech. 8, 742 -- 747 (2013). 28. Jiang, W. et al. Blowing magnetic skyrmion bubbles. Science 349, 283 -- 286 (2015). 29. Woo, S. et al. Observation of room-temperature magnetic skyrmions and their current-driven dynamics in ultrathin metallic ferromagnets. Nat. Mater. 15, 501 -- 506 (2016). 30. Hanneken, C., Kubetzka, A., von Bergmann, K. & Wiesendanger, R. Pinning and movement of individual nanoscale magnetic skyrmions via defects. New J. Phys. 18, 055009 (2016). 31. Jiang, W. et al. Direct observation of the skyrmion Hall effect. Nat. Phys. 13, 162 -- 169 (2017). 32. Litzius, K. et al. Skyrmion hall effect revealed by direct time-resolved x-ray microscopy. Nat. Phys. 13, 170 -- 175 (2017). 20 33. Lima Fernandes, I., Bouaziz, J., Blugel, S. & Lounis, S. Universality of defect-skyrmion interaction profiles. Nat. Commun. 9, 4395 (2018). URL https://doi.org/10.1038/ s41467-018-06827-5. 34. Choi, H. C., Lin, S.-Z. & Zhu, J.-X. Density functional theory study of skyrmion pinning by atomic defects in mnsi. Phys. Rev. B 93, 115112 (2016). URL https://link.aps.org/ doi/10.1103/PhysRevB.93.115112. 35. Maccariello, D. et al. Electrical detection of single magnetic skyrmions in metallic multilayers at room temperature. Nat. Nanotechnol. 13, 233 (2018). 36. Eigler, D. M. & Schweizer, E. K. Positioning single atoms with a scanning tunnelling micro- scope. Nature 344, 524 -- 526 (1990). 37. Persaud, A. et al. Integration of scanning probes and ion beams. Nano Letters 5, 1087 -- 1091 (2005). 38. Romming, N. et al. Writing and deleting single magnetic skyrmions. Science 341, 636 -- 639 (2013). 39. Romming, N., Kubetzka, A., Hanneken, C., von Bergmann, K. & Wiesendanger, R. Field- dependent size and shape of single magnetic skyrmions. Phys. Rev. Lett. 114, 177203 (2015). 40. Dup´e, B., Hoffmann, M., Paillard, C. & Heinze, S. Tailoring magnetic skyrmions in ultra-thin transition metal films. Nat. Commun. 5, 4030 (2014). 21 41. Dzyaloshinsky, I. A thermodynamic theory of weak ferromagnetism of antiferromagnetics. J. Phys. Chem. Sol. 4, 241 -- 255 (1958). 42. Moriya, T. Anisotropic superexchange interaction and weak ferromagnetism. Phys. Rev. 120, 91 -- 98 (1960). 43. Kubetzka, A., Hanneken, C., Wiesendanger, R. & von Bergmann, K. Impact of the skyrmion spin texture on magnetoresistance. Phys. Rev. B 95, 104433 (2017). URL https://link. aps.org/doi/10.1103/PhysRevB.95.104433. 44. Bouhassoune, M., Lima Fernandes, I., Blugel, S. & Lounis, S. Unoccupied surface and inter- face states in Pd thin films deposited on Fe/Ir (111) surface. arXiv preprint arXiv:1901.07029 (2019). 45. Tersoff, J. & Hamann, D. Theory and application for the scanning tunneling microscope. Phys. Rev. Lett. 50, 1998 (1983). 46. Papanikolaou, N., Zeller, R. & Dederichs, P. H. Conceptual improvements of the KKR method. J. Phys. Condens. Matter 14, 2799 (2002). 47. Bauer, D. S. G. Development of a relativistic full-potential first-principles multiple scattering green function method applied to complex magnetic textures of nano structures at surfaces. PhD dissertation at the RWTH-Aachen (2013). 22
1702.02210	3	1702	2018-07-02T19:31:22	Possible hundredfold enhancement in the direct magnetic coupling of a single atomic spin to a circuit resonator	[ "cond-mat.mes-hall", "cond-mat.supr-con", "quant-ph" ]	We report on the challenges and limitations of direct coupling of the magnetic field from a circuit resonator to an electron spin bound to a donor potential. We propose a device consisting of a trilayer lumped-element superconducting resonator and a single donor implanted in enriched $^{28}$Si. The resonator impedance is significantly smaller than the practically achievable limit using prevalent coplanar resonators. Furthermore, the resonator includes a nano-scale spiral inductor to spatially focus the magnetic field from the photons at the location of the implanted donor. The design promises approximately two orders of magnitude increase in the local magnetic field, and thus the spin to photon coupling rate $g$, compared to the estimated coupling rate to the magnetic field of coplanar transmission-line resonators. We show that by using niobium (aluminum) as the resonator's superconductor and a single phosphorous (bismuth) atom as the donor, a coupling rate of $g/2\pi$=0.24 MHz (0.39 MHz) can be achieved in the single photon regime. For this hybrid cavity quantum electrodynamic system, such enhancement in $g$ is sufficient to enter the strong coupling regime.	cond-mat.mes-hall	cond-mat	Possible hundredfold enhancement in the direct magnetic coupling of a single atomic spin to a circuit resonator Bahman Sarabi,1, 2 Peihao Huang,1, 2 and Neil M. Zimmerman1 1National Institute of Standards and Technology, Gaithersburg, MD 20899, USA 2Joint Quantum Institute, University of Maryland, College Park, Maryland 20742, USA (Dated: July 4, 2018) We report on the challenges and limitations of direct coupling of the magnetic field from a circuit resonator to an electron spin bound to a donor potential. We propose a device consisting of a trilayer lumped-element superconducting resonator and a single donor implanted in enriched 28Si. The resonator impedance is significantly smaller than the practically achievable limit using prevalent coplanar resonators. Furthermore, the resonator includes a nano-scale spiral inductor to spatially focus the magnetic field from the photons at the location of the implanted donor. The design promises approximately two orders of magnitude increase in the local magnetic field, and thus the spin to photon coupling rate g, compared to the estimated coupling rate to the magnetic field of coplanar transmission-line resonators. We show that by using niobium (aluminum) as the resonator's superconductor and a single phosphorous (bismuth) atom as the donor, a coupling rate of g/2π=0.24 MHz (0.39 MHz) can be achieved in the single photon regime. For this hybrid cavity quantum electrodynamic system, such enhancement in g is sufficient to enter the strong coupling regime. I. INTRODUCTION Silicon-based spin qubits, including gate-defined quan- tum dot [1–3] and single-atom [4, 5] devices, use the spin degree of freedom to store and process quantum infor- mation, and are promising candidates for future quan- tum electronic circuits. The electronic or nuclear spin is well decoupled from the noisy environment, resulting in extremely long coherence times [6–8] desirable for fault- tolerant quantum computing. Single-atom spin qubits of- fer additional advantages over quantum dot qubits such as longer coherence due to strong confinement potentials, and are expected to have intrinsically better reproducibil- ity. Therefore, it is no surprise that silicon-based single- atom spin qubits hold the record coherence times of any solid state single qubit [9]. However, this attractive iso- lation from sources of decoherence comes at the price of relatively poor coupling to the control and readout units. This causes relatively long qubit initialization times, de- graded readout fidelity, and weak spin-spin coupling for multi-qubit gate operations [10]. One simple way to enhance the coupling rate is to in- crease the ac magnetic field from the external circuit. In this regard, superconducting circuit resonators are at- tractive due to their relatively large quality factors, ease of coupling to other circuits, their capability of generating relatively large ac magnetic fields by carrying relatively large currents, and monolithic integration with semicon- ductor devices. Over the last two decades, superconducting microwave resonators have had extensive applications that range from superconducting qubit initialization, manipulation and readout [11, 12], and inter-qubit coupling [13] to di- electric characterization [14]. One of the most commonly used superconducting quantum computing architectures is one based on cavity quantum electrodynamics (cQED) [15], in which a 2D (circuit based) or 3D cavity is em- ployed to initialize, manipulate and readout the super- conducting qubit. Superconducting circuit cavities have not yet found a similar prevalence in spin qubit circuits due to the fact that the magnetic field of a typical super- conducting resonator and the spin magnetic-dipole mo- ment are relatively small, leading to an insufficient spin- photon coupling strength for practical purposes. The direct magnetic coupling of coplanar resonators (typi- cally coplanar waveguide resonators) to donor electrons in silicon [16–18] and diamond nitrogen-vacancy centers [19, 20] can only achieve a maximum single-spin coupling rate of a few kHz. Several methods have been proposed to enhance the coupling of a single spin to a photon within a supercon- ducting circuit resonator. It is easier to couple a photon to quantum dot spin qubits than to single-atom spins be- cause in the former, the spin dynamics can be translated into an electric dipole interacting with resonator's electric field [21–23]. Such architectures, however, require hy- bridization of the spin states with charge states, coupling charge noise to the spin and thus affecting its coherence. For instance, in two recent experimental demonstrations of strong spin-photon coupling in Si [24, 25], the max- imum value of the figure of merit ratio of the coupling rate to the spin decoherence rate was about five, due in part to the approximately factor of five increase in the dephasing rate [25] arising from the hybridization of spin and charge states. For the single-atom spin qubits, indi- rect coupling to a resonator via superconducting qubits [26, 27] can enhance the coupling, but imposes nonlinear- ity on the circuit complicating cQED analysis, and also introduces loss from Josephson junction tunnel barriers [28, 29] and magnetic flux noise [30]. In this paper, we show that replacing the coplanar transmission line resonator with a lumped-element cir- 8 1 0 2 l u J 2 ] l l a h - s e m . t a m - d n o c [ 3 v 0 1 2 2 0 . 2 0 7 1 : v i X r a 2 FIG. 1. (a) Schematic of the circuit showing the resonator galvanically coupled to the coplanar waveguide (CPW). CGND is the resonator's capacitance to ground. Pin, Pout, Lg and Lp show the microwave input power, output power, geometric inductance and the parasitic inductance of the resonator circuit, respectively. (b) Layout of the device showing resonator's capacitor C and inductor L. Blue and orange show the top and bottom superconducting layers, respectively. The area within the red dashed square is magnified. The spiral inductor within the green dashed square is further magnified to show (c) the single-spiral geometry (red dot represents the spin). Dimensions are w = s = t = din/2 = 30 nm, compatible with standard electron beam lithography, and d0 = ds = 25 nm. The crystalline 28Si (x-28Si) growth is required only in the volume surrounding the spin, and amorphous or polycrystalline Si is acceptable everywhere else. The relatively weak magnetic field Bvia (directions shown represent magnetic fields at the location of the spin) from the via determines the best choice for the direction of the static magnetic field, B0 (cid:107) (Bvia × Bac), where Bac is the magnetic field from the spiral inductor. This ensures that the total ac field is perpendicular to B0. The current flowing through the spiral inductor is shown by green in- or out of plane vectors. cuit resonator that includes a spiral inductor, can lead to a dramatic enhancement in the spin-photon magnetic coupling rate g of approximately two orders of magni- tude. As we will discuss, this improvement is a result of the reduced resonator impedance thanks to the trilayer lumped-element design, as well as employing nano-scale spiral inductor loops which effectively localize the res- onator's magnetic field at the location of the spin, elimi- nating the need for an Oersted line [31] or a micrometer- scale magnet (micromagnet) [24, 32]. We also show that this coupling rate is enough to take the spin-resonator hybrid system to the strong coupling regime, where g is larger than or approximately equal to the resonator de- cay rate κ. In order to achieve this relatively large g, we need a relatively small total inductance Ltot, and con- comitantly small impedance Z = (cid:112)Ltot/C. To reach the desired operation frequency ω0 = (cid:112)1/LtotC, we need a large capacitance C, and thus coplanar inter- digitated capacitances (IDCs) are practically insufficient. Furthermore, IDCs generally introduce relatively large parasitic inductance which could set a limit to the min- imum achievable Z. Therefore, the proposed device ge- ometry, compatible with standard micro- and nanofabri- cation techniques, includes a trilayer (parallel-plate) ca- pacitor with a deposited insulating layer. Trilayer capaci- tors, however, give rise to a lower resonator quality factor Q than that of a typical coplanar geometry. Neverthe- less, as our calculations show, the increase in g resulted from employing a trilayer lumped-element design is large enough to overcome the limited Q, allowing for strong spin-photon coupling. II. SINGLE-ATOM DEVICE The lossless dynamics of a spin- 1 2 system with spin transition frequency ωs coupled with rate g to a cavity resonant at ω0, are described by the Jaynes-Cummings Hamiltonian, HJC = ¯hω0(a†a + 1/2) + (1/2)¯hωs σz + ¯hg(a† σ− + σ+a) [33]. Here, a and σ are photon and spin operators, respectively. The spin-photon magnetic coupling rate is obtained as g = geµBBac,0 (cid:104)g Sxe(cid:105) /¯h where ge (cid:39) 2 is the electron g-factor, µB is the Bohr magneton, and Bac,0 = (cid:104)n = 0 B2 is the root mean square (RMS) local magnetic field at the location of the spin with n = 0 photons on resonance. g(cid:105) and e(cid:105) acn = 0(cid:105)1/2 dinSi substrateBacB0BviaBac28Six-Stark-shift electrodeStark-shift electrodesuperconductorsuperconductora)b)c)GNDGNDGNDCPW28Sipoly-270 nmStark-shiftelectrodeStark-shiftelectrodeStark-shiftelectrodes50 μm2 μmviadsd0Spinggccwts3 FIG. 3. Plots of the (a) total inductance, (b) spin-photon cou- pling rate, (c) vacuum fluctuations current density, and (d) required resonator quality factor Q ≡ ω0/g to allow strong spin-photon coupling versus the number of spiral loops Nloops for different combinations of materials and geometries, i.e. su- perconducting aluminum and Bismuth donor (Al/Bi), super- conducting niobium and phosphorous donor (Nb/P), single- spiral and double-spiral (2S) geometries. as donor (Al/Bi). Fabrication is considered to be slightly better established with Al, whereas Nb has a much higher critical magnetic field of Bc1,Nb = 0.2 T, which is the highest Bc1 among the single-element superconductors. Through the Zeeman effect, for the Nb/P set, the un- coupled electron spin-up (↑(cid:105)) and spin-down (↓(cid:105)) states are split by the photon frequency ω0/2π = 5.6 GHz at the maximum B0 = ¯hω0/geµB (cid:39) Bc1,Nb, not to quench superconductivity. For the Al/Bi set, we consider op- erating at B0 < 10 mT (below the critical magnetic field of Al), and use the splitting of the spin multiplets (ω0/2π = 7.375 GHz) of the Bi donor. These splittings arise from the strong hyperfine interaction between the electron (S = 1/2) and Bi nuclear spin (I = 9/2) [35], leading to a total spin F and its projection mF along B0. By employing the F, mF(cid:105) = 5,−5(cid:105) ↔ 4,−4(cid:105) transition corresponding to the largest Sx matrix element (0.47), and a static magnetic field of B0 = 5 mT, the multiplet degeneracy at B0 = 0 is lifted by more than 20 MHz [36, 37], enough to decouple the nearby transitions. Figure 2 shows an alternative double-spiral (2S) geom- etry where the spiral inductor extends to both supercon- ducting layers. Here we see the following optimum com- bination of materials: i) Amorphous or polycrystalline Si can be deposited on the metal layers; since they have sim- ilar permittivities to crystalline Si, the capacitor size will be similar. ii) Since the spin does not have a metal layer below it, one can still deposit crystalline 28Si and then implant the donor, thus avoiding the fast decoherence FIG. 2. Layout of the double-spiral (2S) device in the vicinity of the nanoscale spiral inductor. Here, ds = 27.5 nm and other dimensions are identical to those in Fig. 1(c). are the ground and excited spin states, respectively, that couple to the microwave field; for spin- 1 Finally, the spin rotation speed can be enhanced with a larger local RMS magnetic field Bac = (cid:104)n B2 any arbitrary n. 2 , (cid:104)g Sxe(cid:105) = 1/2. acn(cid:105)1/2 for The schematic and layout of the proposed circuit are shown in Figs. 1(a) and 1(b-c), respectively, where a LC resonator is coupled to a coplanar waveguide (CPW) using a direct (galvanic) connection through a coupling inductor Lc and the donor is within the res- onator's deliberate (geometric) inductor Lg. The gal- vanic coupling, employed in previous experiments [34], can help to achieve the desired CPW-resonator coupling rates especially when the resonator impedance is signifi- cantly different from the CPW's characteristic impedance Z0 = 50 Ω. The capacitance C is provided using a trilayer capacitor. The RMS current through the in- ductor at an average photon number ¯n on resonance is Iac = (cid:112)(¯n + 1/2)¯hω0/Ltot, where ω0 is the resonance frequency and Ltot denotes the total inductance within the resonator circuit. Since the desired ac magnetic field from the spiral loop(s) is proportional to Iac, it is clear that the inductance (or impedance Z) must be minimized and the photon frequency must be maximized to obtain the maximum magnetic field. Therefore, it is necessary to study the sources of inductance and obtain operation frequency limitations. We study two sets of materials, one that uses niobium as superconductor and phosphorous as donor (Nb/P), and one with aluminum as superconductor and bismuth wtsdinSi substrate28SiBacB0BviaBacpoly-28Six-Stark-shift electrodeStark-shift electrodesuperconductorsuperconductor270 nmviadsd0Spinwhich would otherwise result from the non-crystalline, non-enriched films [38]. From the fabrication point of view, it is easiest to use the same 28Si film as the capaci- tor dielectric (see Figs. 1(c) and 2). It is noteworthy that using a trilayer design makes the resonator quality factor effectively independent from the surrounding material as almost all the electric field energy is confined within the capacitor dielectric. This is useful for the integration of single electron devices that require lossy oxide layers. For all four aforementioned combinations of resonator material and geometry, Ltot is simulated and shown in Fig. 3(a) as a function of number of spiral loops, Nloops (see Appendix for the details of inductance calculations and simulations). Nloops refers to the number of loops in a single spiral layer regardless of the geometry, e.g., Nloops = 2 for both Figs. 1(c) and 2. Increasing Nloops from 1 to 2 increases Bac,0, but, for Nloops > 2, the competing effect of larger Ltot due to larger loop radii suppresses Iac and lowers g. This effect is clearly demon- strated in Fig. 3(b) as the optimum Nloops = 2 for both material sets, where the vacuum fluctuation's coupling rates for the Al/Bi and Nb/P configurations are obtained as g/2π = 0.26 MHz and 0.17 MHz, respectively. For the 2S geometry, coupling rates for the Al/Bi and Nb/P con- figurations approach g/2π = 0.39 MHz and 0.24 MHz, respectively. These values are approximately two orders of magnitude larger than previously proposed architec- tures that use coplanar transmission line resonators [18]. This new regime of g can enable new experiments such as individual spin spectroscopy, and as we report in section IV, it can lead to significant improvements in spin qubit performance in regard with qubit initialization, manipu- lation and readout. Figure 3(c) shows the vacuum fluctuation's current density J = (1/wt)Iac¯n=0 for each configuration, where w and t are the width and thickness of the spiral traces, respectively (see Figs. 1(c) and 2). Clearly, J stays far below the critical current values of Al (Jc ∼ 104 kA/cm2 [39]) and Nb (Jc ∼ 103 kA/cm2 [40]). The condition for the system to enter the strong cou- pling regime is Q >∼ ω0/g in the single photon regime, where Q = ω0/κ and κ are resonator's total quality fac- tor and total photon decay rate, respectively. Thanks to the relatively large spin-photon coupling rate, resonator quality factors in the range of 104 are enough to take the system to the strong coupling regime for all four config- urations (see Fig. 3(d)). In general, in the limit of low drive power and low temperature, trilayer resonators are significantly lossier than coplanar resonators due to the fact that almost all the photon electric energy is stored within the parallel-plate capacitor dielectric which con- tains atomic-scale defects. These defects act as lossy two-level fluctuators at small powers and low tempera- tures, and limit the resonator Q [29]. However, loss tan- gents in the vicinity of tan δ0 (cid:39) 10−5 have been measured for deposited amorphous hydrogenated silicon (a-Si:H) at 4 single photon energies [41] promising resonator Q's ap- proaching 105. More recently, elastic measurements have indicated the absence of tunneling states in a hydrogen- free amorphous silicon film suggesting the possibility of depositing "perfect" silicon [42], promising even higher Q trilayer resonators using silicon as capacitor dielectric. It has been previously shown that the donor can be ionized in the vicinity of metallic or other conductive structures due to energy band bendings [38]. For Al-Si interface, the relatively small work function difference of -30 meV (4.08 eV for Al and 4.05 eV for Si) is smaller than the donor electron binding energy of -46 meV [43], and hence using aluminum is expected to allow the donor bound state. However, the Nb work function of 4.3 eV causes significant band bending which can result in donor ionization. By biasing the microwave transmission line and hence the resonator through the galvanic connec- tion with a DC voltage, the potential energy landscape in the neighborhood of the donor can be modulated to recreate a bound state. If a conduction band electron is required to fill the bound state, solutions such as shining a light pulse using a light emitting diode to create excess electron-hole pairs or using an ohmic path to controllably inject electrons can be employed. From Fig. 3(d) and the data available from silicon- based low-loss deposited dielectrics, we infer that the pro- posed single-atom device can achieve the strong coupling regime. In general, the single-spiral design is expected to be easier to fabricate at the expense of smaller g com- pared to the double-spiral (2S) geometry. III. SIMPLIFIED DEVICE WITH SPIN ENSEMBLE We now discuss a greatly simplified proof-of-principle device and experiment. We consider a low-impedance trilayer resonator on top of a Bi-doped substrate or 28Si layer, and propose to measure the spectrum in the sin- gle photon regime while the electron spin transitions are tuned across the resonator bandwidth. Similar to the single-atom device, tuning of the spins can be performed using magnetic (Zeeman) or electric (Stark, see Fig. 1) fields, or a combination of both. For further simplifica- tion, a single inductor loop can be employed with dimen- sions compatible with standard photolithography. In this simplified experiment with no single-donor implantation or e-beam lithography requirements, one can choose Al as superconductor and thus the kinetic inductance within the circuit is negligible with micrometer-scale dimensions of the inductor loop. When Ns spins are incorporated and exposed to the relatively uniform magnetic field in- side the inductor loop, the collective coupling rate from Ns [17]. This en- the spin ensemble becomes gcol = g hanced coupling rate would yield clearer spin-resonator interaction features in the spectra for this first proof-of- √ principle version of the device compared to the single- atom device. Since the spin ensemble device is micrometer-scale, high-frequency simulations using electromagnetic simu- lation software become feasible (Sonnet R(cid:13) was used for this work). The layout of the simulated device is shown in Fig. 4(a). We employ a galvanic connection to couple this device to the CPW, similar to the single-atom device described in section II. As shown in Fig. 4(b), in the ab- sence of spins, the simulated transmission S21 = Vout/Vin through the CPW shows a characteristic resonance cir- cle with a diameter of 0.5. This indicates that, because of the galvanic coupling and despite Z (cid:39) 0.4 Ω (cid:28) Z0, critical coupling can be achieved, where the resonator's external quality factor Qe ≡ ω0/κe equals its internal quality factor Qi ≡ ω0/κi. Here κe is the resonator's photon loss rate through coupling to the CPW, and κi is the resonator decay rate due to it's internal losses, and we have κ = κe + κi. In this simulation, we have assumed tan δ0 = 10−4 for the capacitor dielectric. We adjust Qe through simulation; we target Qe = Qi in order to optimize the coupling while maintaining res- onator's coherence by avoiding a large κe. This criti- cal CPW-resonator coupling regime can not be practi- cally achieved using typical capacitive coupling geome- tries, and motivated the galvanic coupling scheme. We estimate that in this device, the galvanic connection en- hances κe by approximately one order of magnitude com- pared to capacitive coupling with dimensions compatible with standard photolithography. To obtain gcol, we assume that the current within the loop is split equally and flows on the inner and outer edges of the superconducting loop. This is a more real- istic flow compared to a single current loop or a uniform current density through the inductor cross section. We consider a cylindrical volume V with uniform doping den- sity of 1017 cm−3 inside the inductor loop with a depth of 500 nm from the substrate surface. The x-component of Bac from both loops is then calculated at each individual spin location, giving a gi for each spin i within V . Using an approach described in [44], we find that a histogram of gi shows a relatively sharp peak at gpeak/2π = 2.2 kHz and the number of spins within the full width at half maximum is obtained as 4 × 105, yielding gcol/2π = 1.4 MHz. The value of gpeak is significantly smaller than the range of g shown in Fig. 3(b) for the single-atom de- vice, because the inductor loop is much larger in the spin ensemble device. In order to simulate the device transmission spec- troscopy when coupled to the spin ensemble, we consider the calculated value of gcol and use a theoretical model previously developed for a different quantum device with similar physics [45]. The model is based on the Jaynes- Cummings model [33] and the so-called input-output the- ory [46], with the assumption that the coupled spin en- semble can be treated as an oscillator in the limit of small 5 FIG. 4. (a) Layout of the simulated Al resonator coupled to an ensemble of Bi spins (ground plane surrounding the res- onator and the Stark-shift electrodes are not shown). Here, the dielectric thickness for the capacitor C is assumed to be d = 50 nm. (b) Simulated resonator transmission around its resonance frequency f0 = 7.2 GHz in the absence of spins (no doping). Spectroscopy simulation of the hybrid resonator-spin ensemble device near f0 for (c) zero spin-resonator detuning and (d) with tunable static magnetic field B0 near the degen- eracy field Bd. power and low temperature [47]. The theory also treats the CPW as an extremely low-Q (Qb <∼ 1) cavity with resonance frequency ωb and photon annihilation opera- tor b. This allows accurate simulation of devices with asymmetric transmission, and is not limited to perfectly symmetric transmissions [45]. The Hamiltonian now be- comes H = HJC + ¯hωb b†b + ¯hc(b†a + a†b) where c is the coupling rate between the resonator and cavity b, and the spin-related parameters in HJC are now associ- ated with the spin ensemble. Figures 4(c-d) show the spectroscopy simulation results using tan δ0 = 10−4 and Qi = Qe for the resonator with a symmetric transmission, and a conservative value of T1 = 100 µs for the intrinsic (not Purcell-limited) relaxation time of the spin ensem- ble. In the presence of experimental spectroscopy data (not presented here), one can extract important param- eters of this hybrid spin-resonator system directly using this theory. These parameters include gcol, the resonator a)b)0.51-0.2500.25c)d)-50-60cCPWGND xyzQ and the intrinsic relaxation time T1 for the spin en- semble. From the high-frequency simulation, the geometric in- ductance of the loop is extracted as Ltot = 8.5 pH. De- spite the significantly larger inductor dimensions of the spin ensemble device, its impedance is similar to that of the single-atom device with a nanoscale spiral inductor for Nloops = 2− 3, primarily due to the smaller contribu- tion of the kinetic inductance in the spin ensemble device. Therefore, we conclude that the CPW can also be suffi- ciently coupled to the single-atom device using the same (galvanic) coupling scheme. IV. QUBIT OPERATION The qubit operation parameters of the single-atom de- vice presented in this paper are adopted from a com- monly used cQED approach [15]. To estimate the ini- tialization, manipulation and readout performance, we focus on the Nb/P-2S configuration with g/2π = 0.24 MHz, Qi = 4× 104 and Qe = 4× 105 realizable according to our estimation of capacitive coupling [48], or galvanic coupling. We also assume that the bare cavity resonance frequency is ω0/2π = 5.6 GHz and the Zeeman splitting frequency ωs is tunable around ω0. Initialization: The zero-detuning (∆ ≡ ωs − ω0 = 0) relaxation time limited by the Purcell effect for this strongly-coupled system is obtained as T1,init = Γ−1 P = 2κ−1 = 2.3 µs [49] which is several orders of magnitude smaller than the free spin relaxation time. The linear dependence of ΓP on κ is a result of strong coupling, which distinguishes it from previously measured weakly coupled systems [36] where ΓP ∝ g2/κ. Also, the 2-order of magnitude increase in g in our device contributes to this dramatically fast initialization. √ (cid:112)−A + Manipulation: We can Stark shift the spin states using the electrodes shown in Fig. 2, or use mag- netic field to tune the Zeeman energy. For a demon- stration of the qubit operation parameters of the de- vice, we choose to operate at ∆ = 40g, where the 4π (κ − strong-coupling Purcell rate becomes ΓP /2π = 1 √ A2 + κ2∆2) = 88 Hz with A ≡ ∆2 + 2 4g2 − κ2/4 [49], giving rise to T2,rot = 2/ΓP = 1.8 ms. This relaxation time is not far from the measured Hahn-echo T H 2 = 1.1 ms for a P donor electron spin in enriched 28Si, believed to be limited primarily by the static magnetic field noise and thermal noise, and not due to the proximity to the oxide layers or other amor- phous material [9]. Therefore, it is reasonable to as- sume that the spin T2 time in our device is Purcell lim- ited, hence set by T2,rot. Note that for the Al/Bi set, the direction of the Stark shift must be such that the F, mF(cid:105) = 5,−5(cid:105) ↔ 4,−4(cid:105) transition frequency, al- ready reduced by a relatively small B0 to lift the mul- tiplet degeneracy, is further reduced to avoid exciting 6 the higher frequency multiplet transitions. In this dis- persive regime (∆ (cid:29) g), the microwave drive frequency is ωµw/2π = ωs/2π + (2nlim + 1)g2/2π∆ = 5.614406 GHz where photon number nlim = ∆2/4g2 = 400 sets the maximum limit for the drive power [15], and is lower than the critical photon number ncrit ≈ 1800 corresponding to the spiral inductor's critical current. The qubit rotation speed, at on-resonance photon number nres ≈ 17 × 106 √ nres/π∆ = 29 MHz set by nlim and ωµw, is frot = gκ [50], corresponding to Nπ = 2frotT2,rot > 105 coherent π-rotations. Readout: The spin readout also occurs in the dis- persive regime, where the cavity frequency depends on the spin state, giving rise to a fsep = g2/π∆ = 12 kHz separation frequency. This corresponds to a phase shift of φ = 2 arctan(2g2/κ∆) = 10◦, well above the mea- surement sensitivity usually considered to be 0.1◦, and suggests an extremely high-fidelity readout when the res- onator is driven at ω0. The measurement time to resolve 0)−1 = 3.1 µs where φ is estimated to be Tm = (2κnreadθ2 θ0 ≡ 2g2/κ∆ [15] and nread = 25 is the readout pho- ton number. In order to avoid the highly nonlinear re- sponse regime, the readout power must be limited such that nread <∼ ∆2/4g2. V. CONCLUSION In summary, we have proposed and designed a novel device that enhances the coupling of a single-atom spin to the magnetic field of a circuit resonator by approximately 100 times compared to the previously proposed architec- tures that use coplanar resonators. This dramatic im- provement is a result of using a low impedance, lumped element resonator design and a nanoscale spiral induc- tor geometry. We showed the possibility of entering the strong coupling regime necessary for practical purposes, i.e., spectroscopic measurements and qubit realization. Using the well-established principles of cavity quantum electrodynamics, we showed that this large g can lead to a significantly enhanced spin relaxation rate desired for qubit initialization, tens of megahertz spin rotation speed during manipulation without the need for an Oersted line, and superb dispersive readout sensitivity. Moreover, this architecture can be useful for coupling distant qubits using cavity photons for the realization of multi-qubit gates. We emphasize that this large enhancement in the coupling rate is resulted directly from increasing the ac magnetic field at the spin, rather than by hybridizing the spin and charge states through, e.g., a magnetic field gra- dient generated by local micromagnets. In this way, we believe that a much larger spin-photon coupling rate can be achieved while avoiding a concomitant increase in the charge noise-induced decoherence rate [24, 25]. We also proposed a relatively simple proof-of-principle experiment which does not require single-donor implan- tation or e-beam lithography. This simplified scheme takes advantage of resonator's enhanced coupling to a spin ensemble, expected to result in the direct observa- tion of the vacuum Rabi splitting to confirm strong cou- pling. ACKNOWLEDGMENTS The authors thank G. Bryant, D. Pappas, T. Purdy, M. Stewart, K. Osborn, J. Pomeroy, C. Lobb, A. Morello, C. Richardson, R. Murray and R. Stein for many useful discussions. APPENDIX: INDUCTANCE SIMULATIONS, MAGNETIC FIELD AND COUPLING RATE CALCULATION The total inductance Ltot = Lg + Lp within the res- onator consists of the geometric inductance Lg of the spiral inductor giving rise to the magnetic field Bac that couples to the donor electron spin, and parasitic induc- tance Lp which does not contribute to Bac and only limits it. Lg consists of the trace inductance Lt arising from the length of the spiral trace, and some mutual inductance LM between the loops such that Lg = Lt+LM. The para- sitic inductance Lp includes the kinetic inductance Lk of the spiral, the kinetic inductance LC,k within the capac- itor plates and the geometric self-inductance LC,g of the capacitor such that Lp = Lk +LC,k +LC,g. Since Lp does not create magnetic fields at the location of the spin and only limits Iac through Ltot, we want to minimize it. In order to obtain an accurate estimation of Lg and Lp, we performed calculations as well as software simulations. We study two device geometries, one using a single-layer spiral inductor and the other using a double-layer spiral (2S) inductor shown in Figs. 1(c) and 2, respectively. Lg is approximately calculated, and also separately simulated. In the calculations, for simplicity, we assume that each loop of the spiral inductor is perfectly circu- lar and estimate Lg using the geometric mean distance (GMD) method to the second order in the ratio of the conductor diameter to the loop radius [51]. This indepen- dent loop approximation (ILA) ignores the mutual induc- tance LM within the spiral loops. However, one should note that LM contributes to Bac and is not parasitic. Nevertheless, in addition to the ILA, we also simulated the spiral geometry in FastHenry 3D inductance extrac- tion program [52], which accounts for the mutual induc- tances LM within the spiral geometry. Figure 5 shows a comparison between the simulation and the approximate calculation, where the latter ignores LM. Clearly, LM constitutes a larger portion of Lg with increasing Nloops, but is negligible up to Nloops ≈ 2, 3 where g is optimum (see Fig. 3(b)). The total inductance Ltot in Fig. 3(a) 7 FIG. 5. Comparison between the independent loop approxi- mation (ILA) and FastHenry simulations of the spiral geomet- ric inductance versus for different spiral loop counts Nloops for the single-spiral and double-spiral (2S) geometries. is plotted using Lg from FastHenry simulations. How- ever, for simplicity, the ac field contributed by LM was not taken into account in calculating g, resulting in an underestimated g in Fig. 3(b). The kinetic inductance Lk of the spiral loop is caused by the kinetic energy of the quasi-particles within the superconductor and hence does not create any magnetic fields. In our design, Lk is the largest part of Lp. Using the Cooper-pair density ns and mass mC for a super- conducting material, the length of the superconducting line ls and its cross-sectional area As, one can estimate Lk = mCls/4nse2As [53], where e denotes the electron charge. For the single-layer spiral Nb/P resonator with 1 ≤ Nloops ≤ 5 we obtain 0.6 pH ≤ Lk ≤ 7 pH, linearly proportional to the spiral trace length. The kinetic inductance within the capacitor plates, LC,k ≈ 60 fH, is negligible due to the large cross-sectional area of the plates. The geometric self-inductance of the capacitor, calculated using a stripline model, is LC,g ≈ 62 fH, suppressed by employing a relatively thin capacitor insulating layer (small d0). To confirm this, we simu- lated the resonator geometry including the capacitor and found LC,g = 64 fH, in agreement with the calculated value and also negligible for Nloops >∼ 2. In order to achieve the desired resonance frequencies with the relatively small inductances shown in Fig 3(a), relatively large capacitances are required. By using a capacitor insulator thickness of d0 = 25 nm in the single- atom device, we can keep the square-shaped capacitor dimensions below 200 µm, while significantly suppressing LC,g. For the single-atom device, a simulation of g as a func- tion of the spiral trace width w, spacing s and thick- ness t, was performed. By assuming w = s = t, the results showed that, for the Nb (Al) set, the optimum g is obtained when 30 nm <∼ w = s = t <∼ 35 nm (w = s = t ≈ 20 nm), weakly depending on whether the single-layer spiral or the 2S geometry is used. If e-beam lithography resolution of 20 nm is implemented, the Al set can yield a spin-photon coupling rate of g/2π = 0.45 MHz for Nloops = 2. field from all loops, i.e, Bac = (cid:80) At the location of the spin in the single-atom de- vice, Bac is approximated as the sum of the magnetic loop where µ0Iac/2R(cid:48) loops in + 4d2 s)3/2/2d2 loop ≡ (d2 R(cid:48) in is defined as a characteristic radius which accounts for the spin location with respect to the center of the loops and ds denotes the vertical spin displacement from this center (see Figs. 1(c) and 4). Note that the assumption of current flowing in the center of the spiral conductor underestimates Bac, Bac,0 and g, because in reality the majority of current will flow closer to the superconductor-silicon interface due to the relatively large electric permittivity (r = 12) of silicon. For a better understanding of the dependence of g on the number Nloops of the spiral loops, a naive picture may be helpful to the reader. To the first order for Nloops = 1, loops ln(Nloops), and Bac,0, g ∝ Iac ln(Nloops), but Lp is a weaker function of Nloops than Lg. This suggest that by increasing Nloops, Bac,0 and hence g increase up to a point where Lg ap- proaches Lp, and Iac begins to drop as N α, with α < −1. Employing a lumped element design provides the re- quired flexibility to reach this optimum Lg and the cor- responding Nloops. Iac ∝ 1/(cid:112)Lg + Lp, and Lg ∝ N 2 DISCLAIMER: Certain commercial equipment, in- struments, or materials (or suppliers, or software, ...) are identified in this paper to foster understanding. Such identification does not imply recommendation or en- dorsement by the National Institute of Standards and Technology, nor does it imply that the materials or equip- ment identified are necessarily the best available for the purpose. [1] R. Hanson, L. P. Kouwenhoven, J. R. Petta, S. Tarucha, and L. M. K. Vandersypen, Rev. Mod. Phys. 79, 1217 (2007). [2] D. Loss and D. P. DiVincenzo, Phys. Rev. A 57, 120 (1998). [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] 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). [5] J. J. Pla, K. Y. Tan, J. P. Dehollain, W. H. Lim, J. J. Morton, D. N. Jamieson, A. S. Dzurak, and A. Morello, Nature 489, 541 (2012). [6] K. Saeedi, S. Simmons, J. Z. Salvail, P. Dluhy, H. Rie- mann, N. V. Abrosimov, P. Becker, H.-J. Pohl, J. J. Mor- ton, and M. L. Thewalt, Science 342, 830 (2013). [7] A. M. Tyryshkin, S. Tojo, J. J. Morton, H. Riemann, 8 N. V. Abrosimov, P. Becker, H.-J. Pohl, T. Schenkel, M. L. Thewalt, K. M. Itoh, and S. A. Lyon, Nat. Mater. 11, 143 (2012). [8] M. Steger, K. Saeedi, M. L. W. Thewalt, J. J. L. Morton, and H.-J. H. Riemann, N. V. Abrosimov, P. Becker, Pohl, Science 336, 1280 (2012). [9] J. T. Muhonen, J. P. Dehollain, A. Laucht, F. E. Hud- son, R. Kalra, T. Sekiguchi, K. M. Itoh, D. N. Jamieson, J. C. McCallum, A. S. Dzurak, and A. Morello, Nat. Nanotechnol. 9, 986 (2014). [10] D. D. Awschalom, L. C. Bassett, A. S. Dzurak, E. L. Hu, and J. R. Petta, Science 339, 1174 (2013). [11] A. Wallraff, D. I. Schuster, A. Blais, L. Frunzio, R.-S. Huang, J. Majer, S. Kumar, S. M. Girvin, and R. J. Schoelkopf, Nature 431, 162 (2004). [12] X. Gu, A. F. Kockum, A. Miranowicz, Y. xi Liu, and F. Nori, Phys. Rep. 718-719, 1 (2017). [13] J. Majer, J. Chow, J. Gambetta, J. Koch, B. Johnson, J. Schreier, L. Frunzio, D. Schuster, A. Houck, A. Wall- raff, et al., Nature 449, 443 (2007). [14] B. Sarabi, A. N. Ramanayaka, A. L. Burin, F. C. Well- stood, and K. D. Osborn, Phys. Rev. Lett. 116, 167002 (2016). [15] A. Blais, R.-S. Huang, A. Wallraff, S. M. Girvin, and R. J. Schoelkopf, Phys. Rev. A 69, 062320 (2004). [16] C. Eichler, A. J. Sigillito, S. A. Lyon, and J. R. Petta, Phys. Rev. Lett. 118, 037701 (2017). [17] A. Imamoglu, Phys. Rev. Lett. 102, 083602 (2009). [18] G. Tosi, F. A. Mohiyaddin, H. Huebl, and A. Morello, AIP Adv. 4, 087122 (2014). [19] Y. Kubo, F. R. Ong, P. Bertet, D. Vion, V. Jacques, D. Zheng, A. Dr´eau, J.-F. Roch, A. Auffeves, F. Jelezko, J. Wrachtrup, M. F. Barthe, P. Bergonzo, and D. Esteve, Phys. Rev. Lett. 105, 140502 (2010). [20] R. Amsuss, C. Koller, T. Nobauer, S. Putz, S. Rot- ter, K. Sandner, S. Schneider, M. Schrambock, G. Stein- hauser, H. Ritsch, J. Schmiedmayer, and J. Majer, Phys. Rev. Lett. 107, 060502 (2011). [21] K. Petersson, L. McFaul, M. Schroer, M. Jung, J. Taylor, A. Houck, and J. Petta, Nature 490, 380 (2012). [22] J. J. Viennot, M. C. Dartiailh, A. Cottet, and T. Kontos, Science 349, 408 (2015). [23] X. Hu, Y.-x. Liu, and F. Nori, Phys. Rev. B 86, 035314 (2012). [24] X. Mi, M. Benito, S. Putz, D. M. Zajac, J. M. Taylor, G. Burkard, and J. R. Petta, Nature 555, 599 (2018). [25] N. Samkharadze, G. Zheng, N. Kalhor, D. Brousse, A. Sammak, U. C. Mendes, A. Blais, G. Scappucci, and L. M. K. Vandersypen, Science 359, 1123 (2018). [26] Y. Kubo, C. Grezes, A. Dewes, T. Umeda, J. Isoya, H. Sumiya, N. Morishita, H. Abe, S. Onoda, T. Ohshima, V. Jacques, A. Dr´eau, J.-F. Roch, I. Diniz, A. Auffeves, D. Vion, D. Esteve, and P. Bertet, Phys. Rev. Lett. 107, 220501 (2011). [27] J. Twamley and S. D. Barrett, Phys. Rev. B 81, 241202 (2010). [28] R. W. Simmonds, K. M. Lang, D. A. Hite, S. Nam, D. P. Pappas, and J. M. Martinis, Phys. Rev. Lett. 93, 077003 (2004). [29] J. M. Martinis, K. B. Cooper, R. McDermott, M. Stef- fen, M. Ansmann, K. D. Osborn, K. Cicak, S. Oh, D. P. Pappas, R. W. Simmonds, and C. C. Yu, Phys. Rev. Lett. 95, 210503 (2005). [30] P. Kumar, S. Sendelbach, M. Beck, J. Freeland, Z. Wang, 9 H. Wang, C. Yu, R. Wu, D. Pappas, and R. McDermott, arXiv preprint arXiv:1604.00877 (2016). [31] A. Laucht, J. T. Muhonen, F. A. Mohiyaddin, R. Kalra, J. P. Dehollain, S. Freer, F. E. Hudson, M. Veldhorst, R. Rahman, G. Klimeck, K. M. Itoh, D. N. Jamieson, J. C. McCallum, A. S. Dzurak, and A. Morello, Sci. Adv. 1 (2015). [32] Y. Tokura, W. G. van der Wiel, T. Obata, and S. Tarucha, Phys. Rev. Lett. 96, 047202 (2006). [33] E. T. Jaynes and F. W. Cummings, P. IEEE 51, 89 (1963). [34] M. R. Vissers, J. Hubmayr, M. Sandberg, S. Chaud- huri, C. Bockstiegel, and J. Gao, Appl. Phys. Lett. 107, 062601 (2015). [35] G. Wolfowicz, S. Simmons, A. M. Tyryshkin, R. E. George, H. Riemann, N. V. Abrosimov, P. Becker, H.- J. Pohl, S. A. Lyon, M. L. W. Thewalt, and J. J. L. Morton, Phys. Rev. B 86, 245301 (2012). [36] A. Bienfait, J. J. Pla, Y. Kubo, X. Zhou, M. Stern, C. C. Lo, C. D. Weis, T. Schenkel, D. Vion, D. Esteve, J. J. L. Morton, and P. Bertet, Nature 531, 74 (2016). M. Hofheinz, N. Katz, E. Lucero, C. McKenney, M. Nee- ley, H. Wang, E. M. Weig, A. N. Cleland, and J. M. Martinis, Appl. Phys. Lett. 92, 112903 (2008). [42] X. Liu, D. R. Queen, T. H. Metcalf, J. E. Karel, and F. Hellman, Phys. Rev. Lett. 113, 025503 (2014). [43] C. Jagannath, Z. W. Grabowski, and A. K. Ramdas, Phys. Rev. B 23, 2082 (1981). [44] A. Bienfait, J. Pla, Y. Kubo, M. Stern, X. Zhou, C. Lo, C. Weis, T. Schenkel, M. Thewalt, D. Vion, et al., Nature nanotechnology 11, 253 (2016). [45] B. Sarabi, Cavity quantum electrodynamics of nanoscale two-level systems, Ph.D. thesis, University of Maryland, College Park (2014). [46] M. J. Collett and C. W. Gardiner, Phys. Rev. A 30, 1386 (1984). [47] B. Sarabi, A. N. Ramanayaka, A. L. Burin, F. C. Well- stood, and K. D. Osborn, Appl. Phys. Lett. 106, 172601 (2015). [48] J. M. Martinis, R. Barends, and A. N. Korotkov, arXiv preprint arXiv:1410.3458 (2014). [49] E. A. Sete, J. M. Gambetta, and A. N. Korotkov, Phys. [37] M. H. Mohammady, G. W. Morley, and T. S. Monteiro, Rev. B 89, 104516 (2014). Phys. Rev. Lett. 105, 067602 (2010). [38] M. Fuechsle, J. A. Miwa, S. Mahapatra, H. Ryu, S. Lee, O. Warschkow, L. C. L. Hollenberg, G. Klimeck, and M. Y. Simmons, Nat. Nanotechnol. 7, 242 (2012). [50] S. Haroche, in Fundamental Systems in Quantum Optics, edited by R. J. Dalibard J and Z.-J. J (Elsevier Science Publishers B.V., 1993) Chap. Cavity Quantum Electro- dynamics, p. 767. [39] J. Romijn, T. M. Klapwijk, M. J. Renne, and J. E. Mooij, [51] F. W. Grover, Inductance calculations: working formulas Phys. Rev. B 26, 3648 (1982). and tables (Courier Corporation, 2004). [40] R. Huebener, R. Kampwirth, R. Martin, T. Barbee, and [52] M. Kamon, M. J. Tsuk, and J. K. White, IEEE T. Mi- R. Zubeck, IEEE T. Magn. 11, 344 (1975). [41] A. D. O'Connell, M. Ansmann, R. C. Bialczak, crow. Theory 42, 1750 (1994). [53] M. Tinkham, Introduction to superconductivity (Courier Corporation, 1996).
1301.3573	2	1301	2013-06-05T08:27:38	Symmetry and magnitude of spin-orbit torques in ferromagnetic heterostructures	[ "cond-mat.mes-hall" ]	Current-induced spin torques are of great interest to manipulate the orientation of nanomagnets without applying external magnetic fields. They find direct application in non-volatile data storage and logic devices, and provide insight into fundamental processes related to the interdependence between charge and spin transport. Recent demonstrations of magnetization switching induced by in-plane current injection in ferromagnetic heterostructures have drawn attention to a class of spin torques based on orbital-to-spin momentum transfer, which is alternative to pure spin transfer torque (STT) between noncollinear magnetic layers and amenable to more diversified device functions. Due to the limited number of studies, however, there is still no consensus on the symmetry, magnitude, and origin of spin-orbit torques (SOTs). Here we report on the quantitative vector measurement of SOTs in Pt/Co/AlO trilayers using harmonic analysis of the anomalous and planar Hall effects as a function of the applied current and magnetization direction. We provide an all-purpose scheme to measure the amplitude and direction of SOTs for any arbitrary orientation of the magnetization, including corrections due to the interplay of Hall and thermoelectric effects. Based on general space and time inversion symmetry arguments, we show that asymmetric heterostructures allow for two different SOTs having odd and even behavior with respect to magnetization reversal. Our results reveal a scenario that goes beyond established models of the Rashba and spin Hall contributions to SOTs. The even SOT is STT-like but stronger than expected from the spin Hall effect in Pt. The odd SOT is composed of a constant field-like term and an additional component, which is strongly anisotropic and does not correspond to a simple Rashba field.	cond-mat.mes-hall	cond-mat	1 Symmetry and magnitude of spin-orbit torques in ferromagnetic heterostructures Kevin Garello1, Ioan Mihai Miron2, Can Onur Avci1, Frank Freimuth3, Yuriy Mokrousov3, Stefan Blügel3, Stéphane Auffret2, Olivier Boulle2, Gilles Gaudin2, and Pietro Gambardella1,4,5 1 Catalan Institute of Nanotechnology (ICN), E-08193 Barcelona, Spain 2 SPINTEC, UMR-8191, CEA/CNRS/UJF/GINP, INAC, F-38054 Grenoble, France 3 Peter Grünberg Institut and Institute for Advanced Simulation, Forschungszentrum Jülich and JARA, 52425 Jülich, Germany 4 Institució Catalana de Recerca i Estudis Avançats (ICREA), E-08010 Barcelona, Spain 5 Department of Materials, ETH Zurich, CH-8093 Zurich, Switzerland 2 Current-induced spin torques are of great interest to manipulate the magnetization of thin films and nanostructures. Recent demonstrations of magnetization switching induced by in-plane current injection in heavy metal/ferromagnetic heterostructures have drawn attention to a class of spin torques based on orbital-to-spin momentum transfer, which is alternative to the spin transfer torque between noncollinear magnetic layers and amenable to more diversified device functions. The symmetry, magnitude, and origin of spin-orbit torques (SOTs), however, remain a matter of intense debate. Here we report on the three-dimensional vector measurement of SOTs in AlOx/Co/Pt and MgO/CoFeB/Ta trilayers using harmonic analysis of the anomalous and planar Hall effects as a function of the applied current and magnetization direction. We provide an all-purpose scheme to measure the amplitude and direction of SOTs for any arbitrary orientation of the magnetization, including corrections due to the interplay of Hall and thermoelectric effects. Based on general space and time inversion symmetry arguments, we demonstrate that asymmetric heterostructures allow for two different SOTs having odd and even behavior with respect to magnetization reversal. Our measurements show that such torques include strongly anisotropic field-like and spin transfer-like components, which depend markedly on the annealing conditions and type of heavy metal layer. These results call for SOT models that go beyond the spin Hall and Rashba effects investigated thus far. Spintronic devices rely on the generation of spin torques to control the magnetization of miniaturized memory and logic elements using electric currents.1,2 Conventionally, such torques have been associated with the transfer of spin angular momentum between a “polarizer” and a “free” ferromagnetic layer (FM) separated by a 3 nonmagnetic spacer, mediated by a spin polarized current flowing perpendicular to the two layers.2,3 In recent years, however, experiments4-13 and theory14-27 have pointed out alternative mechanisms to produce spin torques based on the spin-orbit interaction, which mediates the transfer of orbital angular momentum from the lattice to the spin system and do away with the need of a polarizer FM. These mechanisms, which include the spin Hall,28 Rashba,29 and Dresselhaus30 effects, exploit the coupling between electron spin and orbital motion to induce nonequilibrium spin accumulation, which ultimately gives rise to a torque on the magnetization via the usual spin transfer channels between s- and d-electrons.31,32 Henceforth, we refer to the torques induced by such effects as spin-orbit torques (SOTs) to underline their common link to the spin- orbit interaction. Of particular relevance for magnetization switching, experiments on AlOx/Co/Pt heterostructures have shown that current injection in the plane of the layers induces a spin accumulation component transverse to the current, δm┴ ~ z × j,5,6 as well as a longitudinal one that rotates with the magnetization in the plane defined by the current and the z axis of the stack, δm║ ~ (z × j) × m,9,33 where j and m are unit vectors that denote the current density and magnetization direction, respectively. Because of the exchange interaction between s- and d-electrons, these components produce two effective magnetic fields, B┴ ~ δm┴ and B║ ~ δm║, or, equivalently, a field-like torque T┴ ~ m × δm┴ and a spin transfer-like torque T║ ~ m × δm║. If j is injected along x, these torques correspond to T┴ ~ m × y and T║ ~ m × (y × m), respectively. Several studies have shown that T║ is strong enough to reverse the magnetization of high- coercivity FM with both perpendicular9,33,34 and in-plane35 anisotropy for current densities of the order of 107 – 108 A/cm2. The fact that the magnetization of a single-layer FM can be reversibly switched by lateral injection of an initially unpolarized current9 has naturally elicited great 4 interest in SOTs for technological applications. For example, T║ has been proposed9,36,37 and demonstrated33 to induce switching of magnetic tunnel junction devices using a three-terminal configuration, where the read and write current paths are separated to avoid damage of the tunnel barrier. On the theoretical side, however, two apparently contrasting pictures have emerged: one based on the bulk spin Hall effect (SHE) in the heavy metal layer as the sole source of spin accumulation9,20,23-25,33,35 and the other on Rashba-type effective fields and spin-dependent scattering effects that take place at the interface between the heavy metal and the FM.9,21-26 Both pictures lead to qualitatively equivalent expressions for T┴ and T║ (Refs. 23-25) but differ in the relative magnitude of the torques, since a pure SHE implies T║ >> T┴ whereas the opposite is true if only interfacial Rashba contributions are considered.25 On the experimental side, the situation is also controversial: In the same model system, AlOx(1.6 nm)/Co(0.6 nm)/Pt(3 nm), measurements of T┴ range from zero (Ref. 33) to 1 T/108 A/cm2 (Ref. 5,6), whereas estimates of T║ go from 0.017 (Ref. 33) to 0.08 T/108 A/cm2 (Ref. 9). In MgO/CoFeB/Ta, Liu et al. reported that the direction of T║ is reversed with respect to AlOx/Co/Pt, consistently with the opposite sign of the spin Hall angle (θSH) of Ta and Pt.35 However, recent data show that the magnitude and even the sign of both T║ and T┴ depend on the thickness of the Ta layer,38 highlighting how much these torques are sensitive to different effects. This state of affairs, which goes together with the lack of a consistent method of measuring SOTs of arbitrary orientation, makes it hard to optimize their efficacy for specific applications and reach a consensus on the physical origin of the torques. The purpose of this paper is threefold. First, starting from symmetry arguments, we derive general expressions of the spin accumulation and current-induced SOTs in magnetic heterostructures that are independent of specific physical models. Second, we present a self-consistent, sensitive method to perform three-dimensional vector measurements of SOTs using an ac susceptibility technique based on the combination of 5 the 1st and 2nd harmonic contributions of the anomalous Hall (AHE) and planar Hall effects (PHE). Third, we demonstrate unambiguously the existence of two distinct SOTs that have odd and even symmetry with respect to the inversion of the magnetization and include, but are not limited to, T┴ ~ m × y and T║ ~ m × (y × m) (Fig. 1 a-c). We find strongly anisotropic SOT components that have not been observed thus far, which depend on the x and y projections of the magnetization in the plane of the current. T┴ and T║ have comparable magnitude in AlOx/Co/Pt in the low current (low heating) limit and decrease significantly due to interface diffusion upon annealing. Both T┴ and T║ reverse sign and are dominated by anisotropy effects in MgO/CoFeB/Ta. The picture that emerges from this study is that, although SHE contributions to SOTs are important, interfacial effects play a prominent role in determining the magnitude and anisotropy of the torques. Spin-orbit torque symmetry and effective fields SOTs require inversion asymmetry in order to induce measurable effects on the magnetization, which is usually realized by sandwiching a FM between two different layers, as shown in Fig. 1. This is true also for torques produced by the SHE, which would average to zero in a symmetric heterostructure. The minimal requirements imposed by structure inversion asymmetry on a magnetic layer are continuous rotational symmetry around the z axis and mirror symmetry with respect to planes parallel to z. For this geometry we derived the general expressions for δm┴,║ and the resulting SOTs by expanding δm┴,║ in terms of the angular cosines of the magnetization defined with respect to the symmetry axis and current flow directions (see Supplementary Information). We find that the spin accumulation contains magnetization dependent terms that add to the δm┴ ~ y and δm║ ~ y × m components considered thus far, which change the symmetry and amplitude of T┴ and T║. The quantitative significance of 6 these terms, however, must be established by experiment. Our measurements determine a minimal set of terms required to model the action of the field-like and spin transfer – like torques, namely T⊥ = (𝐲 × 𝐦) [𝑇0⊥ + 𝑇2⊥ (𝐳 × 𝐦)2 + 𝑇4⊥ (𝒛 × 𝒎)4 ] + 𝐦 × (𝐳 × 𝐦)(𝐦 ∙ 𝐱) [𝑇2⊥+𝑇4⊥ (𝐳 × 𝐦)2 ] (1) T∥ = 𝐦 × (𝐲 × 𝐦) 𝑇0∥ + (𝐳 × 𝐦)(𝐦 ∙ 𝐱) �𝑇2∥+𝑇4∥ (𝐳 × 𝐦)2 � . (2) m. For the special case 𝑇𝑛⊥ = 𝑇𝑛∥ = 0 for all n ≠ 0, Eqs. 1 and 2 simplify to T⊥ = Note that these torques are, respectively, odd and even with respect to the inversion of 𝑇0⊥ (y × m) and T∥ = 𝑇0∥ m × (y × m), which have been obtained theoretically for several models discussed in the literature.21-26 For the purpose of comparison with the experiment, we consider here the effective magnetic fields B┴ and B║ corresponding to T┴ and T║ obtained above. As the relevant field components are perpendicular to the magnetization, we adopt a spherical coordinate system (Fig. 1d) with basis vectors 𝐞𝜃 = (cos 𝜃 cos 𝜑 , cos 𝜃 sin 𝜑 , − sin 𝜃) and 𝐞𝜑 = (− sin 𝜑 , cos 𝜑 , 0), where 𝐦 = (sin 𝜃 cos 𝜑 , sin 𝜃 sin 𝜑 , cos 𝜃) and B⊥ = −cos 𝜃 sin 𝜑 (𝑇0⊥ + 𝑇2⊥ sin2 𝜃 + 𝑇4⊥ sin4 𝜃)e𝜃 − cos 𝜑 𝑇0⊥ e𝜑 , (3) B∥ = cos 𝜑 �𝑇0∥ + 𝑇2∥ sin2 𝜃 + 𝑇4∥ sin4 𝜃�e𝜃 − cos 𝜃 sin 𝜑 𝑇0∥e𝜑 . (4) Using Eqs. (3) and (4), the action of the current-induced fields on the magnetization can be directly compared to that of a reference external field (Bext) of known magnitude and direction by means of low frequency susceptibility measurements.6,39 7 Hall measurements of current-induced effective fields To quantify and analyze the SOT fields described above, we studied AlOx(2 nm)/Co(0.6 nm)/Pt(3 nm) as model system, deposited on oxidized Si by dc-magnetron sputtering. The AlOx/Co stack was patterned into 1 x 1 and 1 x 0.5 µm2 rectangular dots and the Pt layer etched into a cross, as shown in Fig. 1, which allows for current injection and Hall voltage (VH) detection. We used an ac current of frequency f to modulate the SOT amplitude and induce small oscillations of m about its equilibrium direction, defined by Bext and the magnetic anisotropy of the trilayer. Such oscillations generate a second harmonic contribution to VH, which provides a sensitive way to measure current-induced fields (Supplementary Information). In general, VH depends on mz through the AHE and on the product mxmy through the PHE: 𝑉𝐻 = 𝑅𝐴𝐻𝐸 𝐼 cos 𝜃 + 𝑅𝑃𝐻𝐸 𝐼 sin2 𝜃 sin 2𝜑 , (5) where 𝑅𝐴𝐻𝐸 and 𝑅𝑃𝐻𝐸 are the AHE and PHE resistances, respectively, and I is the injected current. In terms of the total Hall resistance 𝑅𝐻 = 𝑉𝐻 /𝐼 , the first harmonic term 𝑅𝐻𝑓 = 𝑅𝐴𝐻𝐸𝑓 + 𝑅𝑃𝐻𝐸𝑓 independent of modulated fields. The second harmonic term 𝑅𝐻2𝑓 measures the relates to the equilibrium direction of the magnetization and is susceptibility of the magnetization to the current-induced fields and is given by: 𝐵𝜑 𝑅𝐻2𝑓 = ( 𝑅𝐴𝐻𝐸 − 2𝑅𝑃𝐻𝐸 cos 𝜃 sin 2𝜑) 𝑑 cos 𝜃 𝐵𝜃 𝐵𝑒𝑥𝑡 sin 𝜃𝐵 , (6) sin(𝜃𝐵 − 𝜃) + 2𝑅𝑃𝐻𝐸 sin2 𝜃 cos 2𝜑 𝑑𝐵𝑒𝑥𝑡 where Bθ and Bϕ represent the polar and azimuthal components of the total effective field B┴ + B║ induced by the current and θB is the polar angle of Bext (Fig. 1d). Equation Information for more details). If 𝑅𝑃𝐻𝐸 = 0, it is straightforward to evaluate Bθ from Eq. 6 allows us to measure Bθ and Bϕ as a function of θ and ϕ (see Supplementary 𝑑𝐵𝑒𝑥𝑡 = 𝑑𝑅𝐻𝑓𝑑𝐵𝑒𝑥𝑡 . Otherwise, Bθ and Bϕ must be evaluated by 6 by noting that 𝑅𝐴𝐻𝐸 𝑑 cos 𝜃 measuring VH at ϕ = 0º and 90º and fitting 𝑅𝐻2𝑓 using a recursive procedure that accounts 8 for both the AHE and the PHE (𝑅𝐴𝐻𝐸 = 0.72 Ω and 𝑅𝑃𝐻𝐸 = 0.09 Ω for the sample presented in Figs. 1-4). This evaluation procedure has been validated by numerical macrospin simulations as well as by application of external ac fields in phase and antiphase with the current, the amplitude of which was correctly retrieved using Eq. 6 (see Supplementary Information). For full characterization, the sample was mounted on a rotating stage and Bext applied in different directions, defined by polar (θB) and azimuthal (ϕB = ϕ) coordinates. The current was modulated at a frequency f = 10Hz. At extract 𝑅𝐻𝑓 and 𝑅𝐻2𝑓 . The ac current amplitude was varied up to I = 1136 µA, each field point, VH has been measured during 10 s and fast Fourier transformed to corresponding to a current density j = 3.15 x 107A/cm2. Figures 1e and f show 𝑅𝐻𝑓 measured as a function of Bext applied out-of-plane (θB = 0°) and nearly in-plane (θB = 82°), respectively. The curves, proportional to mz, are of 𝑅𝐻𝑓 with increasing in-plane field observed in Fig. 1f is due to the coherent rotation of characteristic of AlOx/Co/Pt layers with strong PMA. The slow and reversible decrease harmonic measurements of 𝑅𝐻2𝑓 for Bext applied at θB = 82°, perpendicular (ϕ = 90º, the Co magnetization towards the hard plane direction. Figure 2 shows the second Fig.2a) and parallel (ϕ = 0º, Fig.2b) to the current. The data are shown after subtraction of sample-dependent contributions to the Hall voltage that are not included in Eq. 5, namely a constant offset due to the voltage probe asymmetry as well as the anomalous small correction to 𝑅𝐻2𝑓 of the order of 0.1 mΩ (Supplementary Information). We note Nernst-Ettinghausen effect (ANE). The ANE can be separately measured, giving a the formation of magnetic domains. According to Eq. 6, 𝑅𝐻2𝑓 is mostly sensitive to the that the choice of θB is not critical as long as Bext is slightly tilted off-plane, to prevent effective field components parallel to eθ, as these affect mz and hence the AHE. significantly weaker. Thus, 𝑅𝐻2𝑓 measured at ϕ = 90º reflects mostly B┴ contributions, Conversely, the components parallel to eϕ are measured through the PHE, which is whereas 𝑅𝐻2𝑓 measured at ϕ = 0º reflects mostly B║ terms. Note that this agrees with the 9 even/odd character of 𝑅𝐻2𝑓 measured at ϕ = 90º/0º with respect to field inversion (Fig. 2a and b), since B┴ and B║ are by definition even and odd with respect to m, opposite to the torques from which they are derived. Field-like and spin transfer-like torque components The effective fields B┴ and B║ derived from 𝑅𝐻2𝑓 are shown in Fig. 2c and d for m // y (ϕ = 90º) and m // x (ϕ = 0º), respectively. These measurements represent a central result of this work, providing a direct quantitative estimate of the SOT fields for different orientations of the magnetization. We find several interesting features that reveal a more complex scenario than previously anticipated. In particular, B┴ depends not only on the current amplitude but also on the applied magnetic field, with a minimum of about 10 mT at Bext = 0 T and increasing up to 20 mT for Bext ≥ 1 T. Since the external field only changes the orientation of the magnetization, this behaviour indicates that B┴ depends strongly on the direction of m. By converting the field closely follows the function − cos 𝜃 (𝑇0⊥ + 𝑇2⊥ sin2 𝜃) with 𝑇0⊥ = -11.6 ± 0.7 mT and dependence into a θ dependence using the AHE, we find that B┴ measured at ϕ = 90º 𝑇2⊥ = -11.2 ± 0.6 mT, as shown in Fig. 2e. This expression is in agreement with Eq. 3, but differs remarkably from that expected from either the Rashba field5,14,24-26 or the field-like component of the SHE torque23,25 reported in the literature, which would require 𝑇2⊥ = 0. We note that 𝑇0⊥ includes the contribution of the Oersted field produced by the current flowing in the Pt layer, which we estimate as 𝜇0𝐼2𝐿 = −0.7 mT (antiparallel to y), where L is the width of the current line and µ0 the vacuum permeability. The dependence of B║ on the magnetization is remarkably different from B┴. approximated by 𝑇0∥ + 𝑇2∥ sin2 𝜃 + 𝑇4∥ sin4 𝜃 , in agreement with Eq. 4, with 𝑇0∥ = 19.0 ± Figure 2f shows that B║ measured at ϕ =0º is weakly dependent on θ, and is well- 10 0.5 mT, 𝑇2∥ = 2 ± 1 mT, and 𝑇4∥ = -1 ± 1 mT. As the higher order coefficients are small and tend to compensate, B║ can be reasonably approximated by a constant value 𝑇0∥ , consistently with previous findings.9,33,34 This behavior is typical of all our AlOx/Co/Pt samples, apart from small changes of the coefficients that we attribute to pattern or material inhomogeneities. measurements for different orientations of m(θ, 0º≤ϕ ≤90º). In such a case, 𝑅𝐻2𝑓 , shown In order to complete the description of B┴ and B║, we performed a series of in Fig. 3a, is given by the linear superposition of two terms 𝑅𝐻2𝑓 (B⊥ ) + 𝑅𝐻2𝑓 (B∥), which of m (see Supplementary Information). Figures 3b and c show 𝑅𝐻2𝑓 (B⊥ ) and 𝑅𝐻2𝑓 (B∥), can be easily separated owing to their different symmetry with respect to the inversion respectively, as a function of ϕ. The lineshape of 𝑅𝐻2𝑓 (B⊥ ) is similar to 𝑅𝐻2𝑓 measured at ϕ = 90º, shown in Fig. 2a, whereas 𝑅𝐻2𝑓 (B∥ ) is similar to 𝑅𝐻2𝑓 measured at ϕ = 0º, shown in Fig. 2b. However, the amplitude of 𝑅𝐻2𝑓 (B⊥ ) increases whereas 𝑅𝐻2𝑓 (B∥) decreases as ϕ goes from 0º to 90°. From these curves we obtain that the polar component of B┴ scales proportionally to sinϕ, whereas the polar component of B║ within the error of our data, the SOT coefficients 𝑇0⊥ , 𝑇2⊥ , and 𝑇0∥ are independent of scales as cosϕ , in agreement with Eqs. (3) and (4), respectively. This implies that, ϕ (Fig. 3d-f), in agreement with the superposition principle for the current and the resulting linear-response torques. Torque to current ratios Figure 4 shows that the amplitude of T┴ and T║ scales linearly with the current up to j = 1.5 x 107 A/cm2. Above this value, we observe a nonlinear increase of the coefficients 𝑇0⊥ , 𝑇2⊥ , and 𝑇0∥ , which we attribute to Joule heating effects. We note that, at the maximum current density employed in this study (3.15 x 107 A/cm2), we observe a 11 reduction of the AHE (-3.5%) and magnetic anisotropy (-13%) as well as an increase of the resistivity of the layers (+13%). All these effects can be attributed to heating and may alter the SOT/current ratio. Our measurements also offer quantitative insight into the magnitude of the different kinds of SOTs. In the following, we express the torques per unit of magnetic moment, thereby assigning them the same units as the effective fields. We discuss first T┴. From the initial slope of the data reported in Fig. 4, we find that the SOT/current ratios corresponding to 𝑇0⊥ and 𝑇2⊥ are -3.2 ± 0.2 and -2.3 ± 0.2 mT per 107 A/cm2, respectively. This corrects our previous estimate of T┴ based on current-induced domain nucleation,5 which largely overestimated 𝑇0⊥ due to heat-assisted magnetization reversal6 and neglect of 𝑇2⊥ . Moreover, we prove beyond doubt that T┴ is not an apparent effect due to spin Hall torque dynamics in the high current regime. This hypothesis was suggested by Liu et al., who reported no evidence of T┴ in AlOx/Co/Pt within a sensitivity of 1.3 mT per 107 A/cm2 (Ref. 33). Our measurements, however, are quasi-static and extend well into the low current regime. We suggest that the negative result of Liu et al. might be partly due to the different preparation of the AlOx/Co/Pt stack (oxidized in air and annealed up to 350ºC) and partly to the lower sensitivity afforded by the dc measurements reported in Ref. 33 (a comparison of ac and dc measurements is reported in the Supplementary Information). By fitting 𝑇0∥ at low current density (j < 1.5 x 107 A/cm2, Fig. 4c), we obtain 𝑇0∥ = 5.0 ± 0.2 mT per 107 A/cm2. Note, however, that the data reported in Fig. 4 (dots and squares) represent a lower bound of the torque/current ratio due to current dispersion in the Hall voltage probes, which reaches up to 23% of the total current in a square Hall cross.40 Measurements of Hall crosses with narrower voltage probes (0.5 µm instead of 1 µm) give consistently higher torque/current ratios, namely 𝑇0⊥ = -4.0 ± 0.3, 𝑇2⊥ = -2.7 ± 0.1, and 𝑇0∥ = 6.9 ± 0.3 mT per 107 A/cm2 (Fig. 4, triangles, and Supplementary 12 Information). This value of 𝑇0∥ is about four times larger than that reported by Liu et al. for AlOx/Co/Pt (1.7 mT per 107 A/cm2),33 and also larger compared to MgO/CoFeB/Ta (4 mT per 107 A/cm2).35 Supposing that T║ originates uniquely from the bulk spin Hall 𝑇0∥ = ℏ2𝑒𝑀𝑡𝐶𝑜 𝜗𝑆𝐻 𝑗 [1 − sech(𝑡𝑃𝑡 /𝜆𝑃𝑡 )], where e is the electron charge, M = 106 A/m, tCo effect in Pt and that the Pt/Co interface is fully transparent, implies = 0.6 nm, and 𝜗𝑆𝐻 is the spin Hall angle of Pt. The last factor takes into account the partial compensation of the spin Hall current due to spin diffusion from the bottom Pt 1.4 nm,41 our measurement of 𝑇0∥ implies 𝜗𝑆𝐻 = 0.16, which is comparable to Ta35 and interface, which depends on the spin diffusion length of Pt (λPt).33 By assuming λPt = more than twice the largest spin Hall angle reported previously for Pt.41 Taking λPt = 10 nm, as discussed in a recent report,42 would push 𝜗𝑆𝐻 to even higher values. We conclude that either 𝜗𝑆𝐻 is much larger than expected or additional spin-orbit and interface-related effects contribute to T║. Torque dependence on interface and material parameters To investigate how the SOTs depend on the quality of the AlOx/Co/Pt interfaces, we measured B┴ and B║ on trilayers annealed to 300 ºC for 30 minutes in vacuum. We corresponding to a reduction of 𝑇0⊥ , 𝑇2⊥ , and 𝑇0∥ by about 17 %, 60%, and 23%, find that annealing induces a significant degradation of the SOT amplitude, respectively (Fig. 5a and b). The resistivity, which is 36 µΩcm in the as deposited samples, increases by about 7%, whereas the AHE goes from 0.80 to 1.14 Ω. This is consistent with previous measurements of annealed AlOx/Co/Pt trilayers,43 where the AHE increase was attributed to the diffusion of Pt atoms into the Co layer. Since annealing above 250 ºC is known to induce mixing of Co and Pt and affect the oxidation of the AlOx/Co interface,43,44 we conclude that both T┴ and T║ are very sensitive to the interface quality of the trilayers. 13 Figure 5c and d show B┴ and B║ measured on MgO(2 nm)/CoFeB(0.9 nm)/Ta(3 nm) layers annealed to 300 ºC in vacuum. By comparison with Fig. 5a and b, it is evident that B┴ and B║ reverse sign compared to AlOx/Co/Pt, consistently with previous studies.35,38 However, the strong θ dependence of both fields, not observed before, fits of B┴ and B║ according to Eqs. 3 and 4 give 𝑇0,2,4⊥ = 4.5±0.1, 5.6±0.2, 5.9±0.3 and reveals that the SOT anisotropy is a general effect that is not unique to AlOx/Co/Pt. The 𝑇0,2,4∥ = -2.4±0.1, 0.4±0.4, -2.0±0.4 mT per 107 A/cm2. Thus, both 2nd and 4th order terms of B┴ and B║ in MgO/CoFeB/Ta. We note also that our value of 𝑇0∥ is in between that of amplitude comparable to the 0th order are required to model the angular dependence reported by Refs. 35,38, whereas the field-like terms are considerably larger than the spin-transfer like ones, contrary to what is found for perpendicular current injection in metallic spin valve systems.45 In conclusion, general symmetry arguments show that T┴ and T║ can have a complex vector dependence on the direction of the magnetization. This work provides the first evidence for this effect as well as a method to measure T┴, T║, and their dependence on the magnetization in vector form. We find that there are significant deviations from the SOT models considered so far based on the Rashba and spin Hall effects. In the case of AlOx/Co/Pt, the largest deviations are observed for T┴ due to terms proportional to 𝑇2⊥ in Eq. 1. Thus, the effective field B┴ generated by the current includes magnetization-dependent components perpendicular to the y axis, whereas the Rashba model can only explain components parallel to y. This suggests that the previous picture of T┴ induced by a Rashba field of constant magnitude9,21-26 or the field-like component of the spin Hall torque23,25 has to be extended by a calculation of the torque based on a realistic description of the electronic structure. In the case of MgO/CoFeB/Ta, both T┴ and T║ present strong anisotropic components, which maximize the torques when the magnetization lies in the plane of the FM. Tuning of the 14 vector properties of SOTs may play a crucial role in developing novel spintronic devices where different magnetic states are induced by specific SOT components. METHODS The samples were fabricated from Al(1.6 nm)/Co(0.6 nm)/Pt(3 nm) and MgO(2)/CoFeB(0.9)/Ta(3) layers deposited on a thermally oxidized silicon wafer by dc magnetron sputtering. The deposition rates were 0.05 nm/s (Co and Al), 0.15 nm/s (Ta) and 0.1 nm/s (Pt, Mg) at an Ar pressure of 2x10-3 mbar. After deposition the Al/Co/Pt films were oxidized by exposure to a radiofrequency (rf) oxygen plasma at a pressure of 3x10-3 mbar and an rf power of 10 W for 29 s. Mg/CoFeB/Ta was naturally oxidized in an oxygen pressure of 150 mbar for 10 s. The AlOx/Co/Pt films were patterned by e- beam lithography and ion beam etching into 1000 and 500 nm square AlOx/Co dots and Pt Hall crosses. The typical resistance of these devices is 3-4 kΩ and is mostly due to the thin Pt contact leads, whereas the resistivity of AlOx/Co/Pt is 36 µΩcm. The MgO/CoFeB/Ta layers were patterned into 1000 nm wide Hall bars with 500 nm voltage branches. The resistivity of our MgO/CoFeB/Ta devices is 184 µΩcm. The Hall voltage measurements were performed at room temperature by using an ac current of amplitude 200 to 1136 µA modulated at f = 10 Hz. 15 References Acknowledgements This work was supported by the European Research Council (StG 203239 NOMAD), the European Commission under the Seventh Framework Programme (GA 318144, SPOT), Ministerio de Economía y Competitividad (ERA-Net EUI2008-03884, MAT2010-15659), Agència de Gestió d'Ajuts Universitaris i de Recerca (2009 SGR 695), and the Agence Nationale de la Recherche (ANR-10-BLANC-1011-3 "SPINHALL"). F.F. and Y.M. acknowledge funding under the HGF-YIG program VH-NG-513. The samples were patterned at the NANOFAB facility of the Institut Néel (CNRS). Author contributions K.G., I.M.M., and P.G. planned the experiment; I.M.M., C.O.A., G.G., and S.A. fabricated the samples. K.G., I.M.M., and C.O.A. performed the measurements; K.G., I.M.M., C.O.A., and P.G. analyzed the data; F.F. derived the general expression for the torques, K.G. and P.G. wrote the manuscript. All authors discussed the results and commented on the manuscript. Author Information The authors declare no competing financial interests. Correspondence and requests for materials should be addressed to K.G. ([email protected]) and P.G. ([email protected]). 16 Figure 1. Torque schematics and magnetization measurements. AlOx/Co/Pt Hall cross with current and voltage leads. The arrows indicate the direction and amplitude of the torques for a, (ϕ = 0º), b, (ϕ = 60º), and c, (ϕ = 90º). The field-like and anisotropic Coordinate system. e, mz measured by the first harmonic Hall resistance 𝑅𝐻𝑓 as a components of T┴ are shown in red and green, respectively. T║ is shown in blue. d, function of applied field Bext parallel to the easy axis (θB = 0º) and f, close to in-plane (θB = 82º, ϕ = 0º). 17 𝑅𝐻2𝑓 measured as a function of Bext applied at θB = 82º, ϕ = 90º and b, θB = 82º, ϕ = 0º. Figure 2. Second harmonic Hall resistance and current-induced spin-orbit fields. a, The amplitude of the ac current is 1.136 mA. c, Effective field B⊥ /cosθ measured at ϕ = 90º as a function of Bext. d, Effective field B∥ measured at ϕ = 0º as a function of Bext. e, B⊥ /cosθ measured at ϕ = 90º as a function of sin2θ. The solid line is a fit to 𝑇0⊥ + 𝑇2⊥ sin2 𝜃 according to Eq. (3). f, B∥ measured at ϕ = 0º as a function of θ. The solid line is a fit to 𝑇0∥ + 𝑇2∥ sin2 𝜃 + 𝑇4∥ sin4 𝜃 according to Eq. (4). Note that, in our units, T⊥ = B⊥/ cos 𝜃 for ϕ = 90º and �T∥ � = B ∥ for ϕ = 0º. 18 Figure 3. Angular dependence of the Hall resistance and SOT coefficients. a, 𝑅𝐻2𝑓 as magnetization. b, Symmetric 𝑅𝐻2𝑓 (B⊥ ) and c, antisymmetric 𝑅𝐻2𝑓 (B∥) components of a function of Bext applied at θB = 82º measured for different in-plane orientations of the 𝑅𝐻2𝑓 . SOT coefficients d, 𝑇0⊥ , e, 𝑇2⊥ , f, 𝑇0∥ as a function of ϕ. The amplitude of the ac current is 1.136 mA. 19 𝑇0⊥ , b, 𝑇2⊥ , c, 𝑇0∥ as a function of j for different samples. Red dots and black squares Figure 4. Dependence of the SOT coefficients on the injected current density. a, refer to symmetric Hall crosses with 1x1 and 0.5x0.5 µm2 dimensions, respectively. The blue triangles refer to an asymmetric Hall cross with a 1 µm wide current line and 0.5 µm wide voltage probes. 20 Figure 5. Effect of thermal annealing and material composition on the current- induced spin-orbit fields. a, Effective field B⊥ /cosθ measured at ϕ = 90º and b, effective field B∥ measured at ϕ = 0º in AlOx/Co/Pt as a function of θ. The measurements refer to a 0.5x0.5 µm2 as deposited sample (black dots) annealed to 300ºC (green triangles). The amplitude of the ac current is 540 µA and 550 µA for the as deposited and annealed samples, respectively. c, B⊥ /cosθ measured at ϕ = 90º and d, B∥ measured at ϕ = 0º for an MgO/CoFeB/Ta Hall bar annealed to 300ºC. The width of the Hall bar is 1 µm and the amplitude of the ac current is 500 µA. The scale on the right hand side of the plots is in mT per 107 A/cm2. 1 2 3 4 5 6 7 8 9 10 11 12 21 Chappert, C., Fert, A. & Van Dau, F. N. The emergence of spin electronics in data storage. Nat. Mater. 6, 813-823, (2007). Brataas, A., Kent, A. D. & Ohno, H. Current-induced torques in magnetic materials. Nat. Mater. 11, 372-381, (2012). Ralph, D. C. & Stiles, M. D. Spin transfer torques. J. Magn. Magn. Mater. 320, 1190-1216, (2008). Chernyshov, A., Overby, M., Liu, X. Y., Furdyna, J. K., Lyanda-Geller, Y. & Rokhinson, L. P. Evidence for reversible control of magnetization in a ferromagnetic material by means of spin-orbit magnetic field. Nat. Phys. 5, 656- 659, (2009). Miron, I. M., Gaudin, G., Auffret, S., Rodmacq, B., Schuhl, A., Pizzini, S., Vogel, J. & Gambardella, P. Current-driven spin torque induced by the Rashba effect in a ferromagnetic metal layer. Nat. Mater. 9, 230-234, (2010). Pi, U. H., Won Kim, K., Bae, J. Y., Lee, S. C., Cho, Y. J., Kim, K. S. & Seo, S. Tilting of the spin orientation induced by Rashba effect in ferromagnetic metal layer. Appl. Phys. Lett. 97, 162507, (2010). Fang, D., Kurebayashi, H., Wunderlich, J., Vyborny, K., Zarbo, L. P., Campion, R. P., Casiraghi, A., Gallagher, B. L., Jungwirth, T. & Ferguson, A. J. Spin- orbit-driven ferromagnetic resonance. Nat Nanotechnol 6, 413-417, (2011). Suzuki, T., Fukami, S., Ishiwata, N., Yamanouchi, M., Ikeda, S., Kasai, N. & Ohno, H. Current-induced effective field in perpendicularly magnetized Ta/CoFeB/MgO wire. Appl. Phys. Lett. 98, 142505, (2011). Miron, I. M., Garello, K., Gaudin, G., Zermatten, P. J., Costache, M. V., Auffret, S., Bandiera, S., Rodmacq, B., Schuhl, A. & Gambardella, P. Perpendicular switching of a single ferromagnetic layer induced by in-plane current injection. Nature 476, 189-193, (2011). Kajiwara, Y., Harii, K., Takahashi, S., Ohe, J., Uchida, K., Mizuguchi, M., Umezawa, H., Kawai, H., Ando, K., Takanashi, K., Maekawa, S. & Saitoh, E. Transmission of electrical signals by spin-wave interconversion in a magnetic insulator. Nature 464, 262-266, (2010). Kurebayashi, H., Dzyapko, O., Demidov, V. E., Fang, D., Ferguson, A. J. & Demokritov, S. O. Controlled enhancement of spin-current emission by three- magnon splitting. Nat. Mater. 10, 660-664, (2011). Liu, L., Moriyama, T., Ralph, D. C. & Buhrman, R. A. Spin-Torque Ferromagnetic Resonance Induced by the Spin Hall Effect. Phys. Rev. Lett. 106, 036601 (2011). 22 13 Demidov, V. E., Urazhdin, S., Ulrichs, H., Tiberkevich, V., Slavin, A., Baither, D., Schmitz, G. & Demokritov, S. O. Magnetic nano-oscillator driven by pure spin current. Nat. Mater. 11, 1028-1031, (2012). 14 Manchon, A. & Zhang, S. Theory of nonequilibrium intrinsic spin torque in a single nanomagnet. Phys. Rev. B 78, 212405 (2008). 15 Obata, K. & Tatara, G. Current-induced domain wall motion in Rashba spin- orbit system. Phys. Rev. B 77, 214429 (2008). 16 Manchon, A. & Zhang, S. Theory of spin torque due to spin-orbit coupling. Phys. Rev. B 79, 094422 (2009). 17 Garate, I. & MacDonald, A. Influence of a transport current on magnetic anisotropy in gyrotropic ferromagnets. Phys. Rev. B 80, 134403 (2009). 18 Matos-Abiague, A. & Rodríguez-Suárez, R. Spin-orbit coupling mediated spin torque in a single ferromagnetic layer. Phys. Rev. B 80, 094424 (2009). 19 20 Haney, P. M. & Stiles, M. D. Current-Induced Torques in the Presence of Spin- Orbit Coupling. Phys. Rev. Lett. 105, 126602 (2010). Vedyayev, A., Strelkov, N., Chshiev, M., Ryzhanova, N. & Dieny, B. Spin Transfer Torques induced by Spin Hall Effect. http://arxiv.org/abs/1108.2589v1, (2011). 21 Wang, X. & Manchon, A. Rashba spin torque in an ultrathin ferromagnetic metal layer. http://arxiv.org/abs/1111.5466 (2011). 22 Wang, X. & Manchon, A. Diffusive spin dynamics in ferromagnetic thin films with a Rashba interaction. Phys Rev Lett 108, 117201, (2012). 23 Manchon, A. Spin Hall effect versus Rashba torque: a Diffusive Approach. http://arxiv.org/abs/1204.4869, (2012). 24 25 26 27 Pesin, D. A. & MacDonald, A. H. Quantum kinetic theory of current-induced torques in Rashba ferromagnets. Phys. Rev. B 86, 014416 (2012). Haney, P. M., Lee, H.-W., Lee, K.-J., Manchon, A. & Stiles, M. D. Current induced torques and interfacial spin-orbit coupling: Semiclassical Modeling. http://arxiv.org/abs/1301.4513v1, (2013). Kim, K.-W., Seo, S.-M., Ryu, J., Lee, K.-J. & Lee, H.-W. Magnetization dynamics induced by in-plane currents in ultrathin magnetic nanostructures with Rashba spin-orbit coupling. Phys. Rev. B 85, 180404 (2012). van der Bijl, E. & Duine, R. A. Current-induced torques in textured Rashba ferromagnets. Phys. Rev. B 86, 094406 (2012). 28 29 30 31 32 33 34 35 36 37 38 23 Dyakonov, M. I. & Perel, V. I. Possibility of Orienting Electron Spins with Current. Jetp Letters-Ussr 13, 467-&, (1971). Bychkov, Y. A. & Rashba, E. I. Properties of a 2d Electron-Gas with Lifted Spectral Degeneracy. JETP Lett. 39, 78-81, (1984). Dresselhaus, G. Spin-Orbit Coupling Effects in Zinc Blende Structures. Physical Review 100, 580-586, (1955). Zhang, S., Levy, P. & Fert, A. Mechanisms of Spin-Polarized Current-Driven Magnetization Switching. Phys. Rev. Lett. 88, 236601 (2002). Tserkovnyak, Y., Brataas, A. & Bauer, G. E. W. Theory of current-driven magnetization dynamics in inhomogeneous ferromagnets. J. Magn. Magn. Mater. 320, 1282-1292, (2008). Liu, L., Lee, O. J., Gudmundsen, T. J., Ralph, D. C. & Buhrman, R. A. Current- Induced Switching of Perpendicularly Magnetized Magnetic Layers Using Spin Torque from the Spin Hall Effect. Phys. Rev. Lett. 109, 096602 (2012). Avci, C. O., Garello, K., Miron, I. M., Gaudin, G., Auffret, S., Boulle, O. & Gambardella, P. Magnetization switching of an MgO/Co/Pt layer by in-plane current injection. Appl. Phys. Lett. 100, 212404 (2012). Liu, L., Pai, C. F., Li, Y., Tseng, H. W., Ralph, D. C. & Buhrman, R. A. Spin- torque switching with the giant spin Hall effect of tantalum. Science 336, 555- 558, (2012). Gaudin, G., Miron, I. M., Gambardella, P. & Schuhl, A. Magnetic memory element. WO 2012/014131. Gaudin, G., Miron, I. M., Gambardella, P. & Schuhl, A. Writable magnetic element. WO 2012/014132. Kim, J., Sinha, J., Hayashi, M., Yamanouchi, M., Fukami, S., Suzuki, T., Mitani, S. & Ohno, H. Layer thickness dependence of the current-induced effective field vector in TaCoFeBMgO. Nat Mater 12, 240-245, (2013). 39 Miron, I., Zermatten, P. J., Gaudin, G., Auffret, S., Rodmacq, B. & Schuhl, A. Domain Wall Spin Torquemeter. Phys. Rev. Lett. 102, 137202 (2009). 40 41 Ibrahim, I. S., Schweigert, V. A. & Peeters, F. M. Diffusive transport in a Hall junction with a microinhomogeneous magnetic field. Phys. Rev. B 57, 15416- 15427, (1998). Liu, L., Buhrman, R. A. & Ralph, D. C. Review and Analysis of Measurements of the Spin Hall Effect in Platinum. arXiv:1111.3702v3, (2011). 24 42 43 Niimi, Y., Wei, D., Idzuchi, H., Wakamura, T., Kato, T. & Otani, Y. Experimental Verification of Comparability between Spin-Orbit and Spin- Diffusion Lengths. Phys. Rev. Lett. 110, 016805 (2013). Rodmacq, B., Manchon, A., Ducruet, C., Auffret, S. & Dieny, B. Influence of thermal annealing on the perpendicular magnetic anisotropy of Pt/Co/AlOx trilayers. Phys. Rev. B 79, 024423 (2009). 44 Wang, Y., Wang, W. X., Wei, H. X., Zhang, B. S., Zhan, W. S. & Han, X. F. Effect of annealing on the magnetic tunnel junction with Co/Pt perpendicular anisotropy ferromagnetic multilayers. J. Appl. Phys. 107, 09c711 (2010). Zimmler, M., Özyilmaz, B., Chen, W., Kent, A., Sun, J., Rooks, M. & Koch, R. Current-induced effective magnetic fields in Co∕Cu∕Co nanopillars. Phys. Rev. B 70, 184438 (2004). 45 SUPPLEMENTARY INFORMATION Symmetry and magnitude of spin-orbit torques in ferromagnetic heterostructures Kevin Garello1, Ioan Mihai Miron2, Can Onur Avci1, Frank Freimuth3, Yuriy Mokrousov3, Stefan Blügel3, Stéphane Auffret2, Olivier Boulle2, Gilles Gaudin2, and Pietro Gambardella1,4,5 1 Catalan Institute of Nanotechnology (ICN), E-08193 Barcelona, Spain 2 SPINTEC, UMR-8191, CEA/CNRS/UJF/GINP, INAC, F-38054 Grenoble, France 3 Peter Grünberg Institut and Institute for Advanced Simulation, Forschungszentrum Jülich and JARA, 52425 Jülich, Germany 4 Institució Catalana de Recerca i Estudis Avançats (ICREA), E-08010 Barcelona, Spain 5 Department of Materials, ETH Zurich, CH-8093 Zurich, Switzerland Table of Contents: S1. General expression for the spin accumulation and SOTs S2. Harmonic analysis of the Hall voltage S3. Separation of AHE and PHE S4. Modulation of the Hall voltage by an external ac field S5. Measurement of current-induced effective fields in the case of nonzero PHE S6. Macrospin simulations S7. Sample-dependent offset and Nernst-Ettingshausen effect S8. Current dispersion in the Hall voltage probes S9. Measurements in the case of nonuniform magnetization S10. Comparison of AC and DC detection methods S11. Dynamic simulations of the my component generated by T∥ S12. Influence of thermal effects S13. Supplementary references 1 S1. General expression for the spin accumulation and SOTs In this section we derive general expressions for the spin accumulation (𝛿𝛿𝒎𝒎), effective fields (𝑩𝑩𝐼𝐼 ), and torques (𝑻𝑻) induced by an electric current in trilayers with structure inversion asymmetry along the stacking direction z. The current is driven by an applied electric field (𝑬𝑬) in the xy plane. We consider only the case of trilayers that exhibit continuous rotational symmetry about the z axis and mirror symmetry for all the planes that are perpendicular to the xy plane, i.e., contain the z axis. The results of this section apply to samples that are either polycrystalline, around the stacking direction it is expected also that 𝛿𝛿𝒎𝒎, 𝑩𝑩𝐼𝐼 , and 𝑻𝑻 vary as the sample is rotated as in our case, or disordered. In fully epitaxial systems that display discrete rotational symmetry while keeping the directions of the electric field and magnetization fixed in space. Consider an applied current that leads to a spin accumulation 𝛿𝛿𝒎𝒎. Explicitly, in the equations below, 𝛿𝛿𝒎𝒎 denotes the induced magnetization associated with the spin accumulation. 𝛿𝛿𝒎𝒎 induces a change of the exchange field (𝛿𝛿𝑩𝑩𝑥𝑥𝑐 ) in the ferromagnet, which acts as an effective magnetic field on the magnetization (𝒎𝒎), given by 𝑩𝑩𝐼𝐼 = 𝛿𝛿𝑩𝑩𝑥𝑥𝑐 = 𝐵𝑥𝑐𝑚 𝛿𝛿𝒎𝒎. The resulting torque is given by 𝑻𝑻 = 𝒎𝒎 × 𝑩𝑩𝐼𝐼 . Since the torque depends only on the component of 𝑩𝑩𝐼𝐼 that is perpendicular to 𝒎𝒎, in the following 𝑩𝑩𝐼𝐼 and 𝛿𝛿𝒎𝒎 will denote the perpendicular components of the effective field and the spin accumulation, respectively. In Fig. S1a and b we consider the case of magnetization in the xz plane and electric field accumulation: A longitudinal one, 𝛿𝛿𝒎𝒎∥ , which lies in the xz plane, and a perpendicular one, in the x direction. We show that symmetry allows for two components for the spin 𝛿𝛿𝒎𝒎⊥ , which points in the y direction. 𝑬𝑬 is invariant under mirror reflection at the xz plane. However, 𝒎𝒎 is inverted, because it is an axial vector. Similarly, the component 𝛿𝛿𝒎𝒎∥ of the axial vector 𝛿𝛿𝒎𝒎 is inverted, but 𝛿𝛿𝒎𝒎⊥ is invariant. Thus, 𝛿𝛿𝒎𝒎∥ has to be an odd function of 𝒎𝒎, while 𝛿𝛿𝒎𝒎⊥ has to be an even function. Mirror reflection at the yz plane followed by a rotation around either 𝛿𝛿𝒎𝒎∥ or 𝛿𝛿𝒎𝒎⊥ , owing to the structure inversion asymmetry. For example, if there was the z axis by 180º leads to the same conclusion. There is no symmetry operation that forbids inversion symmetry, 𝑬𝑬 would change under inversion while 𝒎𝒎 and 𝛿𝛿𝒎𝒎 would remain unchanged. Thus, inversion symmetry would dictate that both 𝑬𝑬 and −𝑬𝑬 lead to the same spin accumulation, meaning that, in such a case, the linear response of the spin accumulation would have to be zero. 2 z z z d E  E a E -δmII c  E  -E× b z -δm⊥ δm⊥ -E mz mz θ θ δmII δm⊥ y y x x -mz -mz Figure S1. Transformation of electric field 𝑬𝑬, magnetization 𝒎𝒎 and spin accumulation 𝛿𝛿𝒎𝒎 δmII leaves 𝑬𝑬 invariant, but inverts 𝒎𝒎 and 𝛿𝛿𝒎𝒎∥ , because the latter two transform like axial vectors. under mirror reflections. a) Magnetization in the xz plane. Mirror reflection at the xz plane b) Same as a, but view on the yz plane. 𝛿𝛿𝒎𝒎⊥ is invariant under reflection at the xz plane. c) Magnetization in the yz plane. Mirror reflection at the yz plane inverts 𝑬𝑬, 𝒎𝒎 and 𝛿𝛿𝒎𝒎⊥ . d) Same as c, but view on the xz plane. The component 𝛿𝛿𝒎𝒎∥ is invariant under reflection at the yz plane. field again along the x direction. The longitudinal spin accumulation, 𝛿𝛿𝒎𝒎∥ , points now in the x In Fig. S1c and d we consider the case of magnetization in the yz plane and the electric direction, while the transverse spin accumulation, 𝛿𝛿𝒎𝒎⊥ , lies in the yz plane. Mirror reflection at the yz plane inverts 𝑬𝑬, 𝒎𝒎, and 𝛿𝛿𝒎𝒎⊥ , while 𝛿𝛿𝒎𝒎∥ is invariant. Within linear response, 𝛿𝛿𝒎𝒎 must change sign upon inversion of 𝑬𝑬. It follows that 𝛿𝛿𝒎𝒎⊥ is again an even function of 𝒎𝒎 because, for inverted electric field and inverted magnetization, 𝛿𝛿𝒎𝒎⊥ is inverted. Likewise, 𝛿𝛿𝒎𝒎∥ is again an odd function of 𝒎𝒎. Mirror reflection at the xz plane followed by a rotation around the z axis by 180º leads to the same conclusion. In the special case of 𝒎𝒎 ∥ 𝑦, 𝛿𝛿𝒎𝒎 vanishes because in this case mirror reflection at the yz plane followed by a rotation around z by 180º leaves 𝒎𝒎 and 𝑬𝑬 invariant, but 𝛿𝛿𝒎𝒎 is inverted. However, if 𝒎𝒎 has nonzero out-of-plane components, there is no symmetry operation present which forbids any of the two components 𝛿𝛿𝒎𝒎∥ or 𝛿𝛿𝒎𝒎⊥ . We will show below that, for 𝒎𝒎 lying in the xz plane, one has 𝛿𝛿𝒎𝒎∥ = 𝐸𝐵𝑥𝑐 �𝒆𝒆𝑦𝑦 × 𝒎𝒎��𝐴𝐴0𝜑𝜑 + 𝐴𝐴2𝜑𝜑 sin2 𝜃𝜃 + 𝐴𝐴4𝜑𝜑 sin4 𝜃𝜃 + ⋯� (1) and 𝛿𝛿𝒎𝒎⊥ = − 𝐸𝐵𝑥𝑐 𝒆𝒆𝑦𝑦 �𝐴𝐴0𝜃𝜃 + 𝐴𝐴2𝜃𝜃 sin2 𝜃𝜃 + 𝐴𝐴4𝜃𝜃 sin4 𝜃𝜃 + ⋯� For 𝒎𝒎 lying in the yz plane one has instead: 𝛿𝛿𝒎𝒎∥ = 𝐸𝐵𝑥𝑐 �𝒆𝒆𝑦𝑦 × 𝒎𝒎��𝐴𝐴0𝜑𝜑 + 𝐵𝐵2𝜃𝜃 sin2 𝜃𝜃 + 𝐵𝐵4𝜃𝜃 sin4 𝜃𝜃 + ⋯� and 𝛿𝛿𝒎𝒎⊥ = − 𝐸𝐵𝑥𝑐 �𝒆𝒆𝑦𝑦 × 𝒎𝒎� × 𝒎𝒎 �𝐴𝐴0𝜃𝜃 − 𝐵𝐵2𝜑𝜑 sin2 𝜃𝜃 − 𝐵𝐵4𝜑𝜑 sin4 𝜃𝜃 − ⋯�. 3 (2) (3) (4) The coefficients 𝐴𝐴2𝜑𝜑 , 𝐴𝐴4𝜑𝜑 , 𝐴𝐴2𝜃𝜃 , 𝐴𝐴4𝜃𝜃 and 𝐵𝐵2𝜑𝜑 , 𝐵𝐵4𝜑𝜑 , 𝐵𝐵2𝜃𝜃 , 𝐵𝐵4𝜃𝜃 describe what we refer to as anisotropy of the SOT. In the absence of anisotropy, one single parameter, 𝐴𝐴0𝜑𝜑 , governs the longitudinal accumulation 𝛿𝛿𝒎𝒎∥ , whereas one single parameter, 𝐴𝐴0𝜃𝜃 , governs the transverse accumulation 𝛿𝛿𝒎𝒎⊥ . To describe the anisotropy one needs four parameters for each order of the expansion, where two 𝐴𝐴 parameters describe the anisotropy of the two spin-accumulation components for the case of magnetization in the xz plane and two 𝐵𝐵 parameters describe the anisotropy for the case of magnetization in the yz plane. Since the trilayers considered in this work exhibit 𝜑𝜑 of the magnetization. continuous rotational symmetry around the z axis, no additional anisotropy arises from the angle Whereas symmetry arguments provide general expressions of 𝛿𝛿𝒎𝒎∥ and 𝛿𝛿𝒎𝒎⊥ , as will be shown below, the same arguments do not provide information about the magnitude of these two be important, and their origin is a matter of debate. As the angle 𝜃𝜃 in Fig. S1 is varied, the terms and the underlying mechanisms. Experiment and theory suggest that both components can magnitudes of 𝛿𝛿𝒎𝒎∥ and 𝛿𝛿𝒎𝒎⊥ are generally expected to change. The resulting anisotropy in the spin accumulation can arise from spin-dependent interface resistances that influence the spin current from the spin Hall effect as it traverses the interface, or from anisotropic relaxation and thus also the Fermi surface change as a function of 𝜃𝜃 due to spin-orbit coupling. This can be times, as pointed out in Ref. 1. From the electronic structure point of view, the band energies understood, e.g., in an sd model, where only the s electrons are subject to Rashba type spin orbit interaction: In the absence of hybridization the spin of the s-bands is determined by the Rashba Due to hybridization, the coupled sd model exhibits a band structure which depends on 𝜃𝜃 . The interaction, while the spin of the d-bands is determined by the direction of the exchange field. symmetries of the trilayers restrict the allowed 𝜃𝜃 dependence. In order to obtain an expression for the spin accumulation valid for any orientation of 𝒎𝒎, 𝒎𝒎 and the z axis (Eqs. 1 and 2) and one perpendicular to it (Eqs. 3 and 4). Since we are one can decompose the electric field into two components, one parallel to the plane spanned by considering the linear response of the spin accumulation to an electric field, 𝛿𝛿𝒎𝒎 is given by a expression for the linear response of axial vectors (such as 𝛿𝛿𝒎𝒎, 𝑩𝑩𝐼𝐼 , and 𝑻𝑻) to an applied electric superposition of these two simple cases in general. In the following, we work out a general field which conforms with the symmetry requirements discussed above. The following concrete, we refer to torque 𝑻𝑻 and torkance 𝒕𝒕. Readers who are more familiar with spin derivation holds for all axial vectors perpendicular to the magnetization; however, to be 4 accumulation may consider 𝑻𝑻 as spin-accumulation and 𝒕𝒕 as spin-accumulation per applied electric field. written as 𝑻𝑻(𝜃𝜃 , 𝜑𝜑) = 𝒕𝒕(𝜃𝜃 , 𝜑𝜑)𝐸𝐸 , where 𝒕𝒕(𝜃𝜃 , 𝜑𝜑) is the torkance and (θ ,ϕ) are the polar and If the electric field is along the x axis, the torque acting on the magnetization can be azimuthal coordinates of the unit magnetization vector 𝒎𝒎 = (sin 𝜃𝜃 cos 𝜑𝜑 , sin 𝜃𝜃 sin 𝜑𝜑 , cos 𝜃𝜃). basis vectors of the spherical coordinate system 𝒆𝒆𝜃𝜃 = �cos 𝜃𝜃 cos 𝜑𝜑 , cos 𝜃𝜃 sin 𝜑𝜑 , – sin 𝜃𝜃� and Since the torkance is perpendicular to the magnetization, it may be expressed in terms of the 𝒆𝒆𝜑𝜑 = (− sin 𝜑𝜑 , cos 𝜑𝜑 , 0) with components 𝑡𝑡𝜃𝜃 (𝜃𝜃 , 𝜑𝜑) and 𝑡𝑡𝜑𝜑 (𝜃𝜃 , 𝜑𝜑): 𝒕𝒕(𝜃𝜃 , 𝜑𝜑) = 𝑡𝑡𝜃𝜃 (𝜃𝜃 , 𝜑𝜑)𝒆𝒆𝜃𝜃 + 𝑡𝑡𝜑𝜑 (𝜃𝜃 , 𝜑𝜑) 𝒆𝒆𝜑𝜑 = ∑ 𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑)𝒆𝒆𝑠𝑠 𝑠𝑠=𝜃𝜃 ,𝜑𝜑 (5) . It is assumed that the magnetic layer exhibits continuous rotational symmetry around the z axis and mirror symmetry with respect to planes perpendicular to the layer plane. We consider first (Fig. S2a), we have 𝑬𝑬 = (cos 𝛾𝛾 , sin 𝛾𝛾 , 0)𝐸𝐸 and the torque can be rewritten as 𝑻𝑻(𝜃𝜃 , 𝜑𝜑) = the consequences of rotational symmetry. If the electric field makes an angle γ with the x axis ∑ 𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑 − 𝛾𝛾)𝐸𝐸𝒆𝒆𝑠𝑠 𝑠𝑠=𝜃𝜃 ,𝜑𝜑 . Since the torque is linear in the electric field, we can decompose 𝑻𝑻(𝜃𝜃 , 𝜑𝜑) into one component due to the electric field 𝐸𝐸 cos 𝛾𝛾 along the x direction plus a second 𝐸𝐸 sin 𝛾𝛾 𝑻𝑻(𝜃𝜃 , 𝜑𝜑) = ∑ ∑ 𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑)𝐸𝐸 cos 𝛾𝛾 𝒆𝒆𝑠𝑠 + 𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑 − 𝜋𝜋/2)𝐸𝐸 sin 𝛾𝛾 𝒆𝒆𝑠𝑠 . The direction: y component due to the electric field along the 𝑠𝑠=𝜃𝜃 ,𝜑𝜑 𝑠𝑠=𝜃𝜃 ,𝜑𝜑 rotation axis thus leads to the condition 𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑 − 𝛾𝛾) = 𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑) cos 𝛾𝛾 + 𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑 − 𝜋𝜋/2) sin 𝛾𝛾, continuous which restricts the allowed ϕ-dependence such that 𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑) = 𝐹𝐹1𝑠𝑠 (𝜃𝜃) cos 𝜑𝜑 + 𝐹𝐹2𝑠𝑠 (𝜃𝜃) sin 𝜑𝜑, (6) where 𝐹𝐹1𝑠𝑠 and 𝐹𝐹2𝑠𝑠 are two functions of θ. As the magnetization and torque are axial vectors, the x and z components change sign S2b). Mirror reflection symmetry thus dictates that 𝑡𝑡𝑥𝑥 (𝜋𝜋 − 𝜃𝜃 , 𝜋𝜋 − 𝜑𝜑) = −𝑡𝑡𝑥𝑥 (𝜃𝜃 , 𝜑𝜑), 𝑡𝑡𝑦𝑦 (𝜋𝜋 − under mirror reflection with respect to the xz plane, while the y component is conserved (Fig. 𝜃𝜃 , 𝜋𝜋 − 𝜑𝜑) = 𝑡𝑡𝑦𝑦 (𝜃𝜃 , 𝜑𝜑) and 𝑡𝑡𝑧𝑧 (𝜋𝜋 − 𝜃𝜃 , 𝜋𝜋 − 𝜑𝜑) = −𝑡𝑡𝑧𝑧 (𝜃𝜃 , 𝜑𝜑), which is equivalent to 𝑡𝑡𝑠𝑠 (𝜋𝜋 − 𝜃𝜃 , 𝜋𝜋 − 𝜑𝜑) = −𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑) . (7) Mirror reflection with respect to the yz plane changes the sign of the y and z components of the S2c), leading to the conditions 𝑡𝑡𝑥𝑥 (𝜋𝜋 − 𝜃𝜃 , −𝜑𝜑) = −𝑡𝑡𝑥𝑥 (𝜃𝜃 , 𝜑𝜑), 𝑡𝑡𝑦𝑦 (𝜋𝜋 − 𝜃𝜃 , −𝜑𝜑) = 𝑡𝑡𝑦𝑦 (𝜃𝜃 , 𝜑𝜑) and magnetization and torkance as well as the sign of the electric field along the x direction (Fig. 𝑡𝑡𝑧𝑧 (𝜋𝜋 − 𝜃𝜃 , −𝜑𝜑) = 𝑡𝑡𝑧𝑧 (𝜃𝜃 , 𝜑𝜑), or, equivalently, to 5 𝑡𝑡𝑠𝑠 (𝜋𝜋 − 𝜃𝜃 , −𝜑𝜑) = 𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑) . (8) Since the angles (−𝜃𝜃 , 𝜑𝜑) and (𝜃𝜃 , 𝜋𝜋 + 𝜑𝜑) are equivalent, an additional condition is given by 𝑡𝑡𝑠𝑠 (−𝜃𝜃 , 𝜑𝜑) = −𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜋𝜋 + 𝜑𝜑) , (9) side in 𝒆𝒆𝑠𝑠 (−𝜃𝜃 , 𝜑𝜑) = −𝒆𝒆𝑠𝑠 (𝜃𝜃 , 𝜋𝜋 + 𝜑𝜑). where the minus sign on the right hand side compensates for the minus sign on the right hand We expand the functions 𝐹𝐹𝑗𝑗𝑠𝑠 in Eq. 6 as a Fourier series, 𝐹𝐹𝑗𝑗𝑠𝑠 (𝜃𝜃) = 𝐴𝐴𝑗𝑗 ,0𝑠𝑠 + ∑ �𝐴𝐴𝑗𝑗 ,𝑛𝑛𝑠𝑠 cos 𝑛𝑛𝜃𝜃 + 𝐵𝐵𝑗𝑗 ,𝑛𝑛𝑠𝑠 sin 𝑛𝑛𝜃𝜃� ∞𝑛𝑛=1 (10) , and note that Eq. 9 is satisfied if 𝐹𝐹𝑗𝑗𝑠𝑠 (−𝜃𝜃) = 𝐹𝐹𝑗𝑗𝑠𝑠 (𝜃𝜃), and Eq. 7 as well as Eq. 8 are satisfied if 𝐹𝐹1𝑠𝑠 (𝜋𝜋 − 𝜃𝜃) = 𝐹𝐹1𝑠𝑠 (𝜃𝜃) and 𝐹𝐹2𝑠𝑠 (𝜋𝜋 − 𝜃𝜃) = −𝐹𝐹2𝑠𝑠 (𝜃𝜃). The condition 𝐹𝐹𝑗𝑗𝑠𝑠 (−𝜃𝜃) = 𝐹𝐹𝑗𝑗𝑠𝑠 (𝜃𝜃) rules out terms proportional to sin 𝑛𝑛𝜃𝜃 in the Fourier expansion of 𝐹𝐹𝑗𝑗𝑠𝑠 (𝜃𝜃). Additionally, cos 𝑛𝑛𝜃𝜃 terms with odd integers n are ruled out in the expansion of 𝐹𝐹1𝑠𝑠 by the condition 𝐹𝐹1𝑠𝑠 (𝜋𝜋 − 𝜃𝜃) = 𝐹𝐹1𝑠𝑠 (𝜃𝜃). For even n, cos 𝑛𝑛𝜃𝜃 can be written as a sum of products of even powers of sin 𝜃𝜃 and cos 𝜃𝜃 due to the relation cos 𝑛𝑛𝜃𝜃 = cos𝑛𝑛 𝜃𝜃 − 𝑛𝑛(𝑛𝑛−1)2 sin2 𝜃𝜃 cos𝑛𝑛−2 𝜃𝜃 + ⋯. Because cos2 𝜃𝜃 = 1 − sin2 𝜃𝜃 , even powers of cos 𝜃𝜃 can be expressed as sums of even powers of sin 𝜃𝜃 . Consequently, 𝐹𝐹1𝑠𝑠 (𝜃𝜃) is a sum of even powers of sin 𝜃𝜃 . The condition 𝐹𝐹2𝑠𝑠 (𝜋𝜋 − 𝜃𝜃) = −𝐹𝐹2𝑠𝑠 (𝜃𝜃) rules out cos 𝑛𝑛𝜃𝜃 with even n. Due to the relations discussed above, it is clear that cos 𝑛𝑛𝜃𝜃 with odd n can be written as cos 𝜃𝜃 times a sum of even powers of sin 𝜃𝜃 . Hence, we arrive at the expansion Figure S2. Symmetry properties of the torque. a, Angle definitions used in the text. b, Effect of mirror reflection at the xz plane on the x and z components of T and m. Since the electric field E is invariant, tx and tz undergo the same sign change as Tx and Tz. c, Effect of mirror reflection at the yz plane on the y and z components of T and m. Since E (along x) changes sign, ty and tz are invariant. 6 and 𝑡𝑡𝑠𝑠 (𝜃𝜃 , 𝜑𝜑) = cos 𝜑𝜑 (𝐴𝐴0𝑠𝑠 + 𝐴𝐴2𝑠𝑠 sin2 𝜃𝜃 + 𝐴𝐴4𝑠𝑠 sin4 𝜃𝜃 + ⋯) + cos 𝜃𝜃 sin 𝜑𝜑 (𝐵𝐵0𝑠𝑠 + 𝐵𝐵2𝑠𝑠 sin2 𝜃𝜃 + 𝐵𝐵4𝑠𝑠 sin4 𝜃𝜃 + ⋯) . (11) The torkance 𝒕𝒕(𝜃𝜃 , 𝜑𝜑) is the sum of the even and odd parts 𝒕𝒕∥ (𝜃𝜃 , 𝜑𝜑) = 𝒕𝒕(𝜃𝜃 ,𝜑𝜑)+𝒕𝒕(𝜋𝜋−𝜃𝜃 ,𝜋𝜋+𝜑𝜑) 2 and 𝒕𝒕⊥ (𝜃𝜃 , 𝜑𝜑) = 𝒕𝒕(𝜃𝜃 ,𝜑𝜑)−𝒕𝒕(𝜋𝜋−𝜃𝜃 ,𝜋𝜋+𝜑𝜑) , respectively. Due to 𝒆𝒆𝜑𝜑 (𝜋𝜋 − 𝜃𝜃 , 𝜋𝜋 + 𝜑𝜑) = −𝒆𝒆𝜑𝜑 (𝜃𝜃 , 𝜑𝜑) and 𝒆𝒆𝜃𝜃 (𝜋𝜋 − 2 𝜃𝜃 , 𝜋𝜋 + 𝜑𝜑) = 𝒆𝒆𝜃𝜃 (𝜃𝜃 , 𝜑𝜑), the expansions of the even and odd parts of the torque are given by 𝒕𝒕∥ (𝜃𝜃 , 𝜑𝜑) = cos 𝜑𝜑 �𝐴𝐴0𝜑𝜑 + 𝐴𝐴2𝜑𝜑 sin2 𝜃𝜃 + 𝐴𝐴4𝜑𝜑 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜑𝜑 + cos 𝜃𝜃 sin 𝜑𝜑 �𝐵𝐵0𝜃𝜃 + 𝐵𝐵2𝜃𝜃 sin2 𝜃𝜃 + 𝐵𝐵4𝜃𝜃 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜃𝜃 (12) 𝒕𝒕⊥ (𝜃𝜃 , 𝜑𝜑) = cos 𝜑𝜑 �𝐴𝐴0𝜃𝜃 + 𝐴𝐴2𝜃𝜃 sin2 𝜃𝜃 + 𝐴𝐴4𝜃𝜃 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜃𝜃 + cos 𝜃𝜃 sin 𝜑𝜑 �𝐵𝐵0𝜑𝜑 + 𝐵𝐵2𝜑𝜑 sin2 𝜃𝜃 + 𝐵𝐵4𝜑𝜑 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜑𝜑 (13) cos 𝜑𝜑 𝒆𝒆𝜃𝜃 (0, 𝜑𝜑) − sin 𝜑𝜑 𝒆𝒆𝜑𝜑 (0, 𝜑𝜑) = 𝒆𝒆𝑥𝑥 and sin 𝜑𝜑 𝒆𝒆𝜃𝜃 (0, 𝜑𝜑) + cos 𝜑𝜑 𝒆𝒆𝜑𝜑 (0, 𝜑𝜑) = 𝒆𝒆𝑦𝑦 , An additional requirement is that the torkance shall be independent of ϕ at θ=0. Since achieved by imposing 𝐴𝐴0𝜑𝜑 = 𝐵𝐵0𝜃𝜃 and 𝐴𝐴0𝜃𝜃 = −𝐵𝐵0𝜑𝜑 , which leads to this is 𝒕𝒕∥ (𝜃𝜃 , 𝜑𝜑) = cos 𝜑𝜑 �𝐴𝐴0𝜑𝜑 + 𝐴𝐴2𝜑𝜑 sin2 𝜃𝜃 + 𝐴𝐴4𝜑𝜑 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜑𝜑 + cos 𝜃𝜃 sin 𝜑𝜑 �𝐴𝐴0𝜑𝜑 + 𝐵𝐵2𝜃𝜃 sin2 𝜃𝜃 + 𝐵𝐵4𝜃𝜃 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜃𝜃 (14) 𝒕𝒕⊥ (𝜃𝜃 , 𝜑𝜑) = cos 𝜑𝜑 �𝐴𝐴0𝜃𝜃 + 𝐴𝐴2𝜃𝜃 sin2 𝜃𝜃 + 𝐴𝐴4𝜃𝜃 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜃𝜃 + cos 𝜃𝜃 sin 𝜑𝜑 �−𝐴𝐴0𝜃𝜃 + 𝐵𝐵2𝜑𝜑 sin2 𝜃𝜃 + 𝐵𝐵4𝜑𝜑 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜑𝜑 (15) It is straightforward to verify that Eqs. 14 and 15 lead to the following expression for the torques: 𝑻𝑻∥ = 𝒎𝒎 × [(𝒆𝒆𝑧𝑧 × 𝑬𝑬) × 𝒎𝒎]�𝐴𝐴0𝜑𝜑 + 𝐵𝐵2𝜃𝜃 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 + 𝐵𝐵4𝜃𝜃 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)4 + … � + (𝒎𝒎 × 𝒆𝒆𝑧𝑧 )(𝒎𝒎 ∙ 𝑬𝑬)��𝐵𝐵2𝜃𝜃 − 𝐴𝐴2𝜑𝜑 � + �𝐵𝐵4𝜃𝜃 − 𝐴𝐴4𝜑𝜑 �(𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 + … � (16) 𝑻𝑻⊥ = (𝒆𝒆𝑧𝑧 × 𝑬𝑬) × 𝒎𝒎 �𝐴𝐴0𝜃𝜃 − 𝐵𝐵2𝜑𝜑 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 − 𝐵𝐵4𝜑𝜑 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)4 − … � + 𝒎𝒎 × [(𝒎𝒎 × 𝒆𝒆𝑧𝑧 )(𝒎𝒎 ∙ 𝑬𝑬)]��𝐴𝐴2𝜃𝜃 + 𝐵𝐵2𝜑𝜑 � + �𝐴𝐴4𝜃𝜃 + 𝐵𝐵4𝜑𝜑 �(𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 + … �. (17) 7 and and and Notably, 𝑻𝑻∥ and 𝑻𝑻⊥ depend explicitly on the three vectors 𝑬𝑬, 𝒎𝒎, and 𝒆𝒆𝑧𝑧 , where the presence of 𝒆𝒆𝑧𝑧 reflects the absence of z reflection symmetry. By choosing 𝑬𝑬 (the electric current) to point in the x direction and 𝑬𝑬 = 1 for simplicity, Eqs. 16 and 17 give 𝑻𝑻∥ = 𝒎𝒎 × �𝒆𝒆𝑦𝑦 × 𝒎𝒎��𝐴𝐴0𝜑𝜑 + 𝐵𝐵2𝜃𝜃 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 + 𝐵𝐵4𝜃𝜃 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)4 + … � + (𝒎𝒎 × 𝒆𝒆𝑧𝑧 )(𝒎𝒎 ∙ 𝒆𝒆𝑥𝑥 )��𝐵𝐵2𝜃𝜃 − 𝐴𝐴2𝜑𝜑 � + �𝐵𝐵4𝜃𝜃 − 𝐴𝐴4𝜑𝜑 �(𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 + … � (18) 𝑻𝑻⊥ = 𝒆𝒆𝑦𝑦 × 𝒎𝒎 �𝐴𝐴0𝜃𝜃 − 𝐵𝐵2𝜑𝜑 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 − 𝐵𝐵4𝜑𝜑 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)4 − … � + 𝒎𝒎 × [(𝒎𝒎 × 𝒆𝒆𝑧𝑧 )(𝒎𝒎 ∙ 𝒆𝒆𝑥𝑥 )]��𝐴𝐴2𝜃𝜃 + 𝐵𝐵2𝜑𝜑 � + �𝐴𝐴4𝜃𝜃 + 𝐵𝐵4𝜑𝜑 �(𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 + … �. (19) form of the torques. Only the coefficients 𝐴𝐴0𝜑𝜑 ≡ 𝑇𝑇0∥ , 𝐴𝐴2𝜑𝜑 ≡ 𝑇𝑇2∥ , 𝐴𝐴4𝜑𝜑 ≡ 𝑇𝑇4∥ and 𝐴𝐴0𝜃𝜃 ≡ 𝑇𝑇0⊥ , 𝐵𝐵2𝜑𝜑 ≡ The previous two equations have been used in the main part of the manuscript to give a vector −𝑇𝑇2⊥ , 𝐵𝐵4𝜑𝜑 ≡ −𝑇𝑇4⊥ , have been retained there, because the others are below the experimental detection limit. We note that the special case of 𝐴𝐴𝑛𝑛𝜑𝜑 = 𝐵𝐵𝑛𝑛𝜃𝜃 = 0 for all n ≠ 0 leads to 𝑻𝑻∥ = 𝐴𝐴0𝜑𝜑 �cos 𝜑𝜑 𝒆𝒆𝜑𝜑 + cos 𝜃𝜃 sin 𝜑𝜑 𝒆𝒆𝜃𝜃 � = 𝐴𝐴0𝜑𝜑 ��𝒆𝒆𝑦𝑦 ∙ 𝒆𝒆𝜑𝜑 � 𝒆𝒆𝜑𝜑 + �𝒆𝒆𝑦𝑦 ∙ 𝒆𝒆𝜃𝜃 �𝒆𝒆𝜃𝜃 � = 𝐴𝐴0𝜑𝜑 �𝒆𝒆𝑦𝑦 − �𝒆𝒆𝑦𝑦 ∙ 𝒎𝒎�𝒎𝒎� = 𝐴𝐴0𝜑𝜑𝒎𝒎 × �𝒆𝒆𝑦𝑦 × 𝒎𝒎� and, likewise, the special case of 𝐴𝐴𝑛𝑛𝜃𝜃 = 𝐵𝐵𝑛𝑛𝜑𝜑 = 0 for all n ≠ 0 simplifies 𝑻𝑻⊥ to 𝑻𝑻⊥ = 𝐴𝐴0𝜃𝜃 �cos 𝜑𝜑 𝒆𝒆𝜃𝜃 − cos 𝜃𝜃 sin 𝜑𝜑 𝒆𝒆𝜑𝜑 � = 𝐴𝐴0𝜃𝜃 �cos 𝜑𝜑 𝒆𝒆𝜑𝜑 + cos 𝜃𝜃 sin 𝜑𝜑 𝒆𝒆𝜃𝜃 � × 𝒎𝒎 = 𝐴𝐴0𝜃𝜃 �𝒎𝒎 × �𝒆𝒆𝑦𝑦 × 𝒎𝒎�� × 𝒎𝒎 = 𝐴𝐴0𝜃𝜃 �𝒆𝒆𝑦𝑦 × 𝒎𝒎� . (21) These reduced expressions have been obtained theoretically for several models discussed in the literature (see main text). magnetic fields 𝑩𝑩𝐼𝐼 (𝜃𝜃 , 𝜑𝜑) associated with the torques. Since 𝑻𝑻 = 𝒎𝒎 × 𝑩𝑩𝐼𝐼 , by multiplying the For the purpose of comparison with the experiment, it is useful to derive the effective previous equation by 𝒎𝒎 and noting that (𝒎𝒎 × 𝑩𝑩𝐼𝐼 ) × 𝒎𝒎 = 𝑩𝑩𝑰𝑰 (𝒎𝒎 ⋅ 𝒎𝒎) − 𝒎𝒎(𝒎𝒎 ⋅ 𝑩𝑩𝑰𝑰 ) = 𝑩𝑩𝑰𝑰 , one has 𝑩𝑩𝐼𝐼 = 𝑻𝑻 × 𝒎𝒎. The effective fields corresponding to 𝑻𝑻∥ and 𝑻𝑻⊥are 𝑩𝑩∥ = �𝒆𝒆𝑦𝑦 × 𝒎𝒎��𝐴𝐴0𝜑𝜑 + 𝐵𝐵2𝜃𝜃 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 + 𝐵𝐵4𝜃𝜃 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)4 + … � + (𝒎𝒎 × 𝒆𝒆𝑧𝑧 ) × 𝒎𝒎(𝒎𝒎 ∙ 𝒆𝒆𝑥𝑥 )��𝐵𝐵2𝜃𝜃 − 𝐴𝐴2𝜑𝜑 � + �𝐵𝐵4𝜃𝜃 − 𝐴𝐴4𝜑𝜑 �(𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 + … � 8 (22) (20) and (24) and 𝑩𝑩⊥ = �𝒆𝒆𝑦𝑦 × 𝒎𝒎� × 𝒎𝒎 �𝐴𝐴0𝜃𝜃 − 𝐵𝐵2𝜑𝜑 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 − 𝐵𝐵4𝜑𝜑 (𝒆𝒆𝑧𝑧 × 𝒎𝒎)4 − … � + (𝒎𝒎 × 𝒆𝒆𝑧𝑧 )(𝒎𝒎 ∙ 𝒆𝒆𝑥𝑥 )��𝐴𝐴2𝜃𝜃 + 𝐵𝐵2𝜑𝜑 � + �𝐴𝐴4𝜃𝜃 + 𝐵𝐵4𝜑𝜑 �(𝒆𝒆𝑧𝑧 × 𝒎𝒎)2 + … �. (23) These expressions are valid also for the spin accumulation, since 𝛿𝛿𝒎𝒎∥ ∼ 𝑩𝑩∥ and 𝛿𝛿𝒎𝒎⊥ ∼ 𝑩𝑩⊥ . In spherical coordinates, using 𝒆𝒆𝜃𝜃 × 𝒎𝒎 = −𝒆𝒆𝜑𝜑 and 𝒆𝒆𝜑𝜑 × 𝒎𝒎 = 𝒆𝒆𝜃𝜃 , we obtain 𝑩𝑩∥ (𝜃𝜃 , 𝜑𝜑) = cos 𝜑𝜑 �𝐴𝐴0𝜑𝜑 + 𝐴𝐴2𝜑𝜑 sin2 𝜃𝜃 + 𝐴𝐴4𝜑𝜑 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜃𝜃 − cos 𝜃𝜃 sin 𝜑𝜑 �𝐴𝐴0𝜑𝜑 + 𝐵𝐵2𝜃𝜃 sin2 𝜃𝜃 + 𝐵𝐵4𝜃𝜃 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜑𝜑 𝑩𝑩⊥ (𝜃𝜃 , 𝜑𝜑) = −cos 𝜑𝜑 �𝐴𝐴0𝜃𝜃 + 𝐴𝐴2𝜃𝜃 sin2 𝜃𝜃 + 𝐴𝐴4𝜃𝜃 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜑𝜑 + cos 𝜃𝜃 sin 𝜑𝜑 �−𝐴𝐴0𝜃𝜃 + 𝐵𝐵2𝜑𝜑 sin2 𝜃𝜃 + 𝐵𝐵4𝜑𝜑 sin4 𝜃𝜃 + ⋯�𝒆𝒆𝜃𝜃 . (25) The polar and azimuthal components of the total effective field 𝑩𝑩𝐼𝐼 = 𝑩𝑩∥ (𝜃𝜃 , 𝜑𝜑) + 𝑩𝑩⊥ (𝜃𝜃 , 𝜑𝜑) are then given by 𝐵𝐵𝜃𝜃 (𝜃𝜃 , 𝜑𝜑) = cos 𝜑𝜑 �𝐴𝐴0𝜑𝜑 + 𝐴𝐴2𝜑𝜑 sin2 𝜃𝜃 + 𝐴𝐴4𝜑𝜑 sin4 𝜃𝜃 + ⋯� + cos 𝜃𝜃 sin 𝜑𝜑 �−𝐴𝐴0𝜃𝜃 + 𝐵𝐵2𝜑𝜑 sin2 𝜃𝜃 + 𝐵𝐵4𝜑𝜑 sin4 𝜃𝜃 + ⋯� 𝐵𝐵𝜑𝜑 (𝜃𝜃 , 𝜑𝜑) = −cos 𝜑𝜑 �𝐴𝐴0𝜃𝜃 + 𝐴𝐴2𝜃𝜃 sin2 𝜃𝜃 + 𝐴𝐴4𝜃𝜃 sin4 𝜃𝜃 + ⋯� − cos 𝜃𝜃 sin 𝜑𝜑 �𝐴𝐴0𝜑𝜑 + 𝐵𝐵2𝜃𝜃 sin2 𝜃𝜃 + 𝐵𝐵4𝜃𝜃 sin4 𝜃𝜃 + ⋯� . (27) the effective field which are perpendicular to 𝒎𝒎. In general, however, even though effective We conclude this section by noting that Eqs. 22, 23 determine only the components of current also induces fields parallel to 𝒎𝒎. While such a radial component of the current-induced fields along the magnetization direction do not produce any torques, it is expected that the magnetic field (𝑩𝑩𝑟𝑟 ) is not accessible in the present experiment, we provide below its most general expression compatible with the symmetry of the system. By extending the analysis for the torque presented above to the effective field, one can show that 𝑩𝑩𝑟𝑟 (𝜃𝜃 , 𝜑𝜑) = [sin 𝜃𝜃 cos 𝜃𝜃 cos 𝜑𝜑 (𝐴𝐴0𝑟𝑟 + 𝐴𝐴2𝑟𝑟 sin2 𝜃𝜃 + 𝐴𝐴4𝑟𝑟 sin4 𝜃𝜃 + ⋯) + sin 𝜃𝜃 sin 𝜑𝜑(𝐵𝐵0𝑟𝑟 + 𝐵𝐵2𝑟𝑟 sin2 𝜃𝜃 + 𝐵𝐵4𝑟𝑟 sin4 𝜃𝜃 + ⋯)]𝒎𝒎 . (28) (26) and 9 (29) S2. Harmonic analysis of the Hall voltage We use the Hall voltage (VH) to measure the behaviour of the magnetization as a function of external field and current-induced torques. In general, one has 𝑉𝐻 = 𝐼𝑅𝐻 = 𝑅𝐴𝐻𝐸 𝐼 cos 𝜃𝜃 + 𝑅𝑃𝐻𝐸 𝐼 sin2 𝜃𝜃 sin 2𝜑𝜑 , where I is the current and RH the Hall resistance due to the anomalous Hall effect (AHE) and planar Hall effect (PHE). We omit here the ordinary Hall effect, which is negligible in The AHE is proportional to 𝑅𝐴𝐻𝐸 𝐼 cos 𝜃𝜃 and the PHE to 𝑅𝑃𝐻𝐸 𝐼 sin2 𝜃𝜃 sin 2𝜑𝜑 , where 𝑅𝐴𝐻𝐸 and ferromagnetic materials, as well as thermoelectric effects, which will be discussed in Section S7. 𝑅𝑃𝐻𝐸 are the AHE and PHE resistances, θ and ϕ the polar and azimuthal angle of the magnetization, respectively, as defined in Fig.1b of the main text. Due to the presence of the magnetization around its equilbrium position (𝜃𝜃0 , 𝜑𝜑0 ), defined by the anisotropy field Bk and effective fields, the injection of a moderate ac current Iac = Iei2πft induces small oscillations of external field Bext. These oscillations modulate RH giving rise to a time-dependent Hall voltage signal. After simplification of the time-dependent phase factor, the dependence of the Hall 𝑉𝐻 (𝐼) ≈ 𝑉𝐻 (𝜃𝜃0 , 𝜑𝜑0 ) + 𝐼 𝑠𝑠𝑉𝐻𝑠𝑠𝐼𝐼 �𝜃𝜃0 ,𝜑𝜑0 . voltage on the current can be expanded to first order as (30) gives 𝑠𝑠𝑉𝐻𝑠𝑠𝐼𝐼 = 𝑅𝐻𝑓 + 𝑅𝐻2𝑓 (𝐼), where the first and second harmonic Hall resistance components are Straightforward differentiation of Eq. 29, keeping into account that both θ and ϕ depend on I, given by 𝑅𝐻𝑓 = 𝑅𝐴𝐻𝐸 cos 𝜃𝜃0 + 𝑅𝑃𝐻𝐸 sin2 𝜃𝜃0 sin 2𝜑𝜑0 (31) 𝑅𝐻2𝑓 = 𝐼(𝑅𝐴𝐻𝐸 − 2𝑅𝑃𝐻𝐸 cos 𝜃𝜃0 sin 2𝜑𝜑0 ) 𝑠𝑠 cos 𝜃𝜃𝑠𝑠𝐼𝐼 �𝜑𝜑0 . (32) �𝜃𝜃0 + 𝐼𝑅𝑃𝐻𝐸 sin2 𝜃𝜃0 𝑠𝑠 sin 2𝜑𝜑𝑠𝑠𝐼𝐼 and We notice that 𝑅𝐻𝑓 is equivalent to the Hall resistance measured in conventional dc measurements, whereas 𝑅𝐻2𝑓 contains two terms that depend explicitly on the current. This dependence can be expressed in terms of the current-induced effective field 𝑩𝑩𝐼𝐼 = 𝑩𝑩∥ + 𝑩𝑩⊥ + 𝑩𝑩𝑂𝑂𝑂𝑂𝑟𝑟𝑠𝑠𝑠𝑠𝑂𝑂𝑠𝑠 by noting that 10 𝑑 cos 𝜃𝜃𝑑𝐼 = 𝑑 cos 𝜃𝜃𝑑𝑩𝑩𝐼𝐼 ∙ 𝑑𝑩𝑩𝐼𝐼𝑑𝐼 = 𝑑 cos 𝜃𝜃 𝑏𝜃𝜃 (33) 𝑑𝐵𝐵𝜃𝜃𝐼𝐼 and 𝑑 sin 2𝜑𝜑𝑑𝐼 ∙ 𝑑𝑩𝑩𝐼𝐼𝑑𝐼 = 𝑑 sin 2𝜑𝜑𝑑𝐵𝐵𝜑𝜑𝐼𝐼 = 𝑑 sin 2𝜑𝜑𝑑𝑩𝑩𝐼𝐼 𝑏𝜑𝜑 , (34) where 𝐵𝐵𝜃𝜃𝐼𝐼 and 𝐵𝐵𝜑𝜑𝐼𝐼 indicate the polar and azimuthal components of BI and bθ and bϕ their derivative with respect to the current. The radial component of BI cannot affect the motion of the magnetization and is thus irrelevant to the discussion of the torques. To measure quantitatively cos 𝜃𝜃 and sin 2𝜑𝜑 that appear in Eqs. 33 and 34. As the magnetic field dependence of cos 𝜃𝜃 and the effective fields bθ and bϕ by means of Eq. 32 we need first to calculate the derivatives of sin 2𝜑𝜑 (proportional to mz and mxmy, respectively) is independent on the nature of the field, we can replace BI by Bext in the derivatives and obtain 𝑑 cos 𝜃𝜃𝑑𝐼 = 𝑑 cos 𝜃𝜃 1 𝑑𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 sin(𝜃𝜃𝐵 − 𝜃𝜃0 ) 𝑏𝜃𝜃 , (35) 𝑏𝜑𝜑 = 2 cos 2𝜑𝜑 𝑑𝜑𝜑𝑑𝐵𝐵𝜑𝜑𝐼𝐼 𝑏𝜑𝜑 ≈ 2 cos 2𝜑𝜑 = 𝑑 sin 2𝜑𝜑𝑑𝐵𝐵𝜑𝜑𝐼𝐼 𝑑 sin 2𝜑𝜑𝑑𝐼 𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 sin 𝜃𝜃𝐵 𝑏𝜑𝜑 , (36) case of uniaxial magnetic anisotropy. Using these expressions, 𝑅𝐻2𝑓 can be written as where θB is the polar angle defining the direction of Bext. Note that the last relation is exact in the 𝑅𝐻2𝑓 = 𝐼(𝑅𝐴𝐻𝐸 − 2𝑅𝑃𝐻𝐸 cos 𝜃𝜃0 sin 2𝜑𝜑0 ) 𝑑 cos 𝜃𝜃 1 𝑑𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 �𝜃𝜃0 sin(𝜃𝜃𝐵 − 𝜃𝜃0 ) 𝑏𝜃𝜃 + 𝐼𝑅𝑃𝐻𝐸 sin2 𝜃𝜃0 2 cos 2𝜑𝜑0 𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 sin 𝜃𝜃𝐵 𝑏𝜑𝜑 . (37) Equation 34 is valid for any arbitrary field BI, Bext, orientation of m, and uniaxial magnetic anisotropy, which makes it very useful in the investigation of SOTs. To make correct use of Eq. 37, it is desirable to measure 𝑅𝑃𝐻𝐸 in order to separate the PHE S3. Separation of AHE and PHE 𝑅𝐻𝑓 as a function of the external field applied at angles ϕ ≠ 0º, 90º. Figure S3a shows 𝑅𝐻𝑓 and AHE contributions to the total Hall resistance. This can be readily achieved by measuring 11 measured at θB = 80°, ϕ = 50° by means of example. Due to the PHE, which is even with respect to magnetization reversal, the endpoints of the hysteresis loop are asymmetric. As the AHE is PHE, 𝑅𝑓𝐴𝐻𝐸 and 𝑅𝑓𝑃𝐻𝐸 , can be separated by antisymmetrization and symmetrization of 𝑅𝐻𝑓 , odd with respect to magnetization reversal, the first harmonic contributions of the AHE and respectively. This is achieved in practice by inverting 𝑅𝐻𝑓 (−𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 → +𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 ) with respect to the origin of the curve and taking the sum or difference with 𝑅𝐻𝑓 (+𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 → −𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 ), as shown in Fig. S3b and c, respectively. Macrospin simulations of the AHE and PHE further demonstrate that Figure S3. Separation of AHE and PHE. a, First harmonic Hall resistance, 𝑅𝐻𝑓 , measured at θB = 80°, ϕ = 50° and I = 600 mA. b, Antisymmetric AHE signal, 𝑅𝐴𝐻𝐸𝑓 Symmetric PHE signal, 𝑅𝑃𝐻𝐸𝑓 . c, measurements (solid lines) of 𝑅𝑃𝐻𝐸𝑓 at different angles ϕ. e, 𝑅𝑓𝑃𝐻𝐸 as a function of . d, Comparison between macrospin simulations (dots) and sin2 𝜃𝜃 showing the expected linear dependence. The slope of this curve gives the PHE resistance, 𝑅𝑃𝐻𝐸 = 0.11 Ω. 12 saturation resistance is deduced from the linear fit of 𝑅𝑓𝑃𝐻𝐸 vs. sin2 𝜃𝜃𝑀 (Fig. S3e). For this this procedure yields correct quantitative results, as shown in Fig. S3d. Finally, the PHE sample, we obtain 𝑅𝑃𝐻𝐸 = 0.11 Ω. The saturation value of the AHE resistance is 𝑅𝐴𝐻𝐸 = 0.81 Ω. Finally, 𝑅𝑓𝐴𝐻𝐸 is employed to calculate the equilibrium angle of the magnetization (θ0) at each value of the applied field using the relationship 𝜃𝜃0 = acos �𝑅𝑓𝐴𝐻𝐸 (𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 ) � . (38) 𝑅𝐴𝐻𝐸 S4. Measurement of an external ac field using the Hall voltage The measurements of the effective fields based on the harmonic analysis of the Hall voltage described above have been quantitatively checked by applying an ac magnetic field of known amplitude and retrieving its value using Eqs. 37 and 38. To this end, we modulated Bext by a small sinusoidal field of amplitude Bac = 10.5 mT at a frequency of 1 Hz. The current frequency polar component of the ac field acting on the magnetization in this case is given by 𝑏𝜃𝜃 𝐼 + was also set to f = 1 Hz in order to match the modulation frequency of the electromagnet. The 𝐵𝐵𝑎𝑐 sin(𝜃𝜃𝐵 − 𝜃𝜃). Since 𝑏𝜃𝜃 and Bac are independent of each other, 𝑅𝐻2𝑓 is given by the sum of 𝑅𝐻2𝑓 (𝑏𝜃𝜃 ) and 𝑅𝐻2𝑓 (𝐵𝐵𝑎𝑐 ). By applying the ac field in-phase or out-of-phase with the current, we measure 𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 , 𝐵𝐵𝑎𝑐 � = 𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 � + 𝑅𝐻2𝑓 (𝐵𝐵𝑎𝑐 ) and 𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 , −𝐵𝐵𝑎𝑐 � = 𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 � − 𝑅𝐻2𝑓 (𝐵𝐵𝑎𝑐 ), respectively (Fig. S4a). This allows us to separate the two contributions 𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 � and 𝑅𝐻2𝑓 (𝐵𝐵𝑎𝑐 ) by taking the sum and difference of the above relationships. As expected, 𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 � coincides with the measurement of 𝑅𝐻2𝑓 in the absence of ac field (Fig. S4b). To calculate the amplitude Bac from the experimental data, we use Eq. 37, which leads to 2𝑅𝐻2𝑓 (𝐵𝐵𝑎𝑐 ) 𝐵𝐵𝑎𝑐 = − 𝑑𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 (39) (𝑅𝐴𝐻𝐸 − 2𝑅𝑃𝐻𝐸 cos 𝜃𝜃0 sin 2𝜑𝜑0 ) 𝑑 cos 𝜃𝜃 𝑠𝑠𝐵𝑒𝑥𝑡 can be evaluated from the numerical derivative of 1𝑅𝐴𝐻𝐸 𝑠𝑠𝑅𝑓𝐴𝐻𝐸 The term 𝑠𝑠 cos 𝜃𝜃 𝑠𝑠𝐵𝑒𝑥𝑡 or, equivalently, from −2𝑅𝐻2𝑓 (𝐵𝐵𝑎𝑐 ), as demonstrated experimentally in Fig. S4c. We present three measurements of 𝑅𝐻2𝑓 (𝐵𝐵𝑎𝑐 ) recorded at ϕ = 90º, 60º, and 0º (Fig. S4d). We note that the curves measured at ϕ = 90º and ϕ = 0º have the same shape, whereas the curve measured at ϕ = 60º is asymmetric due to the contribution of the PHE at this angle. Using Eq. 39, we show that we recover correctly the amplitude Bac = 10.5 mT independently of ϕ (Fig. S4e), confirming the consistency of the 13 Figure S4. Measurement of an external ac field using the Hall voltage. a, 𝑅𝐻2𝑓 measured in- b, Comparison between 𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 �=[𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 , 𝐵𝐵𝑎𝑐 � + 𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 , −𝐵𝐵𝑎𝑐 �]/2 (black line) phase (dashed line) and antiphase (solid line) with an applied ac magnetic field parallel to Bext. and 𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 , 𝐵𝐵𝑎𝑐 = 0� (red line). c, Comparison between the numerical derivative of 𝑅𝐴𝐻𝐸𝑓 𝑅𝐻2𝑓 (𝐵𝐵𝑎𝑐 ) = [𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 , 𝐵𝐵𝑎𝑐 � − 𝑅𝐻2𝑓 �𝑏𝜃𝜃 , 𝑏𝜑𝜑 , −𝐵𝐵𝑎𝑐 �]/2 (red line). d, 𝑅𝐻2𝑓 (𝐵𝐵𝑎𝑐 ) recorded at ϕ = 90º, 60º, and 0º. e, Measured with respect to the applied field (black line) and amplitude of the applied ac field as a function of magnetization tilt obtained from the curves shown in d using Eqs. 38 and 39. harmonic analysis of the AHE and PHE contributions presented in Sects. 2 and 3. The standard deviation of the data in Fig. S4e is 0.6 mT, which corresponds to a relative error of about 6% on our measurements. The external ac field is also useful to determine the direction of 𝑩𝑩⊥ with respect to the Oersted field. In the measurements reported in Fig. S4, Bext is perpendicular to the current and, when positive, it is opposite to the Oersted field generated by a positive current flowing in the Pt observe that 𝑅𝐻2𝑓 increases (decreases) when 𝐵𝐵𝑎𝑐 is in-phase (out-of-phase) with the current, layer. The Oersted field is determined by the conventional right-hand rule. In Fig. S4a we meaning that Bodd and Bac add up for an in-phase measurement, i.e., that 𝑩𝑩⊥ is positive for a positive current. Therefore, we conclude that 𝑩𝑩⊥ is opposite to the Oersted field. 14 S5. Measurement of current-induced effective fields in the case of nonzero PHE Equation 37 relates 𝑅𝐻2𝑓 to the effective field components bθ and bϕ. If the PHE is neglected, bθ reads 𝑏𝜃𝜃 = 𝑅𝐻2𝑓 sin(𝜃𝜃𝐵 − 𝜃𝜃0 ) 𝑑𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 �𝜃𝜃0 , (40) 𝐼𝑅𝐴𝐻𝐸 𝑑 cos 𝜃𝜃 which is readily evaluated by noting that 𝑅𝐴𝐻𝐸 𝑑 cos 𝜃𝜃 𝑑𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 �𝜃𝜃0 = 𝑑𝑅𝐻𝑓𝑑𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 �𝜃𝜃0 . (41) find that neglecting the PHE affects the quantitative determination of bθ and, therefore, of 𝐵𝐵⊥ However, despite the fact that the PHE is significantly smaller than the AHE in AlOx/Co/Pt, we and 𝐵𝐵∥ . In fact, Eq. 40 does not allow for a precise calculation of the current-induced fields and one must resort to the more general Eq. 37. In order to solve Eq. 37 for bθ and bϕ, we thus use a reported in Sect. 3. Starting from 𝑅𝐻2𝑓 measured at ϕ = 0º and 90º (blue curves in Fig. S5a and b, recursive procedure that takes advantage of the indipendent measurement of RPHE and RAHE respectively) we operate as follows: We set 𝑅𝑃𝐻𝐸 = 0 as initial guess and evaluate 𝑏𝜃𝜃0 (𝜑𝜑 = 90°) and 𝑏𝜑𝜑0 (𝜑𝜑 = 0°), shown i) ii) We set 𝑅𝑃𝐻𝐸 to its measured value and evaluate 𝑏𝜃𝜃1 (𝜑𝜑 = 90°) using the previous in black in Fig. S5 a and c, respectively. estimate of 𝑏𝜑𝜑0 (𝜑𝜑 = 0°). Using both further gives 𝑏𝜑𝜑1 (𝜑𝜑 = 0°). iii) We evaluate 𝑏𝜃𝜃1 (𝜑𝜑 = 0°) using 𝑏𝜑𝜑1 (𝜑𝜑 = 0°), which further gives 𝑏𝜑𝜑1 (𝜑𝜑 = 90°), as shown in Fig. S5 b,d, blue curves. Steps ii) and iii) are repeated until we achieve convergence (red curves in Fig. S5). iv) Figures S5c and d show the successive iterations that lead to the final form of 𝑏𝜃𝜃 at ϕ = 90º and 0º, respectively. This procedure is independent on the choice of coefficients used to represent 𝑏𝜃𝜃 and 𝑏𝜑𝜑 . We note that Eqs. 24-27 imply that 𝑏𝜃𝜃 = 𝐵𝐵⊥ at ϕ = 90º and 𝑏𝜃𝜃 = 𝐵𝐵∥ at ϕ = 0º. Recalling that by truncating the angular expansion of the torques to fourth order we have 15 Figure S5. Recursive procedure to determine 𝑩𝑩⊥ and 𝑩𝑩∥ in the case of nonzero PHE. a, 𝑅𝐻2𝑓 measured at ϕ = 90º (black line) and successive iterations that lead to the separation of the pure AHE contribution 𝑅𝐴𝐻𝐸 2𝑓 (𝜑𝜑 = 90°) (red line). b, 𝑅𝐻2𝑓 measured at ϕ = 0º (black line) and successive iterations that lead to 𝑅𝐴𝐻𝐸 2𝑓 (𝜑𝜑 = 0°) (red line). c, Iterations leading to convergent 𝐵𝐵⊥ values at ϕ = 0º. d, Iterations leading to convergent 𝐵𝐵∥ values at ϕ = 0º. 𝐼𝑏𝜃𝜃 (𝜑𝜑 = 0°) = 𝑇𝑇0∥ + 𝑇𝑇2∥ sin2 𝜃𝜃 + 𝑇𝑇4∥ sin4 𝜃𝜃 , 𝐼𝑏𝜃𝜃 (𝜑𝜑 = 90°) = −cos 𝜃𝜃 (𝑇𝑇0⊥ + 𝑇𝑇2⊥ sin2 𝜃𝜃 + 𝑇𝑇4⊥ sin4 𝜃𝜃), (42) 𝐼𝑏𝜑𝜑 (𝜑𝜑 = 0°) = −𝑇𝑇0⊥ , (43) 𝐼𝑏𝜑𝜑 (𝜑𝜑 = 90°) = − cos 𝜃𝜃 𝑇𝑇0∥ , (44) (45) and 𝑇𝑇0,2,4⊥ are obtained by fitting 𝑏𝜃𝜃 (𝜑𝜑 = 0°) and 𝑏𝜃𝜃 (𝜑𝜑 = the values of the coefficients 𝑇𝑇0,2,4∥ 90°) using Eqs. 42 and 43. 16 S6. Macrospin simulations a macrospin model that includes the effects of SOTs. The model allows us to simulate 𝑅𝐻𝑓 and To validate the iteration procedure described above, we performed numerical simulations using 𝑅𝐻2𝑓 as a function of Bext starting from the equilibrium equation for the magnetization: 𝒎𝒎 × 𝑩𝑩𝑂𝑂𝑥𝑥𝑠𝑠 + 𝒎𝒎 × 𝑩𝑩𝑘 + 𝑻𝑻⊥ + 𝑻𝑻∥ = 0 , (46) due to the anisotropy field 𝑩𝑩𝑘 = [𝐵𝐵𝑘1 (𝒎𝒎 ∙ 𝒛) + 𝐵𝐵𝑘2 (𝒎𝒎 ∙ 𝒛)𝟑 ]𝒛, with second and fourth order where the first term represents the torque due to the external field and the second term the torque anisotropy constants 𝐵𝐵𝑘1 = -0.95 T and 𝐵𝐵𝑘2 = -0.2 T, respectively. We assume here the simplest form of 𝑻𝑻⊥ and 𝑻𝑻∥ compatible with the experimental results of AlOx/Co/Pt, namely 𝑻𝑻⊥ = cos 𝜙 𝑇𝑇0⊥ e𝜃𝜃 − cos 𝜃𝜃 sin 𝜙 (𝑇𝑇0⊥ + 𝑇𝑇2⊥ sin2 𝜃𝜃)e𝜙 (47) 𝑻𝑻∥ = cos 𝜃𝜃 sin 𝜙 𝑇𝑇0∥ e𝜃𝜃 + cos 𝜙𝑇𝑇0∥ e𝜙 (48) and and set the SOT coefficients to 𝑇𝑇0⊥ = -12.0 mT, 𝑇𝑇2⊥ = -11.0 mT, and 𝑇𝑇0∥ = 19.0 mT. These each instant t for the angles 𝜃𝜃0 and 𝜑𝜑0 that define the equilibrium direction of m and taking RAHE coefficients are modulated by the current, which is proportional to ei2πft. By solving Eq. 43 at to obtain 𝑅𝐻𝑓 and 𝑅𝐻2𝑓 as a function of Bext. Figure S6a and b show that the simulations faithfully = 0.8 Ω and RPHE = 0.09 Ω, we calculate VH (t) using Eq. 29.Then, VH (t) is Fourier-transformed reproduce the shape of the experimental 𝑅𝐻2𝑓 curves measured at ϕ = 0º and 90º, respectively (see Fig. 2, main text). Further, by applying the iteration steps i-iv described above to the simulated data (Fig. S6c-f), we find a very similar convergence behaviour to that reported for the the values 𝑇𝑇0⊥ = -11.99 mT, 𝑇𝑇2⊥ = -11.01 mT, and 𝑇𝑇0∥ = 18.97 mT, thus confirming the validity of experimental data in Fig. S5. Moreover, after a maximum of seven iteration steps, we recover our analysis. 17 Figure S6. Macrospin simulations of 𝑅𝐻2𝑓 , 𝑩𝑩⊥ , and 𝑩𝑩∥ including both the AHE and PHE contributions to the Hall voltage. a, Simulated 𝑅𝐻2𝑓 signal at ϕ = 90º and b, ϕ = 0º 𝑅𝐴𝐻𝐸 2𝑓 (𝜑𝜑 = 90°) and (d) 𝑅𝐴𝐻𝐸 2𝑓 (𝜑𝜑 = 0°) applied to the simulated signal. Iterative estimate (see text for details). c, d, Iteration procedure to separate the pure AHE components (c) of e, 𝐵𝐵⊥ at ϕ = 0º and f, 𝐵𝐵∥ at ϕ = 0º. Final result for g, 𝐵𝐵⊥ and h, 𝐵𝐵 ∥ derived from the PHE. 𝐵𝐵⊥ and 𝐵𝐵∥ coincide with the input values assumed in the simulations. macrospin simulation of the Hall voltage and analysis of the second harmonic AHE and 18 S7. Sample-dependent offset and Nernst-Ettingshausen effect Aside from the AHE and PHE, the measurement of 𝑅𝐻2𝑓 is sensitive to sample-dependent contributions to the Hall voltage, namely the misalignment of the voltage leads and thermoelectric effects. 2 The current flowing into the Hall cross can generate a temperature gradient due to inhomogeneous heating in correspondence of fabrication defects, generally on the corners of the sample. This gradient induces two types of thermoelectric voltages, the Seebeck effect and the Anomalous Nernst-Ettingshausen effect (ANE). Since heating is modulated by the ac current, we detect both contributions in the 2f component of the Hall once accounted for, 𝑅𝐻2𝑓 presents the same magnetization dependence in all samples. voltage. We note that these effects vary in amplitude and sign from sample to sample and that, The largest sample-dependent contribution to 𝑅𝐻2𝑓 is a constant offset (ROffset) due to the gives a contribution to 𝑅𝐻2𝑓 that is magnetization dependent and induces a small asymmetry in asymmetry of the voltage probes as well as to the Seebeck effect. The ANE, on the other hand, the raw 𝑅𝐻2𝑓 curves. The voltage induced by ANE is perpendicular to the temperature gradient and mz, mimicking the AHE with much smaller amplitude. (Fig. S7a and b). Both the offset and Figure S7. Measurement and subtraction of the Anomalous Nernst effect. a, 𝑅𝐻2𝑓 measured at θΒ = 0º, ϕ = 90º, and I = 635 µA. b, 𝑅𝐻𝑓 simultaneously measured with a. c, d, 𝑅𝐻2𝑓 measured at θΒ = 85º, ϕ = 90º, and I = 635 µA before (c) and after (d) subtraction of the ANE and constant offset. 19 ANE contributions to 𝑅𝐻2𝑓 can be determined by performing a measurement with Bext // z, as shown in Fig. S7a. In this case, the SOT and Oersted fields do not contribute to the second Hence, the residual 𝑅𝐻2𝑓 signal is related uniquely to ROffset and the ANE. As expected for the harmonic signal since the variation of the Hall resistance is symmetric with respect to θ=0°. ANE, 𝑅𝐻2𝑓 is hysteretic and has the same field dependence of 𝑅𝐻𝑓 , which is proportional to mz. extrema of 𝑅𝐻2𝑓 measured for Bext // z. We find RANE = 0.1 mΩ for a current of 635 µA, which is One can easily deduce the amplitude of the ANE, RANE, by taking the difference between the that RANE coincides with the difference of 𝑅𝐻2𝑓 measured at zero field for positive and negative comparable with other values found in the literature. 3 For arbitrary orientation of Bext, we find sweeps of Bext, whereas their average gives ROffset. Finally, both ROffset and RANE can be subtracted from the raw data, giving: 𝑅𝐻2𝑓 = 𝑅𝑟𝑟𝑎𝑤2𝑓 − 𝑅𝐴𝑁𝐸 𝑅𝐻𝑓2𝑅𝐴𝐻𝐸 − 𝑅𝑂𝑂𝑓𝑓𝑠𝑠𝑂𝑂𝑠𝑠 . (49) Figures S7c and d show 𝑅𝐻2𝑓 before and after subtraction of ROffset and ∆RANE, respectively. S8. Current dispersion in the Hall voltage probes The SOT/current ratios reported in the main text are calculated without taking into account the spread of the charge current into the voltage leads of the Hall cross. Depending on the relative width of the Hall voltage probes with respect to the current probes, the actual current giving rise to the SOTs may be smaller than the total current injected into the device. Numerical simulations of the current flow in a Hall cross show that the current density in the central region of the cross reduces by up to 23% (8%) in junctions where the width of the voltage probes is equal (half) the width of the current probes. 4 The dimensions of the Hall cross in our devices are 500 x 500 nm2, 1000 x 1000 nm2, and 1000 x 500 nm2. Figure 3 in the main text shows that the SOT/current ratios measured in the 500 x 500 nm2 and 1000 x 1000 nm2 devices are similar, whereas in the 1000 x 500 nm2 device they are larger by 20-30% relative to the symmetric probes. Figure S8 shows a set of measurements for a 1000 x 500 nm2 device of AlOx/Co/Pt corresponding to a current density j = 2.91x107A/cm2, we obtain 𝑇𝑇0⊥ = -16 ± 1 mT, 𝑇𝑇2⊥ = -12.8 ± recorded at ϕ = 90º, 0º, and θB = 82º. For an ac current amplitude set to I = 1040 µA, 0.8 mT, and 𝑇𝑇0∥ = 25 ± 1 mT. For a 1000 x 1000 nm2 device patterned on the same wafer, we measured 𝑇𝑇0⊥ = -11.4 mT, 𝑇𝑇2⊥ = -9.3 mT, and 𝑇𝑇0∥ = 18.7 mT for the same current amplitude. We 20 measured on a 1000 x 500 nm2 device. a, 𝑅𝐻2𝑓 as a function of Bext applied at θB = 82º, ϕ = Figure S8. Second harmonic Hall resistance and current-induced spin-orbit fields 𝐵𝐵⊥ /cosθ as a function of Bext (ϕ = 90º). d, Effective field B∥ as a function of Bext (ϕ = 0º). 90º and b, θB = 82º, ϕ = 0º. The amplitude of the ac current is 1.136 mA. c, Effective field e, 𝐵𝐵⊥ /cosθ as a function of sin2θ measured at ϕ = 90º. The solid line is a fit to the function 𝑇𝑇0⊥ + 𝑇𝑇2⊥ sin2 𝜃𝜃 with 𝑇𝑇0⊥ = -16.0 mT and 𝑇𝑇2⊥ = -12.8 mT. f, B∥ as a function of θ measured at ϕ = 90º. attribute this effect to the larger current density flowing in the magnetic region of the 1000 x 500 nm2 device due to the reduced width of the voltage probes. S9. Measurements in the case of nonuniform magnetization Our measurements are based on detecting small current-induced oscillations of the Bext. The magnetization curves (𝑅𝐻𝑓 ) measured for 𝜃𝜃𝐵 ≤ 85° show reversible behavior beyond magnetization about its equilibrium direction, which is determined by the anisotropy field and the switching field, consistently with coherent rotation of the magnetization towards Bext. At 21 𝜃𝜃𝐵 > 85°, however, we observe irreversible jumps of the Hall resistance due to the formation of magnetic domains. These jumps are also detected in the 𝑅𝐻2𝑓 curves, as shown in Fig. S9 for a paper are limited to 𝜃𝜃𝐵 ≤ 85°. geometry θB = 87°, ϕ = 90° and I = 635 µA. For this reason, the measurements reported in this nucleation. a, 𝑅𝐻2𝑓 as a function of Bext applied at θB = 87º, ϕ = 90º. The amplitude of the Figure S9. Second harmonic Hall resistance in the presence of magnetic domain ac current is 635 mA. A jump of the signal is observed between 0.5 and 1 T. S10. Comparison of AC and DC detection methods We present here a comparison of our AC detection method with DC measurements of the Hall voltage, analogue to those performed by Liu et al. in Refs. 5 and 6. These authors considered a scalar model where the torques due to the external field, magnetic anisotropy, and current are collinear. In the macrospin approximation, this leads to the following torque equation at equilibrium: 𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 sin(𝜃𝜃0 − 𝜃𝜃𝐻 ) − 𝐵𝐵𝑘 sin 𝜃𝜃0 cos 𝜃𝜃0 + 𝑇𝑇(𝐼) = 0. (50) By estimating the equilibrium magnetization angle 𝜃𝜃0 using the AHE (Eq. 38), one can define two magnetic field values, 𝐵𝐵+ (𝜃𝜃0 ) and 𝐵𝐵− (𝜃𝜃0 ), as the value of 𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 required to produce a given value of 𝜃𝜃0 for positive and negative current, respectively.5 From Eq. 50 one has 𝐵𝐵± (𝜃𝜃0 ) = [𝐵𝐵𝑘 sin 𝜃𝜃0 cos 𝜃𝜃0 ∓ 𝑇𝑇(𝐼)] sin(𝜃𝜃0 − 𝜃𝜃𝐻 ) ⁄ and, finally, 𝑇𝑇(𝐼) = 𝐵−(𝜃𝜃0)−𝐵+(𝜃𝜃0) sin(𝜃𝜃0 − 𝜃𝜃𝐻 ). 2 (51) In practice, 𝐵𝐵± (𝜃𝜃0 ) are calculated by measuring the Hall resistances for positive and negative current [𝑅𝐻 (𝐼± )] as a function of applied field. 22 By assuming the simplest form for the torques, T∥ = 𝑇𝑇0∥ m × (y × m) and T⊥= 𝑇𝑇0⊥ (y × m), Liu et al. used Eq. 51 to measure 𝑇𝑇0∥ and 𝑇𝑇0⊥ for 𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 applied in the xz and yz plane, and 6). For AlOx/Co(0.6 nm)/Pt(2 nm) annealed in vacuum at 350 C, Liu et al. concluded that 𝑇𝑇0∥ respectively (note that, with respect to our notation, the x and y axis are interchanged in Refs. 5 = 0.33±0.06 mT/mA (1.7 mT per 107 A/cm2) and that 𝑇𝑇0⊥= 0, within the sensitivity of the experiment (1.3 mT per 107 A/cm2). Because the SOT amplitudes are generally very sensitive to 𝑇𝑇0∥ . However, Eq. 51 assumes that the magnetization remains confined in the plane defined by the sample growth details, it is not surprising that we obtain different torque values, at least for the external field and z axis, which is not true if both T∥ and T⊥ are present, as can be seen from Eqs. 18, 19 and 24, 25 (see also Fig. 11b). Moreover, if the magnetization deviates from the xz Figure S10. Comparison between AC and DC detection methods. 𝑅𝐻 measured on a Hall cross of 3000 nm (current injection) by 500 nm (Hall voltage) for 𝐼±= ±1.2 mA, θΒ = 86º, and a, ϕ = 90º, b, ϕ = 0º. c, d, Comparison of 𝑅𝐻2𝑓 and Δ𝑅𝐻 /2. e, Comparison of 𝐵𝐵⊥ extracted from 𝑅𝐻2𝑓 (open symbols) and Δ𝑅𝐻 (filled symbols). f, Same for 𝐵𝐵∥ . The data corrected for the PHE are also shown. The integration time used in the AC method was 10s. In the DC method, each field point has been averaged for 10 s. 23 or the yz plane, the PHE contributes to 𝑅𝐻 in a way that is not symmetric for positive and 𝑇𝑇0∥ and 𝑇𝑇0⊥ , have not been taken into account in the analysis of Refs. 5 and 6. negative current. These effects, as well as the presence of additional torque components besides Here, we show that, if the PHE is neglected, the two methods give equivalent results, but Figures S10a and b show 𝑅𝐻 (𝐼+ ) and 𝑅𝐻 (𝐼− ) measured for 𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 applied in the yz (𝜑𝜑 = 90°) and that the AC measurements give a better signal-to-noise ratio compared to DC, as expected. xz (𝜑𝜑 = 0°) plane, respectively. In Fig. S10c and d we compare 𝑅𝐻2𝑓 with the corresponding quantity in a DC measurement, ∆𝑅𝐻 2⁄ = [𝑅𝐻 (𝐼+ ) − 𝑅𝐻 (𝐼− )] 2⁄ . The fields 𝐵𝐵+ (𝜃𝜃0 ) and 𝐵𝐵− (𝜃𝜃0 ) introduced above can be easily derived from ∆𝑅𝐻 . We observe that, apart from the noise level, 𝑅𝐻2𝑓 and ∆𝑅𝐻 present the same dependence on the external field. Figures S10e and f show that, if 𝑅𝑃𝐻𝐸 is set to zero, the current-induced fields 𝐵𝐵⊥ and 𝐵𝐵∥ obtained from the analysis of 𝑅𝐻2𝑓 and ∆𝑅𝐻 using Eqs. 40 and 41 give similar results, although the scattering of the DC data is much larger. The AC data corrected for the PHE are also shown. Clearly, the AC method allows for more sensitive measurements and, therefore, to work in a low current regime where the magnetization behaves coherently and thermal effects are small. S11. Dynamic simulations of the my component generated by T∥ As mentioned above and in the main text, a remarkable difference between our data and those reported by Liu et al. is the null result obtained for T⊥ in Ref. 5. We pointed out the reduced sensitivity of DC measurements and differences in sample preparation as possible clues for such a discrepancy. On the other hand, Liu et al. confuted the interpretation of previous measurements of T⊥ by noting that the spin transfer torque T∥ due to the spin Hall effect may induced a tilt of the magnetization along the y axis, similar to that expected by T⊥ (Ref. 5). This would be the case if �𝑇𝑇0∥ � > 𝐵𝐵𝑘 /2, that is, if 𝑇𝑇0∥ is so large as to overcome the anisotropy field and induce switching. However, the method presented in this work is based on small oscillations (Fig. 4 main text), which is very far from the regime �𝑇𝑇0∥ � > 𝐵𝐵𝑘 /2 hypothized in Ref. 5. We have of the magnetization and the torque amplitudes have been measured in the low current regime carried out dynamic simulations of the x, y, z magnetization components subject to either T⊥+ T∥ or T∥ alone to confirm this point, by numerically solving the LLG equation 𝒎𝒎 = −𝛾𝛾0 �𝒎𝒎 × (𝑩𝑩𝑂𝑂𝑥𝑥𝑠𝑠 + 𝑩𝑩𝑘 ) + T∥ + T⊥ � + 𝛼 𝒎𝒎 × 𝒎𝒎 , where 𝛾𝛾0 is the gyromagnetic ratio and 𝛼 the Gilbert damping. The simulations were performed using the following parameters: 𝐵𝐵𝑘 = 1 T, 𝛼 =0.5, and 24 𝑇𝑇0⊥ = 𝑇𝑇2⊥ = 0 and 𝑇𝑇0∥ = 0.3 T, b, or 𝑇𝑇0⊥ = 𝑇𝑇2⊥ = 0.15 T and 𝑇𝑇0∥ = 0.3 T c, 𝑇𝑇0⊥ = 𝑇𝑇2⊥ = 0 Figure S11. Dynamic simulations of mx, my, and mz for different torque amplitudes. a, and 𝑇𝑇0∥ = 0.6 T. The dashed line represents the injected current. 𝐵𝐵𝑂𝑂𝑥𝑥𝑠𝑠 = 0.1 T, 𝜑𝜑=0º, 𝜃𝜃𝐻 =90º. Figure S11a shows that for 𝑇𝑇0⊥ = 𝑇𝑇2⊥ = 0 and 𝑇𝑇0∥ = 0.3 T < 𝐵𝐵𝑘 /2, no magnetization component appears along the y axis. A nonzero 𝑚𝑦𝑦 only appears if T⊥ is turned on, as in Fig. S11b or when 𝑇𝑇0∥ = 0.6 T > 𝐵𝐵𝑘 /2, as in Fig. S11c. S12. Influence of thermal effects It is well known that heating reduces the magnetic anisotropy field and saturation magnetization of thin films. Therefore ohmic heating induced by the current effectively softens the magnetization of the ferromagnetic layer, increasing the susceptibility of the magnetization to the spin torques. A consequence of this effect is to introduce a spurious increase of the torque amplitudes measured at high current, which may add itself to the intrinsic temperature appear as an odd harmonic component of the Hall resistance, 𝑅𝐻3𝑓 , whereas 𝑅𝐻2𝑓 should reflect dependence of the torques. However, heating effects are proportional to I2 and hence should the presence of spin torque components that are linear with the current. To check the validity of this statement, we implemented Joule heating in our macrospin simulations and compared them modeled by considering a reduced anisotropy field 𝐵𝐵𝑘 (1 − 𝑎𝑘 𝐼2 ) and saturation magnetization with experimental data obtained for two different current amplitudes (Fig. S12). Heating was 𝑀𝑠𝑠 (1 − 𝑎𝑀 𝐼2 ), using the following parameters: 𝐵𝐵𝑘 = 1 T, 𝜇0𝑀𝑠𝑠 = 1 T, 𝑅𝐴𝐻𝐸 = 0.8 Ω, 𝑅𝑃𝐻𝐸 = 0.1 Ω, 𝑇𝑇0⊥ = -0.12 T, 𝑇𝑇2⊥ = -0.11 T, and 𝑇𝑇0∥ = 0.19 T. The coefficients 𝑎𝑘 =0.14 and 𝑎𝑀 =0.035 were derived from the measured current dependence of 𝐵𝐵𝑘 and 𝑅𝐴𝐻𝐸 , respectively. We performed simulations of 𝑅𝐻𝑓 , 𝑅𝐻2𝑓 , and 𝑅𝐻3𝑓 , as shown in Fig. S12. In the top panels, we observe 25 a small decrease of the 𝑅𝐻𝑓 amplitude at remanence, due to the decrease of 𝑀𝑠𝑠 , and at high field due to the reduction of 𝐵𝐵𝑘 . Both the experiments and simulations show a clear 𝑅𝐻3𝑓 signal other hand, the measurements of 𝑅𝐻2𝑓 normalized by the current amplitude nearly superpose (bottom panels) that disappears when heating is turned off in the simulations (blue line). On the (middle panels), meaning that heating effects have only a minor influence on the spin torque measurements in this current regime. This is in agreement with the current dependence of the Above this limit, our simulations show that 𝑅𝐻2𝑓 gradually increases due to the reduction of 𝑀𝑠𝑠 , torque coefficients reported in Fig. 4 of the main text, which is linear below j = 1.5 x 107 A/cm2. whereas 𝑅𝐻3𝑓 is affected by 𝑀𝑠𝑠 (hysteretic part) and 𝐵𝐵𝑘 (high-field part). Finally, we simulated the case where T⊥ is set to zero to check whether a heat-induced modulation of the anisotropy field case, since 𝑅𝐻2𝑓 is zero in such a case (green line). can mimic the action of the field-like torque, as suggested in Ref. 5. This turns out not to be the experimental data. Simulations (left column) and measurements (right column) of 𝑅𝐻𝑓 , 𝑅𝐻2𝑓 , Figure S12. Macrospin simulations of thermal induced effects and comparison with and 𝑅𝐻3𝑓 for ϕ = 90º. The measurements were performed on a 1000 x 1000 nm2 AlOx/Co/Pt device. See text for details. 26 S13. Supplementary references 1 Pauyac, C. O.,Wang, X., Chshiev, M. & Manchon, A. Angular dependence and symmetry of Rashba spin torque in ferromagnetic heterostructures, http://arxiv.org/pdf/1304.4823. 2 Lindberg, O. Hall Effect. Proc. Inst. Radio Engrs. 40, 1414-1419 (1952). 3 Seki, T. et al., Giant spin Hall effect in perpendicularly spin-polarized FePt/Au devices, Nat. Mater. 7, 125-129 (2008). 4 Ibrahim, I. S., Schweigert, V. A. & Peeters, F. M. Diffusive transport in a Hall junction with a microinhomogeneous magnetic field. Phys. Rev. B 57, 15416-15427 (1998). 5 Liu, L., Lee, O. J., Gudmundsen, T. J., Ralph, D. C. & Buhrman, R. A. Current-Induced Switching of Perpendicularly Magnetized Magnetic Layers Using Spin Torque from the Spin Hall Effect. Phys. Rev. Lett. 109, 096602 (2012). 6 Liu, L., Pai, C. F., Li, Y., Tseng, H. W., Ralph, D. C. & Buhrman, R. A. Spin-torque switching with the giant spin Hall effect of tantalum. Science 336, 555-558 (2012). 27
1610.02359	1	1610	2016-10-07T18:00:38	DFT Simulations of Inter-Graphene-Layer Coupling with Rotationally Misaligned hBN Tunnel Barriers in Graphene/hBN/Graphene Tunnel FETs	[ "cond-mat.mes-hall" ]	Van der Waal's heterostrucutures allow for novel devices such as two-dimensional-to-two-dimensional tunnel devices, exemplified by interlayer tunnel FETs. These devices employ channel/tunnel-barrier/channel geometries. However, during layer-by-layer exfoliation of these multi-layer materials, rotational misalignment is the norm and may substantially affect device characteristics. In this work, by using density functional theory methods, we consider a reduction in tunneling due to weakened coupling across the rotationally misaligned interface between the channel layers and the tunnel barrier. As a prototypical system, we simulate the effects of rotational misalignment of the tunnel barrier layer between aligned channel layers in a graphene/hBN/graphene system. We find that rotational misalignment between the channel layers and the tunnel barrier in this van der Waal's heterostructure can significantly reduce coupling between the channels by reducing, specifically, coupling across the interface between the channels and the tunnel barrier. This weakened coupling in graphene/hBN/graphene with hBN misalignment may be relevant to all such van der Waal's heterostructures.	cond-mat.mes-hall	cond-mat	DFT Simulations of Inter-Graphene-Layer Coupling with Rotationally Misaligned hBN Tunnel Barriers in Graphene/hBN/Graphene Tunnel FETs Amithraj Valsaraj1,a, Leonard F. Register a, Emanuel Tutuc a and Sanjay K. Banerjee a a Microelectronics Research Center and Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, TX-78758, USA. Van der Waal's heterostructures allow for novel devices such as two-dimensional-to-two- dimensional tunnel devices, exemplified by interlayer tunnel FETs. These devices employ channel/tunnel-barrier/channel geometries. However, during layer-by-layer exfoliation of these multi-layer materials, rotational misalignment is the norm and may substantially affect device characteristics. In this work, by using density functional theory methods, we consider a reduction in tunneling due to weakened coupling across the rotationally misaligned interface between the channel layers and the tunnel barrier. As a prototypical system, we simulate the effects of rotational misalignment of in a graphene/hBN/graphene system. We find that rotational misalignment between the channel layers and the tunnel barrier in this van der Waal's heterostructure can significantly reduce coupling between the channels by reducing, specifically, coupling across the interface between the channels and the tunnel barrier. This weakened coupling in graphene/hBN/graphene with hBN misalignment may be relevant to all such van der Waal's heterostructures. layer between aligned channel layers the tunnel barrier I. INTRODUCTION The atomically thin two-dimensional (2D) materials translate to excellent electrostatic gate control at nanoscale channel length dimensions1, near-ideal two dimensional carrier behavior2,3, for conventional4 and novel devices applications. Examples of the latter include field effect transistors with gate controlled resonant interlayer tunneling5–8, which we here refer to as a class as interlayer tunnel FETs (ITFETs), and the proposed Bilayer-pseudospin FET (BiSFET)9,10 based on interlayer exciton condensation. Experimental advances in stacking 2D atomic crystals have allowed the fabrication of van der Waal's heterostructures with precisely chosen sequence of atomic layers11,12. Stacking 2D materials in vertical van der Waal's heterostructures allows researchers to probe interesting physical phenomena such as interlayer tunneling between channel layers through a tunnel barrier, as demonstrated in a single layer (SL) graphene/hexagonal boron nitride (hBN)/SL-graphene heterostructure13. Aligning the band structures of the opposing graphene layers in momentum space then allows for resonant interlayer tunneling and associated negative differential resistance7,8,14, which is the foundation for ITFETs. Aligning the band structures in momentum space, of course, requires carefully aligning the crystallographic orientation of the two graphene layers. Achieving such alignment experimentally requires substantial care.6–8,15 During layer-by-layer exfoliation of these van der Waal's materials, rotational misalignment is the norm, and even in such studies with aligned channel layers, the tunnel barrier has not been rotationally aligned. However, misalignment may affect device properties substantially. Prior theoretical studies have calculated the electronic structure of 1 Author to whom correspondence should be addressed. Electronic mail: [email protected]. 1 rotationally misaligned graphene bilayers using a continuum approximation16 and a tight-binding method17. The opening of a band gap in the electronic structure of rotationally misaligned bilayer graphene has been demonstrated due to the formation of interlayer-bonded diamond nanodomains18. Variations in the electronic structure of graphene deposited on rotationally misaligned hBN even have been considered in a study of hydrogen desorption19. Density functional theory (DFT) calculations have been performed on short-period crystalline structures to derive effective Hamiltonians to describe the influence of moiré pattern superlattices on the electronic properties of graphene/graphene and graphene/hBN heterostructures20. However, prior theoretical studies of resonant interlayer tunneling have not considered the effect of rotational misalignment of tunnel barrier layers with respect to the channel layers on the tunneling current.5,14,21 In this work, we consider the effects on interlayer current flow that rotational misalignment of the tunnel barrier with respect to rotationally aligned channel layers may have. Consider for the sake of illustration that the resonant tunneling currents I of interest here could be approximated in the form I ≈ ηe−κt at resonance, where η and κ are constants and t is the tunnel barrier thickness. Misalignment between the graphene layers and the hBN barrier structures, in principle, could alter the evanescence coefficient κ. However, extrapolation of experimental results7 consistent with the apparent value of κ (corresponding to about an order of magnitude reduction in current per hBN layer) in a bilayer graphene/multilayer-hBN/bilayer graphene system to a few and even zero layers still leads to a prediction of weak interlayer tunneling even with the graphene bilayers aligned to each other (confirmed by resonant interlayer tunneling) but not to the hBN tunnel barrier. This limiting result cannot be explained through changes in evanescence alone. Therefore, we also must consider changes in the lead coefficient η due to rotational misalignment, such as would be due to weakened coupling across the rotationally misaligned graphene-hBN interface. To consider these effects, we employ the ab-initio atomistic DFT approach. Direct calculation of current flow in this way, however, is impractical with actual ITFET device and current-flow geometries. However, DFT calculations allow us to track the effects of the hBN on the (nominal) graphene band structure as a measure of the effective coupling across the interface, as a function of rotation angle. Moreover, the inter-channel layer coupling that mediates this current flow also leads to degeneracy breaking in the band structure of the channel layers, so that knowledge of one provides information on the other. We consider a SL-graphene/hBN/SL-graphene systems, although with multi-layer (ML) hBN tunnel barriers in some simulations, largely to minimize the atom count in our DFT simulations with large in-plane unit cells in rotationally misaligned systems. However, the results may be relevant to all such van der Waal's heterostructures. We show that the rotational misalignment between the graphene and hBN layers can significantly degrade the interlayer- graphene-layer coupling, and that this reduction is consistent with the reduction in the coupling across the graphene-hBN interface. In contrast, we find that there is little apparent effect on the evanescence for the coupling within the hBN barrier. That is, our results are consistent with a reduction in the lead coefficient η and not apparently with increases in the evanescence coefficient κ in the above approximation for current. This interface-localized effect would be qualitatively consistent with the weak tunneling in the few hBN layer limit extrapolated from experimental results as noted above. 2 II. METHODS To translating knowledge of degeneracy breaking to information on current flow, consider that in the weak coupling limit-e.g., short of saturating the interlayer current flow to the lead- limited value-the tunneling current I varies as 𝐼 = 𝑔𝑉 4𝜋𝑒 ℏ 2 ∑𝑀𝜇𝜈 𝜇𝜈 [𝑓(𝐸𝜇) − 𝑓(𝐸𝜈)]𝛿(𝐸𝜇 − 𝐸𝜈) (1) interlayer coupling within a first-order Bardeen transfer Hamiltonian approximation.22,23 Here 𝑀𝜇𝜈 represents the interlayer coupling, and is obtained from the overlap of the opposite channel layer wavefunctions across the interlayer tunnel barrier. Here, 𝑔𝑉 is the valley degeneracy of graphene and the summation in Eq. (1) runs over all states μ, ν of the top and bottom graphene layers, respectively. However, we do not need numerical values for I in this work; it is merely the quadratic dependence of interlayer coupling also produces a symmetric/antisymmetric splitting of the otherwise degenerate band structures in the opposite layers under flat band conditions, which is linearly proportional to 𝑀𝜇𝜈 in the weak coupling limit. Thus, by directly calculating the energy splitting due to coupling through a given tunnel barrier, we also can infer the greater effect on interlayer tunneling currents, with the relative change in current being proportional to the square of the relative change in degeneracy breaking in the weak coupling limit. interest. This is of that implemented For aligned and misaligned systems, and for SL and ML hBN barriers, DFT simulations were performed to calculate both the absolute band structures and the band splittings due to inter- graphene-layer coupling in the rotationally aligned and misaligned systems. However, first, the atomic structures were relaxed using the projector-augmented wave method with a plane-wave initio simulation package(VASP)24,25. basis set as Graphene/hBN/graphene supercells were created using the Virtual NanoLab software26. The lattice constants of graphene and hBN are well matched (to within 1.6%), and the considered rotational misalignments between the hBN and the mutually aligned graphene layers are necessarily commensurate for our calculations. Therefore, lattice strain is minimal for the considered supercell structures. Commensurate rotation angles for the graphene/hBN interface are given by,16 the Vienna ab in θ = cos−1 3𝑛2 + 3𝑛 + 0.5 3𝑛2 + 3𝑛 + 1 , (2) where n is a non-negative integer (n=0,1,2,……). The number of atoms in the supercell is given by 3 × (6𝑛2 + 6𝑛 + 2). Figure 1(a) and (b) show the commensurately rotated graphene/SL-hBN/ 3 FIG. 1. Supercells of graphene/SL-hBN/graphene with hBN layers commensurately rotated with respect to mutually aligned graphene layers, at rotation angles of (a) 21.79° and (b) 13.17°. C atoms are colored red, B atoms in blue and N atoms in green online. FIG. 2. Supercells of aligned graphene/hBN/graphene system with (a) SL-hBN and (b) bilayer hBN. C atoms of graphene are placed on top of B atoms of the hBN layer in the energetically most stable configuration. 4 graphene supercells with rotation angles, 21.79° and 13.17°, respectively corresponding to two smallest supercells (𝑛 = 1,2 from Eq. (2)) and visualized here using XCrySDen27. Hexagonal supercells with varying number of aligned hBN layers are shown in Figure 2. In the case of the aligned system, the C atoms of graphene are placed on top of B atoms of the hBN layer, which is the energetically most stable configuration28. With approximately 200 atoms per supercell, use of more computationally intensive hybrid functionals or GW methods is not practical. The local density approximation (LDA)29 was employed for the exchange-correlation (XC) functional as LDA can model materials well with the same orbital character at conduction and valence band edges. In previous work on 2D materials on high-k dielectrics, we had used LDA, generalized gradient approximation (GGA)30 and hybrid HSE0631 for XC functional in simulations of heterostructures with no qualitative differences in the results obtained32,33. Here, we also have re- checked some of the key results using the GGA functional. Van der Waal's non-local dispersive forces were modeled using the DFT-D2 scheme wherein a semi-empirical correction is added to the conventional Kohn-Sham DFT theory34. A k-mesh grid of 7×7×3 for the sampling of the first Brillouin zone of the supercell was selected according to Monkhorst-Pack type meshes with the origin being at the Γ point for all calculations except the band structure calculation (where a fine k-space resolution along major axes was sampled). Atomistic relaxations of supercells were allowed to converge when the Hellmann-Feynman forces on the atoms were less than 0.001 eV/Å. After relaxation, inversion symmetry was enforced on the crystal structure so that the induced degeneracy splitting is only due to coupling between the graphene layers across the hBN layer(s) and not any level of incomplete relaxation. The interlayer degeneracy splitting calculated in this way is only weakly k-dependent within a few hundred meV of the Dirac point. For specificity, we calculated the splitting at ≈100 meV above the Dirac (K) point, avoiding the intersecting band structures about the Dirac point, as illustrated in Figure 3. FIG. 3. An illustration of the method used to calculate the interlayer coupling parameter (𝑀𝜇𝜈) for the sample band structure of rotationally misaligned (𝜃 = 21.79°) graphene/SL-hBN/graphene after relaxation. The (roughly constant) band splitting is measured ≈100 meV above the Dirac (K) point. 5 III RESULTS The interlayer energy splittings with hBN layer number for the fully-aligned graphene/hBN/graphene system calculated by the above outlined procedure were 69.1, 20.4, 2.2, and 0.49 meV for one to four layers of hBN, respectively (which is reasonably in line with the known zero hBN layer energy splitting limit of 370 meV represented by a Bernal stacked graphene bilayer35). For one, two or even more layers, such coupling essentially would short the graphene layers together in the large area systems, in contrast to the experimental results for rotationally misaligned systems7,8. To investigate this discrepancy between theory and experimental results, we first simulated channel/tunnel-barrier/channel heterostructures with a rotationally misaligned tunnel barrier layer between aligned channel layers. For SL-graphene/SL-hBN/SL-graphene system with the hBN layer misaligned by commensurate rotation with mutually aligned graphene layers, such as for a resonant tunneling device), Figure 4 and Table 1 exhibit a substantial reduction in the energy splitting with layer misalignment. As also shown in Table I, such a reduction in interlayer coupling due to this interface-localized effect would drop the estimated interlayer current per unit area by factors of approximately 9, 25 and 53 for the three twist angles of 21.79°, 13.17° and 9.43°, respectively, within the 1st order perturbation limits for energy splitting and tunneling current (Eq. (1)). Thus, despite the growth in primitive unit cell size by factors of 7, 19 and 37 respectively, the expected interlayer current per primitive unit cell for these three rotation angles would remain roughly constant and actually decrease somewhat to 0.80, 0.76 and 0.70 times that for the aligned system. Although, we are not in the perturbational limit for the strongest energy splittings, these results nevertheless suggest a strong effect of rotational misalignment on current flow. FIG. 4. Bar chart of interlayer energy splitting as function of twist angle (𝜃) for graphene/SL- hBN/graphene heterostructures on a linear scale. 6 TABLE I: Variation of interlayer energy splitting as function of commensurate rotation angle for graphene/SL-hBN/graphene, along with number of atoms per unit cell and effect on current in the perturbative limit. Rotation Angle (degrees) Degree of Band Splitting (meV) No. of atoms in Supercell Relative Current Drop (w.r.t. aligned case) Aligned 21.79° 13.17° 9.43° 69.1 23.4 13.8 9.51 6 42 114 222 -- 8.72 25.1 52.8 We then heterostructures considered SL-graphene/ML-hBN/SL-graphene for graphene/hBN interfaces of fixed rotation angle 𝜃 = 21.79°, corresponding to the smallest non- rotationally-aligned supercell from Eq. (2), and varied the number of hBN layers from 1 to 3. After relaxation of the supercell structures, the energy degeneracy splittings were computed and compared with the corresponding values for the aligned systems, and the effect on current in the perturbational limit was estimated, as shown in Table II. Although the reduction in energy degeneracy with increasing number of hBN layers is not entirely logarithmic for the aligned or misaligned systems (Figure 5)-perhaps not unexpectedly with potentially complex tunneling paths among atomic orbitals varying with the number of hBN layers within these atomistic systems-there is no evidence that increasing the number of layers would reduce the tunneling current more rapidly for a misaligned system. For this limited set of data, the effects of increasing tunnel barrier thickness are comparable to or, indeed, even weaker for the misaligned system. That is, to the extent that one may approximate interlayer current flow in the form I ≈ ηe−κt for the sake of discussion, the results of these simulations, Tables I and II and Figures 4 and 5, are consistent with significant reductions in the lead coefficient η and not with reduction in the evanescence coefficient κ with increasing unit cell size. 7 FIG. 5. Bar chart of interlayer energy splitting as function of tunnel barrier hBN layer number for rotationally aligned (blue) and twist angle 𝜃 = 21.79° graphene heterostructures on a log scale. graphene/hBN/ (red) TABLE II: Variation of interlayer energy splitting as function of layer number for aligned and rotationally misaligned graphene/hBN/graphene and effect on current in the perturbative limit. No. of hBN layers Band Splitting for aligned system (meV) 1 2 3 69.1 20.4 2.24 Band Splitting for 𝜃 = 21.79° Relative Current Drop (w.r.t. aligned case) (meV) 23.4 6.90 1.35 8.72 8.74 2.76 In Figure 6, we also show the band-structure near the K point (nominal Dirac point) for both the rotationally aligned SL-graphene/three-layer-hBN/SL-graphene system and the rotationally misaligned (𝜃 = 21.79°) SL-graphene/three-layer-hBN/SL-graphene system. For the aligned system, in addition to the degeneracy breaking due to inter-graphene-layer coupling, a substantial effect on the band structure is observed near the K-point due to the interaction of the graphene layer with the hBN layers as predicted previously28. However, for the misaligned system, there is little apparent effect on band structure near Dirac point, again suggesting weakened coupling across the graphene-hBN interface in the rotationally misaligned case. 8 FIG. 6. A zoomed-in view of the band structure near the K point (nominal Dirac point) for (a) rotationally aligned SL-graphene/three-layer-hBN/SL-graphene system and (b) rotationally misaligned (θ=21.79°) SL-graphene/three-layer-hBN/SL-graphene system. For graphene/SL-hBN/graphene system with misaligned hBN layer between mutually aligned graphene layers, we repeated the simulations with GGA exchange-correlation functional for comparison with the LDA results. The same relaxed crystal structure was used, but separate self-consistent field and band structure simulations were performed using the GGA. Figure 7 shows the band structure and interlayer energy splitting near the Dirac point for graphene/SL- hBN/graphene with hBN layer rotated at an angle 𝜃 = 21.79°, obtained using both the GGA and LDA functionals. As can be seen, the degree of band splitting matches closely. The interlayer energy splitting as function of twist angle for graphene/SL-hBN/graphene calculated using LDA and GGA have been listed in Table III for comparison. The energy splitting obtained using GGA functional is a few meV smaller than the corresponding LDA ones for all three cases. TABLE III: Comparison of interlayer energy splitting obtained using LDA and GGA functionals, as a function of commensurate rotation angle for graphene/SL-hBN/graphene. Rotation Angle Band Splitting from LDA Band Splitting from GGA (degrees) Aligned 21.79° 13.17° (meV) 69.1 23.4 13.8 9 (meV) 64.7 20.6 10.4 FIG. 7. (a) Band structure of SL-graphene/SL-hBN/SL-graphene with hBN layer rotated at 𝜃 = 21.79°, obtained using GGA (black solid lines) and LDA (red dashed lines) exchange-correlation functionals. (b) The zoomed-in interlayer energy splitting near the Dirac point matches closely for both GGA and LDA. Finally, we point out that as the rotation angle decreases, the interlayer coupling and the expected tunneling current does not tend to the value of aligned heterostructure in these calculations, but decreases instead as the unit cell size increases. In practice, for small rotation angles or, more generally, larger unit cell sizes beyond our capabilities to simulate, loss of coherence on unit cells dimensions or other causes would likely lead to saturation of the interlayer current density, which we expect is the experimentally observed condition. Our results only suggest that rotation leads to lower than otherwise expected interlayer tunneling currents, without addressing the limit of this reduction. Moreover, for sufficiently small rotation angles, the physical structure may self-align, at least locally,36 leading back to a strong interlayer coupling and current flow. IV. CONCLUSION In summary, our simulation results suggest that the interlayer tunneling current for graphene/hBN/graphene heterostructures will be affected strongly by the rotational alignment of the hBN interface with respect to the graphene layers. Indeed, for the three commensurate rotation angles considered in our study (21.79°, 13.17° and 9.43°), despite the growth in primitive unit cell size, the expected interlayer current per primitive unit cell for these three rotation angles remained roughly constant and actually decreased somewhat. At some point, as the rotation angle decreases or, more generally, the primitive unit cell size increases beyond our ability to simulate, this trend obviously must fail, perhaps due to loss of coherence on unit cells dimensions and, ultimately, perhaps local formation of commensurately aligned regions separated by non-aligned regions36. However, the essential behavior of interest, reduction of current flow with interlayer rotation 10 should remain. Moreover, we did not see a greater effect on coupling and expected current reduction with increasing number of hBN layers in a misaligned system (21.79°) as compared to the aligned system. That is, to the extent that one may approximate interlayer current flow in the form I ≈ ηe−κt for the sake of discussion, the results of these simulations are consistent with significant reductions in the lead coefficient η and not with reduction in the evanescence coefficient κ with increasing unit cell size. Accordingly, the reduction in expected current appears attributable to, specifically, the coupling across the interface between the channel layer and the tunnel barrier in this atomistic system, consistent with the weak tunneling current observed experimentally for such graphene/hBN/graphene systems7,8. Finally, it is perhaps reasonable to speculate that such effects of rotational misalignment also may be present in other van der Waal's heterostructures not directly simulated here. SUPPLEMENTARY MATERIAL See supplementary material for the complete electronic structures of the studied rotationally misaligned graphene/hBN/graphene heterostructures. ACKNOWLEDGMENTS This work was supported in part through the Southwest Academy of Nanoelectronics (SWAN) by the Nanoelectronic Research Inititiative (NRI), a Semiconductor Research Corporation (SRC) program sponsored by Nano Electronics Research Corporation (NERC) and National Institute of Standards and Technology (NIST). The authors acknowledge the Texas Advanced Computing Center (TACC) at The University of Texas at Austin for providing high performance computing (HPC) resources that have contributed to the research results reported within this paper. REFERENCES 1 D. Jariwala, V.K. Sangwan, L.J. Lauhon, T.J. Marks, and M.C. Hersam, ACS Nano 8, 1102 (2014). 2 K.S. Novoselov, A.K. Geim, S.V. Morozov, D. Jiang, M.I. Katsnelson, I.V. Grigorieva, S.V. Dubonos, and A.A. Firsov, Nature 438, 197 (2005). 3 Y. Zhang, Y.-W. Tan, H.L. Stormer, and P. Kim, Nature 438, 201 (2005). 4 F. Schwierz, Nat. Nanotechnol. 5, 487 (2010). 5 P. Zhao, R.M. Feenstra, G. Gu, and D. Jena, IEEE Trans. Electron Devices 60, 951 (2013). 6 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, Nat. Commun. 4, 1794 (2013). 7 B. Fallahazad, K. Lee, S. Kang, J. Xue, S. Larentis, C. Corbet, K. Kim, H.C.P. Movva, T. Taniguchi, K. Watanabe, L.F. Register, S.K. Banerjee, and E. Tutuc, Nano Lett. 15, 428 (2015). 8 S. Kang, B. Fallahazad, K. Lee, H. Movva, K. Kim, C.M. Corbet, T. Taniguchi, K. Watanabe, L. Colombo, L.F. Register, E. Tutuc, and S.K. Banerjee, IEEE Electron Device Lett. 36, 405 (2015). 9 S.K. Banerjee, L.F. Register, E. Tutuc, D. Reddy, and A.H. MacDonald, IEEE Electron Device Lett. 30, 158 (2009). 10 D. Basu, L.F. Register, D. Reddy, A.H. MacDonald, and S.K. Banerjee, Phys. Rev. B 82, 75409 (2010). 11 11 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, Nat. Nanotechnol. 5, 722 (2010). 12 A.K. Geim and I.V. Grigorieva, Nature 499, 419 (2013). 13 L. Britnell, R.V. Gorbachev, R. Jalil, B.D. Belle, F. Schedin, A. Mishchenko, T. Georgiou, M.I. Katsnelson, L. Eaves, S.V. Morozov, N.M.R. Peres, J. Leist, A.K. Geim, K.S. Novoselov, and L.A. Ponomarenko, Science 335, 947 (2012). 14 R.M. Feenstra, D. Jena, and G. Gu, J. Appl. Phys. 111, 43711 (2012). 15 A. Mishchenko, J.S. Tu, Y. Cao, R.V. Gorbachev, J.R. Wallbank, M.T. Greenaway, V.E. Morozov, S.V. Morozov, M.J. Zhu, S.L. Wong, F. Withers, C.R. Woods, Y.-J. Kim, K. Watanabe, T. Taniguchi, E.E. Vdovin, O. Makarovsky, T.M. Fromhold, V.I. Fal'ko, A.K. Geim, L. Eaves, and K.S. Novoselov, Nat. Nanotechnol. 9, 808 (2014). 16 J.M.B. Lopes dos Santos, N.M.R. Peres, and A.H. Castro Neto, Phys. Rev. Lett. 99, 256802 (2007). 17 S. Shallcross, S. Sharma, E. Kandelaki, and O.A. Pankratov, Phys. Rev. B 81, 165105 (2010). 18 A.R. Muniz and D. Maroudas, Phys. Rev. B 86, 75404 (2012). 19 D.G. Kvashnin, S. Bellucci, and L.A. Chernozatonskii, Phys. Chem. Chem. Phys. 17, 4354 (2015). 20 J. Jung, A. Raoux, Z. Qiao, and A.H. MacDonald, Phys. Rev. B 89, 205414 (2014). 21 D. Reddy, L.F. Register, and S.K. Banerjee, in Device Res. Conf. DRC 2012 70th Annu. (2012), pp. 73–74. 22 J. Bardeen, Phys. Rev. Lett. 6, 57 (1961). 23 S.C. de la Barrera, Q. Gao, and R.M. Feenstra, J. Vac. Sci. Technol. B 32, 04E101 (2014). 24 G. Kresse and J. Furthmüller, Phys. Rev. B 54, 11169 (1996). 25 G. Kresse and J. Furthmüller, Comput. Mater. Sci. 6, 15 (1996). 26 Virtual NanoLab version 2015.1, QuantumWise A/S (www.quantumwise.com). 27 A. Kokalj, Comput. Mater. Sci. 28, 155 (2003). 28 G. Giovannetti, P.A. Khomyakov, G. Brocks, P.J. Kelly, and J. van den Brink, Phys. Rev. B 76, 73103 (2007). 29 J.P. Perdew and Y. Wang, Phys. Rev. B 45, 13244 (1992). 30 J.P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). 31 A.V. Krukau, O.A. Vydrov, A.F. Izmaylov, and G.E. Scuseria, J. Chem. Phys. 125, 224106 (2006). 32 A. Rai, A. Valsaraj, H.C.P. Movva, A. Roy, R. Ghosh, S. Sonde, S. Kang, J. Chang, T. Trivedi, R. Dey, S. Guchhait, S. Larentis, L.F. Register, E. Tutuc, and S.K. Banerjee, Nano Lett. 15, 4329 (2015). 33 A. Valsaraj, J. Chang, A. Rai, L.F. Register, and S.K. Banerjee, 2D Mater. 2, 45009 (2015). 34 S. Grimme, J. Antony, S. Ehrlich, and H. Krieg, J. Chem. Phys. 132, 154104 (2010). 35 K.F. Mak, M.Y. Sfeir, J.A. Misewich, and T.F. Heinz, Proc. Natl. Acad. Sci. 107, 14999 (2010). 36 C.R. Woods, L. Britnell, A. Eckmann, R.S. Ma, J.C. Lu, H.M. Guo, X. Lin, G.L. Yu, Y. Cao, R.V. Gorbachev, A.V. Kretinin, J. Park, L.A. Ponomarenko, M.I. Katsnelson, Y.N. Gornostyrev, K. Watanabe, T. Taniguchi, C. Casiraghi, H.-J. Gao, A.K. Geim, and K.S. Novoselov, Nat. Phys. 10, 451 (2014). 12
1610.03621	2	1610	2017-01-04T17:22:59	Nuclear Quantum-Assisted Magnetometer	[ "cond-mat.mes-hall", "quant-ph" ]	Magnetic sensing and imaging instruments are important tools in biological and material sciences. There is an increasing demand for attaining higher sensitivity and spatial resolution, with implementations using a single qubit offering potential improvements in both directions. In this article we describe a scanning magnetometer based on the nitrogen-vacancy center in diamond as the sensor. By means of a quantum-assisted readout scheme together with advances in photon collection efficiency, our device exhibits an enhancement in signal to noise ratio of close to an order of magnitude compared to the standard fluorescence readout of the nitrogen-vacancy center. This is demonstrated by comparing non-assisted and assisted methods in a $T_1$ relaxation time measurement.	cond-mat.mes-hall	cond-mat	Nuclear Quantum-Assisted Magnetometer Thomas Haberle,1 Thomas Oeckinghaus,1 Dominik Schmid-Lorch,1 Matthias Pfender,1 Felipe F´avaro de Oliveira,1 Seyed Ali Momenzadeh,1 Amit Finkler,1, a) and Jorg Wrachtrup1, 2 1)3. Physikalisches Institut, Universitat Stuttgart, Pfaffenwaldring 57, 70569 Stuttgart, Germany 2)Max Planck Institute for Solid State Research, Heisenbergstr. 1, 70569 Stuttgart, Germany (Dated: 6 March 2018) Magnetic sensing and imaging instruments are important tools in biological and material sciences. There is an increasing demand for attaining higher sensitivity and spatial resolution, with implementations using a single qubit offering potential improvements in both directions. In this article we describe a scanning magnetometer based on the nitrogen-vacancy center in diamond as the sensor. By means of a quantum-assisted readout scheme together with advances in photon collection efficiency, our device exhibits an enhancement in signal to noise ratio of close to an order of magnitude compared to the standard fluorescence readout of the nitrogen- vacancy center. This is demonstrated by comparing non-assisted and assisted methods in a T1 relaxation time measurement. I. INTRODUCTION Spin defects in diamond are novel nanoscale sensors. While the size of a single defect guarantees nanometer spatial resolution in scanning probe arrangements, read- ing out the signal of single defects may significantly in- crease acquisition times. The quality of the experimental readout is measured by its signal to noise ratio (SNR). Since the SNR is primarily photon shot-noise limited, it can be improved by an in- creased fluorescence collection efficiency. With quantum- assisted readout schemes, however, it is possible to en- hance the SNR even further. In this paper we present a quantum metrology and imaging setup, which was opti- mized for achieving the highest SNR by employing such a quantum-assisted readout scheme in combination with improved collection efficiency. We use the negatively- charged nitrogen-vacancy (NV) center in diamond, which is nowadays one of the most prominent candidates for quantum applications, especially at ambient conditions.1 At a finite magnetic field, the NV can be considered as a single qubit, which can be initialized and read-out op- tically, while its spin state can be manipulated with the use of microwave-frequency irradiation. Its key signature is the ability to detect its spin-state by measuring its flu- orescence intensity - further on abbreviated as optically detected magnetic resonance (ODMR).2 For a detailed review of the NV center properties, we refer the reader to Ref. 3. NV ensembles reach magnetic field sensitivities as high as pT/√Hz,4 and single NV centers have been demon- strated to sense a single electron spin at a distance of 50 nm (see Ref. 5). Even the detection of single nuclear spins seems to be within reach.6,7 Moreover, single defect center scanning probe microscopy has reached resolutions in the 10-20 nm range.8–16 Despite such remarkable applications, long integration times are still necessary to reach satisfactory SNRs. In spectroscopy applications it is possible to run measure- ments over several days, since probe and sample form a mechanically static system. However, such readout meth- ods are quite tenuous, inefficient and prevent the observa- tion of even slow dynamics. This is especially critical in the case of scanning probe applications, where it becomes difficult to achieve long-term mechanical stability since the sample and probe are scanned with respect to each other. Additionally, since the measurement has to be re- peated for each pixel, frame acquisition times increase quadratically with increasing pixel density, limiting the resolution for any given scan range. It is thus of paramount importance to improve SNR. Various ideas ranging from enhanced photon collection efficiencies by nano-engineered waveguides17,18 to an im- provement of the readout fidelity of the spin state itself were explored. The latter approach was demonstrated e.g. via charge state conversion19 or nuclear spin-assisted readout schemes such as repetitive readout20 and single shot readout (SSR).21,22 Here, we chose to pursue SSR to improve the readout fidelity, which also allows the use of more sophisticated schemes like quantum memories or error correction. This article consists of three main sections. First, in Sec. II, we shed light on how SNR is improved via SSR. Then, in Sec. III we illustrate the different enhancements and modifications we have made to a commercial plat- form in order to achieve this SNR enhancement. Finally, in Sec. IV, we present results exhibiting a gain of 8.6 in SNR. II. NUCLEAR QUANTUM-ASSISTED ELECTRON SPIN READOUT a)Author to whom correspondence Email: [email protected] should be addressed. NV center based measurements start with the initial- ization of the spin followed by radio frequency control fields to render the NV center's electron spin sensitive to a certain physical quantity. This is then followed by the actual spin readout. Conventionally, the NV center is read out optically by applying a laser pulse. However, the resulting flu- orescence typically yields less than 0.1 significant pho- tons per readout. Thus, since the spin state is destroyed during readout, the whole measurement scheme has to be repeated several hundred thousand times to build up enough statistics. Since a single measurement run can easily take several milliseconds, this renders the measure- ment scheme rather inefficient. Consequently, several approaches have been developed to circumvent this problem. The general idea is to use a different spin readout mechanism which does not de- stroy the spin state, a so-called quantum non-demolition (QND) measurement.23 The readout of a single measure- ment run can then be repeated several times, thus en- hancing the SNR. For example, at low temperature (T < 20 K, see Ref. 24) it is possible to resonantly excite the different sub-levels of the NV center to read out the spin state quite efficiently.25 At ambient conditions, however, it is necessary to use auxiliary memories with longer life time. One way to implement this is to read out the state of the NV center via spin-to-charge conversion.19,26 Spin state dependent ionization and subsequent charge state depen- dent readout allows measurement of the spin state to a certainty only a factor of two away from the spin projec- tion noise level.19 Technically, however, this approach is rather challenging, since the use of lasers with three dif- ferent wavelengths becomes necessary to perform charge conversion and readout. Another alternative is to swap the spin state to a sec- ond spin which can be read out with a QND measure- ment. Here, the number of readout cycles is ultimately limited by the lifetime of the second spin. Therefore, nu- clear spins are prime candidates, since they feature longer lifetimes than electron spins. The NV center has a built- in memory in the form of the nitrogen nucleus, which has a nuclear spin of either I = 1 for 14N or I = 1/2 for 15N. Another candidate would be a 13C isotope (I = 1/2) in the diamond lattice. However, the proportion of NV cen- ters with a 13C nuclear spin at a suitable position is not very high. Two variations of nuclear spin-enhanced readout were demonstrated in the past. The so-called repetitive readout20 and the single shot readout (SSR)22 employed in this work. It is important to note that both meth- ods are projection measurements, i.e. the measurement still needs to be repeated to determine the actual spin state. However, instead of several hundred of thousand repetitions, only a few thousand repetitions suffice. As a figure of merit for the quality of the SSR scheme we use the nuclear spin readout fidelity, F , which reflects the probability, that two subsequent nuclear spin read- outs yield the same result.27 There are two reasons for this not to be the case: Either the nuclear spin flipped of its own accord, or the measurement of the nuclear 2 spin state yielded the wrong result. The probability for a random nuclear spin flip depends on the nuclear spin life- time, and is therefore less likely for short readout times. The correct readout of the nuclear spin state, however, improves with the number of collected photons, which can be increased by a longer readout time. Thus we have two competing mechanisms leading to an optimal read- out time with the highest SSR fidelity (see section IV B). One needs to keep in mind, that the aforementioned nu- clear spin lifetime is the lifetime under laser illumination of the NV center and should not be confused with the nuclear spin lifetime in the dark, which is considerably longer.22,28 In order to improve the lifetime of the nuclear spin it needs to be decoupled from the electron spin. It is there- for necessary to detune the NV center as far as possible from the excited state level anticrossing (at ∼ 50 mT), by applying a strong magnetic bias field.22 However, the increase in nuclear spin lifetime only prevails, if the ap- plied magnetic field is well aligned along the axis of the NV center. Reaching the necessary quality of alignment and stability of the magnetic field poses the true chal- lenge if one wants to establish SSR. The design of the magnet and its thermal stabilization are described in full detail in section III. ) % ( y t i l i e d F R S S 95 90 85 80 75 70 65 60 SNRSSR/SNRStd 6.0 4.0 2.0 1.0 0.2 0.4 0.6 0.8 Sensing Time T (ms) 8.0 4.0 2.0 1.0 0.8 0.4 0.2 0.1 Theoretically predicted improvement FIG. 1. factor of the SSR-assisted SNR vs. the standard SNR (Std), SNRSSR/SNRStd, as a function of SSR fidelity and sensing time T . Isolines of the improvement factor are displayed as guides to the eye. All parameters that are not varied here were taken from the real experiment. The potential improvement of the SNR using SSR was estimated from theory and is displayed in Fig. 1 (full de- tail derivation is given in section VI). Figure 1 shows the relative improvement in SNR between a SSR and a non- SSR measurement as a function of initial SSR fidelity and sensing time. Since SSR has a non-trivial overhead com- pared to the standard readout, it is less efficient for short sensing times but thrives for longer sensing times. From the simulation it becomes apparent, that with an SSR fidelity of e.g. 90 %, the break-even-point would already be reached for a sensing time as short as 30 µs. As shown in section IV, we actually achieve SSR fi- delity higher than 90 %. However, to reach this goal we further need to enhance the photon collection efficiency in addition to improving the magnetic field. This feat was achieved by nano-engineering the dia- mond sample into efficient waveguides in the form of ta- pered nanopillars.17,18 This way, the collection efficiency is typically improved by a factor of up to ten or more compared to unstructured diamond. Therefore, this im- provement alone boosts the SNR by a factor of about three. Combined with SSR, however, it allows us to re- duce the number of readout cycles, which effectively re- duces the need for overly long nuclear spin lifetimes and this finally gives us the high fidelity we are aiming for. III. SYSTEM DESIGN confocal and scanning force microscope The room temperature microscope is built around (at- a toCSFM), combining a confocal fluorescence microscope (CFM) and an atomic force microscope (AFM) from attocube systems AG.29 The attoCSFM has optical access for the confocal part, including a standard objective holder set onto a slip-stick positioner in z direction for the focus. The AFM part is a combination of two positioning and scanning towers. Each tower has an xyz positioning stack of slip stick positioners with 15 mm travelling range and an attached xyz piezo open-loop scanner with 21 × 21 µm2 scan range in x and y and 7 µm in z. One tower is used to hold the diamond sample, while the other is scanning the AFM tip. The AFM's interferometric readout of the cantilever position was replaced with a tuning-fork based readout system, which allows scanning in non-contact mode. In addition, the tuning-fork signal was converted using a home-built transimpedance amplifier, realized with an operational amplifier following the approach of Ref. 30. A third xyz positioning stack is used to position and align a permanent magnet for the creation of the external bias field. The microscope is encased in a housing that is airtight and can support exchange atmosphere or a medium vacuum if desired. The microscope is computer controlled, mostly via a self written software, developed using the Python programming language.31 Full access to the source code allows a flexible working environment with a high degree of automation. A. External Magnetic Field As mentioned earlier, a large magnetic bias field and proper alignment of said field is of central importance to establish efficient SSR. 3 In the presented setup we use a permanent magnet manufactured from NdFeB, since they are commercially available32 and can generate remanent fields as high as 1.3 T. Also, such a magnet does not introduce any ad- ditional heat load due to the presence of an induction current, as in the case for electromagnets. The geometry of the magnet was designed carefully in order to reflect the following aspects. We employ a (100) diamond sample, i.e. the NV center's z-axis is tilted relative to the diamond surface by 54.75°, along which the magnetic field needs to be aligned. The magnet has to fit into the device and leave enough space above the diamond sample to accommodate the tuning fork AFM head for scanning probe measurements. The gradient of the magnetic field should be small both to facilitate alignment and to be less sensitive to drift. At the same time the absolute magnetic field at the site of the NV center should be as large as possible. The final configuration of the magnet is displayed schematically in Fig. 2a, together with the diamond sample membrane and the tuning fork used in scan- ning probe. The magnet is put together from two 10 × 10 × 10 mm3 cubes and four 10 × 2 × 2 mm3 rods. This design was chosen according to a finite-element- analysis of the stray magnetic field, shown in Fig. 2b. FIG. 2. Strong permanent magnet in a scanning probe micro- scope setup. (a) In the middle of the picture is the diamond membrane with nanopillars (everything except the nanopil- lars is to scale) resting on the microwave stripline above the objective. The tuning-fork holder with preamplifier circuit can be seen on the right. The magnet formed by two 1 cm cube permanent magnets and four smaller rods is supported and approaches the diamond from the left. (b) Close-up of a z-x cross-section of a finite-element analysis for the stray magnetic field from the magnet displayed in (a), area indi- cated by dotted rectangle. In the upper left corner the very end of the magnet is indicated. The magnetic field strength is only plotted in regions, where the alignment on the NV center's axis is better than 99 %. With this magnet design aligned fields as high as 0.7 T can be achieved. In order to align the magnetic field, the permanent magnet is mounted onto an xyz slip stick positioner from attocube. The first step in alignment is to scan the mag- net in the xy-plane above the NV center while recording the fluorescence. Misalignment affects the electron spin as it causes mixing of the electron spin sub-levels of the NV center and thus reduces the fluorescence count-rate.33 After this coarse alignment, the xy-scan is repeated in a focus area of highest fluorescence, with the SSR fidelity being evaluated at every pixel. The maximum is then chosen as the final position of the magnet (see Fig. 6b). The absolute strength of the magnetic field is then de- termined by the distance between the magnet and the diamond sample. With the additional constraint of the xy-alignment, this means the previously chosen z-value defines the final magnetic field strength. This way, we can reach values of up to 0.7 T. Once the magnet is brought into proper position and provided that the system has a good temperature sta- bility, the magnetic field strength and alignment are re- markably stable (see Fig. 3). In the next section we show how the necessary stability of our system was realized. B. Temperature Stabilization Drift poses a problem for most cutting-edge mi- croscopy systems, especially at ambient conditions. Our system combines two microscopy techniques, namely con- focal and scanning probe microscopy. The as-purchased instrument is constructed exclusively from titanium, which is non-magnetic and features a relatively low ther- mal expansion coefficient of around 8.6 × 10−6 K−1.34 It is further stabilized by keeping the inside of the micro- scope at a slightly elevated temperature relative to the environment via a proportional-integral (PI) feedback loop. This way drift-rates of less than 5 nm/h measured by the AFM were achieved, which is good enough from the microscopy point of view. However, the stability of the transition frequency of the NV center is not sufficient for SSR, where the two hyperfine lines of the 15N nucleus need to be addressed individually (see section III D and Fig. 5b). This means that the peak-to-peak frequency drift needs to be smaller than the full width at half maximum (FWHM) of the hyperfine resolving ODMR line of about 0.7 MHz, which was originally not the case (see Fig. 3, green line). It is possible to compensate for the frequency drifts by fre- quent ODMR measurements, but these take time and thereby reduce the achieved SNR. The frequency drift is caused by a change in magnetic field strength at the site of the NV center, which can be due to a drift of the relative position or due to thermally induced changes of the magnetization of the permanent magnet. From the finite-element-analysis simulation, a magnet position yielding a field of 398 mT (value used in experiments) corresponds to a magnetic field gradi- ent of 0.15 mT/µm. The above mentioned maximum value of 0.7 MHz in frequency drift would then trans- late into a necessary temperature stability of 0.39 K,35 which is easily realized. The temperature stability of the magnetization, however, is given by 0.11 %/K.36 This value translates into a necessary temperature stability of 0.06 K peak-to-peak for a frequency stability better than 4 0.7 MHz. The most straightforward way to achieve this high level of stability was to add a second temperature stabiliza- tion layer. The microscope was encased in a styrofoam box of a thickness of 60 mm, with a lining of acoustical foam to simultaneously reduce acoustic coupling to the scanning probe system. The breadboard, which carries the microscope inside this second layer, was tempera- ture stabilized by a second PI loop. The setpoint for the breadboard temperature was chosen to lie between the setpoint of the inside of the microscope and the ambient temperature in the laboratory.37 ) z H M ( y c n e u q e r F e v i l a e R 1.5 1.0 0.5 0.0 −0.5 −1.0 −1.5 Single Layer Double Layer 0 2 4 6 Time (h) 8 10 FIG. 3. Drift of the NV center's transition frequency as a function of time with single layer (green) and double layer (red) thermal stabilization. With the latter, the frequency shifts are reduced to about 0.5 MHz peak-to-peak in the course of 10 h. With both layers of the temperature control active, the stability in the magnetic field and therefore the sta- bility of the NV center's ODMR transition frequencies is improved to fluctuations of about 0.5 MHz peak-to-peak in the course of 10 h, as demonstrated in Fig. 3 (corre- sponding to 18 µT or temperature changes of the magnet of approximately 0.04 K). C. Spin manipulation setup The NV center's electron spin is manipulated via mi- crowave (MW) radiation in the GHz regime, while the 15N nuclear spin is manipulated with radiofrequency (RF) fields in the MHz regime (γ15N = −4.3 MHz/T, γe− = 28 GHz/T). The spin manipulation setup can be separated into five categories as color-indicated in Fig. 4. The control part (orange) is comprised of the primary control PC, which is used to set the frequencies of the RF and MW sources and analyze the readout data from a field-programmable gate array (FPGA) used for data acquisition (red). Ad- ditionally, it feeds the Arbitrary Waveform Generator (AWG) (Tektronix AWG520) with the desired measure- ment pulse sequence. The AWG falls into two categories, on the one hand it is the first stage of the MW path (blue) which synthesizes the MW pulses. On the other hand the AWG is the central device that synchronizes the different paths of the measurement system. For this purpose it is sending triggers to the RF function generator in the RF path (cyan), to the data acquisition FPGA and to the optics via the acousto-optic modulator (AOM) (green) to pulse the laser. The optics path itself consists of a standard confocal setup. PC Control RF FG FPGA AWG Data Acq. AOM Optics RF MW FG MW RF path MW path Diamond 50 Ω Experiment FIG. 4. Diagram of the spin manipulation setup, separated into five subcategories. Control (orange), Radio-Frequency path (cyan), Data Acquisition (red), Microwave path (blue), Optics (green) and finally the actual Experiment at the di- amond sample (purple). For a detailed description see main text. The MW path is comprised of the AWG giving out two 90° phase shifted waveforms in the 100 MHz range, which are fed into an IQ mixer (Marki MLIQ-0218L) in single sideband modulation mode, together with the high frequency carrier in the GHz range, which is generated by a R&S SMR20 signal generator. The modulated high frequency signal is finally sent to a TriQuint RM022020 Spatium amplifier. The RF path begins with a Rigol DG1032Z as the RF function generator, which is gated by the AWG. The RF pulses are then amplified with a Mini-Circuits LZY-22+ amplifier. Subsequently they are sent through a home built 1 MHz high pass filter to prevent previously ob- served low frequency ringing. Finally, the RF signal is combined with the MW sig- nal via an AMTI Microwave Circuits D1G018G3 diplexer before being sent to the diamond sample via a coplanar 5 waveguide stripline (purple). To prevent reflections of the transmitted power, the system is terminated with 50 Ω. Data acquisition is done via a TimeTagger38 based on an FPGA, which receives the photon clicks from the avalanche photodiodes (APDs) and all necessary triggers from the AWG to bin them accordingly in a time-resolved fashion. This data is then provided to the computer (PC) for analysis. D. SSR Implementation With the capability to manipulate both the electron spin and the nuclear spin, it is now possible to imple- ment SSR-assisted measurements. Figure 5 depicts such a measurement sequence, which consists of three parts: 1. During the sensing part of length T , an arbitrary measurement sequence is run on the NV center's electron spin (e.g. a Hahn echo sequence or a re- laxometry sequence.39). 2. In a non-SSR assisted measurement this would be followed by optical readout and destruction of the electron spin state. In the SSR-assisted case the spin information is in- stead transferred onto the nuclear spin via a con- trolled NOT (CeNOTn) gate in the form of an elec- tron spin selective π-pulse on the nuclear spin.40 3. Finally, the nuclear spin state is read out via the actual SSR. The readout of the nuclear spin state via SSR is real- ized with a CnNOTe gate. Such a gate is a nuclear spin state selective π pulse - it flips the electron spin condi- tional on the nuclear spin state. The CnNOTe gate is then followed by a laser pulse to read out and repolar- ize the electron spin. The recorded count-rate is reduced if the nuclear spin resides in the probed state, and left untouched otherwise. This change in count-rate can be used to determine the nuclear spin state.22 However, the readout relies heavily on a stable underlying count-rate, which is not always the case. Specifically, with surface- sensitive shallow NV centers the count-rate tends to fluc- tuate and change over time.41 To circumvent this prob- lem, we probe both nuclear spin states virtually in a si- multaenous manner by alternating the MW frequency be- tween the two hyperfine lines (MW1 and MW2 in Fig. 5). This readout is repeated N times with the photon num- bers being counted separately for the two transitions and finally subtracted from each other. The sign of the re- sult reflects the nuclear spin state. Since fluctuations in the count-rate affect both transitions in the same way, they will cancel out during the subtraction, making the readout more robust. PSfrag replacements Laser MW1 MW2 MW RF (a) (b) π ( N ( π π 2 π π π { Sensing ≡ T {{ Transfer SSR FIG. 5. (a) SSR assisted measurement scheme. The measure- ment is divided into three parts, consisting of the Sensing part, conducted on the electron spin during the sensing time T us- ing strong, broad microwave pulses (MW). This is followed by the Transfer of the electron spin state onto the nuclear spin via a CNOT gate realized by a radio-frequency (RF) pulse on the nuclear spin. Finally the SSR of the nuclear spin is conducted via N repetitions of CNOT gates on the electron spin (soft, therefore narrow MW1 and MW2 pulses, see also (b)) with subsequent readout laser pulse. The readout is con- ducted on both 15N transitions, for robustness reasons. (b) Typical ODMR of the ms = 0i → ms = −1i transition showing the hyperfine splitting due to the 15N nucleus. The two probing frequencies MW1 and MW2 are indicated. IV. PERFORMANCE All of the following measurements were conducted at a magnetic field of 398 mT, which corresponds to a transi- tion frequency between the ms = 0i and ms = −1i sub levels of 8.274 GHz. The magnetic field was aligned prior to the measurements via fluorescence and SSR fidelity, as described in section III A. A. Enhanced Photon Collection with Nanopillars A single NV center in a standard diamond membrane with shallow implantation yields in our case a count-rate of around 150 kHz.42 We compare this to the count-rate from a single NV center situated inside a tapered nanopillar, which showed a photon flux of more than 760 kHz. This is an improve- ment of a factor of about 5 in count-rate, which corre- sponds to a two-fold improvement in the SNR. 6 B. Single Shot Readout The MW power applied during the CnNOTe was cho- sen to result in a π pulse length of about 1 µs, which is a good value to have a high selectivity of the gate within a reasonable time. The length of the laser pulses for the SSR was chosen to be 200 ns following Ref. 22, since it is a good trade-off between high contrast in electron spin readout and re-polarization rate. Additionally, a short laser illumination is also of benefit concerning the nu- clear spin lifetime, because the lifetime decreases under laser illumination. The number of SSR repetitions, N , was optimized ex- perimentally by recording the SSR fidelity while sweeping the repetitions. Figure 6a displays the SSR fidelity as a function of SSR repetitions with a maximum of F = 92 % at about N = 400. (a) ) % ( y t i l e d F i 95 90 85 80 75 70 0.0 0.5 1.0 1.5 2.0 103 SSR Repetitions (µ) ) m µ ( y 160 120 80 40 0 0 40 80 120160 x (µm) 92 88 84 80 76 ) % ( y t i l e d F i FIG. 6. Optimization of single-shot readout scheme. (a) Fi- delity as a function of the number of repetitions (laser pulses) in an SSR pulse sequence. The maximum fidelity achieved in the setup was 92 %. (b) Magnetic field alignment using the SSR fidelity. With the optimal number of repetitions, a second, finer magnetic field alignment based on SSR fidelity was per- formed (see Fig. 6b). Once aligned, small frequency de- viations can be corrected by "refocusing", i.e. measuring the ODMR frequency again. C. Benchmark Experiment For benchmark purposes, we compare measurements with and without the aid of SSR. Since SSR works in principle with any measurement scheme conducted on the NV center, we chose the relaxometry or T1 measure- ment scheme, which is simple to implement and inter- pret. Additionally it is operating well within the regime of long sensing times, where the SSR can demonstrate its capabilities. The T1 decay was measured by letting the polarized electron spin of the NV center evolve during different free evolution times, τ , followed by the readout. In our ex- ample the values for τ were chosen to range from 1 µs up to 50 ms with a logarithmic spacing. Every value is mea- sured twice, with the second measurement being followed by a MW π pulse, which inverts the spin state. This in- version is used as a reference measurement to compensate for charge state effects and count rate fluctuations.14 The full measurement scheme is conducted twice, once with the standard readout (simple laser pulse) and once with SSR (composed of the RF pulse and the actual SSR). To compare the measurements, each run was inte- grated for the same amount of time. Figure 7 displays the results for an integration time of 1 h. The non-SSR measurement managed to complete 11200 sweeps during the integration time, the SSR mea- surement only 8690. However, since the SSR is much more efficient, the result in Fig. 7 clearly shows an im- proved SNR for SSR compared to standard readout. The results were fitted with an exponential function to deter- mine the characteristic T1 decay time and the SNR of the measurement in a quantitative way. (a) t s a r t n o C d e z i l a m r o N 1.2 1.0 0.8 0.6 0.4 0.2 0.0 (µ) R N S 3000 2500 2000 1500 1000 500 0 SSR Non-SSR 10-3 10-2 10-1 100 101 τ (ms) SSR Non-SSR 23 µs 10-3 10-2 10-1 100 101 τ (ms) FIG. 7. (a) Comparison between a T1 measurement with SSR (blue) and with standard readout (red) for an integration time of 1 h. Both datasets were fitted with an exponential function. The standard deviation between the fit and the data was de- termined and is plotted as the lightly shaded uncertainty re- gions, standard readout, red, SSR, blue. (b) Comparison of SNR as a function of τ for the experimental parameters, de- rived from theory. The break-even point where SSR becomes more efficient is reached for τ = 23 µs (green line). Table I summarizes the results from the two measure- ments. The estimated T1 time is practically the same, especially if one takes the errors into account. The more interesting aspect is the difference in SNR for the two 7 measurements, which is a factor of 8.6 larger in the SSR measurement. This is also in a reasonable agreement with theory (see Sections VI A and VI B). The SNR of the measurement is even slightly higher than what the theory predicts. TABLE I. Results from the T1 measurement data analysis for an integration time of 1 h. Both measurement schemes yield consistent T1 times, but the SSR aided scheme has a factor 8.6 higher SNR. This also fits reasonably well to the estimate from theory, both in the individual SNR values and in the improvement factor of 6.8. T1 SNR SNR from Theory Non-SSR (3.8 ± 1.0) ms SSR (3.5 ± 0.1) ms 9.4 10.3 80.5 70.3 As previously mentioned, the SSR measurement is more efficient than the standard readout for longer sens- ing times, due to its overhead. With our experimental parameters the break-even point can be assessed to a sensing time of τ = 23 µs (see Fig. 7b). In our case, for any sensing time longer than that, SSR will perform bet- ter than the standard readout. D. Performance Summary In summary, a significant gain is obtained both via the SSR-assisted readout and the tapered nanopillars. From the nanopillars we get an improvement of the SNR by a factor of 2.3. In the exemplary T1 measurement we showed a further improvement via SSR of a factor of 8.6. In combination, this means that compared to a non-pillar sample operated with standard readout, we have an SNR which is a factor of 19.3 higher. In terms of acquisition time this corresponds to a reduction of a factor of 373 to reach the same SNR. In other words, the one hour integrated SSR measurement shown in Fig. 7 would have taken more than 15 days on a non-enhanced setup to reach the same SNR. We would like to point out, that the two improvements are not fully independent of each other, as the SSR strongly benefits from a high count- rate, which allows a higher fidelity in the first place. V. OUTLOOK The tremendous reduction in acquisition time will make previously time-consuming measurements possible. This is particularly significant in scanning probe applica- tions, where the integration time per pixel reaches up to minutes,14–16,39,43 rendering high-resolution scans practi- cally impossible. Additionally, more information can be acquired in the same time, e.g. full spectra at each pixel instead of just a few data points. The ability to engage SSR and manipulate nuclear spins along with the NV center's electron spin, grants ac- cess to a very interesting hybrid quantum system between sensitive sensor and long-lived memory. This allows cut- ting edge quantum metrology applications, such as nu- clear magnetic resonance (NMR) with a vastly enhanced magnetic field sensitivity as has been recently shown in Ref. 44. These results were extended to other nuclear spins with further improved NMR resolving even chemi- cal shifts, as shown in Ref. 45. With a clear set of guidelines to estimate from which sensing times it is preferable to use SSR, the presented system is now in a state where these techniques can be applied in scanning probe microscopy. VI. ACKNOWLEDGEMENTS We would like to thank Johannes Greiner, Florestan Ziem and Nabeel Aslam for fruitful discussions. We ac- knowledge support from the EU FP7, grant No. 611143 (DIADEMS). F.F.O. acknowledges CNPq for the finan- cial support through the project No. 204246/2013-0. A.F. acknowledges support from the Alexander von Hum- boldt Foundation. APPENDICES A. SNR of standard readout The SNR of the standard measurement is defined by shot noise. To estimate the SNR for given parameters, we are looking at the fluorescence answer of the readout laser pulse. The fluorescence answer of the readout laser pulse is divided into a detection and a reference window, where the detection window is normalized to the refer- ence window. This way the measurement becomes more robust against count-rate fluctuations. The actual signal is then found by repeating the mea- surement with a final π pulse at the end to invert the electron spin state. Thereby, additional artefacts can be prevented, e.g. from a transition to the neutral charge state NV0. With the two detection window counts, n(0,π), and the respective reference window counts, n(0,π),r, the signal, Sstd, is given by: Sstd = n0 n0,r − nπ nπ,r . The error, ∆Sstd can then be calculated by: n0,r ∆Sstd ="(cid:18) 1 n0 n2 0,r nπ,r ∆n0(cid:19)2 ∆n0,r!2 +(cid:18) 1 +(cid:18) nπ n2 + ∆nπ(cid:19)2 ∆nπ,r(cid:19)2  π,r 1/2 . (1) σ =s p (1 − p) Nsweeps . 8 This expression can be simplified, since several of the variables above have the same values, at least to a first approximation. The photon numbers can be simplified to n0,r = nπ,r = n n0 = c0n nπ = cπn, and the errors by ∆n0,r = ∆nπ,r = ∆n = √n ∆n0 = c0∆n ∆nπ = cπ∆n. Here, c0 is the count-rate enhancement of the ms = 0i state and cπ is the enhancement of the ms = −1i state.21 Inserting this into equation (1) yields ∆Sstd =r 2 n The SNR is given by Sstd ∆Sstd : SNRstd =r n 2 (c2 0 + c2 π). . c0 − cπ pc2 0 + c2 π This calculation was also conducted with taking back- ground fluorescence into account. However, the differ- ence in the results was practically non-existent, which is why it was left out to keep the calculation clearer and easier to follow. We can now insert the numbers from the measure- ment displayed in Fig. 7. There we had c0 = 1.12 and cπ = 0.73, n can be calculated from the number of sweeps 11200, the length of the acquisition window 300 ns and the count-rate 760 kHz. leads theoretical value This, then, to of a SNRstd1hour = 10.3. B. SNR of SSR SSR is a projective measurement, where every single measurement yields either nuclear spin up or down, say 0 or 1, respectively. The measurement is therefore fol- lowing a binomial distribution. Once again, two mea- surements, one with inverting π pulse and one without are conducted. The signal for the SSR is given by the difference of the spin non-flip probabilities for the mea- surement, p0, and the inverted measurement pπ: SSSR = p0 − pπ. The standard deviation for Nsweeps repetitions of the SSR is given by: The error ∆SSSR can be derived as: ∆SSSR =s 1 Nsweeps For the SNR this yields SNRSSR = SSSR ∆SSSR = (p0 (1 − p0) + pπ (1 − pπ)). (p0 − pπ)pNsweeps pp0 (1 − p0) + pπ (1 − pπ) . In the measurement the values p0 = 0.85 and pπ = 0.40 were observed. These values agree with Ref. 46. With Nsweeps = 8690, this leads to a theoretical prediction of SNRSSR1hour = 70.3. The estimated improvement factor of the SSR mea- surement compared to the standard measurement is therefore SNRSSR SNRstd = 6.8. This value is even slightly below the measured value of 8.6. 1 shows Figure the SNR improvement factor SNRSSR/SNRstd dependent on the SSR fidelity and the sensing time, as it was derived from theory. For the cal- culation of SNRstd the same values were taken as in sec- tion VI A. For SNRSSR the fidelity value was varied and the contrast was calculated based on the NV center re- siding 30% of the time in the NV0 state.46 For both the number of sweeps was calculated using the varied sensing time. C. Diamond Sample We use commercially available electronic-grade chem- ical vapor deposition (CVD) diamond.47 This diamond is then cut to a thin (∼ 30 µm) membrane,48 implanted at ∼ 5 keV with 15N+ ions and annealed at a temper- ature of about 950 ◦C. Subsequently, an array of a few thousand nanopillars is fabricated using the same recipe as in Ref. 18. This gives us the capability to select an NV center with the correct orientation with respect to the magnetic field (see Sec. III A) while also having a T2 (Hahn) of (108.1 ± 1.4) µs and T1 time of (3.5 ± 0.1) ms. D. Thermal Stability 9 and Jelezko Wrachtrup, 1F. J. physica status solidi (a) 203, 3207 (2006). 2A. Gruber, A. Drabenstedt, C. Tietz, L. Fleury, J. Wrachtrup, and C. v. Borczyskowski, Science 276, 2012 (1997). 3M. W. Doherty, N. B. Manson, P. Delaney, F. Jelezko, J. Wrachtrup, and L. C. Hollenberg, Phys. Rep. 528, 1 (2013). 4T. Wolf, P. Neumann, K. Nakamura, H. Sumiya, T. Ohshima, J. Isoya, and J. Wrachtrup, Phys. Rev. X 5, 041001 (2015). 5M. S. Grinolds, S. Hong, P. Maletinsky, L. Luan, M. D. Lukin, R. L. Walsworth, and A. Yacoby, Nat. Phys. 9, 215 (2013). 6C. Muller, X. Kong, J.-M. Cai, K. Melentijevi´c, A. Stacey, M. Markham, D. Twitchen, J. Isoya, S. Pezzagna, J. Meijer, J. F. Du, M. B. Plenio, B. Naydenov, L. P. McGuinness, and F. Jelezko, Nat. Commun. 5, 4703 (2014). 7A. O. Sushkov, I. Lovchinsky, N. Chisholm, R. L. Walsworth, H. Park, and M. D. Lukin, Phys. Rev. Lett. 113, 197601 (2014). 8B. Berman, J. Appl. Phys. 97, 014903 (2005). 9C. L. Degen, Appl. Phys. Lett. 92, 243111 (2008). 10J. R. Maze, P. L. Stanwix, J. S. Hodges, S. Hong, J. M. Taylor, P. Cappellaro, L. Jiang, M. V. G. Dutt, E. Togan, A. S. Zibrov, A. Yacoby, R. L. Walsworth, and M. D. Lukin, Nature 455, 644 (2008). Chernobrod and M. G. P. 11G. Balasubramanian, I. Y. Chan, R. Kolesov, M. Al-Hmoud, J. Tisler, C. Shin, C. Kim, A. Wojcik, P. R. Hemmer, A. Krueger, T. Hanke, A. Leitenstorfer, R. Bratschitsch, F. Jelezko, and J. Wrachtrup, Nature 455, 648 (2008). 12J. M. Taylor, P. Cappellaro, L. Childress, L. Jiang, D. Budker, P. R. Hemmer, A. Yacoby, R. Walsworth, and M. D. Lukin, Nat. Phys. 4, 810 (2008). 13L. Thiel, D. Rohner, M. Ganzhorn, P. Appel, E. Neu, and P. Maletinsky, B. Muller, R. Kleiner, D. Koelle, Nat. Nanotech. 11, 677 (2016). 14T. Haberle, D. Schmid-Lorch, F. Reinhard, and J. Wrachtrup, Nat. Nanotech. 10, 125 (2015). 15D. Rugar, H. J. Mamin, M. H. Sherwood, M. Kim, C. T. Rettner, K. Ohno, and D. D. Awschalom, Nat. Nanotech. 10, 120 (2015). 16S. J. DeVience, L. M. Pham, I. Lovchinsky, A. O. Sushkov, N. Bar-Gill, C. Belthangady, F. Casola, M. Corbett, H. Zhang, M. Lukin, H. Park, A. Yacoby, and R. L. Walsworth, Nat. Nanotech. 10, 129 (2015). 17B. J. Hausmann, M. Khan, Y. Zhang, T. M. Babinec, K. Mar- and M. Loncar, tinick, M. McCutcheon, P. R. Hemmer, Diamond Relat. Mater. 19, 621 (2010). 18S. A. Momenzadeh, R. J. Stohr, F. F. de Oliveira, A. Brun- ner, A. Denisenko, S. Yang, F. Reinhard, and J. Wrachtrup, Nano Lett. 15, 165 (2014). 19B. J. Shields, Q. P. Unterreithmeier, N. P. de Leon, H. Park, and M. D. Lukin, Phys. Rev. Lett. 114, 136402 (2015). 20L. Jiang, J. S. Hodges, J. R. Maze, P. Maurer, J. M. Taylor, D. G. Cory, P. R. Hemmer, R. L. Walsworth, A. Yacoby, A. S. Zibrov, and M. D. Lukin, Science 326, 267 (2009). 21M. Steiner, P. Neumann, J. Beck, F. Jelezko, and J. Wrachtrup, Phys. Rev. B 81, 035205 (2010). 22P. Neumann, J. Beck, M. Steiner, F. Rempp, H. Fedder, P. R. Hemmer, J. Wrachtrup, and F. Jelezko, Science 329, 542 (2010). Khalili, Braginsky 23V. and Y. B. F. The temperature controller employed in our setup is a Lakeshore Model 335 Cryogenic Temperature Controller. It features two separate, PI controllable channels, each with a heating power of up to 50 W. The controller is configured and read out via a software written in the pro- gramming language Python. This allows logging of the temperature and heating power development over time, which in turn can be used to determine when the system has reached stable operation and measurements can be commenced, usually after 0.5 to 1.5 h. Rev. Mod. Phys. 68, 1 (1996). 24M. L. Goldman, A. Sipahigil, M. W. Doherty, N. Y. S. D. Bennett, M. Markham, D. J. Twitchen, and M. D. Lukin, Yao, N. B. Manson, A. Kubanek, Phys. Rev. Lett. 114, 145502 (2015). 25L. Robledo, L. Childress, H. Bernien, B. Hensen, P. F. A. Alke- made, and R. Hanson, Nature 477, 574 (2011). 26D. A. Hopper, R. R. Grote, A. L. Exarhos, and L. C. Bassett, Phys. Rev. B 94, 241201 (2016). 27This definition slightly differs from previously used definitions,46 but is more straightforward for our purposes. 28M. Pfender, N. Aslam, H. Sumiya, S. Ondoa, P. Neumann, and J. Wrachtrup, arXiv , 1610.05675 J. Isoya, C. Meriles, (2016). 29Attocube - combined atomic force & confocal microscope. 30C. L. Jahncke, O. Brandt, K. E. Fellows, and H. D. Hallen, Rev. Sci. Instrum. 75, 2759 (2004). 31Python Software Foundation. 32HKCM Engineering e.K. 33J.-P. Tetienne, L. Rondin, P. Spinicelli, M. Chipaux, Jacques, J.-F. Roch, and V. T. Debuisschert, New J. Phys. 14, 103033 (2012). 34P. Hidnert, J. Res. Natl. Bur. Stand. 30, 101 (1943). 35Based on 0.7 MHz corresponding to 166 nm position change, which corresponds to a 2.84 × 10−6 relative change in the 5 cm titanium rod holding the magnet. 36B. Ma, J. Herchenroeder, B. Smith, M. Suda, D. Brown, and Z. Chen, J. Magn. Magn. Mater. 239, 418 (2002). 37The temperature stability in the laboratory itself is approxi- mately 0.5 K peak-to-peak. 38Swabian Instruments Time Tagger 20. 39D. Schmid-Lorch, T. Haberle, F. Reinhard, A. Zappe, and J. Wrachtrup, M. Slota, L. Bogani, A. Finkler, 10 Nano Lett. 15, 4942 (2015). 40For the applied field of 398 mT the π-pulse length on the 15N was measured to be 35.5 µs at a resonance frequency of 1.717 MHz. 41C. Bradac, T. Gaebel, N. Naidoo, M. J. Sellars, J. Twamley, L. J. Brown, A. S. Barnard, T. Plakhotnik, A. V. Zvyagin, and J. R. Rabeau, Nat. Nanotech. 5, 345 (2010). 42All count-rates reported in this article were measured at satura- tion. 43M. Pelliccione, B. A. Myers, L. M. A. Pascal, A. Das, and A. C. Bleszynski Jayich, Phys. Rev. Applied 2, 054014 (2014). 44S. Zaiser, T. Rendler, I. Jakobi, T. Wolf, S.-Y. Lee, S. Wag- and ner, V. Bergholm, T. Schulte-Herbrugen, P. Neumann, J. Wrachtrup, Nat. Commun. 7, 12279 (2016). 45N. Aslam, M. Pfender, P. Neumann, R. Reuter, A. Zappe, F. F. de Oliveira, A. Denisenko, J. Isoya, and J. Wrachtrup, manuscript in preparation (2017). J. Beck, M. 46G. Waldherr, Steiner, A. Gali, T. Frauenheim, F. Jelezko, Phys. Rev. Lett. 106, 157601 (2011). 47Element Six, item no. 145-500-0390. 48Applied Diamond, Inc. P. Neumann, and J. Wrachtrup,
1712.03387	1	1712	2017-12-09T14:09:58	Periodic chiral magnetic domains in single-crystal nickel nanowires	[ "cond-mat.mes-hall" ]	We report on experimental and computational investigations of the domain structure of ~0.2 x 0.2 x 8 {\mu}m single-crystal Ni nanowires (NWs). The Ni NWs were grown by a thermal chemical vapor deposition technique that results in highly-oriented single-crystal structures on amorphous SiOx coated Si substrates. Magnetoresistance measurements of the Ni NWs suggest the average magnetization points largely off the NW long axis at zero field. X-ray photoemission electron microscopy images show a well-defined periodic magnetization pattern along the surface of the nanowires with a period of {\lambda} = 250 nm. Finite element micromagnetic simulations reveal that an oscillatory magnetization configuration with a period closely matching experimental observation ({\lambda} = 240 nm) is obtainable at remanence. This magnetization configuration involves a periodic array of alternating chirality vortex domains distributed along the length of the NW. Vortex formation is attributable to the cubic anisotropy of the single crystal Ni NW system and its reduced structural dimensions. The periodic alternating chirality vortex state is a topologically protected metastable state, analogous to an array of 360{\deg} domain walls in a thin strip. Simulations show that other remanent states are also possible, depending on the field history. Effects of material properties and strain on the vortex pattern are investigated. It is shown that at reduced cubic anisotropy vortices are no longer stable, while negative uniaxial anisotropy and magnetoelastic effects in the presence of compressive biaxial strain contribute to vortex formation.	cond-mat.mes-hall	cond-mat	Periodic chiral magnetic domains in single-crystal nickel nanowires Jimmy J. Kan1,, Marko V. Lubarda1,3 and Eric E. Fullerton1 ,, Keith T. Chan1,†, Vojtech Uhlir1, Andreas Scholl2, Vitaliy Lomakin1, 1. Center for Memory and Recording Research, University of California – San Diego, La Jolla, California 92093, USA 2. Advanced Light Source, Lawrence Berkeley National Laboratory (LBNL), 1 Cyclotron Road, Berkeley, California 94720, USA 3. Faculty of Polytechnics, University of Donja Gorica, Oktoih 1, 81000 Podgorica, Montenegro ABSTRACT We report on experimental and computational investigations of the domain structure of ~0.2 x 0.2 x 8 μm single-crystal Ni nanowires (NWs). The Ni NWs were grown by a thermal chemical vapor deposition technique that results in highly-oriented single-crystal structures on amorphous SiOx coated Si substrates. Magnetoresistance measurements of the Ni NWs suggest the average magnetization points largely off the NW long axis at zero field. X-ray photoemission electron microscopy images show a well-defined periodic magnetization pattern along the surface of the nanowires with a period of λ = 250 nm. Finite element micromagnetic simulations reveal that an oscillatory magnetization configuration with a period closely matching experimental observation (λ = 240 nm) is obtainable at remanence. This magnetization configuration involves a periodic array of alternating chirality vortex domains distributed along the length of the NW. Vortex formation is attributable to the cubic anisotropy of the single crystal Ni NW system and its reduced structural dimensions. The periodic alternating chirality vortex state is a topologically protected metastable state, analogous to an array of 360° domain walls in a thin strip. Simulations show that other remanent states are also possible, depending on the field history. Effects of material properties and strain on the vortex pattern are investigated. It is shown that at reduced cubic anisotropy vortices are no longer stable, while negative uniaxial anisotropy and magnetoelastic effects in the presence of compressive biaxial strain contribute to vortex formation. The investigation of magnetism in mesoscale structures has attracted considerable interest in recent years1-6. As the structural dimensions of materials are reduced down to typical length scales associated with ferromagnetic ordering, competitions between several magnetic interactions arise and can result in the formation of new, intricate magnetic configurations7-11. These new structures could offer a pathway towards future applications in high-density data storage12-16, compact magnetic sensors17, 18, high frequency nanoscale oscillators19, 20, and magnetic logic21-24. For example, a competition between the exchange interaction, magnetic anisotropy, stray field energy, and geometric confinement leads to the emergence of vortex structures in magnetic disks25, 26 and stripe domains in thin films27, 28. Magnetic vortices are flux closure states that exist in micron and sub-micron diameter magnetically-soft disks. The vortex state is characterized by the circulation of the magnetization, either counterclockwise or clockwise (c=+1 or -1), and the polarity of the core (the magnetic singularity at the center of the disk), which points either up or down (p=+1 or -1) perpendicular to the disk plane. The combination of circulation and polarity defines the vortex chirality either right handed (cp=+1) or left handed (cp=-1). Stripe domains conversely arise in thin films with perpendicular magnetic anisotropy as originally described by Kittel29-31. The width of the meandering domains is determined from a balance between exchange energy (i.e. domain wall energy) and dipole energy and varies with film thickness. The morphology of the domains depends on the field history. In this paper, we describe the domain structure that arises in single-crystal nickel nanowires (NWs). In these NWs, the combination of the well-defined magneto-crystalline anisotropy and NW shape results in the emergence of a periodic vortex domain structure where the vortex chirality alternates handedness along the primary axis of the wire. Previous investigations of magnetic NWs have typically been carried out by studies on amorphous or polycrystalline wires grown by nanoporous template electrodeposition or self-assembly32-35. These methods of synthesis induce crystalline grains in the microstructure that can cause local variations in the magnetic anisotropy and moment. Thus it is impractical to explore the full magneto-crystalline anisotropy (MCA) of such structures, and as a result, these previous studies have largely varying conclusions depending on the differences in crystalline quality36-40. For these types of cylindrical nanowires, a quasi- 1D model where the magnetic response is dominated by the shape anisotropy is typically sufficient to explain the magnetic properties. Because of the high aspect ratio, the static and magnetic reversal properties can be modeled using a Stoner-Wohlfarth or curling-mode model assuming a single, strong shape anisotropy along the wire axis41, 42. This strong shape anisotropy leads to magnetic reversal that is characterized by domain nucleation and propagation from the wire ends. The Ni NWs described in this study are synthesized by a thermal chemical vapor deposition (CVD) technique that results in highly-oriented, single-crystal Ni NWs grown on amorphous SiOx coated Si substrates. Details of this growth technique have been published previously43. The resulting NWs have square cross sections with about 200 nm width, and lengths of up to 8 μm. Transmission electron microscopy images and electron diffraction patterns in Fig. 1 show the long/growth axis of the NWs are along the <001> direction and the side of the NWs are bounding surfaces that are atomically smooth. Analysis by coherent x-ray diffraction has shown the NWs to be single-crystalline44,45. Figure 1: Transmission electron microscopy images and electron diffraction patterns of a Ni NW after transfer to TEM grid. For nickel single crystals, the MCA easy axis is the <111> direction46. At any point inside the crystal, there are eight equivalent <111> easy directions at angles of approximately 55° relative to the wire axis as shown in Fig. 2a. The MCA is in competition with the magnetic shape anisotropy, which prefers the magnetization to lie along the nanowire's long axis (<001> directions). Previous measurements of the anisotropic magnetoresistance (AMR) response of the Ni NWs at T = 10 K reflect this competition of magnetic energies43. In AMR measurements, the electrical resistance of a ferromagnetic structure depends on the relative angle between the direction of current flow and average magnetization direction. The value of the resistivity is given by 𝜌(𝜃) = 𝜌∥ − (𝜌∥ − 𝜌⊥) sin2(𝜃) where 𝜌∥ and 𝜌⊥ represent the resistivities for configurations fully magnetized parallel and perpendicular to the wire axis, respectively. As shown in Fig. 2b, the resistance at remanence is intermediate between 𝑅∥ and 𝑅⊥, suggesting that the average magnetization at H = 0 lies off-axis. The AMR curves of the Ni NWs contain only small regions of hysteresis, implying that this intermediate magnetization state is reachable regardless of initial magnetic conditions. The shape of these field angle and magnitude dependent AMR curves suggest a competition between the shape anisotropy and MCA. Figure 2: a) Diagram of Ni NW structure with example crystalline axes indicated. 8 equivalent <111> axes exist. b) Anisotropic magnetoresistance behavior of Ni NW with varying external applied field orientations, and AMR calculated from micromagnetic modeling. The limitation of AMR measurements is that they only measure the average overall magnetization of the NWs. In order to visualize the space resolved micromagnetic configurations, we have performed magnetic and structural imaging using x-ray photoemission electron microscopy (PEEM). For PEEM imaging, the NWs were transferred onto silicon oxide substrates by a contact transfer technique. To prevent substrate charging during imaging, substrates were coated with 10-nm Pd prior to NW transfer. Imaging was performed at room temperature on EPU beamline 11.0.1 at the Advanced Light Source (Lawrence Berkeley National Lab) using PEEM3. PEEM3 has an energy range of 150-2000 eV with a spatial resolution down to 30 nm, and offers full polarization control. Magnetic contrast in photoemission experiments is provided by an asymmetry in the absorption cross sections for left and right circularly polarized x-rays for ferromagnetic materials. This dichroism allows for the direct imaging of the local magnetization vector when coupled to a spatially sensitive detector. A schematic of the imaging experiment is shown in Fig. 3a. Incident x-rays are turned to the Ni L3 and L2 absorption edges at 853 eV and 871 eV, respectively, and arrive at the sample surface with an angle of approximately 30° relative to the substrate surface. To observe the off-axis magnetization components suggested by AMR measurements, the NW is oriented at an almost orthogonal angle relative to the incoming x-ray direction (~82°). This almost orthogonal configuration maximizes the magnetic contrast as XMCD is sensitive to the magnetization component along the x-ray propagation axis. X-ray absorption spectra (XAS) and XMCD from the wire are shown in Fig. 3b. The L2 absorption edge exhibits a slight multiplet splitting, indicating a possible surface oxidation of the NW. Besides this slight surface oxidation, the XAS behavior is typical of pure Ni without impurities. Figure 3: Diagram of NiNW structure with example crystalline axes indicated. 8 equivalent <111> axes are oriented ~53° relative to the <001> axes. X-ray magnetic circular dichroism (XMCD) is defined as the ratio of 𝑋𝐴𝑆(𝑅𝐶𝑃)/𝑋𝐴𝑆(𝐿𝐶𝑃) where RCP and LCP are right circular and left circular polarized light respectively. To observe the magnetic configuration of the Ni NW, XMCD images are recorded at the Ni L3 edge (853 eV). XAS and XMCD images from a Ni NW are shown in Fig. 4. The brightly lit section of the image in Fig. 4a corresponds to the structure of the wire, whereas the dark section is a shadow arising from the obscured x-ray flux due to the physical size/height of the NW. The probing depth of the PEEM technique is limited by the inelastic mean free path of photoelectrons, which is 1.6 nm. As a consequence, XMCD is a surface sensitive technique that can only provide information about the magnetization of the top 5 nm of the sample surface. Though the electron inelastic mean free path is small, the x-ray attenuation length from absorption is one order of magnitude larger, approximately 40 nm at this absorption edge. Due to the small thickness of the NWs, some residual flux of x-rays will traverse the thickness of the wire and contribute to photoexcitation of electrons from the conductive substrate, forming a "shadow" region. The integrated absorption through the body of the nanostructure is also governed by XMCD effects and can thus give an indication of the bulk magnetization configuration by examining the XMCD of the shadow47 (even though no magnetic material is physically present in that region). Figure 4b shows an image of the wire and shadow with XMCD contrast. The bright and dark regions indicate local magnetization components parallel (+y) and anti-parallel (-y) to the x-ray direction. Figure 4: a) XAS image of a Ni NW. b) XMCD image of the Ni NW with magnetic contrast. c) Horizontal line scans of wire and shadow images with XMCD contrast. Inset) Description of axes and experimental geometry The XMCD images show a clear periodic domain pattern down the length of the NW. A line intensity scan along the NW and shadow XMCD images in Fig. 4c shows well-formed oscillations in the magnetization with a period of λ ~ 250nm. Due to the XMCD absorption within the NW as described above, the also-periodic magnetization configuration of the shadow confirms that the domain pattern is not only existing on the surface of the NW, but persists in the NW bulk. The shadow profile is almost out-of-phase with the surface profile, suggesting that the bulk component roughly follows the surface magnetization. For example, if the local surface magnetization in one area of the wire is oriented collinear to the x-ray flux direction (+y), the surface XMCD image would be locally bright in this area due to strong absorption. If the local core magnetization is similarly oriented along (+y), then a strong absorption of the x-rays would reduce the transmitted flux to the substrate surface, resulting in a dark area. From these images, it is apparent that the bulk and surface magnetizations of the NW are mutually periodic along the length of the NW. To better understand the images and the underlying magnetic configuration, finite element micromagnetic simulations were performed to computationally investigate the response of the single crystal Ni NW. The simulation results were obtained by numerically solving the Landau-Lifshitz-Gilbert (LLG) equation using the FastMag Micromagnetic Simulator48 for a NW with the magnetic properties of bulk Ni fcc crystals. AMR loops, magnetization hysteresis loops, and domain images were computed for a 200 x 200 x 8000 nm NW discretized using tetrahedral elements of 10 nm nominal edge size. The values of material parameters of the NW used for the first set of simulations were: saturation magnetization MS = 480 emu/cm3, exchange stiffness constant A=1.0 μerg/cm, first and second order cubic anisotropies K1 = -1.20 Merg/cm3, K2 = 0.41 Merg/cm3, respectively, which correspond to values expected at T = 10 K46. Negative first order anisotropy (K1 < 0) implies that the easy axis of magnetization is as along the <111> directions. Figure 2b shows the field and angular dependence of the AMR calculated based on micromagnetic modeling, which demonstrates good agreement with the experimental measurements. The key features of intermediate resistance remanent state, high saturation field, and low hysteresis are reproduced in simulations. Figure 5. Magnetization configurations of Ni NW at T = 10 K obtained by micromagnetic simulations. (a) Saturated state with magnetization uniformly pointing in x-direction in the presence of saturating magnetic field of 10 kOe. (b) Oscillatory precursor state with period of λ = 240 nm which develops when field is gradually reduced from saturating value (10 kOe) to 2.5 kOe. (c) Periodic alternating chirality vortex state with period of λ = 240 nm that evolves from precursor state if applied field is abruptly removed. (d) Uniform chirality vortex state obtained from precursor state following gradual reduction of applied field from 2.5 kOe to zero. Figure 5 illustrates magnetization configurations of the cubic anisotropy Ni NW obtained by micromagnetic simulations under different applied magnetic field conditions at T = 10 K. A uniform 10 kOe magnetic field applied perpendicular to the wire long axis uniformly saturates the NW magnetization in the (+y) direction as shown in Fig. 5a. As the field is reduced to the vicinity of 2.5 kOe, a precursor oscillatory state develops (Fig. 5b) with average magnetization pointing between the (+y) direction of the applied field and the long axis direction (either (+z) or (-z)) favored by shape anisotropy, with a modulation along the x- direction due to the influence of cubic anisotropy and magnetostatics which promotes a periodic texture. The period of the oscillation in the precursor state is 240 nm, which is in good agreement with our experimentally observed oscillation period. The precursor oscillatory state shown in Fig. 5b is the ground state of the Ni NW magnetization in the presence of a 2.5 kOe magnetic field applied in the y-direction. Two possible magnetic configurations can be obtained from this point. If the field is abruptly removed, the oscillatory precursor configuration is out of equilibrium, and multiple vortices nucleate across the NW with a chirality and period reflecting the magnetization modulation of the precursor state (Fig. 5c). The magnetic configuration thus obtained is consistent with the experimentally obtained PEEM images (Fig. 4) in terms of texture and periodicity (λ = 240 nm). Conversely, if the field is gradually reduced from 2.5 kOe toward zero, a single vortex nucleates near an end of the NW at around 2.0 kOe, and expands across the full length of the structure before any other nucleation events take place. As a result, the NW is left in a uniform vortex state with polarity either in the (+z) or (–z) direction as shown in Fig. 5d, corresponding to the z-component of magnetization of the precursor state. The formation of a vortex domain in the Ni NW reduces the magnetostatic energy and aligns spins closer to the <111> axes to reduce the MCA energy. Figure 6: Exploded slices of the micromagnetic configuration of Ni NW, incrementing in distance along the wire axis showing a half period. Arrows clarify the chirality of vortex domains existing within the NW. Cross sections along the length of the NW in the periodically alternating chirality vortex state. An x-y projection of the spins is shown, but all arrows have a component along the wire axis (towards the reader). The transition between left and right handed configurations involves simultaneous expulsion and nucleation of opposite chirality vortex cores at opposite corners of the NW. The magnetic state obtained from the precursor by abruptly removing the field can be described as a periodic array of alternating chirality vortices of same polarity, shown in Fig. 6. The magnetic state is particularly interesting due to the exotic spin configuration which is topologically protected, because once the vortices form, they are effectively locked in, as the collapse of a vortex domain of chirality χ and merger of two adjacent vortices of chirality -χ requires the expulsion of a vortex core of chirality χ, and injection of a core of opposite chirality (-χ), which is energetically costly. In this respect, the alternating chirality vortex domain structure is analogous to an array of 360° DWs in a thin strip, where expulsion of a domain necessitates an energetically expensive out-of-plane rotation of magnetic moments49-52. Energy calculations show that the alternating chirality vortex state has higher energy than the uniform vortex state, which is the ground state for the modeled cubic NW system at zero field. The presence of multiple domain walls in the former configuration leads to greater overall exchange and anisotropy energy, just as for uniaxial anisotropy systems. The fact that the system relaxes to such a metastable state instead of the ground state when the field is abruptly cut off after reaching the vicinity of 2.5 kOe can be explained by the proximity of the precursor state and the alternating chirality vortex state in configuration space. Namely, when the field is abruptly removed, the out-of-equilibrium magnetization configuration finds and settles in a local energy minimum of the newly formed energy landscape (at zero field) that corresponds to the stable state configurationally most similar to the precursor state. The stability of the alternating chirality vortex state is a consequence of the energy barrier preventing the collapse of topologically protected alternating chirality vortex domains interior to the NW, and is also due to a particular magnetization configuration at the NW ends, which effectively clogs the NW, preventing vortex domains from escaping the structure at its two extremities. The configuration at the ends is given in Fig. 7, showing a significant portion of the local magnetization oriented in the opposite z-direction to the average magnetization of the NW, i.e., opposite to the polarity direction of the vortices. This configuration is attributed to the interplay of the magnetostatic, cubic anisotropy, and exchange interactions at the discontinuity, which leads to a magnetization configuration at the NW ends that differs from the magnetization pattern interior to the NW. As a result, the vortices cannot escape at the ends without a reconfiguration of magnetization at one or both extremities. The energy barrier associated with such a reconfiguration contributes to the stability of the alternating chirality vortex state by effectively blocking the escape routes of vortices at the two discontinuities. Figure 7: Magnetization configuration at NW end. An alternative state, intermediate between the uniform vortex ground state and the periodic alternating chirality vortex state, is obtained once the magnetization relaxes to equilibrium following abrupt removal of field at saturation (above 10 kOe) in the y-direction, or upon relaxation to equilibrium from a completely random initial state that is obtained by specifying the magnetization vector at each node of the finite element mesh to be in an arbitrary direction at initialization. In both cases, the system is in a highly out-of-equilibrium state as it begins to relax. Consequently, the vortices that first nucleate along the NW (at arbitrary locations) quickly expand on account of the non-equilibrated surrounding magnetization, locally minimizing the energy and ultimately leaving the NW in a stable non-periodic alternating chirality vortex state shown in Fig. 8. The obtained (meta)stable configurations depicted in Fig. 5 c,d and Fig. 8 indicate that the relaxation process and the final equilibrated magnetization configuration of the modeled Ni NW are dependent on field history. Figure 8: Non-periodic alternating chirality vortex state at T = 10 K obtained when the saturating field is abruptly cut of at saturation (~10 kOe). The anisotropy of single crystal Ni strongly depends on temperature46. To study the magnetic response of Ni NW at room temperature, simulations were performed with values of material parameters corresponding to T = 293 K, i.e.: MS = 450 emu/cm3, A=1.0 μerg/cm, K1 = -0.057 Merg/cm3 and K2 = -0.023 Merg/cm3. Due to the much lower magnetic anisotropy energy density at room temperature, the shape anisotropy of the slender NW dominates over cubic anisotropy, and the obtained magnetization at ground state (in the absence of field) is seen to be largely uniform and in the direction of the NW long axis, with moderate nonuniformity at the NW extremities. The absence of periodic texture in the magnetization of the modeled system at T = 293 K is at variance with the oscillations observed experimentally at room temperature using PEEM. The discrepancy suggests a disparity between material properties of the experimentally grown and the modeled Ni NW. By varying the values of cubic anisotropy energy densities K1 and K2, introducing negative uniaxial anisotropy, or including magnetoelastic effects in the model53, the interplay between the MCA and shape anisotropy can be tuned to produce a variety of magnetic configurations. Figures 9a,b show a vortex domain formation which develops at room temperature when a negative uniaxial anisotropy is introduced to the system, with anisotropy energy densities Ku = -0.01 Merg/cm3 and Ku = -1.0 Merg/cm3, respectively. Negative uniaxial anisotropy favors magnetization alignment in the plane of the NW cross section, promoting the formation of vortex domains54. Figure 9: Alternating chirality vortex domains obtained at room temperature, in the presence of negative uniaxial anisotropy, with energy density constants (a) Ku = -0.01 Merg/cm3 and (b) Ku = -1.0 Merg/cm3. Vortex formation is also seen to result with the introduction of the magnetoelastic interaction53, assuming uniform compressive biaxial strain in the xy basal plane. Such a strain could arise from the synthesis process of the single crystal Ni NWs which grow vertically from the amorphous SiOx coated Si substrate43, 55. Figure 10 shows the remanent configuration of the modeled Ni NW assuming magnetostrictive constants λ111 = -24×10-6 and λ100 = -46×10-6, Young's modulus E = 200 GPa, and strain ε = -0.05. For uniform biaxial compressive strain and magnetostrictive properties corresponding to nickel56, the <111> directions are the easy magnetization directions, thus favoring vortex formation over uniform magnetization along the NW long axis. Figure 10: Alternating chirality vortex domains obtained at room temperature in the presence of uniform biaxial compressive strain ε = 0.05, assuming magnetostrictive constants λ111 = -24×10-6 and λ100 = -46×10- 6, Young's modulus E = 200 GPa, typical of nickel. Figure 11 shows the spontaneously periodic magnetization pattern which develops in the presence of negative uniaxial anisotropy and in the absence of magnetoelastic effects, when the saturation magnetization is reduced from 450 emu/cm3 to 250 emu/cm3. In this case, vortices are no longer energetically favorable, and the magnetization in the cross sections of the NW is largely uniform (Fig. 11). Due to the accumulation of effective magnetic surfaces charges at the sides of the NW to which the magnetization is normal, a periodic texture develops resulting in flux closure and reduction in magnetostatic energy57 that is analogous to stripe domains in magnetic thin films. Results shown in Figs. 10-12 indicate that the magnetic response of the studied Ni NW is strongly dependent on material properties of the system. Figure 11. Periodic array of domains obtained in the presence of negative uniaxial anisotropy (Ku = -1.0 Merg/cm3) at reduced saturation magnetization of Ms = 250 emu/cm3. In summary, we have investigated the magnetization configurations arising in single-crystal Ni NWs of dimensions 200 x 200 x 8000 nm. Utilizing magneto-transport and magnetic microscopy techniques, we revealed that the magnetization state of the Ni NW system can exist as a series of alternating domains that are periodically distributed along the nanowire axis, with a period of λ ~ 250 nm. Micromagnetic simulations reveal that an array of alternating chirality topologically protected vortex domains with a period closely matching experimental observation (λ ~ 240 nm) is a possible remanent state. Vortex formation in single crystal Ni NWs is attributed to cubic anisotropy of the material system, which favors magnetization to point in the easy <111> directions, and its reduced structural dimensions. Simulations show that other remanent states are also possible depending on the field history. The ground state is found to correspond to a single uniform vortex domain extending along the length of the NW. Because of a much smaller cubic anisotropy of Ni at room temperature, shape anisotropy dominates MCA, resulting in largely uniform NW magnetization oriented longitudinally along the NW. With introduction of negative uniaxial anisotropy or in the presence of magnetoelastic effects under the assumption of uniform compressive biaxial strain in the basal plane, vortices are seen to reappear at T = 293 K. Rapid quenching of the NW temperature during the synthesis process and strain buildup could bias NW magnetization toward the periodic alternating chirality vortex metastable state observed in simulations. The alternating chirality topologically protected vortex domains could be found interesting for employment in magnetic memory and logic technologies where bits of information are represented by magnetic domains that can be stored in structures of reduced dimensions and manipulated and transmitted by applied fields, strains, and/or spin polarized currents. This work was supported by the National Science Foundation through grant award number: DMR-0906957. This research used resources of the Advanced Light Source, which is a DOE Office of Science User Facility under contract number: DE-AC02-05CH11231. Current affiliation: Anzu Partners, LLC, La Jolla, California, 92037, USA † Current affiliation: Global Forecasting, Gartner, Inc., Stamford, Connecticut 06902, USA 1. 2. 3. R. Skomski, J. Phys.-Condes. Matter 15 (20), R841-R896 (2003). E. E. Fullerton and J. R. Childress, Proc. IEEE 104 (10), 1787-1795 (2016). N. A. Frey, S. Peng, K. Cheng and S. H. Sun, Chemical Society Reviews 38 (9), 2532-2542 (2009). 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. S. D. Bader, Reviews of Modern Physics 78 (1), 1-15 (2006). S. D. Bader and S. S. P. Parkin, in Annual Review of Condensed Matter Physics, Vol 1, edited by J. S. Langer (2010), Vol. 1, pp. 71-88. J. A. Katine and E. E. Fullerton, J. Magn. Magn. Mater. 320 (7), 1217-1226 (2008). S. Heinze, K. von Bergmann, M. Menzel, J. Brede, A. Kubetzka, R. Wiesendanger, G. Bihlmayer and S. Blugel, Nat. Phys. 7 (9), 713-718 (2011). H. B. Braun, Adv. Phys. 61 (1), 1-116 (2012). K. Y. Guslienko, J. Nanosci. Nanotechnol. 8 (6), 2745-2760 (2008). M. Klaui, C. A. F. Vaz, J. Rothman, J. A. C. Bland, W. Wernsdorfer, G. Faini and E. Cambril, Phys. Rev. Lett. 90 (9) (2003). S. Muhlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R. Georgii and P. Boni, Science 323 (5916), 915-919 (2009). A. Moser, K. Takano, D. T. Margulies, M. Albrecht, Y. Sonobe, Y. I. S. Sun and E. E. Fullerton, J. Phys. D: Appl. Phys. 35, R157 (2002). A. Fert, V. Cros and J. Sampaio, Nature Nanotechnology 8 (3), 152-156 (2013). I. R. McFadyen, E. E. Fullerton and M. J. Carey, MRS Bull. 31 (5), 379-383 (2006). S. Ikeda, J. Hayakawa, Y. M. Lee, F. Matsukura, Y. Ohno, T. Hanyu and H. Ohno, IEEE Trans. Electron Devices 54 (5), 991-1002 (2007). S. S. P. Parkin, M. Hayashi and L. Thomas, Science 320 (5873), 4 (2008). S. Maat, Marley, A. C. , in Handbook of Spintronics, edited by Y. Xu, Awschalom, D. D., Nitta, J. (Springer Science + Business Media, 2014), pp. 977-1028. S. M. A. Shibli, K. S. Beenakumari and N. D. Suma, Biosens. Bioelectron. 22 (5), 633-638 (2006). D. Houssameddine, U. Ebels, B. Delaet, B. Rodmacq, I. Firastrau, F. Ponthenier, M. Brunet, C. Thirion, J. P. Michel, L. Prejbeanu-Buda, M. C. Cyrille, O. Redon and B. Dieny, Nature Materials 6 (6), 447-453 (2007). F. B. Mancoff, N. D. Rizzo, B. N. Engel and S. Tehrani, Nature 437 (7057), 393-395 (2005). D. A. Allwood, G. Xiong, C. C. Faulkner, D. Atkinson, D. Petit and R. P. Cowburn, Science 309 (5741), 1688-1692 (2005). T. Schneider, A. A. Serga, B. Leven, B. Hillebrands, R. L. Stamps and M. P. Kostylev, Appl. Phys. Lett. 92 (2) (2008). J. Jaworowicz, N. Vernier, J. Ferre, A. Maziewski, D. Stanescu, D. Ravelosona, A. S. Jacqueline, C. Chappert, B. Rodmacq and B. Dieny, Nanotechnology 20 (21) (2009). X. C. Zhang, M. Ezawa and Y. Zhou, Sci Rep 5 (2015). R. P. Cowburn, D. K. Koltsov, A. O. Adeyeye, M. E. Welland and D. M. Tricker, Phys. Rev. Lett. 83 (5), 1042-1045 (1999). T. Shinjo, T. Okuno, R. Hassdorf, K. Shigeto and T. Ono, Science 289 (5481), 930-932 (2000). O. Hellwig, A. Berger, J. B. Kortright and E. E. Fullerton, J. Magn. Magn. Mater. 319 (1-2), 13-55 (2007). M. Hehn, S. Padovani, K. Ounadjela and J. P. Bucher, Phys. Rev. B 54 (5), 3428-3433 (1996). B. Kaplan and G. A. Gehring, J. Magn. Magn. Mater. 128 (1-2), 111-116 (1993). C. Kittel, Physical Review 70 (11-1), 965-971 (1946). Z. Malek and V. Kambersky, Czechosl. Journ. Phys. 8 (416) (1958). X. Y. Li, H. Wang, K. N. Xie, Q. Long, X. F. Lai and L. Liao, Beilstein Journal of Nanotechnology 6, 2123-2128 (2015). T. Ohgai, Nanowires - Recent Advances, 101-125 (2012). K. Park and J. S. Lee, Sci Rep 6 (2016). T. Ohgai, L. Gravier, X. Hoffer, M. Lindeberg, K. Hjort, R. Spohr and J. P. Ansermet, Journal of Physics D-Applied Physics 36 (24), 3109-3114 (2003). L. Sun, P. C. Searson and C. L. Chien, Appl. Phys. Lett. 79 (26), 4429-4431 (2001). Y. Rheem, B. Y. Yoo, W. P. Beyermann and N. V. Myung, Nanotechnology 18 (1) (2007). A. S. Samardak, E. V. Sukovatitsina, A. V. Ognev, L. A. Chebotkevich, R. Mahmoodi, S. M. Peighambari, M. G. Hosseini, F. Nasirpouri and Iop, in Iv Nanotechnology International Forum (2012), Vol. 345. W. O. Rosa, L. G. Vivas, K. R. Pirota, A. Asenjo and M. Vazquez, J. Magn. Magn. Mater. 324 (22), 3679-3682 (2012). P. K. Choudhury, S. Banerjee, S. Ramaprabhu, K. P. Ramesh and R. Menon, J. Nanosci. Nanotechnol. 13 (12), 8162-8166 (2013). E. C. Stoner and E. P. Wohlfarth, Phil. Trans. R. Soc. London A 240 (826), 599 (1948). F. T. Parker, M. W. Foster, D. T. Margulies and A. E. Berkowitz, Phys. Rev. B 47 (13), 7885-7891 (1993). K. T. Chan, J. J. Kan, C. Doran, L. Ouyang, D. J. Smith and E. E. Fullerton, Philos. Mag. 92 (17), 2173-2186 (2012). K. T. Chan, J. J. Kan, C. Doran, O. Y. Lu, D. J. Smith and E. E. Fullerton, Nano Letters 10 (12), 5070-5075 (2010). J. W. Kim, A. Ulvestad, S. Manna, R. Harder, E. Fohtung, A. Singer, L. Boucheron, E. E. Fullerton and O. G. Shpyrko, Journal of Applied Physics 120 (16) (2016). W. Martienssen, Warlimont, H., in Springer Handbooks (Springer-Verlag Berlin Heidelberg, 2005). S. Jamet, S. Da Col, N. Rougemaille, A. Wartelle, A. Locatelli, T. O. Mentes, B. S. Burgos, R. Afid, L. Cagnon, S. Bochmann, J. Bachmann, O. Fruchart and J. C. Toussaint, Phys. Rev. B 92 (14) (2015). R. Chang, S. Li, M. V. Lubarda, B. Livshitz and V. Lomakin, Journal of Applied Physics 109 (3), 07D358 (2011). J. S. Zhang, S. A. Siddiqui, P. Ho, J. A. Currivan-Incorvia, L. Tryputen, E. Lage, D. C. Bono, M. A. Baldo and C. A. Ross, New J. Phys. 18 (2016). M. Hehn, D. Lacour, F. Montaigne, J. Briones, R. Belkhou, S. El Moussaoui, F. Maccherozzi and N. Rougemaille, Appl. Phys. Lett. 92 (7) (2008). A. Pushp, T. Phung, C. Rettner, B. P. Hughes, S. H. Yang, L. Thomas and S. S. P. Parkin, Nat. Phys. 9 (8), 505-511 (2013). L. Thomas, M. Hayashi, R. Moriya, C. Rettner and S. Parkin, Nature Communications 3 (2012). J. Dean, M. T. Bryan, G. Hrkac, A. Goncharov, C. L. Freeman, M. A. Bashir, T. Schrefl and D. A. Allwood, Journal of Applied Physics 108 (7) (2010). H. Y. Kwon, S. P. Kang, Y. Z. Wu and C. Won, Journal of Applied Physics 113 (13) (2013). E. Fohtung, J. W. Kim, K. T. Chan, R. Harder, E. E. Fullerton and O. G. Shpyrko, Appl. Phys. Lett. 101 (3) (2012). C. Y. Liang, S. M. Keller, A. E. Sepulveda, A. Bur, W. Y. Sun, K. Wetzlar and G. P. Carman, Nanotechnology 25 (43) (2014). A. Hubert, Schafer, R., Magnetic Domains: The Analysis of Magnetic Microstructures. (Springer- Verlag Berlin Heidelberg, 1998). 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53. 54. 55. 56. 57.
1510.05325	1	1510	2015-10-19T00:00:04	Minimum Anisotropy of a Magnetic Nanoparticle out of Equilibrium	[ "cond-mat.mes-hall" ]	In this article we study magnetotransport in single nanoparticles of Ni, Py=Ni$_{0.8}$Fe$_{0.2}$, Co, and Fe, with volumes $15\pm 6$nm$^3$, using sequential electron tunneling at 4.2K temperature. We measure current versus magnetic field in the ensembles of nominally the same samples, and obtain the abundances of magnetic hysteresis. The hysteresis abundance varies among the metals as Ni:Py:Co:Fe=4\,:50\,:100\,:100(\%), in good correlation with the magnetostatic and magnetocrystalline anisotropy. The abrupt change in the hysteresis abundance among these metals suggests a concept of minimum magnetic anisotropy required for magnetic hysteresis, which is found to be $\approx 13$meV. The minimum anisotropy is explained in terms of the residual magnetization noise arising from the spin-orbit torques generated by sequential electron tunneling. The magnetic hysteresis abundances are weakly dependent on the tunneling current through the nanoparticle, which we attribute to current dependent damping.	cond-mat.mes-hall	cond-mat	Minimum Anisotropy of a Magnetic Nanoparticle out of Equilibrium W. Jiang,1 P. Gartland,1, ∗ and D. Davidovi´c1, † 1School of Physics, Georgia Institute of Technology, 837 State Street, Atlanta, Georgia 30332, USA (Dated: April 19, 2021) Abstract In this article we study magnetotransport in single nanoparticles of Ni, Py=Ni0.8Fe0.2, Co, and Fe, with volumes 15 ± 6nm3, using sequential electron tunneling at 4.2K temperature. We measure current versus magnetic field in the ensembles of nominally the same samples, and obtain the abundances of magnetic hysteresis. The hysteresis abundance varies among the metals as Ni:Py:Co:Fe=4 :50 :100 :100(%), in good correlation with the magnetostatic and magnetocrystalline anisotropy. The abrupt change in the hysteresis abundance among these metals suggests a concept of minimum magnetic anisotropy required for magnetic hysteresis, which is found to be ≈ 13meV. The minimum anisotropy is explained in terms of the residual magnetization noise arising from the spin-orbit torques generated by sequential electron tunneling. The magnetic hysteresis abundances are weakly dependent on the tunneling current through the nanoparticle, which we attribute to current dependent damping. 5 1 0 2 t c O 9 1 ] l l a h - s e m . t a m - d n o c [ 1 v 5 2 3 5 0 . 0 1 5 1 : v i X r a 1 Magnetic anisotropy in ferromagnets is vital in magneto-electronic applications, such as giant magnetoresitance1,2 and spin-transfer torque.3 -- 5 For example, in some applications, a strong spin-orbit anisotropy is desired in order to establish a hard or fixed reference magnetic layer, while in other applications, it is beneficial to use a weaker anisotropy in order to fabricate an easily manipulated, soft or free magnetic layer. The ability to tune the degree of anisotropy for various applications is therefore of utmost importance. In thermal equilibrium, the minimum anisotropy necessary for magnetic hysteresis is temper- ature dependent6 -- 8. In this article, we address the minimum anisotropy in the case of a voltage-biased metallic ferromagnetic nanoparticle First studies of discrete levels and mag- netic hysteresis in metallic ferromagnetic nanoparticles have been done on Co nanoparti- cles9 -- 11. Here we discuss the magnetic hysteresis abundances in single electron tunneling devices containing similarly sized single nanoparticles of Ni, Py=Ni0.8Fe0.2, Co, and Fe. At 4.2K temperature. the probability that a given nanoparticle sample will display magnetic hysteresis in current versus magnetic field, at any bias voltage, was found to vary as follows: Ni:Py:Co:Fe=0.04 :0.5 :1 :1. The very small (high) probability of magnetic hysteresis in the Ni (Fe and Co) nanoparticles suggests a concept of minimum magnetic anisotropy necessary for magnetic hysteresis, comparable to the average magnetic anisotropy of Py nanoparticle. The minimum magnetic anisotropy is explained here in terms of the fluctuating spin-orbit torques exerted on the magnetization by sequential electron tunneling. These torques lead to the saturation of the effective magnetic temperature at low temperatures. In order for the nanoparticle to exhibit magnetic hysteresis at 4.2K, the blocking temperature must be larger than the residual temperature. The magnetic hysteresis abundances are found to be independent of the tunneling current through the sample, which suggests that the damping is proportional to the tunneling current. I. EXPERIMENT Our samples consist of similarly-sized ferromagnetic nanoparticles tunnel-coupled to two Al leads via amorphous aluminum oxide barriers. First, a polymethilmetachryllate bridge is defined by electron-beam lithography on a SiO2 substrate using a technique developed previously, as sketched in Fig. 1A. Next, we deposit 10nm of Al along direction 1. Then, we switch the deposition to direction 2, and deposit 1.5nm of Al2O3, 0.5-1.2nm of ferromagnetic 2 FIG. 1: A: Sketch showing the sample fabrication process. B: Scanning electron micrograph of a typical sample. material, 1.5nm of Al2O3, topped off by 10nm of Al, followed by liftoff in acetone. The tunnel junction is formed by the small overlap between the two leads as shown in the circled part in Fig. 1B. The nanoparticles are embedded in the matrix of Al2O3 in the overlap. The nominal thickness of deposited Co, Ni, and Py is 0.5-0.6nm. At that thickness, the deposited metals form isolated nanoparticles approximately 1 − 5nm in diameter. We find that if we deposit Fe at the nominally thickness 0.6nm, then the resulting samples are generally insulating. Thus, the deposited Fe thickness is increased to 1 − 1.2nm, which yields samples in the same resistance range as in samples of Co, Ni, and Py. We suppose that because Fe can be easily oxidized, Fe nanoparticles are surrounded by iron oxide shells. Thus, our sample characterization suggests that it is appropriate to attribute the difference among Co, Ni, and Py samples to intrinsic material effects rather than size discrepancies, while, in Fe nanoparticles, the comparison is complicated by the uncertainty in the size of the metallic core. Still, we find the comparison with Fe to be fair, because of the wide range of nanoparticle diameters involved. We obtain the transmission electron microscope (TEM) image of the deposited pure aluminum oxide surface, and the aluminum oxide surface topped with nominally 1.2nm of Fe, 0.55nm of Ni, and 0.6nm of Co, as shown in Fig. 2. The deposition is done immediately prior to loading the sample in the TEM. Pure aluminum oxide surface appears completely amorphous, with no visible signs of crystalline structure. In comparison, single crystal structure can be identified in the TEM images for Fe, Ni, and Co. From the TEM image, the areal coverage of Ni nanoparticles is 44% and the nanoparticle density is 3.6 · 104µm−2. 3 FIG. 2: EDS spectra and TEM images of aluminum oxide surface topped with 0.6nm of Co (top row), 0.6nm of Ni (second row), 1.2nm of Fe (third row), and control Al2O3 on TEM grid (bottom row). 4 Assuming that the nanoparticles have pancake shape, the average area and the height of the particles are 0.44/3.6 · 104µm−2 ≈ 12nm2 and 0.55nm/0.44 = 1.25nm, respectively. The standard deviation of the nanoparticles area is 40% of the average area, which is estimated by the shape analysis in 120 Ni neighboring nanoparticles. Thus, the volume of the Ni nanoparticles is 15 ± 6nm3, where 6nm3 is the standard deviation. The volume distribution in Co nanoparticles is similar to that in Ni. We take the energy dispersive x-ray spectroscopy (EDS) for the samples we imaged, which confirms the materials deposited on the aluminum oxide surface. II. MEASUREMENTS FIG. 3: Current versus voltage in four representative samples at 4.2K temperature.A.Co, B.Ni, C.Fe nanoparticles, and D.Leaky pure Al2O3 tunneling junctions. 5 The IV curves are measured using an Ithaco model 1211 current preamplifier and are reproducible with voltage sweeps. Figures 3 A, B, and C display the IV curves of three rep- resentative tunneling junctions with embedded nanoparticles of Co, Fe, and Ni, respectively, in samples immersed in liquid Helium at 4.2K. The IV curves display Coulomb blockade (CB) which confirms electron tunneling via metallic nanoparticles, but no discrete levels are resolved. We also measure tunneling junctions containing only the aluminum oxide, without embedded metallic nanoparticles. Those junctions are generally insulating but some are not, e.g., there may be leakage. The IV curves of those leaky junctions are linear, as shown in Fig. 3D, as expected for simple tunnel junctions. We show the IV curve in a pure alu- minum oxide junction, to demonstrate that the CB in the samples with embedded metallic nanoparticles originate from tunneling via those nanoparticles. The issue here is that, as will be shown immediately below, some of the samples with embedded nanoparticles do not display any hysteresis in current versus magnetic field at 4.2K. Since those samples also display Coulomb blockade in the I-V curve, we can conclude that the absence of hysteresis is intrinsic to the nanoparticles, rather than an artifact from tunneling through a possibly leaky aluminum oxide. Current versus applied magnetic field (parallel to the film plane) is obtained by measur- ing the current at a fixed bias voltage while sweeping the magnetic field slowly. Fig. 4A-D show the representative magnetic hysteresis loops for Co, Fe, Ni, and Py samples at 4.2K, respectively. As seen in Fig. 4C and D, the Ni-nanoparticle and one Py nanoparticle have no magnetic hysteresis. All the samples that lack hysteresis do so at the lowest resolved tunnel- ing current. The statistics of the presence of hysteresis for different materials are displayed in Table. 1. All the Co (over 50) and Fe (6), one half of the Py (out of 10), and only 2 of the 46 measured Ni samples, with the switching field 0.05T and 0.12T , display magnetic hysteresis. The abrupt change in the hysteresis abundance between Co, Fe, Py, and Ni suggests a con- cept of minimum magnetic anisotropy necessary for magnetic hysteresis, akin to the concept of Mott's maximum metallic resistivity. Note that the Ni samples and the non-hysteretic Py samples still display significant magneto-resistance. Among the hysteretic samples, the average magnetic switching field varies as Ni:Py:Co:Fe=0.085 :0.114 :0.233 :0.257 (Tesla). S0[K1(m2 The magnetic energy EM ( ~m) of the Fe/Ni/Py and Co nanoparticles can be written as z + Ks ~m N ~m], respectively, where S0¯h is the total spin in the nanoparticle, ~m is the magnetization unit z + Ks ~m N ~m] and S0[K1m2 x) + K2m2 z + K2m4 xm2 y + m2 ym2 z + m2 zm2 xm2 ym2 6 FIG. 4: Examples of tunneling current versus magnetic field at 4.2K, for samples with A.Co, B.Fe, C.Ni, and D.Py nanoparticles. In D, two pairs of curves correspond to two different samples. Red(black) lines correspond to decreasing(increasing) magnetic field. vector, and N is the demagnetization tensor which is set to have 3 eigenvalues equal to 0.2, 0.3, and 0.5. In the calculations, Euler angles between the principal axes of shape anisotropy and magnetocrystalline anisotropy axes are all equal to π/5. The results of the calculation of the energy barrier EB are shown in table 1, assuming the nanoparticle average volume obtained in Sec. 2. The error bar reflects the standard deviation in the nanoparticle volume. The abrupt change in magnetic hysteresis abundance is monotonic with the calculated EB. Since 50% of Py nanoparticles display hysteresis, it follows that the minimum anisotropy energy barrier required for magnetic hysteresis in our samples at 4.2K is ≈13meV. This energy barrier is too large to explain our findings in terms of the reduction in the blocking temperature among these metals. Experimentally, the magnetometry on similarly-sized 7 TABLE I: Hysteresis percentage versus magnetic anisotropy in different materials12,13 Material Hysteresis(%) Ks(µeV/spin) K1(µeV/spin) K2(µeV/spin) EB(meV) Co14 Fe15 Py16 Ni17 100 100 50 4 105 128 67.1 37.9 64.1 4.0 0 -24.4 8.3 1.3 0 4.1 97 ± 39 51 ± 20 13 ± 5 7.7 ± 3 nanoparticles show that the blocking temperature varies between 13 − 30K for Co and 6 − 20K for Ni.18 -- 22 Though Co nanoparticles appear to have higher blocking temperature than Ni nanoparticles, the difference in blocking temperature is not sufficient to explain the vast contrast in the hysteresis abundance in our nanoparticles under electron transport. Theoretically, the Arrhenius flipping rate can be estimated as νat exp(−EB/kBT ). Assuming νat = 1010Hz, we obtain the magnetization flipping time of 100 hours, much longer than the time it takes to measure a magnetic hysteresis loop. We conclude that the breakdown in magnetic hysteresis we observe reflects the effect of sequential electron tunneling through the nanoparticle on magnetization. This effect will be discussed in Sec. 4. III. NUMERICAL SIMULATIONS The magnetic Hamiltonian for Ni and Fe nanoparticles can be written as H(n, ~Sn) = S0[ K1,n 2 (α2β 2 + β 2α2 + α2γ2 + γ2α2 + β 2γ2 + γ2β 2) + K2,n 6 (α2β 2γ2 +α2γ2β 2 + β 2α2γ2 + β 2γ2α2 + γ2β 2α2 + γ2α2β 2)] − Ks ~Sn N ~Sn Co has uniaxial magnetocrystalline anisotropy, thus the magnetic Hamiltonian is H(n, ~Sn) = −S0[K1,nγ2 + K2,nγ4] − Ks ~Sn N ~Sn Here, n is the number of electrons in the nanoparticle while S0¯h is the total spin. K1,n, K2,n, and Ks represent magnetocrystalline anisotropy constants and the shape anisotropy constant per spin, respectively. α,β,γ=Sx,Sy,Sz/S0. N is the demagnetization tensor. The magnetocrystalline anisotropy constants per unit volume K1,V and K2,V are obtained from Refs. 14 -- 17 . We obtain 2S0/Na from Ref.12, where Na is the total number of atoms in 8 the nanoparticle. Then, S0/V = ρS0/NaMA, where ρ is the mass density and MA is the atomic mass. The average value of the magnetocrystalline anisotropy constants per spin (K1, K2) are (K1,V , K2,V )V /S0, respectively, while, Ks = µ0M 2 s V /2S0, where Ms is the satu- ration magnetization obtained in Ref. 13,16. Because of spin-orbit anisotropy fluctuations, (K1,n, K2,n) fluctuate around (K1, K2), respectively, according to the number of electrons n. We expect that the average magnetic anisotropy has strong material dependence, while the mesoscopic fluctuations in the total magnetic anisotropy energy due to single electron tunneling on/off, are independent of the material.23 -- 25 In Co nanoparticles, the change in total magnetic energy after electron tunneling-on is S0∆K ∼ 0.4meV.24 In Ni nanoparticles of the same size at 4.2K or below, S0 and K are both ≈ 1/3 of the values in Co nanoparticles. So, the relative fluctuations in magnetic anisotropy ∆K/K in Ni nanoparticles are enhanced 9 − 10 times compared to Co nanoparticles with the same size. At the same time, the mag- netic energy barrier in Ni nanoparticles is suppressed because of the cubic symmetry and weak shape anisotropy. As a result, the magnetization in Ni nanoparticles is significantly more susceptible to perturbation by electron transport compared to Co or Fe nanoparticles. The numerical simulation of the magnetization dynamics is based on the master equation modified from that in Ref.26. In the sequential tunneling regime, the number of electrons in the nanoparticle hops between n and n + 1. Assuming the electron tunnels through only one minority single electron state j, which reduces the spin of the nanoparticle by 1/2 after the electron tunneling-on event (including more electron states does not affect the result in a major way), the master equation can be written as ∂Pα ∂t =X α′ X l=L,R X σ=↑,↓ Γlσ n< α′cjσα >2 [−(1 − fl(∆EM ))Pα + fl(∆EM )Pα′] + < α′c† jσα >2 [−fl(−∆EM )Pα + (1 − fl(−∆EM ))Pα′]o + ǫ < α′Sx − iǫSyα >2 [ρB(ǫ∆EM )nB(∆EM )Pα′ + ρB(ǫ∆EM )nB(−∆EM )Pα] ΓB X α′,ǫ=± The first part of the master equation describes the magnetic tunneling transitions. Here, α > and α′ > represent magnetic eigenstates of the nanoparticle , i.e., the eigenstate of H(n, ~Sn) with n or n + 1 electrons. α > and α′ > can be obtained as superpositions of the eigenstates of S 2 and Sz, which are the pure spin-states S0, m >. cjσ (c† jσ) is the annihilation (1) 9 FIG. 5: Simulations of the time evolution of the magnetization vector in a Ni nanoparticle. The blue-yellow-red scale indicates the magnetic anisotropy energy landscape from low to high versus the polar angle Θ and the azimuthal angle Φ in a Ni nanoparticle. The value of t is the past time counted from the beginning of electron tunneling. The white dots signify possible angle positions at the specified time for each simulation. As time progresses, the likely magnetization position is spread out over phase space and approaches a random isotropic distribution. (creation) operator for an electron with spin σ on level j. Γlσ denotes the tunneling rate to level j through the leads l = L, R for electron with spin σ and fl is the Fermi distribution in the leads. ∆EM = Eα − Eα′. < α′c† jσα > is the tunneling transition matrix element for transition between the initial state α > and the final state α′ >. The second part of the master equation describes the magnetic damping due the coupling to the bosonic bath. ΓB is the rate related to the damping rate 1/T1. ρB(∆EM ) is the spectral density of the boson which is set to be constant because it varies very slowly26. nB is the Bose-distribution function. Fig. 5 shows the motion of the magnetization statistical distribution versus time. The magnetic anisotropy energy vs (Θ, Φ) represents the relation between magnetic energy and magnetization direction, where Θ, Φ are the two polar angles. Each magnetic state corre- sponds to a contour on the magnetic anisotropy energy landscape. Classically, one would consider the magnetic state as the magnetization precessing on the contour corresponding to that state. A random dot is selected on a contour to indicate the presence of the mag- 10 netic state related to that contour. The number of dots on each contour is decided by the probability of the corresponding magnetic state. The plotted dots together represent the distribution of the magnetization over Θ, Φ-space at that time. In the simulation for Fig. 5, S0 = 100. ΓL,Rσ = 6 × 107 which corresponds to a current about 5pA. ΓB = 2 × 103 leads to a relaxation time of ∼ 2.5µs. (K1,n, K2,n) = 1.25(K1, K2) and (K1,n+1, K2,n+1) = 0.75(K1, K2). Ks is taken from Table. 1. N is set to have 3 eigenval- ues equal to 0.2, 0.3, and 0.5. Varying N and the Euler angles, which were earlier defined, does not affect the qualitative result. The magnetic field is set to 0.001T to eliminate Kramers degeneracy. We set the nanoparticle to be initially at the ground state with n electrons. Then we iterate the master equation for 40µs. The magnetic state distribution of the nanoparticle gradually spreads from the ground state to other states and eventually becomes isotropic as shown in Fig. 5. IV. DISCUSSION AND CONCLUSIONS The transfer of a single electron into the magnetic nanoparticle creates a fluctuation in the spin-orbit energy of the nanoparticle.23 -- 25,27 Such a fluctuation in turn creates a spin- orbit torque that is exerted on the magnetization. In the previous section, we show how a fluctuating spin-orbit torque can lead to isotropic distribution of the magnetization. The fluctuating spin-orbit torques are mesoscopic effects and do not depend significantly on the material of the nanoparticle.23,25 But, the nonfluctuating magnetic anisotropy, such as mag- netocrystalline and magnetostatic shape anisotropy, depends strongly on the material. As the magnetic anisotropy of the nanoparticle is reduced, the strength of the fluctuating spin- orbit torques relative to the deterministic torques will increase, creating a noise floor which sets the limit on magnetic anisotropy below which magnetic hysteresis cannot be observed in sequential electron tunneling. Such a noise floor is reflected by the abrupt change in magnetic hysteresis abundance in similarly sized Ni, Py, Fe, and Co nanoparticles. By con- trast, in thermal equilibrium, magnetometery of the ensembles of similarly sized Co and Ni nanoparticles show much less dramatic change in the blocking temperature.18 -- 22 The mini- mum magnetic anisotropy also explains prior measurements of voltage biased single magnetic molecules, in a double tunneling barrier, which showed no signs of magnetic hysteresis, even at temperatures much lower than the blocking temperature,28,29 notwithstanding that the 11 magnetometry of ensembles of such molecules showed magnetic hysteresis below the blocking temperature.30,31 It may be surprising to the reader that the minimum magnetic anisotropy is found to be independent of the size of the tunneling current through the nanoparticle. The tunneling current we use in the measurements of current versus magnetic field, varies between 1pA and 100pA. A Co nanoparticle at 100pA is likely to exhibit magnetic hysteresis, while a Ni nanoparticle at 1pA is highly unlikely to do so. The ratio of the applied tunneling currents is one order of magnitude larger than the ratio of the energy barriers between the average Co and Ni nanoparticle. Since the data presented in this paper were gathered, we have studded single Ni nanoparticles at mK-temperature, and discovered that 2 out of the 5 measured Ni nanoparticles display magnetic hysteresis at the onset voltage for sequential electron tunneling.32 The hysteresis in current versus magnetic field was abruptly suppressed in the bias voltage range starting just above the lowest discrete energy level for single-electron tunneling. The abrupt suppression of magnetic hysteresis versus bias voltage was explained in terms of the magnetization blockade, which was caused by the bias voltage dependent damping rate.32,33 In the magnetization blockade regime, the ordinary spin-transfer33 or the spin-orbit torques32 are damped because of the small bias energy available for single-electron tunneling. In the voltage region where the magnetization is blocked, the spin-transfer and damping rates are both proportional to the electron tunneling rate, regardless of which spin- transfer mechanism is at play (e.g. the ordinary spin-transfer or the spin-orbit torques). A change in the bias-voltage will change the damping rate,32 but the effective magnetic tem- perature, controlled by the ratio of the damping rate and the spin-transfer rate,33 will be independent of the electron tunneling rate. It is reasonable to assume that the magnetic nanoparticle will exhibit magnetic hysteresis in our experimental times scales, if the flipping time given by the Arrhenius law, based on the attempt frequency and the ratio of the energy barrier and the effective magnetic temperature, is longer than the hysteresis measurement time. Since neither the energy barrier nor the effective magnetic temperature depend on the tunneling current, this explains, at least in principle, why magnetic hysteresis abundances in our samples are so weakly dependent on the tunneling current. The characteristic tempera- ture above which the two hysteretic Ni nanoparticles stop displaying magnetic hysteresis is 2−3K.32 That characteristic temperature corresponds to the magnetization blockade energy, which is comparable to the single-electron anisotropy. We can conclude that the minimum 12 magnetic anisotropy is the limiting anisotropy of the nanoparticle below which magnetic hysteresis cannot be guaranteed. If the magnetic hysteresis does occur in the nanoparticle with magnetic anisotropy smaller than the minimum magnetic anisotropy, it will do so below 2-3K temperature, the characteristic temperature of the magnetization blockade. In summary, we have performed magnetoresistance measurements on a variety of ferro- magnetic materials 1-5 nm in diameter at 4.2K, and found an abrupt change in magnetic hysteresis abundance between Ni, Py, Fe, and Co nanoparticles. This abruptness leads to the conclusion that there is a minimum magnetic anisotropy energy in a metallic ferromagnetic nanoparticle or a magnetic molecule out of equilibrium, required to guarantee magnetic hys- teresis at low temperatures. The size of the tunneling current does not affect the minimum magnetic anisotropy. Our finding has an implication for the miniaturization of magnetic random access memory. It demonstrates a limit below which reliable reading of soft-layer magnetization cannot be predictable below the blocking temperature. Research supported by the U.S. Department of Energy, Office of Basic Energy Sciences, Division of Materials Sciences and Engineering under Award DE-FG02-06ER46281. We thank Dr. Ding from the Microscopy Center, School of Materials Science and Engineering at Georgia Institute of Technology for his help in taking the TEM image of the Ni nanoparticles. ∗ Electronic address: [email protected] † Electronic address: [email protected]. 1 M. N. Baibich, J. M. Broto, A. Fert, F. N. Van Dau, F. Petroff, P. Etienne, G. Creuzet, A. Friederich, and J. Chazelas, Phys. Rev. Lett. 61, 2472 (1988). 2 P. Grunberg, R. Schreiber, Y. Pang, M. B. Brodsky, and H. Sowers, Phys. Rev. Lett. 57, 2442 (1986). 3 J. C. Slonczewski, Journal of Magnetism and Manetic Materials 159, L1 (1996). 4 L. Berger, Phys. Rev. B 54, 9353 (1996). 5 D. C. Ralph and M. D. Styles, J. Magn. Magn. Mater. 320, 1190 (2008). 6 L. N´eel, Ann. G´eophys 5, 99 (1949). 7 W. F. Brown, Phys. Rev. 130, 1677 (1963). 8 W. Wernsdorfer, E. B. Orozco, K. Hasselbach, A. Benoit, B. Barbara, N. Demoncy, A. Loiseau, 13 H. Pascard, and D. Mailly, Phys. Rev. Lett. 78, 1791 (1997). 9 S. Gueron, M. M. Deshmukh, E. B. Myers, and D. C. Ralph, Phys. Rev. Lett. 83, 4148 (1999). 10 M. M. Deshmukh, S. Kleff, S. Gueron, E. Bonet, A. N. Pasupathy, J. von Delft, and D. C. Ralph, Phys. Rev. Lett 87, 226801 (2001). 11 W. Jiang, P. Gartland, and D. Davidovic, Journal of Applied Physics 113, 223703 (2013). 12 D. Papaconstantopoulos, Handbook of the band structure of elemental solids (Plenum Press New York, 1986). 13 C. Kittel, Introduction to Solid State Physics (Wiley, 2004), ISBN 9780471415268, URL http://books.google.com/books?id=kym4QgAACAAJ. 14 D. Paige, B. Szpunar, and B. Tanner, J. Magn. Magn. Mater. 44, 239 (1984), ISSN 0304-8853, URL http://www.sciencedirect.com/science/article/pii/0304885384902488. 15 C. Tung, I. Said, and G. E. Everett, J. Appl. Phys 53, 2044 (1982). 16 E. Martinez, L. Lopez-Diaz, L. Torres, and C. Garcia- Cervera, Physica B: Condensed Matter 372, 286 (2006), URL http://www.sciencedirect.com/science/article/pii/S092145260501104X. 17 C. Gadsden and M. Heath, Solid State Commun. 20, 951 (1976), ISSN 0038-1098, URL http://www.sciencedirect.com/science/article/pii/0038109876904804. 18 A. Sarkar, S. Kapoor, G. Yashwant, H. G. Salunke, and T. Mukherjee, The Journal of Physical Chemistry B 109, 7203 (2005), URL http://pubs.acs.org/doi/abs/10.1021/jp044521a. 19 F. Luis, F. Petroff, J. M. Torres, L. M. Garcia, J. Bartolome, J. Carrey, and A. Vaures, Phys. Rev. Lett. 88, 217205 (2002), URL http://link.aps.org/doi/10.1103/PhysRevLett.88.217205. 20 M. Yoon, Y. Kim, Y. Kim, V. Volkov, H. Song, Y. Park, and I.-W. Park, Materials Chemistry and Physics 91, 104 (2005), ISSN 0254-0584, URL http://www.sciencedirect.com/science/article/pii/S0254058404006054. 21 N. Galvez, P. Sanchez, J. M. Dominguez-Vera, A. Soriano-Portillo, M. Clemente-Leon, and E. Coronado, J. Mater. Chem. 16, 2757 (2006), URL http://dx.doi.org/10.1039/B604860A. 22 F. C. Fonseca, G. F. Goya, R. F. Jardim, R. Muccillo, N. L. V. Carreno, E. Longo, and E. R. Leite, Phys. Rev. B 66, 104406 (2002), URL http://link.aps.org/doi/10.1103/PhysRevB.66.104406. 23 A. Cehovin, C. M. Canali, and A. H. MacDonald, Phys. Rev. B 66, 094430 (2002). 14 24 G. Usaj and H. U. Baranger, Europhys. Lett. 72, 110 (2005). 25 P. W. Brouwer and D. A. Gorokhov, Phys. Rev. Lett. 95, 017202 (2005). 26 O. Parcollet and X. Waintal, Phys. Rev. B 73, 144420 (2006), URL http://link.aps.org/doi/10.1103/PhysRevB.73.144420. 27 S. Kleff, J. v. Delft, M. M. Deshmukh, and D. C. Ralph, Phys. Rev. B 64, 220401 (2001). 28 E. Burzur´ı, A. S. Zyazin, A. Cornia, and H. S. J. van der Zant, Phys. Rev. Lett. 109, 147203 (2012), URL http://link.aps.org/doi/10.1103/PhysRevLett.109.147203. 29 M.-H. Jo, J. E. Grose, K. Baheti, M. M. Deshmukh, J. J. Sokol, E. M. Rumberger, D. N. Hendrickson, J. R. Long, H. Park, and D. C. Ralph, Nano Letters 6, 2014 (2006). 30 L. THOMAS, F. LIONTI, R. BALLOU, D. GATTESCHI, R. SESSOLI, and B. BARBARA, Nature 383, 145 (1996). 31 J. R. Friedman, M. P. Sarachik, J. Tejada, and R. Ziolo, Phys. Rev. Lett. 76, 3830 (1996). 32 P. Gartland, W. Jiang, and D. Davidovic, Phys. Rev. B 91, 235408 (2015). 33 X. Waintal and P. W. Brouwer, Phys. Rev. Lett. 91, 247201 (2003). 15
1809.03728	1	1809	2018-09-11T08:20:37	Atomically thin p-n junctions based on two-dimensional materials	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	Recent research in two-dimensional (2D) materials has boosted a renovated interest in the p-n junction, one of the oldest electrical components which can be used in electronics and optoelectronics. 2D materials offer remarkable flexibility to design novel p-n junction device architectures, not possible with conventional bulk semiconductors. In this Review we thoroughly describe the different 2D p-n junction geometries studied so far, focusing on vertical (out-of-plane) and lateral (in-plane) 2D junctions and on mixed-dimensional junctions. We discuss the assembly methods developed to fabricate 2D p-n junctions making a distinction between top-down and bottom-up approaches. We also revise the literature studying the different applications of these atomically thin p-n junctions in electronic and optoelectronic devices. We discuss experiments on 2D p-n junctions used as current rectifiers, photodetectors, solar cells and light emitting devices. The important electronics and optoelectronics parameters of the discussed devices are listed in a table to facilitate their comparison. We conclude the Review with a critical discussion about the future outlook and challenges of this incipient research field.	cond-mat.mes-hall	cond-mat	This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E Atomically thin p-n junctions based on two-dimen- sional materials Riccardo Frisenda1, Aday J. Molina-Mendoza2, Thomas Mueller2, Andres Castella- nos-Gomez3 and Herre S. J. van der Zant1,4,5 1 Instituto Madrileño de Estudios Avanzados en Nanociencia (IMDEA-Nanociencia), Campus de Can- toblanco, E-28049 Madrid, Spain. 2 Institute of Photonics, Vienna University of Technology, Guhausstrae 27-29, 1040 Vienna, Austria. 3 Materials Science Factory, Instituto de Ciencia de Materiales de Madrid (ICMM-CSIC), E-28049, Madrid, Spain. 4 Kavli Institute of Nanoscience, Delft University of Technology, Lorentzweg 1, 2628 CJ Delft, The Nether- lands. 5 Departamento de Fisica de la Materia Condensada, Universidad Autonoma de Madrid, Campus de Can- toblanco, E-28049 Madrid, Spain. E-mail: [email protected], [email protected], [email protected] ABSTRACT: Recent research in two-dimensional (2D) materials has boosted a renovated interest in the p- n junction, one of the oldest electrical components which can be used in electronics and optoelectronics. 2D materials offer remarkable flexibility to design novel p-n junction device architectures, not possible with conventional bulk semiconductors. In this Review we thoroughly describe the different 2D p-n junction ge- ometries studied so far, focusing on vertical (out-of-plane) and lateral (in-plane) 2D junctions and on mixed- dimensional junctions. We discuss the assembly methods developed to fabricate 2D p-n junctions making a distinction between top-down and bottom-up approaches. We also revise the literature studying the different applications of these atomically thin p-n junctions in electronic and optoelectronic devices. We discuss ex- periments on 2D p-n junctions used as current rectifiers, photodetectors, solar cells and light emitting de- vices. The important electronics and optoelectronics parameters of the discussed devices are listed in a table to facilitate their comparison. We conclude the Review with a critical discussion about the future outlook and challenges of this incipient research field. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E 1. Introduction The p-n junction has become an important component in modern electronics since its serendipitous discov- ery by Russel Ohl almost 80 years ago.1 This kind of device can be created by joining together two semicon- ductors of different type: a p-type semiconductor containing an excess of holes and n-type one with an excess of electrons. As a result an intrinsic electric field at the interface between them is generated and it can be used to rectify currents or to separate photogenerated electron-hole pairs. In bulk semiconductors the typi- cal way to create a p-n junction is to dope two different parts of a single crystal with different ions or dopants forming a three-dimensional (3D) p-n junction. The resulting face to face arrangement constitutes the only possible device architecture considered for bulk semiconducting materials. If one scales down the dimensions of the semiconductors involved, by passing from 3D to two-dimensional (2D) materials, new and exciting possibilities arise. The design of p-n junctions offers now more possibilities and freedom: p-n junctions in the 2D case can be constructed following two main architectures: a lateral junction, in which the two 2D materials are joined at the same plane (creating a one-dimensional interface between the two materials) and a vertical junction, in which the 2D materials are stacked face to face, thereby exhibiting a two-dimensional overlap. Moreover, the ultra-thin nature of 2D materials gives rise to novel properties compared to 3D semiconductors.2-8 An example is the thickness-dependency of the bandgap for some materials which enables even more possibilities to create different p-n junctions concepts. Figure 1. Scheme of the different p-n junctions architectures based on 2D semiconductors. Us- ing two-dimensional (2D) materials a large variety of p-n junctions can be produced. These can be junc- tions based on a single 2D material (homostruc- tures), on the junction between two different 2D materials (heterostructures) or junctions based on the combination of a 2D material with a material with higher or lower dimensionality (mixed-dimen- sional structures). Different concepts of p-n junc- tions within these categories are shown. Figure 1 shows a schematic of eight kinds of p-n junctions based on 2D materials which one can encounter in the present literature. We separate homojunctions (based on a single 2D material), from heterojunctions (formed by joining two different 2D materials) and mixed dimensional junctions (based on the combination of a 2D material with 0D, 1D or 3D materials. More specifically, the eight kinds of p-n junctions are: This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E 2D homostructures 1. Thickness-based junctions, in which the p- and n-regions are formed by two regions of the same material with different thicknesses. 2. Electrostatically doped junctions, in which the doping in different regions of the same 2D material is controlled by local electrostatic gates. 3. Chemical doping, in which the doping of a region in a 2D material is modified by the adsorption of molecules, nanoparticles or quantum dots onto the surface of the material. 4. Elemental doping, in which two flakes of the same 2D material with different doping are stacked one on top of the other, forming an out-of-plane junction. 2D heterostructures 5. Vertical heterojunctions, in which two different 2D materials are stacked one on top of the other and the junction is formed in the out-of-plane direction. 6. Lateral heterojunctions, in which two 2D materials are joined in the same plane along a one-dimen- sional interface. Mixed-dimensional 7. 2D-0D and 2D-1D p-n junctions, in which a molecular crystal or a nanotube film is in contact with a 2D material. 8. 2D-3D p-n junctions, in which a 2D material is in contact with a bulk 3D semiconductor. 2. Semiconducting 2D materials building blocks and tools The different p-n junction architectures introduced in Section 1 can be realized thanks to the discovery of 2D materials. In this section we will describe the atomic and electronic structure of some of the main 2D materials which can be used to fabricate p-n junctions. We will then introduce the production and isolation methods for these materials discussing top-down approaches such as the deterministic transfer and bottom- up methods such as CVD growth. 2.1. Atomic and electronic structure 2D materials can be extracted out of layered materials by mechanical or chemical exfoliation.9-12 The atoms in these materials are arranged in layers with strong in-plane bonds (typically covalent bonds) and weak bonds between the different layers (generally van der Waals interactions). The layers can be composed of a single atomic plane, like in the case of graphene or hexagonal boron nitride (h-BN), or they can be made out of multiple atomic planes (e.g. in MoS2 each monolayer is composed of three atomic layers). Many families of layered materials are known and in the following we will discuss the properties of some important exam- ples of 2D materials.10, 13-15 Figure 2a shows the crystal structure of a monolayer of graphene, MoS2 and h-BN. Graphene and h-BN share the same hexagonal crystal structure but differ strongly in their electronic structure. Graphene is a zero-gap semiconductor with a linear dispersion close to its neutrality point.16 The doping in graphene can be con- trolled electrostatically and both n-type and p-type can be realized. Among the other methods used to dope graphene one can find chemical doping, substitutional doping and irradiation methods. Moreover many This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E groups dedicated efforts to open a bandgap in graphene using external electric fields, geometric confinement or hydrogen adsorption.17-21 In contrast, h-BN is an insulator with a bandgap of ~6 eV that makes it highly transparent to visible light.22 Among the semiconducting 2D materials the most studied ones are probably MoS2 and other members of the transition metal dichalcogenide (TMDC) family that are composed of a layer with the transition metal atoms sandwiched between two layers of chalcogenides atoms.2, 23-29 MoS2 is a semiconductor with an indi- rect bandgap of 1.2 eV in its bulk or multilayer form but it exhibits a direct bandgap of 1.9 eV when thinned down to a monolayer.30-32 Monolayer WS2, MoSe2 and WSe2 share similar electronic properties, having direct bandgaps of energy between 1 and 2 eV. While MoS2 and WS2 are typically n-doped, WSe2 is ambipolar. An- other semiconducting 2D material largely used in p-n junctions is black phosphorous (BP). This layered al- lotrope of phosphorous is a narrow-bandgap semiconductor with a bandgap of 0.34 eV in bulk that increases to 1.5 eV when BP is thinned one monolayer thick.33-35 Among the materials described BP is by far the most unstable as it degrades quickly when exposed to air and thus it requires to work in a high-vacuum environ- ment or to employ encapsulation protocols (i.e., sandwiching the 2D material in between two protecting insulating layers) to prevent its degradation.36-40 In addition to BP and graphene, the elemental (Xenes) fam- ily comprises also other materials such as silicene, germanene and stanene.41, 42 Other important families of semiconducting layered materials are monochalcogenides such as SnS or GeSe,43, 44 trichalcogenides such as TiS3 or HfS3 45, 46 and metal halides such as PbI2 or CrI3.47, 48 Layered oxides, nitrides and carbides are also actively researched families of 2D materials.49-51 Table 1 reports some of the important material properties of selected 2D materials useful for the construction of p-n junctions. Figure 2. Two dimensional materials employed as building blocks to fabricate p-n junctions. a) A large number of two-dimensional (2D) layered materials with varying chemical composition, atomic struc- tures and electronic properties are available. Starting from the top a zero-gap semiconductor, graphene, a semiconductor, MoS2, and an insulator, h-BN are displayed. b) Schematic band diagram of important mate- rials for 2D based p-n junctions showing the approximate band alignment between the different 2D materials according to the available literature values of the work function and band gap. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E Material Bandgap (eV) Electron affinity (eV) Doping type 1L-MoS2 1.89 ML-MoS2 1.2 1L-MoSe2 1.64 1L-WS2 1.96 1L-WSe2 1L-BP 1.6 1.5 ML-BP 0.35 4.3 4.0 3.9 3.9 3.6 3.9 4.1 n n n n Ambipolar Ambipolar Ambipolar 2.2. Isolation of 2D materials Table 1. Bandgap energy, electron affinity and dop- ing type of some layered materials. Bandgap ener- gies and electron affinities are taken from Refs. 52-55. While 2D materials supported on bulk substrates are known since at least 50 years, their isolation and inte- gration in devices is more recent. In 2004 Geim and Novoselov 9 demonstrated the first electronic device based on a 2D material, fabricated by isolating single-layer graphene from bulk graphite using mechanical cleaving. In this section we will discuss the different approaches developed to produce 2D materials and to integrate them in p-n junctions, including top-down approaches (which are very powerful in creating differ- ent and novel p-n junctions) and bottom-up approaches suitable for large-scale integration. 2.2.1. Top-down approach: deterministic transfer Atomically thin 2D materials can be isolated from bulk layered crystals by mechanical cleaving. The presence of strong bonds in the plane of the layers and weak bonds between them allows these materials to cleave perfectly along the atomic plane. As mentioned above, the first technique developed is mechanical cleavage, also known as the "Scotch tape method". In this method a piece of tape is pressed against the surface of a natural or artificial bulk layered crystal and then it is peeled off, cleaving the crystal and leaving debris and flakes of the material onto the tape surface. The flakes can then be transferred to an arbitrary surface by pressing on it the tape containing the flakes and subsequently releasing the tape gently. The main drawback of the Scotch tape method is that it produces flakes with different sizes and thicknesses randomly distributed over the sample substrate and only a small fraction of these flakes are atomically thin. This limitation can partially be circumvented by the use of optical identification methods to find atomically thin crystals from the crowd of thicker, bulky flakes.56-58 Nevertheless, the combination of the Scotch tape method with optical identification methods alone cannot provide a reliable way to fabricate p-n junctions by artificial stacking of 2D crystals. Since 2010, the research on those artificial stacks of 2D materials has grown exponentially driven by the development of different transfer techniques that allow to place 2D materials on a desired location with an unprecedented degree of control and accuracy.59-63 A typical deterministic place- ment setup used to transfer and artificially stack 2D crystals is schematically depicted in Fig. 3a. Such a setup This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E is usually based either on a zoom lens or on an optical microscope equipped with long working distance objectives. Two manually actuated micropositioners are used to move the target surface and the flake to be transferred. Fig. 3b shows the steps to fabricate a vertical heterojunction using the deterministic transfer method. Figure 3c shows an optical image of a vertical homojunction formed by a n-doped MoS2 flake stacked onto a p-doped MoS2 one. The top image shows an intermediate fabrication step and the bottom image is the final device. Figure 3. Top-down and bottom-up approaches to fabricate p-n junctions. a) Schematic diagram of a setup employed for the all-dry transfer of 2D crystals. b) Schematic diagram of the fabrication of a vertical heterojunction by a top-down approach. c) Sequential optical images of the fabrication of a p-n junction by vertical stacking of a p-type MoS2 flake onto a n-type MoS2 one. d) Schematic diagram of the synthesis process of MoS2/WS2 vertical or lateral heterostructures with a bottom-up method. e) Optical image of a vertically stacked WS2/MoS2 heterojunction synthesized at 850 °C. f) Optical image of a lateral WS2/MoS2 heterojunc- tion grown at 650 °C. Panel (c) readapted from Ref. 64 with permission from John Wiley and Sons and panels (d), (e) and (f) readapted from Ref. 65 with permission from Springer Nature. 2.2.2. Bottom-up approach: CVD growth Many 2D materials can be directly grown in the form of single- or few-layer nanosheets on various sub- strates.66 The production process usually involves a thermal chemical vapour deposition (CVD) process, in which the vapour-phase reactants are generated by thermally evaporation of specific source materials.67-69 This growth method is very promising for the production and large-scale integration of CVD grown 2D p-n junctions and already many vertically stacked and lateral p-n junctions have been demonstrated. The family of TMDCs is particularly interesting for CVD growth because of the relatively small lattice mismatches be- tween its various members and the availability of different doping types. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E In 2014 Gong et al.65 demonstrated a single-step vapour phase growth process for the creation of lateral or vertical monolayer MoS2/WS2 heterostructures controlled by the growth temperature. Figure 3d shows a schematic of the synthesis, which involves sulphur powder, MoO3 as a source of molybdenum and a mixed powder of W and Te for the tungsten. The difference in nucleation and growth rates between MoS2 and WS2 allows for the sequential growth of MoS2 and WS2 instead of the formation of a MoxW1-xS2 alloy. By carrying out the process at a temperature of ~850 ℃ the authors produced vertically stacked WS2-MoS2 bilayers, similar to the structure shown in Fig. 3e, whereas at ~650 ℃ in-plane lateral heterojunctions were created shown in Fig. 3f. In another study Duan et al 70 demonstrated the growth of lateral MoS2-MoSe2 and WS2- WSe2 heterojunctions by switching in situ the vapour-phase reactants to enable lateral epitaxial growth of single- or few-layer TMDC heterostructures. Finally, the growth of MoSe2-WSe2 lateral heterojunctions was demonstrated by Huang and coauthors.71 Recently a different approach was demonstrated by Zheng et al., which used pulsed laser deposition to achieve a single-step growth of a lateral p-n junction between layered In2Se3 and nonlayered CuInSe2.72 The biggest challenge in CVD process of heterojunctions is probably the interplay between the many degrees of freedom of the system (such as temperature, flow rate, substrate, lattice mismatch and others) which makes it difficult to ensure the reproducibility of the final process out- come in different laboratories and therefore it requires the full optimization of the growth recipe in every new growth setup. 3. p-n junctions based on 2D materials Using to the top-down and bottom-up approaches described in the previous section, a large number of p-n junction devices based on 2D materials has been demonstrated in literature. In this section, we will review these results by passing from homojunctions to heterojunctions and ending with the mixed-dimensional junctions. 3.1. Homostructures Homojunctions devices are p-n junctions based on a single 2D material. Figure 4a-c show three examples of such junctions: (1) based on quantum-confinement effects, (2) electrostatic gating and (3) on chemical dop- ing to obtain a spatial variation of the doping profile. A fourth kind of homojunction is based on elemental doping of 2D materials. 3.1.1. Thickness modulation Figure 4a shows a 2D homojunction based on thickness modulation. In ultra-thin materials, the bandgap energy becomes a thickness-dependent quantity because of q uantum-confinement effects. This permits the creation of a p-n junction in which the p and n regions are made of the same material, just with different thickness. One example shown in Fig. 4d is a p-n junction based on a single WSe2 flake.73 To fabricate this device Xu and coauthors started from a bilayer WSe2 flake and then partially thinned the flake to a monolayer with an Ar plasma. Metallic contacts to the monolayer and bilayer regions were subsequently defined. To fabricate this kind of homojunctions one can also take advantage of the exfoliation process itself as it gener- ally produces flakes that are already composed of different thickness regions.74 This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E 3.1.2. Electrostatic doping The reduced thickness of 2D materials usually comes along with a large electric field effect tunability. Figure 4b shows a schematic diagram of an electrostatically defined 2D p-n junction and Fig. 4e shows the optical picture of the actual device. In this case WSe2 is used because of its ambipolar nature (by tuning the polarity of the gate voltage one can make it either n- or p-type). Several groups reported this kind of device.75-79 Here, we will discuss the implementation by Pospischil and coauthors. In their work, two metallic gates separated by 460 nm are defined onto a SiO2/Si substrate and subsequently covered by a 100 nm thick Si3N4 gate die- lectric. A flake of mechanically exfoliated WSe2 was then deterministically transferred, partly covering both prepatterned gates electrodes, and source and drain electrodes were defined afterwards. By independently controlling the two split gates the doping in the two regions of the WSe2 flake located above the local gates can be controlled. Similar devices based on this split-gate geometry were fabricated with BP80 and gra- phene.81, 82 A different geometry was recently used by Li et al.83 in which a single local graphene gate elec- trode, insulated by a thin h-BN flake, is partly covered by a WSe2 flake forming the active channel connected to source and drain electrodes. A complementary approach is to use ionic gating of the 2D material instead of the solid state split gates.84-87 Figure 4. Different examples of p-n junctions based on 2D materials homojunctions. a-c) Schematic diagram of a p-n junction based on (a) MoS2 with different thickness, (b) ambipolar WSe2 with two split gate This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E electrodes and (c) local molecular doping of partially h-BN encapsulated MoS2. d) Optical image of a thick- ness-modulated p-n junction based on single-layer and bilayer WSe2. e) Optical image of a split-gate homo- junction as schematically drawn in panel (b). f) Optical image of a chemically doped MoS2 p-n junction as schematically illustrated in panel (c). The device is covered with molecules and part of the MoS2 is protected by a h-BN flake. Panel (d) readaptedfrom Ref. 73 with permission from IOP Publishing, panel (e) readapted from Ref. 77 with permission from Springer Nature and panel (f) readapted from Ref. 88 with permission from American Chemical Society (copyright 2014). 3.1.3. Chemical doping In addition to making 2D materials sensitive to external electric fields, the reduced thickness gives them a high sensitivity to the environment surrounding their surface. Molecules physisorbed or chemisorbed onto the surface can influence transport in the 2D material for example by introducing doping effects89-94 and a p- n junction can thus be fabricated by locally doping the material. This principle has been used to fabricate chemically doped p-n junctions with graphene,95, 96 TMDCs and black phosphorous.97, 98 A chemically doped MoS2 p-n junction is shown in Fig. 4f reported by Choi and coauthors.88 In this work a solution of AuCl3 was deposited onto the surface of a MoS2 flake partially covered by an h-BN flake. AuCl3 is known to induce p- type doping in MoS2 thereby creating a lateral p-n junction, where the n region is provided by the natural n- doped MoS2. On the other hand, a common molecule used to induce strong n-doping in MoS2 is benzene viologen (BV). Li and coauthors99 used BV and AuCl3 to create an out-of-plane p-n junction (a device in which charge transport is out-of-plane) where one face of a few layer MoS2 flake was p-doped with AuCl3 and the other one with BV. 3.1.4. Elemental doping A different approach to dope 2D materials is elemental doping, a technique that has proven to be successful for controlling carrier types in bulk materials. In bulk TMDCs elemental doping with Nb (five valence elec- trons), Fe or Re (seven valence electrons) has been used as a substitutional p-type or n-type dopant to re- place the metallic atoms such as Mo or W (six valence electrons).100 For 2D layered materials this technique was initially demonstrated for MoS2 by Suh and coauthors.101 Substitutional doping with Nb was used to convert MoS2 from n-type to p-type and a vertical p-n homojunction was fabricated by stacking Nb doped MoS2 onto undoped MoS2. Similarly, Jin et al.102 demonstrated a p-n homojunction by stacking undoped MoSe2 onto doped MoSe2 with Nb atoms. Figure 3b shows an optical image of a similar device fabricated by stacking p-type MoS2 doped with Nb onto n-type MoS2 doped with Fe atoms.64, 103 Recently, in a different approach phosphorous atoms were used to substitute the surficial sulfur atoms of MoS2, leading to a p-type doping of MoS2.104, 105 3.2. Heterostructures The combination of two different 2D materials to form a heterostructure is one of the most promising strat- egies because of the large variety of bandgap energies and doping types available among the different 2D materials. Figure 5a-b show schematic diagrams of a vertical and a lateral heterojunction. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E 3.2.1. Vertical junctions Vertical heterostructures represent a popular architecture in 2D materials. Due to a surface free of dangling- bonds and interlayer van der Waals interactions it is possible to stack 2D materials on top of each other without constraint on the lattice constants.6, 106-108 The first example of a van der Waals heterostructure in literature was demonstrated by Dean et al.60 that stacked graphene on top of an h-BN flake. Soon thereafter it was realized that by stacking an n-type 2D material onto a p-type 2D material, using the deterministic transfer method, atomically thin p-n junctions could be created.109, 110 An example of an atomically thin ver- tical p-n junction111 based on single-layer MoS2 and WSe2 is shown in Fig. 5c. The authors fabricated the heterojunction by deterministic placement of individual mechanically exfoliated monolayers of both mate- rials on a SiO2/Si substrate and deposited Pd contacts to inject electrons and holes into the n-MoS2 and p- WSe2 layers, respectively. In another work Lee and coauthors112 introduced graphene contacts for the MoS2 and WSe2 monolayers to improve the collection of charges, realizing a graphene sandwiched 2D p-n junction. The freedom of stacking 2D materials on top of each other permitted the fabrication of many different verti- cal stacks by this top-down approach. Among the different devices we find junctions between two TMDCs, such as WSe2-MoSe2 or MoTe2-MoS2, and junctions of a TMDC with a different 2D layered material, BP-MoS2, GaTe-MoS2 or with ultrathin membranes such as InAs-WSe2. Figure 5d shows a BP-MoS2 vertical p-n junc- tion113 fabricated by transferring a flake of few-layer BP onto a monolayer MoS2 grown by CVD onto a SiO2/Si substrate. More recently vertical p-n junctions were directly fabricated by epitaxial growth without the need of mechanical assembly. Figure 5. P-n junctions based on heterojunctions between two different 2D materials. a-b) Schematic diagram of a vertical (a) and lateral (b) p-n junction. c) Optical image of a vertical p-n junction formed by stacking a monolayer MoS2 flake onto a monolayer WSe2 flake. d) Optical image of a vertical MoS2-black This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E phosphorous device. The dark purple region is monolayer MoS2, while the blue flake is few-layer black phos- phorus. The light purple region is SiO2. e) Optical image of a lateral WSe2-WS2 p-n junction. The white dashed contour highlights an insulating Al2O3 layer, which allows for each electrode to contact only the WSe2 or WS2 respectively. f) Optical image of a WSe2-MoS2 p-n junction device with different electrodes. Panel (c) rea- dapted from Ref. 111 with permission from American Chemical Society (copyright 2014), panel (d) readapted from Ref. 113 with permission from American Chemical Society (copyright 2014), panel (e) readapted from Ref. 70 with permission from Springer Nature and panel (f) readapted from Ref. 114 with permission from The American Association for the Advancement of Science. 3.2.2. Lateral junctions While in literature one can find many examples of vertical 2D junctions, the same is not true for lateral het- erojunctions. In this case a direct mechanical assembly is not possible and one has to rely on bottom-up fabrication methods, as discussed in section 2.2.2. Figure 5e shows a lateral junction device fabricated by Duan et al.70 in which separate contacts are made to WS2 and WSe2. The white dashed rectangle in the image outlines a 50 nm thick Al2O3 layer deposited onto the WSe2 to insulate the WS2 contact electrodes. Figure 5f shows a different device, fabricated by Li et al.,114 in which different metals respectively Ti and Pd are used to contact the MoS2 and WSe2 regions of the grown flake. 3.3. Hybrid and mixed-dimensional Apart from a purely 2D architecture, many efforts were dedicated to produce hybrid p-n junctions where a 2D material is in contact with a material of lower dimensionality, such as in the case of a molecular crystal or a nanotubes film, or higher dimensionality, like in the case of a bulk semiconductor such as Si. Figure 6a- b show a schematic diagram of a 2D-0D and a 2D-3D p-n junction. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E Figure 6. P-n junctions produced by heterostructures with mixed dimensionality. a-b) Schematic of mixed-dimensional 2D-0D (a) and 2D-3D (b) heterojunctions. c) False colour SEM image of single wall carbon nanotubes (CNTs) single layer MoS2 p-n junction. d) Optical image of a bilayer MoS2 flake on top of a SiO2/Si substrate to fabricate a 2D-0D heterojunction by depositing a CuPc molecular crystal through a shadow mask. e) Optical image of a 2D-3D device in an intermediate state of fabrication. A monolayer MoS2 is placed across the sidewall of a square window etched into a SiO2 layer exposing the underlying p-doped silicon. Panel (c) readapted from Ref. 115 with permission from National Academy of Sciences, panel (d) readapted from Ref. 116 with permission from Royal Society of Chemistry and panel (e) readapted from Ref. 117 with permission from American Chemical Society. Molecular organic crystals and nanotube films have surfaces with saturated bonds and are therefore free from dangling bonds. They typically interact via van der Waals forces and this allows the integration of these low-dimensional materials with 2D materials in van der Waals heterostructures.8, 118 Figure 3c shows a col- ored scanning electron microscopy (SEM) image of a 1D-2D junction created by transferring a layer of sorted single-walled carbon nanotubes (SWCNTs) onto a single-layer MoS2 flake. A different kind of 1D-2D junctions have been demonstrated by joining a single nanowire and a 2D material. Among the works we find junctions between MoS2, WSe2 and BP and a ZnO nanowire.119-121 Figure 3d shows a microscope picture of a 0D-2D junction between a bilayer MoS2 flake and a 30 nm thick Cu-phthalocyanine (CuPC) thin film, thermally evap- orated through a window opened in a PMMA layer. Bulk materials such as Si or GaAs, which are largely used in conventional p-n junctions and electronics, can be used in conjunction with 2D materials to create novel p-n 2D-3D junctions. Lopez-Sanchez and coau- thors117 realized a heterojunctions composed of n-type monolayer MoS2 and p-type silicon. Figure 6e shows This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E this MoS2-Si device fabricated by transferring a MoS2 monolayer flake onto a prepatterned highly doped p- type Si substrate covered with SiO2 containing a window through which the underlying Si is exposed. Similar devices based on MoS2 and Si have been thoroughly studied in literature.122-124 A different geometry was shown by Wang et al. 125, who deposited vertically standing MoS2 onto a silicon substrate with a scalable sputtering method. Using different materials, Gehring et al.126 demonstrated a 2D-3D junction composed of a few tens of nm thick black phosphorus flake on top of a highly n-doped GaAs substrate. Multilayer 4. Electrical properties 4.1. The p-n junction as a rectifier The first use of a p-n junction that we will discuss is that of a current rectifier. The built-in electric field at the p-n interface allows the flow of charge carriers in one direction (forward bias) while blocking the current in the other direction (backward bias). Many 2D based p-n junctions have been used as rectifiers, showing excellent performances and novel functionalities. One such example is the control of the rectification ratio with a gate field. Deng and coauthors reported gate-tunable rectifying current-voltage characteristics (IVs) in a monolayer MoS2-BP p-n junction (see Fig. 5d). Figure 7a shows IVs recorded at various gate voltages between -30 V and 50 V. At negative gate voltages they observe a reduction of both the forward current (p- type connected to positive voltage and n-type to negative voltage) and of the reverse current (p-type con- nected to negative voltage and n-type to positive voltage). Increasing the gate voltage leads to an increase of both the forward and the reverse current in the device. The rectification ratio, defined as the ratio of the forward/reverse current, increases as the back gate voltage decreases as can be seen in Fig. 7b. At a gate voltage of -30 V, a rectification ratio of 105 is obtained. This modulation can be achieved because the band alignment between MoS2 and black phosphorus at the p-n junction interface can be tuned by the gate voltage. Moreover, the gate voltage also modulates the sheet resistance and the contact resistance of the MoS2 and BP respectively. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E Figure 7. Atomically thin diodes based on 2D p-n junctions. a) Gate tunable current-voltage (IV) charac- teristics of a BP-MoS2 vertical heterojunction. b) Rectification ratio at VSD = ±2 V as a function of gate voltage of the device of panel (a). c) IV characteristics of a double-gated WSe2 homojunction device in different local gate configurations. d) Schematic diagram of the band structure and the state of the local gates of the device in panel (c) in PP, PN, NP and NN configuration. Panels (a) and (b) readapted from Ref. 113 with permission from American Chemical Society (copyright 2014) and panel (c) readapted from Ref. 79 with permission from American Chemical Society (copyright 2014). A second kind of devices, that can be called "reconfigurable diode", employs electrostatic doped p-n junctions based on a split gate configuration, which allows inducing locally hole- and electron-doping in different parts of the channel; a new feature introduced by the 2D nature of the devices.76-78 Figure 7c shows for example the IV characteristics of a double-gated monolayer WSe2 homojunction studied by Groenendijk and coau- thors.79 By tuning the Fermi energy both hole- and electron-doping can be readily accessed due to the ambi- polarity of WSe2. Two of the four IVs of Fig. 7c are linear (NN and PP configuration) while the other two are highly non-linear (PN and NP configuration) displaying rectifying behavior, whose direction can be con- trolled by the gate bias polarity. The voltages applied to the local gates used to achieve these four configura- tions are shown in Fig. 7d together with a simplified band diagram of the device. 4.2. Physical mechanism of the electrical transport Although the shape and features of the IV characteristics of 2D p-n junctions are similar to those of a con- ventional p-n junction, the underlying physical mechanism of rectification can be very different. When con- sidering the two model junctions depicted in Fig. 8a one can see that reducing the thickness of the p- and n- This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E type materials modifies the interface between the two materials. In a bulk p-n junction, charge transfer at the interface between the two materials creates a depletion region from which all the free charges are re- moved. In the case of atomically thin junctions (for example in a vertical junction between monolayer MoS2 and monolayer WSe2) such a depletion region cannot be formed because of the reduced thickness. Simplified diagrams of the band profiles for a multilayer and a monolayer junction, taken along the thickness direction, are shown in Fig. 8a. While the bands of the p- and n-type materials bend in the depletion region in the bulk case (right panels), a sharp discontinuity of the bands is present at the monolayer interface (left panels). The depletion region in a bulk p-n junction is modified by the application of a voltage as schematically de- picted in Fig. 8b. While under forward voltage the size of the depletion region is reduced, under reverse bias its size increases. With increasing forward voltage, the depletion zone eventually becomes thin enough that the built-in electric field cannot counteract charge carrier motion across the p-n junction; this leads to an increase in current. In the case of an atomically thin junction under forward bias the current is governed by tunneling-mediated interlayer recombination between majority carriers at the bottom (top) of the conduc- tion (valence) band of the n-type (p-type) material. This interlayer recombination (Fig. 8c) can be described by two physical mechanisms or a combination of both: Langevin recombination, which is mediated by Cou- lomb interaction and describes the direct recombination of an electron and a hole, or Shockley -- Read -- Hall (SRH) recombination, which is mediated by inelastic tunnelling of majority carriers into trap states in the gap.127-130 These two processes can be present both at the same time in a 2D p-n junction and each of these processes predicts a different dependence of the electron-hole recombination ratio on the majority carriers density. The rectifying IVs characteristics can then be explained by an increase of the interlayer recombina- tion rate under forward bias and the photocurrent generated in a 2D junction and its dependence on a gate field can be modeled by the two recombination processes.112 This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E Figure 8. P-n junctions in bulk and ultra-thin materials. a) Schematic of the devices (top) and band profiles (bottom) in a multilayer-multilayer p- n junction (right) and monolayer-monolayer (left). In the bottom panels, energy is depicted on the ver- tical axis and the device thickness on the x-axis. b) Schematic of the bands in a bulk p-n junction under equilibrium (VSD=0 V) and under forward and re- verse bias. c-d) Schematic diagrams of interlayer re- combination (c) in a monolayer-monolayer p-n junction and of (d) the exciton dissociation process. 5. Optoelectronic properties 5.1. Response to illumination: photodetectors and solar cells P-n junctions constitute the central building blocks of photodetectors, solar cells and light emitting diodes (LEDs). Their main application is thus in optoelectronics. Typically, there are two operating modes for p-n junctions: photovoltaic mode (PV), in which the p-n junction is not biased, and photoconductive mode (PC), where the p-n junction works under reverse external bias.131, 132 The PV effect forms the basis for the solar cells. The PC mode is used in photodetection applications and has the advantages of having a lower capaci- tance that improves the response time and the presence of an external electric field facilitates the separation of electro-hole pairs improving the responsivity. Some of the most studied vertical p-n junctions are heterostructures made from WSe2 and MoS2.109, 111, 112 In the work by Furchi et al.111 the authors transferred a mechanically exfoliated monolayer of WSe2 on top of a monolayer MoS2, both on a SiO2/Si substrate, with source and drain electrodes placed in contact with each This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E of the materials. Due to the ambipolar nature of WSe2, this device could be tuned into n-n junction or p-n junction regimes by means of the back gate voltage and, in the p-n configuration, a rectification ratio of ∼100 was achieved. By illuminating the device in the p-n configuration with a white light source, a photovoltaic effect was observed, as shown in the IVs of Figure 9a measured at increasing optical powers, with an external quantum efficiency (EQE) of ∼ 1.5 % and a power conversion efficiency of ∼ 0.2 %. In this vertical p-n junc- tion, photons are absorbed in both materials generating electron-hole pairs followed by relaxation of the photogenerated carriers and charge transfer between the layers (see Fig. 8c-d). The carriers have to laterally diffuse to the electrodes and electron-hole recombination may occur reducing the efficiency of the solar cell. Lee et al.112 circumvented this issue by employing graphene electrodes, sandwiching the semiconducting monolayers, and enhancing the responsivity of the devices by a factor of ∼5. This increase in responsivity comes mainly from the smaller travel distance that the photoexcited carriers have to travel to reach the electrodes leading to a more efficient carrier extraction in the vertical direction as shown in Fig. 9b. In order to increase the EQE of the devices, solar cells involving materials with different thicknesses were investi- gated, finding EQEs of 2.4 %, 12 % and 34 % for monolayer, bilayer and multilayer p-n junctions respectively. This improvement is due not only to the enhanced light absorption in the multilayer devices, but also due to the exponential suppression of direct electron tunnelling between the two graphene electrodes. In an opti- mized (∼15 nm thick) MoS2-WSe2 multilayer heterostructure, Wong et al.133 demonstrated internal photo- carrier collection efficiencies exceeding 70 % and power conversion efficiencies of up to 3.4 % at 633 nm wavelength. Furthermore, other combinations of p- and n-doped materials have been explored to realize vertical solar cells, although p-doped 2D materials are less frequent than n-doped materials. Among the naturally p-doped 2D semiconductors we list black phosphorus, franckeite and MoTe2. Deng et al. investigated a BP-MoS2 ver- tical p-n heterojunction working in the visible range of the electromagnetic spectrum with a peak EQE of 0.3 %,113 a value lower than that of WSe2-MoS2 solar cells. Nevertheless, one of the most attractive features of BP is its low bandgap, allowing light absorption also in the near-infrared part of the electromagnetic spec- trum. MoS2-BP junctions have shown EQEs up to ∼20 % under illumination with 1.55 µm wavelength light, employing few-layer flakes of both BP and MoS2.134 More recently, Molina-Mendoza et al. investigated the capability of franckeite as near-infrared solar cell in a p-n junction formed by multilayer franckeite-mono- layer MoS2 able to generate an electrical power of ∼1 pW under infrared illumination with wavelength of 940 nm as shown in Figure 9d.135 Also, few-layer MoTe2-few-layer MoS2 136 and monolayer MoTe2-MoS2 137 p-n junctions have shown promising results in light detection and energy harvesting in the near-infrared. The junction between monolayer MoTe2-MoS2 shows photoresponse at 1550 nm, due to interlayer transi- tions between the two materials.137 However, for the individual device made of a pure MoS2 or MoTe2 mon- olayer, no photoresponse is observed at 1550 nm. Lateral 2D junctions are also very promising for optoelectronic applications. In the case of heterojunctions, several combinations of materials have already been achieved and studied as energy harvesting devices. CVD grown WSe2-MoS2 was used by Li et al. to build a p-n junction which showed an open-circuit voltage of 220 meV under white light illumination.114 In another work, the authors found for the same system (Figure 9d) a power conversion efficiency of 2.6 % and an increased open-circuit voltage of 390 meV.138 Devices based This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E on laterally grown MoSe2-MoS2 and WS2-WSe2 heterostructures have also been studied by Duan et al, where the authors found that the WS2-WSe2 heterostructures show rectifying behaviour and a photovoltaic effect, yielding an open-circuit voltage of 470 meV and an EQE of ∼ 10%.70 With respect to homostructures, the most studied materials are WSe2 and BP due to their ambipolar nature that allows for electrostatic doping of different region of the semiconductor channel. Three different groups independently reported a solar cell built by inducing p and n doping in a single-layer WSe2 flake by means of local back-gate electrodes (see Fig. 4b).76-78 Figure 9e shows IVs of this device under illumination in different configurations of the local gates. The IVs shows that photogeneration in PV mode occurs only in the NP and PN configurations, with clear short-circuit current and open-circuit voltage. The inset of Fig. 9e shows the photovoltaic power that can be extracted in the PN configuration. Baugher et al. and Pospischil et al. reported an external quantum efficiency for their solar cells of 0.2 % and 0.5 %, respectively. Alternatively, Buscema et al. reported a similar device structure employing BP as the semiconductor channel,80 which permitted to extend the range of power generation for wavelengths up to 940 nm. On the other hand, the WSe2 devices are limited to the visible region of the electromagnetic spectrum. Figure 9. Optoelectronic properties of 2D p-n junctions. a) Current-voltage (IV) characteristics of a WSe2- MoS2 vertical heterojunction at different illumination powers. b) Photoresponsivity versus voltage traces of WSe2-MoS2 vertical heterojunctions sandwiched between two graphene electrodes for three devices with different thickness of the WSe2 and MoS2 flakes. The x-axis of each trace is normalized by the thickness of the corresponding heterojunction. c) Gate tunable IV characteristics of a WSe2-MoS2 vertical heterojunction This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E kept under illumination. d) IV characteristics of a franckeite-MoS2 vertical heterojunction in dark and under infrared illumination. e) IV characteristics of a double-gated WSe2 device under optical illumination. The biasing conditions are: PN (Vgl = -40 V, Vgr = 40 V), NP (Vgl = 40 V, Vgr = -40 V), NN (Vgl = Vgr = 40 V), PP (Vgl = Vgr = -40 V). When operated as a diode (PN and NP), electrical power (Pel) can be extracted. Inset: Pel versus voltage in the PN configuration. Panel (a) readapted from Ref. 111 with permission from American Chemical Society, panels (b) and (c) readapted from Ref. 112 with permission from Springer Nature, panel (d) readapted from Ref. 135, panel (e) readapted from Ref. 77 with permission from Springer Nature. 5.2. Scanning photocurrent studies The experiments discussed above were performed under global illumination, that is by illuminating the de- vice with a spot larger than its size. We will now turn to local illumination and specifically to scanning pho- tocurrent (SPC) studies. By scanning a diffraction-limited light spot over a device while measuring the cur- rent one can construct SPC maps that yield information on the spatial profile of the bands or the electric fields present in the device.139-144 This information is crucial to understand the underlying mechanisms rul- ing the photocurrent generation. The ultrathin nature of 2D p-n junctions allows the light to easily reach the p-n interface both in lateral and in vertical architectures.64, 78, 145, 146 Figure 10a shows an optical image of a vertical p-n junction composed of elemental doped p-MoS2 and n-MoS2.64 By recording a SPC map at zero bias the authors studied the local generation of the short circuit current in the device arising from the PV effect. Figure 10b shows such a map recorded with a laser of 532 nm and with highlighted contours of the n- and p-type flakes. A negative photocurrent (blue/red regions) is generated in the region where the two MoS2 flakes overlap as a result of the strong electric field in the depletion region that separates the photocarriers. Compared to global illumination, the SPC map highlights that photocurrent generation can be non-homoge- neous across the overlapping region. In Fig. 10b, a hotspot is clearly present, indicating that the interaction between the two flakes is not spatially homogenous, most likely due to the presence of interlayer adsorbates trapped between the layers during the assembly of the device. SPC studies have also been conducted in lateral devices such as the one depicted in Fig. 10c. Ross et al. com- bined SPC maps with photoluminescence (PL) maps to study an electrostatically gated WSe2 device.78 The zero bias SPC map, shown in Fig. 10d, has been recorded at 100 K with a 660 nm laser and shows that the photocurrent generation occurs at the interface between the p- and the n-doped regions. Figure 10e presents the corresponding map of the integrated PL intensity, which is homogenous across the whole WSe2 flake, indicating that the luminescence is not quenched by the underlying gates. More revealingly, Fig. 10f shows a color map of the peak photoluminescence photon energy, exhibiting two distinct regions clearly correlated with the n-doped (blue) and p-doped (red) regions of the WSe2 flake. Above the gate held at Vgl = +8.0 V the presence of negatively charged X- trions (two electrons and one hole) dominates, while above the other gate, held at Vgr = -8.0 V, the higher-energy positively charged X+ trions (two holes and one electron) are more frequent, implying an excess of holes. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E Figure 10. Scanning photocurrent studies of 2D p-n junctions. a) Optical image of the device. b) Photo- current image of the MoS2 p-n junction from panel (a) with zero bias voltage applied (map of the short circuit current). c) Microscope image of a monolayer WSe2 electrostatic p-n junction device. The source and drain contacts are white and the two bottom gates are red. d) Corresponding scanning photocurrent image show- ing pronounced photocurrent generation localized at the junction. The black dashed lines outline the back gates whereas the white dashed line present the WSe2 flake. e) Integrated photoluminescence map. f) Pho- toluminescence peak energy map showing p and n regions as a result of the different energies of oppositely charged excitons. X-(X+) represents the negatively (positively) charged exciton found in the n (p) region. Pan- els (a) and (b) readapted from Ref. 64 with permission from John Wiley and Sons and panels (c), (d), (e) and (f) readapted from Ref. 78 with permission from Springer Nature. 5.3. Electroluminescence: light emitting diodes Electroluminescence is the result of radiative recombination of electrons and holes in a material, usually a semiconductor, which release their energy as photons. The most basic requirement for efficient light emis- sion is a direct optical transition in the semiconductor, as in the case of monolayer semiconducting TMDCs. Electrically driven light emission produced by a unipolar current in single layer MoS2 FETs has been reported This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E to take place near the contact electrodes via impact excitation147 or in a suspended sheet by Joule heating.148 However, effective emission requires the injection of both electrons and holes, which is typically achieved using a p-n junction. LEDs based on 2D materials have been realized in different ways, employing different materials and device architectures. For example, the ambipolar nature of single layer WSe2 has been exploited to generate light by employing local gates to electrostatically dope different regions in a semiconductor channel, forming a lateral p-n junction device.76-78 This kind of device (Figs. 4a and 4d) exhibited electroluminescence efficien- cies (ratio between emitted optical power and electrical input power) ranging between 0.1 and 1%, with energy emission peaks at 752 nm (see Fig. 11a, Ref. 77) and 751 nm (Fig. 11b, Ref. 76) at room temperature. The difference between the energy of the emission peaks is attributed to different dielectric environments that can influence the exciton binding energy. Zhang et al. have employed mono- and few-layer WSe2 in p-i-n (i, intrinsic) junctions with an electric double- layer transistor (EDLT) architecture,149 where the doping of the semiconductor channel is modified by tun- ing the voltage between the drain and source electrodes with respect to the gate electrode, enabling the accumulation of opposite charge carriers close to each of the electrodes. This device configuration circum- vents the requirement of using monolayers for light emission due to the breakdown of the inversion sym- metry in few-layer flakes under high-gate fields, and it also enables the emission of circularly polarized light as shown in Figure 11c. A similar device structure has been used to induce light emission by means of liquid- gated transistors employing mono- and bilayer WS2,150 CVD-MoS2 151 or bulk ReS2.152 Concerning electroluminescent heterojunctions and mixed dimensional junctions many examples have been reported in literature. Lopez-Sanchez et al. demonstrated the light emission from a monolayer MoS2-highly p-doped Si hybrid junction (see Fig. 6e).117 Figure 11d shows a gray-scale optical picture of the electrolumi- nescence generated from the device, which exhibits peak emission at ∼2 eV. This emission energy is blue- shifted by almost 200 meV from the monolayer MoS2 exciton energy, which is due to the influence of the dielectric environment. The authors also reported that devices with unencapsulated MoS2 show a significant reduction of current and emitted light intensity after a few days in ambient conditions. On the other hand, by encapsulating the MoS2 monolayer with 30 nm thick HfO2 or Al2O3 the stability of the device can be greatly extended. In a different study Li et al. reported EL from a p-i-n junction between multilayer MoS2 and a GaN substrate, using Al2O3 as insulator.153 Furthermore, LEDs based on 2D heterostructures have been reported. Cheng et al. measured the light emis- sion from a WSe2/MoS2 vertical heterojunction.154 In this case, both monolayer-WSe2/few-layer-MoS2 and bilayer-WSe2/few-layer-MoS2 were investigated (Fig. 11e), revealing different features in the emission spec- trum related to excitonic peaks A (Fig. 11f) and B (not shown here), hot electron luminescence peaks A' and B' (Fig. 11f) and indirect bandgap emission (not shown here). Table 2 lists some photonic properties of elec- troluminescent 2D p-n junctions. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E Peak emis- sion (nm) Threshold cur- rent (nA) Materials p/n WSe2 WSe2 WSe2 WSe2 Si/MoS2 WSe2/MoS2 WS2 MoS2 752 751 750 740 620 792 630 650 Notes - Estimated electroluminescence effi- ciency ηEL ≈ 0.1% - Refer- ence 77 76 78 4 50 5 1000 Emission of circularly polarized light 149 109 150000 - 40000 Stable after encapsulation with 30 nm thick oxide - - Large-area MoS2 monolayer grown by chemical vapor deposition 117 154 150 151 Table 2. Photonic properties of on 2D p-n junctions. The threshold current is the lowest current able to produce light emission from the device. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E Figure 11. Light emission from 2D p-n junctions. a) Electroluminescence (EL) emission spectra of a mon- olayer WSe2 electrostatic p-n junction recorded for constant currents of 50, 100 and 200 nA. The curves are offset for clarity. The green curve demonstrates that no light emission is obtained under unipolar (n-type) conduction. b) Left axis: external quantum efficiency (EQE) as a function of wavelength at a constant laser power of 2 mW in a monolayer WSe2 device. Peaks in the EQE correspond to exciton transitions A, B and A', as labelled. Right axis: electroluminescence intensity from a monolayer WSe2 electrostatic p-n junction. The traces are offset vertically for clarity. Inset: Diagram of the band structure around the K and Q points, with arrows indicating the lowest-energy exciton transitions for monolayer WSe2. c) Circularly polarized EL spec- trum of a WSe2 electric double-layer transistor for different voltages. Inset: Diagram of the band structure around the K and K' points. d) Intensity map showing the electroluminescent emission of a MoS2/Si hybrid This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E p-n junction. The entire surface of the heterojunction is emitting light. e) False color EL image of a MoS2-WSe2 heterojunction device under an injection current of 100 μA. f) EL spectra of a monolayer WSe2/MoS2 hetero- junction at different injection current. Panel (a) readapted from Ref. 77 with permission from Springer Nature, panel (b) readapted from Ref. 76 with permission from Springer Nature, panel (c) readapted from Ref. 149 with permission from The American Association for the Advancement of Science, panel (d) readapted from Ref. 117 with permission from American Chemical Society, and panels (e) and (f) readapted from Ref. 154 with permission from American Chemical Society. 6. Comparison between devices In this section we show three tables that list the different p-n junctions based on 2D materials found in liter- ature. Table 3 contains the 2D homojunctions, Table 4 the 2D heterojunctions and Table 5 the mixed-dimen- sional junctions. For each device we report some of the relevant parameters such as the materials composing the device and the thickness. We also list figure of merits for electronic or optoelectronic applications such as the rectification ratio, the responsivity or the open circuit voltage. In the "Materials" column, in the case of heterojunctions and mixed-dimensional junctions the two materials used are listed with the p-type as first and the n-type as second material. In the "Thickness" column the thickness of the junction is given either in nm or in number of layers (indicated by the code nL where n is the number of layers, ML stays for multilayer). Thickness Rectification (nm) ratio VOC (V) Reference 2L/1L 10 ML/1L 1000 MoSe2 4/28 100000 BP WSe2 WSe2 WSe2 WSe2 BP WSe2 11L/6L 600 5.0 1L 1L 1L 6.5 1L 10000 50 500 100000 11 - - 0 0.24 0.21 0.83 0.03 0.85 0.72 0.05 0.7 73 74 155 156 83 78 77 76 80 79 Materials p/n WSe2 MoS2 Homo- junc- tions s s e n k c i h T c i t a t s o r t c e l E l a c i m e h C l a t n e m e l E This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E MoS2 MoS2 MoS2 BP BP MoSe2 MoS2 BP MoS2 3 60 7 3 10 6 15 8.5 10 100 100 10 0.6 0.5 - 100000 0.45 50 1000000 - 5600 250 - 0.35 0.45 0.14 0.58 99 88 88 98 97 157 103 158 64 Table 3. p-n junctions based on 2D homojunctions. Hetero- junc- tions Materials p/n Thickness Rectification (nm) ratio VOC (V) Reference WSe2/MoS2 1L/1L WSe2/MoS2 1L/1L WSe2/MoS2 2L/13L 50 50 15 0.55 0.5 0.27 BP/MoS2 11/1L 100000 0.3 WSe2/MoS2 5/10 WSe2/MoSe2 1L/1L 50 50 0.28 0.055 WS2/MoS2 ML/ML 10000 0.25 GaTe/MoS2 20 400000 0.22 InAs/WSe2 >20 1000000 - BP/MoS2 15 >100 MoTe2/MoS2 1L/1L 10 MoTe2/MoS2 4L/4L 4000 0.36 0.15 0.3 WSe2/MoSe2 3L/3L 10000 0.46 111 112 154 113 159 160 161 162 163 164 137 136 165 l a c i t r e V l a r e t a L This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E WSe2/BP 12/20 2500 franckeite/MoS2 25/1.4 400 0.29 0.08 ReSe2/MoS2 60/7 60000 0.1 ReS2/ReSe2 64/48 3150 0.18 GaSe/InSe 19/13 200000 GeSe/MoS2 WSe2/MoS2 MoTe2/SnSe2 - 9 31 >100 1000 1000 SnSe/MoS2 28/7 100000 SnS/WS2 200/0.7 15 - - - - - - WSe2/SnS2 1L/1L 10000000 0.03 CuO/MoS2 150/1L WS2/MoS2 WSe2/WS2 In2Se3/CuInSe2 WSe2/MoS2 1L ML 14 1L 10 100 - 10 10 - 0.12 0.47 0.03 0.22 166 135 167 168 169 170 171 171 172 173 174 175 65 70 72 114 Table 4. p-n junctions based on 2D heterojunctions. Hetero- junc- tions Materials p/n Thickness Rectification (nm) ratio VOC (V) Reference D 1 - D 2 , D 0 - D 2 SWCNT/MOS2 -/1L 10000 Rubrene/MoS2 300/5 100000 Pentacene/MoS2 40/2L 5 C8-BTBT/MoS2 5 100000 CuPC/MoS2 20/1L 10000 - - 0.3 0.5 0.6 115 176 177 178 116 D 3 - D 2 This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E Si/MoS2 Ge/MoS2 Si/MoS2 BP/GaAs WS2/GaN Si/MoS2 Si/Bi2Se3 LSMO/MoS2 1L 2L 1L 15 400 12.5 - 1L - 2 100 120 1000 20 50 1000 0.41 - 0.58 0.55 - - 0.24 0.4 122 179 117 126 180 181 182 183 Table 5. p-n junctions based on mixed dimensional heterojunctions. 7. Future perspective and challenges Despite the large number of experiments discussed in this Review, many challenges still need to be overcome for the integration of 2D p-n junctions in mass-produced electronic components. The two most important challenges are the large-scale fabrication of 2D p-n junctions and the environmental degradation of 2D ma- terials. 7.1. Large-scale fabrication and environmental degradation The first main challenge is related to the large-scale production of tailored van der Waals heterostructures with well-controlled interfaces. Even if the deterministic placement methods are very successful for labora- tory-scale experiments, they are not suited for commercial applications. Growing methods like the CVD growth described in Section 2.2.2 have already proven to be capable of growing high-quality 2D materials including lateral and vertical heterostructures at a laboratory-scale. Van der Waals epitaxial methods hold even more promises to the synthesis of high-quality 2D heterostructures. Up-scaling of these growth meth- ods is possible and in the years to come the realization of higher quality devices can be expected.184 A second, promising strategy to up-scale the production of 2D p-n junctions is to combine the growth of single 2D materials (such as MoS2) with various doping techniques (mostly chemical or electrostatic). A second challenge is the environmental degradation of many of the known 2D materials. For example, when exposed to air, black phosphorous in its ultrathin form tends to uptake moisture which degrades the elec- tronic properties of the material.37-40 In the case of BP the most accepted mechanism for the degradation involves the reaction of the material with oxygen which changes the material properties.185, 186 One way pre- vent this degradation is by encapsulating the air-sensitive material between two flakes of h-BN under oxy- gen- and moisture-free conditions.187-189 One active area of 2D materials research is therefore dedicated to up-scale the encapsulation methods. A different approach that is currently pursued is the active search for new 2D materials, which do not present degradation problems, that could come either from synthesis (for example TiS3) or from natural sources (e.g. franckeite).46, 135, 190, 191 This active search already helped to in- crease the number of available 2D materials from just a handful to more than twenty in less than ten years. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E 7.2. Future perspectives Apart from conventional optoelectronic applications, 2D p-n junctions hold still many unexplored applica- tions and fundamental questions. For example, thermoelectric applications of 2D p-n junctions have not been thoroughly investigated yet. The traditional Peltier device, a component largely used in electronics for cool- ing (and less commonly for heating), is based on a p-type and a n-type semiconductor thermally connected in parallel and electrically in series. Van der Waals heterostructures could be used to fabricate atomically thin cooling (or heating) elements combined with the other attractive properties of 2D materials, such as the high transparence or the flexibility. Another interesting route is the study of novel p-n junction geometries (like for example circular p-n junctions recently demonstrated in graphene192, 193) or new devices based on 2D p-n junctions, such as memories or logic gates.83, 194 The real possibilities still hidden in 2D materials are probably many more than the ones discussed and 2D p-n junctions hold many promises for commercial applications. These 2D junctions are especially interesting building blocks for flexible and transparent electronic components, such as solar cells or light emitting di- odes.195, 196 Another application that can benefit from the ultrathin nature of 2D p-n junctions is in light sens- ing and harvesting applications or for nanophotonics.14, 197 As we discussed in section 5, 2D p-n junctions can be used as photodetectors and many materials combinations are available that can be used to design devices sensible to wavelengths ranging from the infrared to the ultraviolet have already been demonstrated.134, 135, 137 8. Conclusions In recent years we have witnessed the production of novel p-n junctions that take advantage of the ultrathin nature of 2D materials. Both bottom-up and top-down production methods have proven capable of creating p-n junctions of high quality with exquisite optoelectronic properties. In this Review we have revised the recent progress on 2D p-n junctions, discussing the most used materials and fabrication methods and exam- ples of 2D p-n junctions from literature belonging to eight different junction architectures. With these archi- tectures different applications have been realized and we have discussed experiments that use 2D p-n junc- tions respectively as rectifiers, as photodetectors/solar cells and as light emitting diodes. Finally a compari- son between the important figures of merit of the various devices from literature is made. By no means, this field is closed. 2D materials continue to offer many opportunities to fabricate novel p-n junctions with out- standing properties that open up exciting scientific routes both in terms of fundamental questions and in term of applications. ACKNOWLEDGEMENTS AC-G acknowledges funding from the European Commission under the Graphene Flagship, contract CNECTICT-604391. RF acknowledges support from the Netherlands Organization for Scientific Research (NWO) through the research program Rubicon with project number 680-50-1515. This project further re- ceived funding from the European Union's Horizon 2020 research and innovation program under grant agreement No. 696656 (Graphene Flagship). This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E COMPETING INTERESTS The authors declare no competing financial interests. REFERENCES M. Riordan and L. Hoddeson, IEEE spectrum, 1997, 34, 46-51. Q. H. Wang, K. Kalantar-Zadeh, A. Kis, J. N. Coleman and M. S. Strano, Nature Nanotechnology, 2012, 7, 699-712. S. J. Kim, K. Choi, B. Lee, Y. Kim and B. H. Hong, Annual Review of Materials Research, 2015, 45, 63-84. H. Schmidt, F. Giustiniano and G. Eda, Chemical Society Reviews, 2015, 44, 7715-7736. S. Das, J. A. Robinson, M. Dubey, H. Terrones and M. Terrones, Annual Review of Materials Research, 2015, 45, 1-27. K. Novoselov, A. Mishchenko, A. Carvalho and A. C. Neto, Science, 2016, 353, aac9439. Y. Liu, N. O. Weiss, X. Duan, H.-C. Cheng, Y. Huang and X. Duan, Nature Reviews Materials, 2016, 1, 16042. D. Jariwala, T. J. Marks and M. C. Hersam, Nature materials, 2016, 16, 170-181. K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva and A. A. Firsov, Science, 2004, 306, 666-669. V. Nicolosi, M. Chhowalla, M. G. Kanatzidis, M. S. Strano and J. N. Coleman, Science, 2013, 340, 1226419. K. Novoselov, D. Jiang, F. Schedin, T. Booth, V. Khotkevich, S. Morozov and A. Geim, Proceedings of the National Academy of Sciences of the United States of America, 2005, 102, 10451-10453. K. Novoselov and A. C. Neto, Physica Scripta, 2012, 2012, 014006. P. Miró, M. Audiffred and T. Heine, Chemical Society Reviews, 2014, 43, 6537-6554. A. Castellanos-Gomez, Nature Photonics, 2016, 10, 202-204. R. Roldan, L. Chirolli, E. Prada, J. A. Silva-Guillen, P. San-Jose and F. Guinea, Chemical Society Reviews, 2017, 46, 4387-4399. B. Partoens and F. Peeters, Physical Review B, 2006, 74, 075404. M. Y. Han, B. Özyilmaz, Y. Zhang and P. Kim, Physical review letters, 2007, 98, 206805. K. F. Mak, C. H. Lui, J. Shan and T. F. Heinz, Physical review letters, 2009, 102, 256405. D. C. Elias, R. R. Nair, T. Mohiuddin, S. Morozov, P. Blake, M. Halsall, A. Ferrari, D. Boukhvalov, M. Katsnelson and A. Geim, Science, 2009, 323, 610-613. F. Xia, D. B. Farmer, Y.-m. Lin and P. Avouris, Nano letters, 2010, 10, 715-718. R. Balog, B. Jørgensen, L. Nilsson, M. Andersen, E. Rienks, M. Bianchi, M. Fanetti, E. Laegsgaard, A. Baraldi and S. Lizzit, Nature materials, 2010, 9, 315-319. Y. Lin and J. W. Connell, Nanoscale, 2012, 4, 6908-6939. M. Chhowalla, H. S. Shin, G. Eda, L.-J. Li, K. P. Loh and H. Zhang, Nature chemistry, 2013, 5, 263. D. Jariwala, V. K. Sangwan, L. J. Lauhon, T. J. Marks and M. C. Hersam, ACS nano, 2014, 8, 1102-1120. X. Duan, C. Wang, A. Pan, R. Yu and X. Duan, Chemical Society Reviews, 2015, 44, 8859-8876. O. V. Yazyev and A. Kis, Materials Today, 2015, 18, 20-30. W. Choi, N. Choudhary, G. H. Han, J. Park, D. Akinwande and Y. H. Lee, Materials Today, 2017. S. Manzeli, D. Ovchinnikov, D. Pasquier, O. V. Yazyev and A. Kis, Nature Reviews Materials, 2017, 2, 17033. H. Li, J. Wu, Z. Yin and H. Zhang, Accounts of chemical research, 2014, 47, 1067-1075. K. F. Mak, C. Lee, J. Hone, J. Shan and T. F. Heinz, Physical Review Letters, 2010, 105, 136805. A. Splendiani, L. Sun, Y. Zhang, T. Li, J. Kim, C.-Y. Chim, G. Galli and F. Wang, Nano Letters, 2010, 10, 1271-1275. B. Radisavljevic, A. Radenovic, J. Brivio, V. Giacometti and A. Kis, Nature Nanotechnology, 2011, 6, 147-150. F. Xia, H. Wang and Y. Jia, Nature communications, 2014, 5, 4458. J. Yang, R. Xu, J. Pei, Y. W. Myint, F. Wang, Z. Wang, S. Zhang, Z. Yu and Y. Lu, Light: Science & Applications, 2015, 4, e312. R. Gusmao, Z. Sofer and M. Pumera, Angewandte Chemie, 2017. A. Castellanos-Gomez, L. Vicarelli, E. Prada, J. O. Island, K. Narasimha-Acharya, S. I. Blanter, D. J. Groenendijk, M. Buscema, G. A. Steele and J. Alvarez, 2D Materials, 2014, 1, 025001. J. O. Island, G. A. Steele, H. S. van der Zant and A. Castellanos-Gomez, 2D Materials, 2015, 2, 011002. J. D. Wood, S. A. Wells, D. Jariwala, K.-S. Chen, E. Cho, V. K. Sangwan, X. Liu, L. J. Lauhon, T. J. Marks and M. C. Hersam, Nano letters, 2014, 14, 6964-6970. J.-S. Kim, Y. Liu, W. Zhu, S. Kim, D. Wu, L. Tao, A. Dodabalapur, K. Lai and D. Akinwande, Scientific reports, 2015, 5. A. Favron, E. Gaufrès, F. Fossard, A.-L. Phaneuf-L'Heureux, N. Y. Tang, P. L. Lévesque, A. Loiseau, R. Leonelli, S. Francoeur and R. Martel, Nature materials, 2015, 14, 826-832. A. Molle, J. Goldberger, M. Houssa, Y. Xu, S.-C. Zhang and D. Akinwande, Nature materials, 2017. S. Balendhran, S. Walia, H. Nili, S. Sriram and M. Bhaskaran, small, 2015, 11, 640-652. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E L.-D. Zhao, S.-H. Lo, Y. Zhang, H. Sun, G. Tan, C. Uher, C. Wolverton, V. P. Dravid and M. G. Kanatzidis, Nature, 2014, 508, 373-377. C. W. Li, J. Hong, A. F. May, D. Bansal, S. Chi, T. Hong, G. Ehlers and O. Delaire, Nature Physics, 2015, 11, 1063-1069. S. Srivastava and B. Avasthi, Journal of materials science, 1992, 27, 3693-3705. J. O. Island, A. J. Molina-Mendoza, M. Barawi, R. Biele, E. Flores, J. M. Clamagirand, J. R. Ares, C. Sánchez, H. S. van der Zant and R. D'Agosta, 2D Materials, 2017, 4, 022003. 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, 2017, 546, 270-273. M. A. McGuire, Crystals, 2017, 7, 121. M. Osada and T. Sasaki, Journal of Materials Chemistry, 2009, 19, 2503-2511. R. Ma and T. Sasaki, Accounts of chemical research, 2014, 48, 136-143. B. Anasori, M. R. Lukatskaya and Y. Gogotsi, Nature Reviews Materials, 2017, 2, 16098. J. Kang, S. Tongay, J. Zhou, J. Li and J. Wu, Applied Physics Letters, 2013, 102, 012111. Y. Liu, P. Stradins and S.-H. Wei, Science advances, 2016, 2, e1600069. V. Tran, R. Soklaski, Y. Liang and L. Yang, Physical Review B, 2014, 89, 235319. Y. Cai, G. Zhang and Y.-W. Zhang, Scientific reports, 2014, 4, 6677. P. Blake, E. Hill, A. Castro Neto, K. Novoselov, D. Jiang, R. Yang, T. Booth and A. Geim, Applied Physics Letters, 2007, 91, 063124. Z. Ni, H. Wang, J. Kasim, H. Fan, T. Yu, Y. Wu, Y. Feng and Z. Shen, Nano letters, 2007, 7, 2758-2763. S. Roddaro, P. Pingue, V. Piazza, V. Pellegrini and F. Beltram, Nano letters, 2007, 7, 2707-2710. R. Frisenda, E. Navarro-Moratalla, P. Gant, D. P. De Lara, P. Jarillo-Herrero, R. V. Gorbachev and A. Castellanos-Gomez, Chemical Society Reviews, 2018, 47, 53-68. C. R. Dean, A. F. Young, I. Meric, C. Lee, L. Wang, S. Sorgenfrei, K. Watanabe, T. Taniguchi, P. Kim and K. L. Shepard, Nature nanotechnology, 2010, 5, 722-726. P. Zomer, S. Dash, N. Tombros and B. Van Wees, Applied Physics Letters, 2011, 99, 232104. A. Castellanos-Gomez, M. Buscema, R. Molenaar, V. Singh, L. Janssen, H. S. van der Zant and G. A. Steele, 2D Materials, 2014, 1, 011002. L. Wang, I. Meric, P. Huang, Q. Gao, Y. Gao, H. Tran, T. Taniguchi, K. Watanabe, L. Campos and D. Muller, Science, 2013, 342, 614-617. C. Reuter, R. Frisenda, D. Y. Lin, T. S. Ko, D. Perez de Lara and A. Castellanos‐Gomez, Small Methods, 2017, 1, 1700119. Y. Gong, J. Lin, X. Wang, G. Shi, S. Lei, Z. Lin, X. Zou, G. Ye, R. Vajtai and B. I. Yakobson, Nature materials, 2014, 13, 1135- 1142. J. Yu, J. Li, W. Zhang and H. Chang, Chemical Science, 2015, 6, 6705-6716. S. L. Wong, H. Liu and D. Chi, Progress in Crystal Growth and Characterization of Materials, 2016, 62, 9-28. R. Dong and I. Kuljanishvili, Journal of Vacuum Science & Technology B, Nanotechnology and Microelectronics: Materials, Processing, Measurement, and Phenomena, 2017, 35, 030803. Y. Cui, B. Li, J. Li and Z. Wei, Science China Physics, Mechanics & Astronomy, 2018, 61, 016801. X. Duan, C. Wang, J. C. Shaw, R. Cheng, Y. Chen, H. Li, X. Wu, Y. Tang, Q. Zhang and A. Pan, Nature nanotechnology, 2014, 9, 1024-1030. C. Huang, S. Wu, A. M. Sanchez, J. J. Peters, R. Beanland, J. S. Ross, P. Rivera, W. Yao, D. H. Cobden and X. Xu, Nature materials, 2014, 13, 1096-1101. Z. Zheng, J. Yao and G. Yang, ACS Applied Materials & Interfaces, 2017, 9, 7288-7296. Z.-Q. Xu, Y. Zhang, Z. Wang, Y. Shen, W. Huang, X. Xia, W. Yu, Y. Xue, L. Sun and C. Zheng, 2D Materials, 2016, 3, 041001. M. Sun, D. Xie, Y. Sun, W. Li, C. Teng and J. Xu, Scientific Reports, 2017, 7, 4505. R. Bratschitsch, Nature nanotechnology, 2014, 9, 247-248. B. W. Baugher, H. O. Churchill, Y. Yang and P. Jarillo-Herrero, Nature nanotechnology, 2014, 9, 262-267. A. Pospischil, M. M. Furchi and T. Mueller, Nature nanotechnology, 2014, 9, 257-261. J. S. Ross, P. Klement, A. M. Jones, N. J. Ghimire, J. Yan, D. Mandrus, T. Taniguchi, K. Watanabe, K. Kitamura and W. Yao, Nature nanotechnology, 2014, 9, 268-272. D. J. Groenendijk, M. Buscema, G. A. Steele, S. Michaelis de Vasconcellos, R. Bratschitsch, H. S. van der Zant and A. Castellanos- Gomez, Nano letters, 2014, 14, 5846-5852. M. Buscema, D. J. Groenendijk, G. A. Steele, H. S. J. van der Zant and A. Castellanos-Gomez, Nature Communications, 2014, 5, 4651. N. M. Gabor, J. C. Song, Q. Ma, N. L. Nair, T. Taychatanapat, K. Watanabe, T. Taniguchi, L. S. Levitov and P. Jarillo-Herrero, Science, 2011, 334, 648-652. H.-Y. Chiu, V. Perebeinos, Y.-M. Lin and P. Avouris, Nano letters, 2010, 10, 4634-4639. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53. 54. 55. 56. 57. 58. 59. 60. 61. 62. 63. 64. 65. 66. 67. 68. 69. 70. 71. 72. 73. 74. 75. 76. 77. 78. 79. 80. 81. 82. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E D. Li, M. Chen, Z. Sun, P. Yu, Z. Liu, P. M. Ajayan and Z. Zhang, Nature Nanotechnology, 2017. J. Ye, S. Inoue, K. Kobayashi, Y. Kasahara, H. Yuan, H. Shimotani and Y. Iwasa, Nature materials, 2010, 9. Y. Zhang, J. Ye, Y. Matsuhashi and Y. Iwasa, Nano letters, 2012, 12, 1136-1140. Y. Zhang, J. Ye, Y. Yomogida, T. Takenobu and Y. Iwasa, Nano letters, 2013, 13, 3023-3028. A. A. El Yumin, J. Yang, Q. Chen, O. Zheliuk and J. Ye, physica status solidi (b), 2017, 254, 1700180. M. S. Choi, D. Qu, D. Lee, X. Liu, K. Watanabe, T. Taniguchi and W. J. Yoo, ACS nano, 2014, 8, 9332-9340. S. Mouri, Y. Miyauchi and K. Matsuda, Nano letters, 2013, 13, 5944-5948. J. D. Lin, C. Han, F. Wang, R. Wang, D. Xiang, S. Qin, X. A. Zhang, L. Wang, H. Zhang, A. T. Wee and W. Chen, ACS Nano, 2014, 8, 5323-5329. C. R. Ryder, J. D. Wood, S. A. Wells, Y. Yang, D. Jariwala, T. J. Marks, G. C. Schatz and M. C. Hersam, Nature chemistry, 2016, 8, 597-602. G. Abellán, V. Lloret, U. Mundloch, M. Marcia, C. Neiss, A. Görling, M. Varela, F. Hauke and A. Hirsch, Angewandte Chemie International Edition, 2016, 55, 14557-14562. A. J. Molina-Mendoza, L. Vaquero-Garzon, S. Leret, L. de Juan-Fernández, E. M. Pérez and A. Castellanos-Gomez, Chemical Communications, 2016, 52, 14365-14368. X. He, W. Chow, F. Liu, B. Tay and Z. Liu, Small, 2017, 13, 1602558. E. C. Peters, E. J. Lee, M. Burghard and K. Kern, Applied Physics Letters, 2010, 97, 193102. D. B. Farmer, Y.-M. Lin, A. Afzali-Ardakani and P. Avouris, Applied Physics Letters, 2009, 94, 213106. X. Yu, S. Zhang, H. Zeng and Q. J. Wang, Nano Energy, 2016, 25, 34-41. G. Wang, L. Bao, T. Pei, R. Ma, Y.-Y. Zhang, L. Sun, G. Zhang, H. Yang, J. Li and C. Gu, Nano letters, 2016, 16, 6870-6878. H.-M. Li, D. Lee, D. Qu, X. Liu, J. Ryu, A. Seabaugh and W. J. Yoo, Nature communications, 2015, 6. J. Legma, G. Vacquier, H. Traore and A. Casalot, Materials Science and Engineering: B, 1991, 8, 167-174. J. Suh, T. E. Park, D. Y. Lin, D. Fu, J. Park, H. J. Jung, Y. Chen, C. Ko, C. Jang, Y. Sun, R. Sinclair, J. Chang, S. Tongay and J. Wu, Nano Lett, 2014, 14, 6976-6982. Y. Jin, D. H. Keum, S. J. An, J. Kim, H. S. Lee and Y. H. Lee, Adv Mater, 2015, 27, 5534-5540. S. A. Svatek, E. Antolín, D.-Y. Lin, R. Frisenda, C. Reuter, A. J. Molina-Mendoza, M. Muñoz, N. Agrait, T.-S. Ko, D. P. de Lara and A. Castellanos-Gomez, Journal of Materials Chemistry C, 2017, 5, 854-861. A. Nipane, D. Karmakar, N. Kaushik, S. Karande and S. Lodha, ACS nano, 2016, 10, 2128-2137. E. Kim, C. Ko, K. Kim, Y. Chen, J. Suh, S. G. Ryu, K. Wu, X. Meng, A. Suslu and S. Tongay, Advanced Materials, 2016, 28, 341-346. A. K. Geim and I. V. Grigorieva, Nature, 2013, 499, 419-425. M.-Y. Li, C.-H. Chen, Y. Shi and L.-J. Li, Materials Today, 2016, 19, 322-335. A. Pant, Z. Mutlu, D. Wickramaratne, H. Cai, R. K. Lake, C. Ozkan and S. Tongay, Nanoscale, 2016, 8, 3870-3887. H. Fang, C. Battaglia, C. Carraro, S. Nemsak, B. Ozdol, J. S. Kang, H. A. Bechtel, S. B. Desai, F. Kronast and A. A. Unal, Proceedings of the National Academy of Sciences, 2014, 111, 6198-6202. X. Hong, J. Kim, S.-F. Shi, Y. Zhang, C. Jin, Y. Sun, S. Tongay, J. Wu, Y. Zhang and F. Wang, Nature nanotechnology, 2014, 9, 682-686. M. M. Furchi, A. Pospischil, F. Libisch, J. Burgdorfer and T. Mueller, Nano letters, 2014, 14, 4785-4791. C. H. Lee, G. H. Lee, A. M. van der Zande, W. Chen, Y. Li, M. Han, X. Cui, G. Arefe, C. Nuckolls, T. F. Heinz, J. Guo, J. Hone and P. Kim, Nat Nanotechnol, 2014, 9, 676-681. Y. Deng, Z. Luo, N. J. Conrad, H. Liu, Y. Gong, S. Najmaei, P. M. Ajayan, J. Lou, X. Xu and P. D. Ye, ACS nano, 2014, 8, 8292- 8299. M.-Y. Li, Y. Shi, C.-C. Cheng, L.-S. Lu, Y.-C. Lin, H.-L. Tang, M.-L. Tsai, C.-W. Chu, K.-H. Wei and J.-H. He, Science, 2015, 349, 524-528. D. Jariwala, V. K. Sangwan, C.-C. Wu, P. L. Prabhumirashi, M. L. Geier, T. J. Marks, L. J. Lauhon and M. C. Hersam, Proceedings of the National Academy of Sciences, 2013, 110, 18076-18080. S. Vélez, J. Island, M. Buscema, O. Txoperena, S. Parui, G. A. Steele, F. Casanova, H. S. van der Zant, A. Castellanos-Gomez and L. E. Hueso, Nanoscale, 2015, 7, 15442-15449. O. Lopez-Sanchez, E. Alarcon Llado, V. Koman, A. Fontcuberta i Morral, A. Radenovic and A. Kis, Acs Nano, 2014, 8, 3042- 3048. H. Huang, Y. Huang, S. Wang, M. Zhu, H. Xie, L. Zhang, X. Zheng, Q. Xie, D. Niu and Y. Gao, Crystals, 2016, 6, 113. S. H. H. Shokouh, A. Pezeshki, A. Raza, S. Raza, H. S. Lee, S. W. Min, P. J. Jeon, J. M. Shin and S. Im, Advanced Materials, 2015, 27, 150-156. P. J. Jeon, Y. T. Lee, J. Y. Lim, J. S. Kim, D. K. Hwang and S. Im, Nano letters, 2016, 16, 1293-1298. S. H. Hosseini Shokouh, A. Pezeshki, S. R. A. Raza, K. Choi, S.-W. Min, P. J. Jeon, H. S. Lee and S. Im, ACS nano, 2014, 8, 5174-5181. 83. 84. 85. 86. 87. 88. 89. 90. 91. 92. 93. 94. 95. 96. 97. 98. 99. 100. 101. 102. 103. 104. 105. 106. 107. 108. 109. 110. 111. 112. 113. 114. 115. 116. 117. 118. 119. 120. 121. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E M.-L. Tsai, S.-H. Su, J.-K. Chang, D.-S. Tsai, C.-H. Chen, C.-I. Wu, L.-J. Li, L.-J. Chen and J.-H. He, ACS nano, 2014, 8, 8317- 8322. L. Hao, Y. Liu, W. Gao, Z. Han, Q. Xue, H. Zeng, Z. Wu, J. Zhu and W. Zhang, Journal of applied physics, 2015, 117, 114502. V. Dhyani and S. Das, Scientific Reports, 2017, 7. L. Wang, J. Jie, Z. Shao, Q. Zhang, X. Zhang, Y. Wang, Z. Sun and S. T. Lee, Advanced Functional Materials, 2015, 25, 2910- 2919. P. Gehring, R. Urcuyo, D. L. Duong, M. Burghard and K. Kern, Applied Physics Letters, 2015, 106, 233110. P. Langevin, Ann. Chim. Phys, 1903, 28, 122. W. Shockley and W. Read Jr, Physical review, 1952, 87, 835. R. N. Hall, Physical Review, 1952, 87, 387-387. N. Greenham and P. Bobbert, Physical Review B, 2003, 68, 245301. M. M. Furchi, D. K. Polyushkin, A. Pospischil and T. Mueller, Nano Lett, 2014, 14, 6165-6170. J. O. Island, S. I. Blanter, M. Buscema, H. S. J. van der Zant and A. Castellanos-Gomez, Nano Letters, 2015, 15, 7853-7858. J. Wong, D. Jariwala, G. Tagliabue, K. Tat, A. R. Davoyan, M. C. Sherrott and H. A. Atwater, ACS nano, 2017, 11, 7230-7240. L. Ye, H. Li, Z. Chen and J. Xu, ACS Photonics, 2016, 3, 692-699. A. J. Molina-Mendoza, E. Giovanelli, W. S. Paz, M. A. Niño, J. O. Island, C. Evangeli, L. Aballe, M. Foerster, H. S. Van Der Zant and G. Rubio-Bollinger, Nature Communications, 2017, 8. A. Pezeshki, S. H. H. Shokouh, T. Nazari, K. Oh and S. Im, Advanced Materials, 2016, 28, 3216-3222. K. Zhang, T. Zhang, G. Cheng, T. Li, S. Wang, W. Wei, X. Zhou, W. Yu, Y. Sun and P. Wang, ACS nano, 2016, 10, 3852-3858. M. L. Tsai, M. Y. Li, J. R. D. Retamal, K. T. Lam, Y. C. Lin, K. Suenaga, L. J. Chen, G. Liang, L. J. Li and H. He Jr, Advanced Materials, 2017, 29, 1701168. Y. Ahn, A. Tsen, B. Kim, Y. W. Park and J. Park, Nano letters, 2007, 7, 3320-3323. F. Xia, T. Mueller, R. Golizadeh-Mojarad, M. Freitag, Y.-m. Lin, J. Tsang, V. Perebeinos and P. Avouris, Nano Letters, 2009, 9, 1039-1044. H. Yuan, X. Liu, F. Afshinmanesh, W. Li, G. Xu, J. Sun, B. Lian, A. G. Curto, G. Ye and Y. Hikita, Nature nanotechnology, 2015, 10, 707-713. G. Rao, M. Freitag, H.-Y. Chiu, R. S. Sundaram and P. Avouris, ACS nano, 2011, 5, 5848-5854. M. Tosun, D. Fu, S. B. Desai, C. Ko, J. S. Kang, D.-H. Lien, M. Najmzadeh, S. Tongay, J. Wu and A. Javey, Scientific reports, 2015, 5, 10990. S. Kallatt, G. Umesh, N. Bhat and K. Majumdar, Nanoscale, 2016, 8, 15213-15222. M. C. Lemme, F. H. Koppens, A. L. Falk, M. S. Rudner, H. Park, L. S. Levitov and C. M. Marcus, Nano letters, 2011, 11, 4134- 4137. S. L. Howell, D. Jariwala, C.-C. Wu, K.-S. Chen, V. K. Sangwan, J. Kang, T. J. Marks, M. C. Hersam and L. J. Lauhon, Nano Letters, 2015, 15, 2278-2284. R. Sundaram, M. Engel, A. Lombardo, R. Krupke, A. Ferrari, P. Avouris and M. Steiner, Nano letters, 2013, 13, 1416-1421. L. Dobusch, S. Schuler, V. Perebeinos and T. Mueller, Advanced Materials, 2017, 29, 1701304. Y. Zhang, T. Oka, R. Suzuki, J. Ye and Y. Iwasa, Science, 2014, 344, 725-728. S. Jo, N. Ubrig, H. Berger, A. B. Kuzmenko and A. F. Morpurgo, Nano letters, 2014, 14, 2019-2025. E. Ponomarev, I. Gutiérrez-Lezama, N. Ubrig and A. F. Morpurgo, Nano letters, 2015, 15, 8289-8294. I. Gutiérrez-Lezama, B. A. Reddy, N. Ubrig and A. F. Morpurgo, 2D Materials, 2016, 3, 045016. D. Li, R. Cheng, H. Zhou, C. Wang, A. Yin, Y. Chen, N. O. Weiss, Y. Huang and X. Duan, Nature communications, 2015, 6. R. Cheng, D. Li, H. Zhou, C. Wang, A. Yin, S. Jiang, Y. Liu, Y. Chen, Y. Huang and X. Duan, Nano letters, 2014, 14, 5590-5597. Y. Yang, N. Huo and J. Li, Journal of Materials Chemistry C, 2017, 5, 7051-7056. L. Wang, L. Huang, W. C. Tan, X. Feng, L. Chen and K.-W. Ang, Advanced Electronic Materials, 2018, 4, 1700442. Y. Jin, D. H. Keum, S. J. An, J. Kim, H. S. Lee and Y. H. Lee, Advanced Materials, 2015, 27, 5534-5540. Y. Liu, Y. Cai, G. Zhang, Y. W. Zhang and K. W. Ang, Advanced Functional Materials, 2017, 27, 1604368. J. Ahn, P. J. Jeon, S. R. A. Raza, A. Pezeshki, S.-W. Min, D. K. Hwang and S. Im, 2D Materials, 2016, 3, 045011. Y. Gong, S. Lei, G. Ye, B. Li, Y. He, K. Keyshar, X. Zhang, Q. Wang, J. Lou and Z. Liu, Nano letters, 2015, 15, 6135-6141. N. Huo, J. Kang, Z. Wei, S. S. Li, J. Li and S. H. Wei, Advanced Functional Materials, 2014, 24, 7025-7031. F. Wang, Z. Wang, K. Xu, F. Wang, Q. Wang, Y. Huang, L. Yin and J. He, Nano letters, 2015, 15, 7558-7566. S. Chuang, R. Kapadia, H. Fang, T. Chia Chang, W.-C. Yen, Y.-L. Chueh and A. Javey, Applied Physics Letters, 2013, 102, 242101. T. Hong, B. Chamlagain, T. Wang, H.-J. Chuang, Z. Zhou and Y.-Q. Xu, Nanoscale, 2015, 7, 18537-18541. N. Flöry, A. Jain, P. Bharadwaj, M. Parzefall, T. Taniguchi, K. Watanabe and L. Novotny, Applied Physics Letters, 2015, 107, 123106. P. Chen, T. T. Zhang, J. Xiang, H. Yu, S. Wu, X. Lu, G. Wang, F. Wen, Z. Liu and R. Yang, Nanoscale, 2016, 8, 3254-3258. 122. 123. 124. 125. 126. 127. 128. 129. 130. 131. 132. 133. 134. 135. 136. 137. 138. 139. 140. 141. 142. 143. 144. 145. 146. 147. 148. 149. 150. 151. 152. 153. 154. 155. 156. 157. 158. 159. 160. 161. 162. 163. 164. 165. 166. This is the post-peer reviewed version of the following article: R. Frisenda et al., "Atomically thin p-n junctions based on two-dimensional materials" Chemical Society Reviews 47, 3339-3358 (2018) https://doi.org/10.1039/C7CS00880E X. Wang, L. Huang, Y. Peng, N. Huo, K. Wu, C. Xia, Z. Wei, S. Tongay and J. Li, Nano Research, 2016, 9, 507-516. A.-J. Cho, S. D. Namgung, H. Kim and J.-Y. Kwon, APL Materials, 2017, 5, 076101. F. Yan, L. Zhao, A. Patanè, P. a. Hu, X. Wei, D. Zhang, Q. Lv, Q. Feng, C. Shen and K. Chang, Nanotechnology, 2017. W. C. Yap, Z. Yang, M. Mehboudi, J.-A. Yan, S. Barraza-Lopez and W. Zhu, Nano Research, 2017, 11, 420-430. C. Li, X. Yan, X. Song, W. Bao, S. Ding, D. W. Zhang and P. Zhou, Nanotechnology, 2017, 28, 415201. S. Yang, M. Wu, B. Wang, L.-D. Zhao, L. Huang, C. Jiang and S.-H. Wei, ACS applied materials & interfaces, 2017, 9, 42149- 42155. A. S. Aji, M. Izumoto, K. Suenaga, K. Yamamoto, H. Nakashima and H. Ago, Physical Chemistry Chemical Physics, 2018, 20, 889-897. T. Yang, B. Zheng, Z. Wang, T. Xu, C. Pan, J. Zou, X. Zhang, Z. Qi, H. Liu and Y. Feng, Nature communications, 2017, 8, 1906. K. Zhang, M. Peng, W. Wu, J. Guo, G. Gao, Y. Liu, J. Kou, R. Wen, Y. Lei and A. Yu, Materials Horizons, 2017, 4, 274-280. F. Liu, W. L. Chow, X. He, P. Hu, S. Zheng, X. Wang, J. Zhou, Q. Fu, W. Fu and P. Yu, Advanced Functional Materials, 2015, 25, 5865-5871. D. Jariwala, S. L. Howell, K.-S. Chen, J. Kang, V. K. Sangwan, S. A. Filippone, R. Turrisi, T. J. Marks, L. J. Lauhon and M. C. Hersam, Nano letters, 2015, 16, 497-503. D. He, Y. Pan, H. Nan, S. Gu, Z. Yang, B. Wu, X. Luo, B. Xu, Y. Zhang and Y. Li, Applied Physics Letters, 2015, 107, 183103. D. Sarkar, X. Xie, W. Liu, W. Cao, J. Kang, Y. Gong, S. Kraemer, P. M. Ajayan and K. Banerjee, Nature, 2015, 526, 91-95. Y. Yu, P. W. Fong, S. Wang and C. Surya, Scientific reports, 2016, 6. C. Yim, M. O'brien, N. McEvoy, S. Riazimehr, H. Schäfer-Eberwein, A. Bablich, R. Pawar, G. Iannaccone, C. Downing and G. Fiori, Scientific reports, 2014, 4, 5458. Z. Wang, M. Li, L. Yang, Z. Zhang and X. P. Gao, Nano Research, 2017, 10, 1872-1879. Y. Niu, R. Frisenda, S. A. Svatek, G. Orfila, F. Gallego, P. Gant, N. Agraït, C. Leon, A. Rivera-Calzada and D. P. De Lara, 2D Materials, 2017, 4, 034002. K. Kang, S. Xie, L. Huang, Y. Han, P. Y. Huang, K. F. Mak, C.-J. Kim, D. Muller and J. Park, Nature, 2015, 520, 656. Y. Huang, J. Qiao, K. He, S. Bliznakov, E. Sutter, X. Chen, D. Luo, F. Meng, D. Su and J. Decker, Chemistry of Materials, 2016, 28, 8330-8339. F. Alsaffar, S. Alodan, A. Alrasheed, A. Alhussain, N. Alrubaiq, A. Abbas and M. R. Amer, Scientific Reports, 2017, 7, 44540. R. A. Doganov, E. C. O'Farrell, S. P. Koenig, Y. Yeo, A. Ziletti, A. Carvalho, D. K. Campbell, D. F. Coker, K. Watanabe and T. Taniguchi, Nature communications, 2015, 6. A. Avsar, I. J. Vera-Marun, J. Y. Tan, K. Watanabe, T. Taniguchi, A. H. Castro Neto and B. Ozyilmaz, ACS nano, 2015, 9, 4138- 4145. Y. Cao, A. Mishchenko, G. Yu, E. Khestanova, A. Rooney, E. Prestat, A. Kretinin, P. Blake, M. Shalom and C. Woods, Nano letters, 2015, 15, 4914-4921. M. Velický, P. S. Toth, A. M. Rakowski, A. P. Rooney, A. Kozikov, C. R. Woods, A. Mishchenko, L. Fumagalli, J. Yin and V. Zólyomi, Nature Communications, 2017, 8, 14410. G. Prando, Nature Nanotechnology, 2017, 12, 191-191. Y. Jiang, J. Mao, D. Moldovan, M. R. Masir, G. Li, K. Watanabe, T. Taniguchi, F. M. Peeters and E. Y. Andrei, Nature Nanotechnology, 2017, 12, 1045. R. Heinisch, F. Bronold and H. Fehske, Physical Review B, 2013, 87, 155409. M. Huang, S. Li, Z. Zhang, X. Xiong, X. Li and Y. Wu, Nature nanotechnology, 2017, 12, 1148-1154. D. Akinwande, N. Petrone and J. Hone, Nature communications, 2014, 5, 5678. G. Fiori, F. Bonaccorso, G. Iannaccone, T. Palacios, D. Neumaier, A. Seabaugh, S. K. Banerjee and L. Colombo, Nature Nanotechnology, 2014, 9, 768-779. F. Xia, H. Wang, D. Xiao, M. Dubey and A. Ramasubramaniam, Nature Photonics, 2014, 8, 899. 167. 168. 169. 170. 171. 172. 173. 174. 175. 176. 177. 178. 179. 180. 181. 182. 183. 184. 185. 186. 187. 188. 189. 190. 191. 192. 193. 194. 195. 196. 197.
1312.5751	1	1312	2013-12-19T21:04:23	Reply to Comment by Vanderbilt, Souza and Haldane: arXiv:1312.4200	[ "cond-mat.mes-hall" ]	We reassert our statement that the non-quantized Weyl node contribution to the anomalous Hall conductivity is not a Fermi surface property, contrary to the claim by Vanderbilt, Souza and Haldane.	cond-mat.mes-hall	cond-mat	Reply to Comment by Vanderbilt, Souza and Haldane: arXiv:1312.4200 1Department of Physics and Astronomy, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada Y. Chen,1 D.L. Bergman,2 and A.A. Burkov1 2Department of Physics, California Institute of Technology, 1200 E. California Blvd, MC114-36, Pasadena, California 91125, USA (Dated: July 5, 2019) We reassert our statement that the non-quantized Weyl node contribution to the anomalous Hall conductivity is not a Fermi surface property, contrary to the claim by Vanderbilt, Souza and Haldane. We have recently pointed out1 that the common state- ment in the literature that "intrinsic anomalous Hall conductivity is entirely a Fermi surface property mod- ulo a quantized contribution from filled bands" is incor- rect, since it ignores non-quantized Weyl node contri- bution from filled bands. Vanderbilt, Souza and Haldane (VSH)2 assert that Weyl node contribution is included in the Fermi surface contribution and our claim stems from a mathematical error in the identification of the Fermi surface contribution. Here we show that no mathemati- cal errors are involved in our claim, and our disagreement is based on physics, namely the fact that Weyl node con- tribution is associated with Fermi arc edge states and not with bulk states on the Fermi surface. This fact is ignored by VSH. To understand the nature of our disagreement with VSH, let us consider the following toy model Hamiltonian of a metallic ferromagnet with two Weyl nodes, i.e. the smallest possible number: Ht(k) = kxσx + kyσy + mt(kz)σz, (1) where σ are Pauli matrices, describing the two bands that touch at the Weyl nodes and will use ¯h = 1 units everywhere. The index t = 1, 2 labels two distinct Dirac fermions. The "mass term" m1(kz) is such that it changes sign from positive to negative at kz = ±k0, which are the Weyl node locations, while mass term m2(kz) is always positive. This is the simplest possible model of the electronic structure of a three-dimensional ferromag- net with Weyl nodes. The eigenvalues and corresponding eigenfunctions of Ht are given by ǫstk = sǫtk = sqk2 s1 + s ustki = 1 √2  x + k2 y + m2 t (kz), mt(kz) ǫtk , sk+s1 − s mt(kz) ǫtk   ,(2) where s = ±. The components of the Berry connection field and the Berry curvature are easily calculated and given by Ast x (k) = −ihustk∂kxustki = − Ast y (k) = −ihustk∂kyustki = ky 2ǫtk[ǫtk + smt(kz)] , kx 2ǫtk[ǫtk + smt(kz)] . (3) and Ωst z (k) = ∂kx Ast y (k) − ∂ky Ast x (k) = s mt(kz) 2ǫ3 tk ; (4) Let us take the Fermi energy to be somewhere in the up- per s = + bands, i.e. ǫF > 0. The anomalous Hall con- ductivity can now be calculated as integral of the Berry curvature over all occupied states. Let us separately con- sider the contributions of the fully occupied s = − bands and the partially occupied s = + bands. We have σ− xy = e2Xt Z d3k (2π)3 Ω−t z (k) = − and e2 8π2 Xt Z dkz sign[mt(kz)] = − e2k0 2π2 , (5) σ+ xy = e2Xt Z d3k (2π)3 Ω+t z (k)nF (ǫtk) e2 = 8π2 Xt Z dkz [sign[mt(kz)] − mt(kz)/ǫF ] × Θ(ǫF − mt(kz)). (6) where Θ(x) is the Heaviside step function. The σ− xy con- tribution, coming from completely filled bands, is pre- cisely what we call the Weyl node contribution. Note that σ+ xy is a continuous function of the Fermi energy xy only in the and cancels the Weyl node contribution σ− limit ǫF → ∞, i.e. when the upper bands are filled. Let us now focus on the contribution of the unfilled bands, σ+ xy. We first note that the circulation of the Berry connection field along a section of the Fermi surface (i.e. the total Berry phase accumulated along this section), corresponding to a specific value of kz, is given by I dk · A+t(k) = π [1 − mt(kz)/ǫF ] = π [sign[mt(kz)] − mt(kz)/ǫF ] + π[1 − sign[mt(kz)]]. (7) Note that the last term in Eq. (7) is either equal to 0 or 2π, which is a reflection of a 2π ambiguity of the defini- tion of Berry phase. The essence of VSH's argument is that one can use the above 2π ambiguity of the Berry phase to include the Weyl node contribution σ− contribution as xy into the Fermi surface σF S xy = σ+ xy − e2 8π2 Xt Z dkz sign[mt(kz)]. (8) xy and everything is By construction, this cancels out σ− now a Fermi surface contribution. While this is formally correct, we see absolutely no physical reason for doing xy = σ+ this and instead identify σF S xy,1 which is equivalent to ignoring the second term in Eq. (7). This is just as correct formally as what VSH propose, but in addition is also correct physically. The reason is that the con- tribution of filled bands σ− xy can not be associated with Fermi surface and instead comes from chiral Fermi arc edge states, as is now well-understood in the context of extensive work on Weyl semimetals.3,4 We thus conclude that the criticism of our work by VSH is groundless. 2 1 Y. Chen, D.L. Bergman, and A.A. Burkov, Phys. Rev. B 3 X. Wan, A.M. Turner, A. Vishwanath, and S.Y. Savrasov, 88, 125110 (2013). 2 D. Vanderbilt, arXiv:1312.4200. I. Souza, and F.D.M. Haldane, 4 A.A. Burkov and L. Balents, Phys. Rev. Lett. 107, 127205 Phys. Rev. B 83, 205101 (2011). (2011).
1209.4382	2	1209	2012-12-11T13:35:46	Spin-orbit coupling assisted by flexural phonons in graphene	[ "cond-mat.mes-hall" ]	We analyze the couplings between spins and phonons in graphene. We present a complete analysis of the possible couplings between spins and flexural, out of plane, vibrations. From tight-binding models we obtain analytical and numerical estimates of their strength. We show that dynamical effects, induced by quantum and thermal fluctuations, significantly enhance the spin-orbit gap.	cond-mat.mes-hall	cond-mat	Spin-orbit coupling assisted by flexural phonons in graphene 1Instituto de Ciencia de Materiales de Madrid. CSIC. Sor Juana In´es de la Cruz 3. 28049 Madrid. Spain. H. Ochoa,1 A. H. Castro Neto,2,3 V. I. Fal'ko,4,5 F. Guinea1 2Graphene Research Centre and Department of Physics, National University of Singapore, 6 Science Drive 2, 117546, Singapore. 3Department of Physics, Boston University, 590 Commonwealth Ave., Boston MA 02215, USA. 4Physics Department, Lancaster University, Lancaster, LA1 4YB, UK. 5DPMC, University of Geneva, 24 Quai Ernest-Ansermet, CH1211 Geneve 4, Switzerland We analyze the couplings between spins and phonons in graphene. We present a complete analysis of the possible couplings between spins and flexural, out of plane, vibrations. From tight-binding models we obtain analytical and numerical estimates of their strength. We show that dynamical effects, induced by quantum and thermal fluctuations, significantly enhance the spin-orbit gap. I. INTRODUCTION Graphene combines in an unique way structural and electronic properties not found in other materials1,2. Its special electronic structure made graphene the initial model for a two dimensional topological insulator.3 -- 5 Nu- merical calculations suggested that the gap which is the hallmark of a topological insulator is in the micro Kelvin range, too small to be experimentally observed.6 -- 9 Re- cently, different approaches based on heavy adatoms de- position have been proposed in order to engineer a topo- logical state.10,11 Experimentally, these proposals are dif- ficult to be implemented, since the presence of adatoms implies a z → −z (mirror) symmetry breaking which in- duces a Rashba-like spin-orbit coupling (SOC).12 These calculations are also the main guide for the interpretation of spin transport experiments. The calculations assumed lattice made up of ions of infinite mass, and ignored the lattice degrees of freedom. In the present work, we analyze the coupling of spins to lattice vibrations in graphene, with emphasis on flex- ural modes, which are unique to a two dimensional mem- brane, and have lower frequencies than conventional in- plane phonons. Flexural phonons have a great impact on the SOC since out-of-plane distortions of the lattice hybridize π orbitals with higher orbitals of carbon, lead- ing to a first order contribution in the spin-orbit inter- action strength, contrary to in-plane distortions, whose contribution is at least quadratic, as it happens with the intrinsic SOC.6,8 The main results of our paper are: i) A complete characterization of all possible couplings to flexural modes, including analytical expressions for each of them, and ii) An analysis of the effect of dynamic flex- ural modes on the Kane-Mele spin-orbit coupling, which defines the range of parameters where the topological in- sulator features of graphene can be observed. This anal- ysis is relevant in order to study the effect of ripples in spin transport.13 Our analysis is also applicable to carbon nanotubes,14 an issue that has become more interesting since carbon nanotubes have been recently proposed as a possible platform for hosting Majorana fermions.15,16 We first present a symmetry analysis of the possible couplings between spins and lattice vibrations. This study can also be applied to static out of plane defor- mations. Next, in Section III, we make numerical es- timates of the couplings, based on simple tight binding models. Section IV discusses the change induced in the spin-orbit couplings due to the presence of lattice excita- tions, due to quantum and thermal fluctuation. Finally, Section V contains a summary of the more relevant re- sults. A number of mathematical details are included in the Appendices. II. SYMMETRY ANALYSIS AND COUPLINGS The low energy sector of the electronic spectrum lies on the two inequivalent corners of the Brillouin zone K±, known as valleys or Dirac points. In our convention 3a), 0(cid:1), where a is the carbon-carbon K± = ±(cid:0)4π/(3 √ distance, see Fig. 1. The electronic Hamiltonian reads: H = −ivF (cid:126)Σ · (cid:126)∂ + ∆I Σz ⊗ sz + ∆R (Σx ⊗ sy − Σy ⊗ sx) (1) where (cid:126)Σ = (Σx, Σy) and Σz are 4× 4 matrices associated to the sublattice degree of freedom. This Hamiltonian operate in a space of 8-component Bloch functions Ψ = (ψA,K+,↑, ψB,K+,↑, ψB,K−,↑,−ψA,K−,↑, ψA,K+,↓, ψB,K+,↓, ψB,K−,↓,−ψA,K−,↓)T . The first term corresponds to the massless Dirac Hamiltonian, the second term is the Kane-Mele mass which arises due to the SO interaction, and the last one is a Rashba-like coupling which is present in the case of a mirror symmetry breaking. Both ∆I and ∆R are weak, 1 − 15 µeV in the former case according to previous estimates.6,8,9 We analyze first the SO-coupling assisted electron in- teraction with flexural phonons from a symmetry group theory approach.17 Before we present the exhaustive analysis, we disclose the results in the following lines. The spin-phonon coupling Hamiltonian reads: Hs−ph = HA1 + HB2 + HG(cid:48) (2) Irrep z → −z symmetric A1 A2 B2 Σz ⊗ sz Λz ⊗ sz Σx ⊗ sz (cid:18) −Σy ⊗ sz (cid:18) Λx ⊗ sz Λy ⊗ sz (cid:19) (cid:19) E1 E2 E(cid:48) 1 G(cid:48) z → −z asymmetric Σx ⊗ sy − Σy ⊗ sx Σx ⊗ sx + Σy ⊗ sy Σz ⊗ sx (cid:18) −Σz ⊗ sy (cid:18) Σx ⊗ sy + Σy ⊗ sx (cid:19)  Λx ⊗ sx Σx ⊗ sx − Σy ⊗ sy Λx ⊗ sy −Λy ⊗ sy −Λy ⊗ sx , (cid:19) (cid:18) −Λz ⊗ sy  Λz ⊗ sx TABLE I: Classification of the possible SO coupling terms according to how they transform under the symmetry oper- ations of C(cid:48)(cid:48) 6v. Note that all these operators are even under time reversal operation t → −t. where: +g2[−Λz ⊗ sy HA1 = g1 (Σx ⊗ sy − Σy ⊗ sx) ∂i∂iuA1+ (cid:0)∂2 +g3[(Σx ⊗ sy + Σy ⊗ sx)(cid:0)∂2 (cid:1) + 2Λz ⊗ sx∂x∂yuA1]+ (cid:1) + xuA1 − ∂2 xuA1 − ∂2 yuA1 yuA1 +2 (Σx ⊗ sx − Σy ⊗ sy) ∂x∂yuA1 ] (3) HB2 = g4Σz ⊗ sz(uB2)2 HG(cid:48) = g5[(Λx ⊗ sx)u1 + (Λx ⊗ sy)u2− −(Λy ⊗ sy)u3 − (Λy ⊗ sx)u4] (4) (5) in order The main ingredients these couplings, electronic operators and symmetry-adapted phonon fields, are summarized in Tab. II respectively. Next, we construct these couplings step by step. to construct I and Tab. Electronic operators The point group of the graphene crystal is C6v, which contains 12 elements: the identity, five rotations and six reflections in planes perpendicular to the crystal plane. Instead of dealing with degenerate states at two inequiv- alent points one can enlarge the unit cell in order to contain six atoms, in such a way that K± are mapped onto the Γ point (see Fig. 1 b)). From the point of view of the lattice symmetries, this means that the two elementary translations (ta1, ta2) are factorized out of the translation group and added to the point group C6v, which becomes C(cid:48)(cid:48) If we do not consider the spin degree of freedom, the π 6v = C6v + ta1 × C6v + ta2 × C6v. Irrep Z phonon mode (z → −z asymmetric) 2 (cid:19) A1 B2 G(cid:48) TABLE II: Classification of flexural (Z) phonon modes accord- ing to how they transform under the symmetry operations of C(cid:48)(cid:48) 6v. The numbers indicate the out-of-plane displacement of the atom within the 6-atoms unit cell. electronic states at K± transforms according to the 4- dimensional G(cid:48) irreducible representation of C(cid:48)(cid:48) 6v. Then, in order to write down the low-energy electronic Hamil- tonian we have to consider the 16 Hermitian operators acting in a 4-dimensional space. We define two different sets of 4 × 4 hermitian matrices {Σi}, {Λi} associated to sublattice and valley degrees of freedom respectively, so the set {I, Σi, Λi, Σi · Λj} provides a representation of the algebra of generators of U (4). In the basis in- troduced before Σi and Λi matrices are odd under the time reversal operation t → −t, see Appendix A. If we consider the spin degree of freedom then the π electronic states at K± transform according to the 8-dimensional representation G(cid:48) × D1/2, where D1/2 is the spinor rep- resentation associated to the spinorial part of the wave function. Thus, we introduce a set of Pauli matrices {si} associated to the spin degree of freedom, also odd under time reversal operation. Importantly, we are now intro- 11111116(cid:180)111(cid:45)1(cid:45)1(cid:45)116(cid:180)01(cid:45)101(cid:45)112(cid:180)2(cid:45)1(cid:45)1(cid:45)211112(cid:180)01(cid:45)10(cid:45)1112(cid:180)2(cid:45)1(cid:45)12(cid:45)1(cid:45)1112(cid:180)∗ Note that A/B, K−(cid:105) = (A/B, K+(cid:105)) . However, we must consider the real linear combinations of the vec- tors of Eq. (7) which transforms according to G(cid:48) in order to construct the couplings. These are: 3 1(cid:105) = 3(cid:105) = i 2 2(cid:105) = i 2 4(cid:105) = [−AK+(cid:105) + AK−(cid:105) − BK+(cid:105) + BK−(cid:105)] [AK+(cid:105) + AK−(cid:105) − BK+(cid:105) − BK−(cid:105)] 1 2 [−AK+(cid:105) + AK−(cid:105) + BK+(cid:105) − BK−(cid:105)] [AK+(cid:105) + AK−(cid:105) + BK+(cid:105) + BK−(cid:105)] 1 2 (8) The polarization vectors of Eq. (6), together with he vec- tors of Eq. (8), form a symmetry adapted basis, in such a way that the displacement vector of a flexural mode can be written as h(cid:105) = uA1 A1(cid:105) + uB2 B2(cid:105) + u1 1(cid:105) + u2 2(cid:105) + u3 3(cid:105) + u4 4(cid:105), where ui are the symmetry adapted (real) displacement fields. The results of this analysis are sum- marized in Tab. II. Spin-phonon couplings This analysis allows us to identify the SO assisted elec- tron coupling with flexural phonons at the center and the corners of the Brillouin zone. The spin-phonon in- teraction Hamiltonian can be expanded in powers of the phonon displacement fields and their derivatives, in such a way that the displacement fields (and the derivatives) are paired with the electronic operators corresponding to the same irreducible representation, and taking into ac- count that these combinations must be even under the operation z → −z. The couplings of Eqs. (3)-(5) corre- spond to the leading terms in such expansion. Since a uniform translation of the crystal cannot af- it is clear that it can cou- fect the electron motion, ple to acoustic phonons (at Γ, A1 phonons) only through spatial derivatives of the corresponding dis- placement field. Moreover, if we consider the graphene as a continuum surface, only out-of-plane distortions which generate extrinsic curvature can couple to elec- tron spin through the SO interaction. This implies that the leading term must depend on second deriva- tives ∂i∂juA1. Since ∂i∂iuA1 transforms according to A1, (cid:1) forms a doublet which and (cid:0)∂2 xuA1 − ∂2 yuA1 , 2∂x∂yuA1 transforms according to E2, in principle three different couplings are allowed by the symmetries. We obtain Eq. 3. In the case of optical phonons at the center of the Bril- louin zone, note that there is no term which transforms according to the B2 irreducible representation in the col- umn of z → −z asymmetric operators, which means that the leading term must be quadratic on the phonon dis- placement fields. Since B2 × B2 = A1 we obtain a Kane- Mele-like coupling term, Eq. 4. Finally, in the case of flexural phonons at the corners of the Brillouin zone, where both acoustic and optical FIG. 1: a) Real space lattice and Brillouin zone of flat graphene with 2 atoms per unit cell. b) The same with 6 atoms per unit cell. Note that K± are now equivalent to Γ. ducing a pseudovector in the 3-dimensional space, mean- ing that the operators which contain sz are even under a reflection in the graphene plane (mirror symmetry oper- ation, z → −z), whereas the operators which contain the in-plane components are odd. The result of this sym- metry based approach is summarized in Tab. I, where all the possible time reversal symmetric terms which in- volve spin operators are classified according to how they transforms under the symmetry operations of C(cid:48)(cid:48) 6v. These matrices correspond to all the allowed SO coupling terms within the low-energy description. Phonon modes In the lattice with 6 atoms per unit cell, the am- plitude of a flexural phonon mode at Γ (≡ K±) is given by a 6-component vector whose entries are asso- ciated to the displacements of each sublattice atoms: h(cid:105) = (hA1, hB1, hA2, hB2, hA3, hB3). This vector belongs to a 6-dimensional representation of C(cid:48)(cid:48) 6v which can be reduced as A1 + B2 + G(cid:48). The polarization vectors as- sociated to the 1-dimensional irreducible representations correspond to the acoustic (ZA) and optical (ZO) modes at the original Γ point (with 2 atoms per unit cell), whose polarization vectors in the 6 atoms basis read: A1(cid:105) = 1√ 6 (1, 1, 1, 1, 1, 1) B2(cid:105) = 1√ 6 (1,−1, 1,−1, 1,−1) (6) The 4-dimensional irreducible representation corresponds to the 4 degenerate modes at K± points, whose polariza- tion vectors read: A, K+(cid:105) = B, K+(cid:105) = A, K−(cid:105) = B, K−(cid:105) = 1√ 3 1√ 3 1√ 3 1√ 3 (1, 0, ei 2π 3 , 0, e−i 2π 3 , 0) (0, 1, 0, ei 2π 3 , 0, e−i 2π 3 ) (1, 0, e−i 2π 3 , 0, ei 2π 3 , 0) (0, 1, 0, e−i 2π 3 , 0, ei 2π 3 ) (7) a1a2A1A2K+K-a)b)c)K+K-a1a2A1A2K+K-a)b)c)K+K-Vssσ (cid:17) pi · (cid:126)δ tAB s,s tAB s,pi tAB pi,pj (cid:104) + (cid:16) (cid:16) (cid:17)(cid:16) (pi · pj) −(cid:16) pi · (cid:126)δ pj · (cid:126)δ pi · (cid:126)δ Vspσ (cid:17) (cid:17)(cid:16) (cid:17)(cid:105) Vppσ+ pj · (cid:126)δ Vppπ TABLE III: Two-center matrix elements in the Slater-Koster approximation as function of the tight-binding parameters (see the text). pi represents a unitary vector in the direc- tion of maximum amplitude of the orbital pi, and (cid:126)δ is the vector which connects neighboring sites A and B. branches are degenerate, the coupling reads as Eq. 5. III. TIGHT-BINDING MODEL  (cid:88) (cid:88) (cid:88) λ,λ(cid:48) The strength of these couplings can be estimated from a tight-binding model. We assume the convention of Fig 1. We choose the simplest tight-binding model with 4 orbitals {s, px, py, pz} per carbon atom and only nearest neighbors hoppings. We neglect the effect of d orbitals9 because all these couplings are first order in the SO inter- action constant, as we are going to see. The Hamiltonian can be written as: HT B = † tλc i,λci,λ + tij λ,λ(cid:48)c † i,λcj,λ(cid:48) + H.C. i,λ <i,j> (9) where the latin indices label the sites of the carbon atoms and λ = s, px, py, pz labels the orbitals considered in the calculation. For the on-site energies we take ts = s, tpi = p. The two-center matrix elements can be com- puted within the Slater-Koster approximation as it is in- III. We take (in eV):18,19 s = −7.3, dicated in Tab. p = 0, Vssσ = −3.63, Vspσ = 4.2, and Vppσ = 5.38, and Vppπ = −2.24. We describe the spin-obit interaction in terms of the Hamiltonian HSO = ∆(cid:126)L · (cid:126)s, where (cid:126)L and (cid:126)s correspond to the orbital angular momentum and spin operators respectively. We take ∆ = 20 meV.20 Our aim is to estimate the strength of the effective couplings within the low energy sector of the electronic spectrum. We can define the π (pz, low energy sector) and σ (s, px, py, high energy sector) orbital subspaces at least locally (see the discussion next and Fig. 2). Then, the electronic Hamiltonian can be written in the block form: (cid:18) Hπ Hπσ Hσπ Hσ (cid:19) H = (10) 4 FIG. 2: Flat graphene and bent graphene. The arrows repre- sent the px (black), py (red), and pz (blue) orbitals. The local basis of atomic p orbitals in the bent graphene is uniquely de- fined by the isomorphism between them. (11) We project out σ orbitals by a Schrieffer-Wolf transformation.21 We take the Green function G = −1, evaluate the block Gπ associated to the low- ( − H) energy sector, and use it in order to identify the low- π,σ = energy effective Hamiltonian. ( − Hπ,σ) −1, then we can write: If we define G(0) (cid:18) Gπ Gπσ Gσπ Gσ (cid:19)  (cid:17)−1 (cid:16)G(0) (cid:20)(cid:16)G(0) (cid:17)−1 Hσπ π = Hπσ (cid:16)G(0) (cid:17)−1 σ −1 (cid:21)−1 We obtain Gπ = π = Hπ + HπσG(0) G−1 0) the effective Hamiltonian reads: , so − σ Hσπ. In the low energy sector ( ≈ + HπσG(0) σ Hσπ π Hef f π ≈ Hπ − HπσH−1 σ Hσπ (12) spσ Both the SO interaction and the out-of-plane distor- tions enter in the π − σ mixing blocks. In the ab- sence of out-of-plane distortions the only coupling al- lowed by the symmetries has the structure of a Kane- Mele mass. Our estimation from the tight-binding model is ∆f lat have ∆f lat (cid:1) in agreement with Ref. 8. We I = s∆2/(cid:0)18V 2 I ≈ 9 µeV. In the case of B2 phonons we have to add to this anal- ysis the effect of a vertical displacement of one sublat- tice respect to the other. If the calculation is performed with the 6 atoms unit cell one can identify both the cou- pling with flexural phonons at K± (G(cid:48)). In the case of A1 phonons, since the coupling depends on the second derivatives of the phonon field, the calculation is not so straightforward. Phonons at Γ As we mentioned before, the coupling with acoustic phonons can be inferred from the effect of extrinsic cur- vature of the graphene sample. The approach that we present here is quite similar to the calculation of the SO coupling in carbon nanotubes. The crucial fact is how to choose the basis of π and σ orbitals. In the carbon nanotubes calculation the σ orbitals are chosen in such a way that they follow the shape of the nanotube, and the π orbital is chosen in the radial direction. Here we do essentially the same, but we try to formalize it a little bit. In the low energy limit the graphene sample can be described within a continuum theory. We assume that the position of the carbon atoms lie on a smooth surface, which is valid in the long-wavelength limit. Moreover, we assume that this surface is isometric to a plane. This means that a diffeomorphism f from a flat graphene sur- face (Sf lat) to a curved graphene surface (Scurv) exists in such a way that the metric on the curved surface is pull-backed to the flat one. Thus, the isomorphism f de- fines an unique way to introduce a local basis for the px and py atomic orbitals. Consider graphene in a flat configuration. We intro- duce unitary vectors px, py in the direction of maximum amplitude of the orbitals px, py respectively, see Fig. 2. From a geometrical point of view, these vectors are el- ements of the tangent bundle associated to Sf lat.22 At the same time, the axis of maximum localization of the pz orbital verifies pz = px × py. Now consider a curved graphene which is related to the flat graphene by a iso- morphism f . The push-forward of f maps the tangent bundle of Sf lat to the tangent bundle of Scurv, which means that the vectors px, py at any position of the bent graphene are uniquely determined by the action of the push-forward of the isomorphism on the original px, py defined in the flat configuration. And of course, pz in the bent graphene surface is given by the vectorial product of the new px, py. More physically, what we are doing is the following. Consider a sheet of paper. Consider a point on it and write down a small arrow. Now deform the sheet of paper smoothly (folding is not differentiable) without breaking it. The new surface of your paper is then isometric to the original plane. Consider the previ- ous point in the new curved sheet of paper. The vector tangent to the arrow at that point, which is uniquely de- fined, is actually the push-forwarded original vector on the flat sheet of paper. This way of introducing the local basis in the bent graphene has two advantages: 1) we re- cover "smoothly" the original basis when we restore the curved graphene to the original flat configuration; 2) we keep the notion of parallelism by imposing f to be an isomorphism instead of just a diffeomorphism. Note that this apparent restriction does not affect the estimation of the SO assisted electron coupling with flexural phonons, since we are introducing extrinsic curvature and setting the Gaussian curvature to zero. We know how to choose the orbital basis and then cal- culate the two-center matrix elements, at least locally, using the Koster-Slater parametrization. For simplicity, we consider a curved graphene surface with a constant curvature along a given direction (a cylinder), so essen- tially the same problem as a carbon nanotube. Thus, we 5 FIG. 3: a) Definition of the angle φ. b) Sketch for the calcu- lation of the new hoppings between pz and pi orbitals. have two parameters, the radius of curvature R and the angle φ between the direction of curvature and the x-axis (essentially the chiral angle in a nanotube), see Fig. 3 a). The new hoppings between π and σ orbitals can be cal- culated following the prescription of Tab. III. These are function of the angle θ defined in Fig. 3 b). Assuming that R (cid:29) a, we have to the leading term in a/R: θ ≈ δx cos(φ) + δy sin(φ) R (13) where δx,y are the components of the vector (cid:126)δ which con- nects nearest neighbors. After a straightforward calcula- tion, the block Hamiltonian that mixes π and σ states at K± can be written as the matrix:  Hσπ = 3a 8R 0 −Vspσe−iτ 2φ iτ(cid:0)V1 + eiτ 2φV2 0 0 −V1 + eiτ 2φV2 (cid:1) −Vspσeiτ 2φ iτ(cid:0)V1 + e−iτ 2φV2 0 (cid:1) 0 V1 − e−iτ 2φV2 0  where V1 = Vppσ + Vppπ, V2 = (Vppσ + 3Vppπ) /2, and τ = ±1 labels the valley K±. (14) This expression is exact to the leading order in a/R as- suming a constant R and φ along the graphene surface. Now we perform a local approximation, which is valid at long wavelengths: we assume that R and φ depends slightly on the position. Hence, they can be related with the second derivatives of the height profile, since the sec- ond fundamental form (F) of a surface in the Monge's parametrization (r = (x, y, h(x, y))) reads: (cid:18) ∂x∂xh ∂x∂yh ∂x∂yh ∂y∂yh (cid:19) F = 1√ 1 + ∂ih∂ih (15) If we neglect quadratic terms on h, then F is just the tensor of second derivatives. At the same time, note that within the continuum description of graphene as a mem- brane, the height profile h should be identified with the φa)b)pzδθpiflexural acoustic phonon field at long wavelengths uA1 . Then, the local R and φ can be related with the second derivatives of uA1 as: ∂x∂xuA1 ≈ −R−1 cos2 φ ∂y∂yuA1 ≈ −R−1 sin2 φ ∂x∂yuA1 ≈ −R−1 sin φ cos φ (16) By projecting out the σ electronic states as Eq. 12 indicates, we arrive at the electron-phonon coupling of Eq. 3 with: g1 = as∆ (Vppσ + Vppπ) 12V 2 spσ ≈ 3 meV · A g2 = aVppπ∆ 2 (Vppσ − Vppπ) ≈ 4 meV · A (17) Importantly, within the present Koster-Slater approxi- mation we have to extend the tight-binding calculation to second-nearest neighbors in order to obtain a non-zero g3 coupling. In that case we obtain: (cid:16) (cid:17) g3 = a∆ (3Vppπ + Vppσ) ppσ + V (2) V (2) ppπ 8 (Vppσ − Vppπ)2 (18) ppσ and V (2) where V (2) hopping parameters. ppπ are new second-nearest neighbors In order to estimate the coupling with optical phonons we have to consider the effect of a vertical displacement of one sublattice with respect to the other, similarly to the case of silicene.23 Processes which involve only one phonon give a vanishing contribution as expected from symmetry considerations. We can repeat the same scheme as before by considering virtual processes medi- σπ , note that Hσ ated by two phonons (∝ Hf lex also contains the SO interaction). We identify the Kane- Mele like coupling with flexural optical phonons at Γ, whose strength reads: σ Hf lex πσ H−1 g4 = s∆ (Vppπ − Vppσ)2 22 9a2V 4 spσ ≈ 20 meV · A−2 (19) 6 hopping is taken to zero. These parameters can be esti- mated from the Slater-Koster parameters as: s − p Von = 2Vspσ − 2Vppσ 3 √ Vssσ − 2 Vhop = (20) 3 This model was employed in Ref. 6 in order to estimate the SOC in graphene and carbon nanotubes. Within this model for flat graphene, the Kane-Mele coupling reads ∆f lat hop), in agreement with Ref. 6. Note also that this estimation is numerically very close to the one of Ref. 8. I = 3Von∆2/(4V 2 We perform the calculation in the unit cell with 6 atoms. We can estimate the terms that mix π and σ states by considering the vertical displace- ment of one lattice respect to the other, as we mentioned before. By projecting out the σ orbitals we arrive to a 6x6 effective Hamiltonian for the π electronic states at the new Γ, which can be seen as a matrix expressed in the monoelectronic basis (ΓA1π(cid:105),ΓA2π(cid:105),ΓA3π(cid:105),ΓB1π(cid:105),ΓB2π(cid:105),ΓB3π(cid:105)). In order to identify the effective Hamiltonian in the this matrix in the monoelectronic basis associated to the lattice with 2 atoms per unit cell, say let's (ΓAπ(cid:105),ΓBπ(cid:105),K+Aπ(cid:105),K+Bπ(cid:105),K−Aπ(cid:105),K−Bπ(cid:105)). Both basis are related by the unitary transformation: low energy sector we have to express  (21)  U = 1√ 3 3 1 0 1 1 0 ei 2π 1 0 e−i 2π 0 0 1 0 0 1 0 1 0 3 0 0 0 1 ei 2π e−i 2π 3 3 3 1 e−i 2π ei 2π 3 0 0 0 0 0 0 1 e−i 2π ei 2π 3 3 By doing so, we identify the strength of the coupling with phonons at the corner of the Brillouin zone: 3Von∆(cid:0)√ √ g5 = (cid:1) 2Vppσ − √ √ 2 2aV 2 hop 2Vppπ + Vspσ ≈ 4 meV · A−1 (22) Phonons at K± IV. ENHANCEMENT OF KANE-MELE MASS We estimate now the coupling with phonons at the cor- ners of the Brillouin zone. We simplify the tight-binding model in order to treat the problem analytically. We are going to consider the model described in Ref. 24 and adapted in Ref. 25 in order to describe the acoustic phonon modes in graphite. We consider two parameters: Von, which is the on-site energy of the σ orbitals, and Vhop, which is the hopping between σ orbitals at nearest neighbors when the orbitals are maximally localized in the direction which links the two atoms, otherwise the One of the most interesting consequences of this anal- ysis is the effect of the coupling of Eq. 4 in the electronic spectrum. The contribution of flexural phonons to the Kane-Mele coupling can be written as: (cid:10)(uB2)2(cid:11) ∆ph = g4 (23) where the brackets express the thermal average over the entire Brillouin zone. The flexural optical (ZO) mode strictly at Γ transforms according to B2. A rough estimate consists on neglecting V. CONCLUSIONS 7 We have studied the possible SO-mediated electron in- teraction with flexural phonons allowed by the symme- tries of the lattice and estimated the strength of these couplings from a tight-binding model. This analysis can be used in order to study the SO coupling in carbon nan- otubes. We find that the quadratic coupling with the phonons at Γ, particularly with the ZO branch, renormalizes the Kane-Mele mass in a remarkable way. From our theory, we predict an enhancement of two orders of magnitude in comparison to previous estimations for flat graphene, putting this gap close to the present ex- perimental limits.27 Note that the frequency of flexu- ral modes depends on the coupling to the substrate, if any, and can be tuned by applied strains.28 On general grounds, it can be expected that a compressive strain will lower the frequency of these modes, enhancing the spin-orbit coupling. Our theory is also relevant for spin transport experi- ments, particularly in suspended samples,29 where charge transport is ultimately limited by flexural phonons.30 VI. ACKNOWLEDGEMENTS This work has been funded by the MICINN, Spain, (FIS2008-00124, FIS2011-23713, CONSOLIDER CSD2007-00010), and ERC, grant 290846. AHCN acknowledges DOE grant DE-FG02-08ER46512, ONR grant MURI N00014-09-1-1063, and the NRF-CRP award "Novel 2D materials with tailored properties: be- yond graphene" (R-144-000-295-281). H. O. acknowl- edges financial support through grant JAE-Pre (CSIC, Spain). The authors acknowledge the hospitality of the KITP, Santa Barbara. This work is partially supported by the National Science Foundation under Grant No. NSF PHY11-25915. Appendix A: Choice of the spinor basis With this appendix we try to avoid possible confusions due to notation and choice of spinor basis in the main text, in particular when compared to Ref. 4. When discussing the symmetries of the graphene lat- tice from the perspective of group theory it is common to introduce electronic operators without specifying their explicit expression in a certain basis, only the algebraic rules which determine how they transform under the symmetry operations of the enlarged point group C(cid:48)(cid:48) 6v. For instance, the two 4 × 4 matrices which transform ac- cording to the vector irreducible representation E1 are denoted by Σx, Σy. The matrix Σz is defined in order to complete the Pauli-matrix algebra, and it transforms according to A2. On the other hand, the matrices which FIG. 4: Effective Kane-Mele mass induced by the coupling with flexural phonons. In red (lower curve) the estimation neglecting the acoustic branch and the dispersion of the op- tical one. In blue (upper curve) the calculation within the model described in Appendix B. the contribution from the acoustic branch and the disper- Γ ≈ 110 meV,26 tem- sion of the optical mode. Since ωZO perature plays no role, see the red curve in Fig. 4. How- ≈ ever, the zero-point motion contribution ∆0 = 0.03 meV is non-negligible (here M is the mass of the carbon atom). 2M ωZO g4 Γ This is a very crude approximation, since the identifi- cation of uB2 with the ZO mode is strictly true at the Γ point. The entire Brillouin zone contributes to the aver- age, so away from Γ both flexural acoustic (ZA) and op- tical branches enter. We can use the symmetry-adapted basis A1(cid:105) and B2(cid:105) in order to describe the polarizations of the ν = ZA,ZO phonons: ν(cid:105) = ην A1 (q)A1(cid:105) + ην B2 (q)B2(cid:105) (24) Note that, because of time reversal symmetry, η (q) = ∗ [η (−q)] . Thus, we have: (cid:12)(cid:12)ην B2 (q)(cid:12)(cid:12)2(cid:68)(cid:12)(cid:12)uν q (cid:12)(cid:12)2(cid:69) (25) T (cid:10)(uB2 )2(cid:11) ≈ 1 (cid:88) (cid:88) N q∈BZ ν where: N(cid:88) i=1 uν q = 1√ N uν (Ri) e−iq·Ri (26) is the Fourier transform of the phonon displacement field in branch ν, and the brackets denote thermal average. We need a model in order to describe the deviations of the polarizations vectors and the frequencies of both branches in the entire Brillouin zone. In Appendix B we describe the simpler nearest-neighbor forces model that one can consider in order to describe the dynamics of flexural phonons. We compute the induced Kane-Mele mass from Eqs. (23) and (25) within this model. The results are shown in Fig. 4 (blue curve). Remarkably, the Kane-Mele gap induced by phonons 2∆ph is of the order of 0.1 meV. q1,q2zzsz100200300400500T (K)0.020.040.060.08(cid:54) (meV)q,qzzszσz ⊗ τz ⊗ sz Irrep z → −z symmetric A1 A2 B2 (cid:18) −σy ⊗ sz τz ⊗ sz (cid:19) σx ⊗ τz ⊗ sz (cid:18) −σy ⊗ τy ⊗ sz σy ⊗ τx ⊗ sz (cid:19) E1 E2 E(cid:48) 1 G(cid:48) (cid:19) σz ⊗ τz ⊗ sx z → −z asymmetric σx ⊗ τz ⊗ sy − σy ⊗ sx σx ⊗ τz ⊗ sx + σy ⊗ sy σx ⊗ τz ⊗ sx − σy ⊗ sy (cid:19) (cid:18) −σz ⊗ τz ⊗ sy (cid:18) σx ⊗ τz ⊗ sy + σy ⊗ sx (cid:18) −τz ⊗ sy (cid:19)   −σy ⊗ τy ⊗ sx −σy ⊗ τy ⊗ sy σy ⊗ τx ⊗ sy σy ⊗ τx ⊗ sx τz ⊗ sx 8 The notation is simplified essentially because in the basis the time reversal operation is implemented by T = iΣy ⊗ Λy ⊗ syK, in such a way that the three sets of matrices are odd under the action of T : T−−→ ΣyΣ∗ T−−→ ΛyΛ∗ T−−−→ sys∗ i Σy = −Σi i Λy = −Λi i sy = −si (A2) Σi Λi si and then the possible SO terms are constructed from products of a spin matrix si with Σi or Λi. Appendix B: Model for flexural phonons TABLE IV: Classification of the possible SO coupling terms according to how they transform under the symmetry opera- tions of the lattice and reflection in the graphene plane. transform according to the vector irreducible representa- tion E(cid:48) 2 are denoted by Λx, Λy, and as before Λz is de- fined in order to complete the Pauli-matrix algebra and transforms according to B1. E = α a2 We consider the simpler nearest-neighbor forces model where the elastic energy of the lattice can be written as:  hAi − 1 3 (cid:88) i (cid:88) (cid:104)ij(cid:105) 2 hBi − 1 3 (cid:88) (cid:104)ij(cid:105) hBj + 2 hAj (B1) A more physical construction would be the following. As we mentioned in the text, in order to write down the low-energy electronic Hamiltonian (neglecting the spin) we have to consider the 16 Hermitian operators acting in a 4-dimensional space. We could define two differ- ent sets of 2 × 2 Pauli matrices {σi}, {τi} associated to the physical sublattice and valley degrees of freedom re- spectively, so the set {I, σi, τi, σi ⊗ τj} provides a rep- resentation of the algebra of generators of U (4). These operators act in a space of 4-component Bloch functions Ψ = (ψA,K+, ψB,K+, ψA,K−,↑, ψB,K− )T . Then, we intro- duce the Pauli matrices associated to spin, so we have to double the space of Bloch functions in order to contain the two spin projections. The possible SO couplings in this basis are summarized in Tab. IV. In the first row we can recognize the Kane-Mele and Rashba-like couplings as discussed in Ref. 4. Note that in this basis the time reversal operation is implemented by the anti-unitary op- erator T = iτx ⊗ syK, where K denotes complex conju- gation. In the main text we employ a different basis, where the order of the projection of the Bloch func- tions at each sublattice in different valleys is in- verted, and also a minus sign is introduced: Ψ = (ψA,K+, ψB,K+, ψB,K− ,−ψA,K− )T . This basis is very convenient because the notation is simplified. In this basis the operators Σi, Λi are related with the matrices acting in subalattice and valley indices as: Here a is the carbon-carbon distance and α is a constant with units of energy which can be related with the bend- ing rigidity (κ) of graphene in a continuum description,31 as we are going to see next. This model leads to the dy- namical matrix: (cid:33) f (q)2 3 + ∗ −2f (q) 3 −2f (q) f (q)2 3 + 3 (B2) (cid:32) 2α 3a2 D (q) = where f (q) = (cid:80) nearest-neighbors. The frequencies read: (cid:115) α eiq·(cid:126)δα, and the sum is extended to (cid:18) (cid:19) f (q)2 (B3) ω± = 2α 3M a2 3 + 3 2 with polarization vectors ±(cid:105) = 1√ . Here M is the mass of the carbon atom. The two branches of Eq. (B3) are plotted in Fig. 5. The upper branch ω+ must be identified with the optical one, whereas ω− corresponds to the acoustic one. ural phonons.26 At q ∼ Γ we have: The model reproduces very well the dispersion of flex- ± 2f (q) (cid:16) f (q) f (q) ,∓1 (cid:17)T ω+ ≡ ωZO (cid:114) (cid:114) 8α (cid:114) α M a2 − q ≈ q ≈ ω− ≡ ωZA αa2 8M q2 8M a−2 q2 (B4) Σi = σi ⊗ I Λi = I ⊗ τi Note that the dispersion relation of ZA phonons is quadratic, as expected from symmetry considerations. On the other hand, both branches are degenerate at K± (A1) theory of elasticity31 we deduce the relation: √ α = 6 3κ (B5) 9 We set the value of α from the frequency of the flexural Γ ≈ 110 meV.26 We obtain optical phonon at Γ, ωZO α = 8.5 eV. By using the relation of Eq. (B5) we obtain κ ≈ 0.8 eV, which is a very reasonable value for the bending rigidity of graphene. This agreement confirms the reliability of the model. We compute the induced Kane-Mele mass from Eqs. (23) and (25) within this model. Note that: (cid:12)(cid:12)ηZO (cid:12)(cid:12)ηZA B2 B2 (q)(cid:12)(cid:12)2 (q)(cid:12)(cid:12)2 (cid:18) (cid:60)f (q) (cid:18) f (q) 1 − (cid:60)f (q) f (q) 1 + (cid:19) (cid:19) = = 1 2 1 2 (B6) FIG. 5: Dispersion of flexural phonons computed within the nearest-neighbor force model described in the text with α = 8.5 eV. In red (upper curve) the dispersion for the optical branch, in blue (lower curve) the acoustic branch. K± = ωZO (ωZA Γ /2), as it is also expected from sym- metry arguments. When we compare this model with the K± = ωZO where (cid:60)f denotes the real part of f . The results are shown in Fig. 4 (blue curve). 1 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jaing, Y. Zhang, S. V. Dubonos, I. V. Grogorieva, 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 F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). 4 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). 5 M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). Lett. 108, 196804 (2012). 17 E. McCann and V. I. Falko, Phys. Rev. Lett. 108, 166606 (2012). 18 D. Tom´anek and S. G. Louie, Phys. Rev. B 37, 8327 (1988). 19 D. Tom´anek and M. A. Schluter, Phys. Rev. Lett. 67, 2331 (1991). 20 J. Serrano, M. Cardona, and T. Ruf, Solid State Commun. 113, 411 (2000). 21 J. R. Schrieffer and P. A. Wolf, Phys. Rev. 149, 491 (1966). 22 M. Nakahara, Geometry, Topology and Physics (Adam Hilger, Bristol, 1990). 6 D. Huertas-Hernando, F. Guinea, and A. Brataas, Phys. 23 N. D. Drummond, V. Z´olyomi, and V. I. Fal'ko, Phys. Rev. Rev. B 74, 155426 (2006). B 85, 075423 (2012). 7 Y. Yao, F. Ye, X.-L. Qi, S.-C. Zhang, and Z. Fang, Phys. 24 M. F. Thorpe and D. Weaire, Phys. Rev. Lett. 27, 1581 Rev. B 75, 041401 (2007). 8 H. Min, J. E. Hill, N. A. Sinitsyn, B. R. Sahu, L. Kleinman, and A. H. MacDonald, Phys. Rev. B 74, 165310 (2006). 9 M. Gmitra, S. Konschuh, C. Ertler, C. Ambrosch-Draxl, and J. Fabian, Phys. Rev. B 80, 235431 (2009). 10 C. Weeks, J. Hu, J. Alicea, M. Franz, and R. Wu, Phys. Rev. X 1, 021001 (2011). 11 K.-H. Jin and S.-H. Jhi, arXiv:1206.3608v1 [cond-mat.mes- hall] (2012). (1971). 25 F. Guinea, J. Phys. C 14, 3345 (1981). 26 L. Wirtz and A. Rubio, Solid State Comm. 131, 141 (2004). 27 A. S. Mayorov, D. C. Elias, I. S. Mukhin, S. V. Morozov, L. A. Ponomarenko, K. S. Novoselov, A. K. Geim, and R. V. Gorbachev, arXiv:1206.3848v1 [cond-mat.mes-hall] (2012). 28 P. L. de Andr´es, F. Guinea, and M. I. Katsnelson, Phys. 12 A. H. Castro-Neto and F. Guinea, Phys. Rev. Lett. 103, Rev. B 86, 144103 (2012). 026804 (2009). 13 S. Fratini, D. Gos´albez-Mart´ınez, P. M. C´amara, and J. Fern´andez-Rossier, arXiv:1202.6216v1 [cond-mat.mes- hall] (2012). 14 T. Ando, Journal of the Physical Society of Japan 69, 1757 (2000). 15 J. D. Sau and S. Tewari, arXiv:1111.5622v1 [cond- mat.supr-con] (2011). 16 J. Klinovaja, S. Gangadharaiah, and D. Loss, Phys. Rev. 29 M. H. D. Guimaraes, A. Veligura, P. J. Zomer, T. Maassen, I. J. Vera-Marun, N. Tombros, and B. J. van Wees, Nano Lett. 12, 3512 (2012). 30 E. V. Castro, H. Ochoa, M. I. Katsnelson, R. V. Gor- bachev, D. C. Elias, K. S. Novoselov, A. K. Geim, and F. Guinea, Phys. Rev. Lett. 105, 266601 (2010). 31 L. D. Landau and E. M. Lifshitz, Theory of Elasticity (Pergamon Press, Oxford, 1959). (cid:71)MK(cid:71)0.020.040.060.080.10(cid:209)ΩeV
1305.0133	3	1305	2013-11-06T15:29:50	Ballistic spin transport in exciton gases	[ "cond-mat.mes-hall" ]	Traditional spintronics relies on spin transport by charge carriers, such as electrons in semiconductor crystals. This brings several complications: the Pauli principle prevents the carriers from moving with the same speed; Coulomb repulsion leads to rapid dephasing of electron flows. Spin-optronics is a valuable alternative to traditional spintronics. In spin-optronic devices the spin currents are carried by electrically neutral bosonic quasi-particles: excitons or exciton-polaritons. They can form highly coherent quantum liquids and carry spins over macroscopic distances. The price to pay is a finite life-time of the bosonic spin carriers. We present the theory of exciton ballistic spin transport which may be applied to a range of systems where bosonic spin transport has been reported, in particular, to indirect excitons in coupled GaAs/AlGaAs quantum wells. We describe the effect of spin-orbit interaction of electrons and holes on the exciton spin, account for the Zeeman effect induced by external magnetic fields, long range and short range exchange splittings of the exciton resonances. We also consider exciton transport in the non-linear regime and discuss the definitions of exciton spin current, polarization current and spin conductivity.	cond-mat.mes-hall	cond-mat	Ballistic spin transport in exciton gases A. V. Kavokin,1, 2 M. Vladimirova,1 B. Jouault,1 T. C. H. Liew,3 J.R. Leonard,4 and L.V. Butov4 1Laboratoire Charles Coulomb, CNRS-Universite de Montpellier II, Pl. Eugene de Bataillon, 34095 Montpellier Cedex, France 2Physics and Astronomy School, University of Southampton, Highfield, Southampton, SO171BJ, UK 3Mediterranean Institute of Fundamental Physics, 31, via Appia Nuova, Roma, 00040, Italy 4Department of Physics, University of California at San Diego, La Jolla, CA 92093-0319, USA Traditional spintronics relies on spin transport by charge carriers, such as electrons in semicon- ductor crystals. The challenges for the realization of long-range electron spin transport include rapid spin relaxation due to electron scattering. Scattering and, in turn, spin relaxation can be effectively suppressed in excitonic devices where the spin currents are carried by electrically neutral bosonic quasi-particles: excitons or exciton-polaritons. They can form coherent quantum liquids that carry spins over macroscopic distances. The price to pay is a finite life-time of the bosonic spin carriers. We present the theory of exciton ballistic spin transport which may be applied to a range of systems supporting bosonic spin transport, in particular, to indirect excitons in coupled quantum wells. We describe the effect of spin-orbit interaction for the electron and the hole on the exciton spin, account for the Zeeman effect induced by external magnetic fields and long range and short range exchange splittings of the exciton resonances. We also consider exciton transport in the non-linear regime and discuss the definitions of the exciton spin current, polarization current and spin conductivity. PACS numbers: 71.35.-y, 03.75.Kk, 03.75.Mn, 73.63.Hs, 78.55.Cr I. INTRODUCTION Excitons are electrically neutral and have finite life- times. These are two obstacles which make the develop- ment of excitonic spintronics, or spin-optronics challeng- ing. How possibly one can explore the current, which is carried by neutral particles, and whose amplitude changes with distance and time? - is a fair question to ask. While electrons and holes have been consid- ered as perfectly valid spin carriers, and exotic effects like the spin Hall effect1 have been intensively studied for them2 -- 6, the spin currents carried by excitons7 -- 9 and exciton-polaritons10,11 over tens or even hundreds of mi- crometers remained relatively less explored. There ex- isted a huge imbalance of theoretical works on fermionic and bosonic spin transport. This is changing now. A number of phenomena have been observed and studied in the field of bosonic spin currents recently7−26. To sum- marize tens of publications in one sentence: bosonic sys- tems bring new quantum coherent effects to the physics of spin transport. For instance, stimulation23 -- 25 and amplification26 of spin currents are possible in exciton and exciton-polariton Bose gases. Bosonic spintronics or spin-optronics operates with electrically neutral spin car- riers which makes control of spin currents carried by exci- tons a non-trivial task. Fortunately, the exciton density replaces charge in many aspects: the density currents may be efficiently controlled by stationary or dynamic potential gradients as demonstrated in recent works27,28. Combined with evident advantages of bosonic amplifica- tion and low dephasing, this makes spin-optronics a valu- able alternative to fermionic spintronics. Besides bosonic effects, exciton spin transport has another important spe- cific feature: it is dissipative by its nature, as the spin carriers have a finite (and short for excitons in regular materials) life-time. In continuous wave optical experi- ments stationary spin textures can appear: excitons are injected in the structure, they propagate ballistically or diffusively, and eventually disappear by radiative recom- bination. Their polarization properties and spin are in- herited by the emitted photons, that is why the polariza- tion patterns observed in near field photoluminescence experiments directly characterize exciton spin currents in the plane of the structure. The goal of this work is to define what the exciton spin, magnetization and polarization currents are, and to explain how they can be described within the most frequently used spin density matrix (DM) approach and mean-field 29,30 approximation. We consider a specific system, namely a planar zinc-blend semiconductor struc- ture containing quantum wells, where excitons can be formed. This choice is motivated by recent experimental results in GaAs/AlGaAs based coupled quantum wells. We limit the scope of this paper to heavy-hole excitons, however, our approach can be easily extended to light- hole excitons or excitons in quantum wells of a different symmetry. We do not speak here about the large vari- ety of recent experimental results and application of the formalism presented here to the description of one partic- ular experiment, as this would make this paper too long and too specific. For a direct comparison of theoretical simulations with the experimental data we address the reader to Ref. 8. The approaches formulated here are suitable for the description of a variety of excitonic spin effects in quantum wells. The paper is organized as follows. In Section II we introduce the spin DM formalism accounting for the dif- ferent mechanisms of spin re-orientation and the relation to electron and hole spin currents. In Section III we present numerical results obtained within the spin DM formalism and analyze them. In Section IV we study the non-linear spin dynamics of propagating excitons using the Gross-Pitaevskii (GP) equations. The next three sec- tions of the paper are devoted to exciton spin currents and polarization currents. Conclusions are perspectives are given in Section VIII. II. THE SPIN MATRIX FORMALISM FOR PROPAGATING EXCITONS In zinc-blend semiconductor quantum wells (e.g. in the most popular GaAs/AlGaAs system), the lowest energy exciton states are formed by electrons with spin projec- tions on the structure axis of +1/2 and -1/2 and heavy holes whose quasi-spin (sum of spin and orbital momen- tum) projection to the structure axis is +3/2 or -3/2. Consequently, the exciton spin defined as the sum of the electron spin and heavy-hole quasi-spin may have one of four projections on the structure axis: +1,-1,+2,-2 31. These states are usually nearly degenerate, while there may be some splitting between them due to the short and long-range exchange interactions. Only the states with quasi-spin projections ±1 are coupled to the light, these are so-called bright states. The states with quasi- spin projections ±2 are called dark states. It is important to note that the present formalism ad- dresses the spin part of the exciton wavefunction, which is a product of electron and hole spin functions. For ex- ample, the probability to find the exciton in the spin state +1 is given by a product of probabilities to find an elec- tron in the spin state -1/2 and the heavy hole in the spin state +3/2. The four component exciton wave-function is: Ψ = (Ψ+1, Ψ−1, Ψ+2, Ψ−2) =(cid:16)Ψe,− 1 2 Ψh,+ 3 2 , Ψe,+ 1 2 Ψh,− 3 2 , Ψe,+ 1 2 Ψh,+ 3 2 , Ψe,− 1 2 Ψh,− 3 2(cid:17) (1) and Ψe,− 1 where Ψe,+ 1 tron spinor wavefunction; Ψh,+ 3 ponents of the heavy-hole spinor wavefunction. are the components of the elec- are the com- and Ψh,− 3 2 2 2 2 To describe the dynamics of the system we will first define the Hamiltonian (Section II A) describing the dif- ferent physical mechanisms of spin evolution. We then introduce the spin DM (Section II B) for the description of exciton spin states. We relate its components to the observable Stokes' vectors of light emitted by bright exci- tons and use the Liouville equation to describe its evolu- tion in time. From the DM one can describe the polariza- tion state of excitons and thus exciton spin currents. It is also instructive to consider the relationship with electron and hole spin currents (Section II C). A. Hamiltonian 2 Here we derive the exciton Hamiltonian in the basis of +1,-1,+2,-2 states, accounting for the spin-orbit inter- action (Dresselhaus and Rashba effects)32,37, long- and short-range exchange interactions33 and Zeeman effect, but neglecting exciton-exciton interactions, which will be discussed in the Section IV and neglecting magnetic field effect on center-of-mass motion and internal structure of exciton34 -- 36. We consider excitons propagating ballisti- cally in the plane of a quantum well. We shall character- ize them by a fixed wave-vector, kex. We represent the full exciton Hamiltonian Htot ex as a sum of three parts de- scribing the spin-orbit and Zeeman effects on electrons, He ex, and holes, Hh ex, and the exchange induced splittings of exciton states, Hex ex: Htot ex = He ex + Hex ex. ex + Hh (2) 1. Dresselhaus and Zeeman terms We recall that the Rashba-Dresselhaus effect is a momentum-dependent splitting of spin bands in two- dimensional semiconductor systems. It originates from a combined effect of the atomic spin orbit coupling and asymmetry of the potential in the direction perpendicu- lar to the two-dimensional plane. This asymmetry comes either from the applied bias (which is described by the Rashba term in the Hamiltonian) or from the intrinsic asymmetry of the crystal lattice (described by the Dres- selhaus term in the Hamiltonian). We shall separately consider both the Dresselhaus term (in this sub-section) and the Rashba term (in the next sub-section). In order to build the 4×4 matrix Hamiltonian for exci- tons, we start with simpler 2× 2 Hamiltonians describing the spin-orbit and Zeeman effects for electrons and holes. The electron Hamiltonian in the basis of (+1/2,-1/2) spin states is: He = βe (ke,x σx − ke,y σy) − 1 2 geµBBσ. (3) Here ge is the electron g-factor, µB is the Bohr magne- ton, B is a magnetic field, σ is the Pauli matrix vector, and βe is the Dresselhaus constant describing spin-orbit interactions of electrons. The Pauli matrix operators are: σz =(cid:20) 1 0 0 −1(cid:21) , i 0 (cid:21) , σy =(cid:20) 0 −i 1 0(cid:21) . σx =(cid:20) 0 1 (4) Rewriting Eq. (3) and retaining only z-component of the magnetic field, which corresponds to the Faraday geom- etry, one can obtain: He =(cid:20) − 1 =(cid:20) − 1 2 geµBB βe (ke,x − ike,y) βe (ke,x + ike,y) 1 2 geµBB 2 geµBB βekeeiφ βekee−iφ 2 geµBB (cid:21) , 1 (cid:21) (5) where φ is the angle between kex and the chosen x-axis. The exciton Hamiltonian needs to be written in the basis of (+1,-1,+2,-2) exciton spin states, which correspond to (-1/2, +1/2, +1/2, -1/2) electron spin states. The elec- tron spin-flip couples +1 and +2 states as well as -1 and -2 states. For each of these two couples of states we apply the Hamiltonian [Eq. (5)], which results in the following electronic contribution to the 4 × 4 exciton Hamiltonian:  (6) ex = He  − 1 2 geµBB βekee−iφ − 1 2 geµBB 1 2 geµBB 1 2 geµBB βekee−iφ βekeeiφ βekeeiφ 0 0 0 0 0 0 0 0 2. Rashba terms We note that another possible spin-orbit contribution to the Hamiltonian may come from the Rashba effect, which takes place in biased quantum wells. The Rashba term to be added in Eq. (3) is αe (σxke,y − σyke,x), where αe is a constant proportional to the Rashba field. The contribution of the Rashba term to the electron Hamil- tonian, in the basis of (+1/2,-1/2) electron spin states, can be re-written: e =(cid:20) H′ 0 iαekee−iφ −iαekeeiφ 0 (cid:21) . (7) Using the same procedure as for Dresselhaus terms, this gives an additional contribution to the exciton Hamilto- nian, in the basis of (-1/2, +1/2, +1/2, -1/2) electron spin states: ex = He′  0 0 iαekee−iφ 0 0 0 0 −iαekeeiφ −iαekeeiφ 0 0 0 iαekee−iφ 0 0 0  (8) Unless stated explicitly, we will for simplicity omit the Rashba terms in the rest of this paper and consider only the Dresselhaus terms. Note that there are no linear in wave-vector Rashba terms for heavy holes in zinc-blend quantum wells grown along the (001)-axis. 3. Heavy-hole contribution (Faraday geometry) The heavy hole contribution to the Hamiltonian can be calculated with the reasoning similar to the electron case. The heavy-hole Hamiltonian written in the basis of (+3/2,-3/2) states is: Hh = βh (kh,x σx + kh,y σy) − 1 2 ghµBB σz. (9) Here gh is the heavy-hole g-factor and βh is the Dressel- haus constant for heavy holes32,37. Note that the Dressel- haus Hamiltonian is different for heavy holes formed by 3 p-orbital states and for conduction band electrons formed by s-orbital electronic states in a zinc-blend crystal lat- tice. The resulting from Dresselhaus coupling effective magnetic fields acting upon electron and heavy hole spins are oriented differently as well. Re-writing Eq. (9), we obtain: Hh =(cid:20) βh (kh,x + ikh,y) − 1 2 ghµBB =(cid:20) − 1 2 ghµBB βhkhe−iφ βhkheiφ 2 ghµBB (cid:21) , 1 βh (kh,x − ikh,y) 1 2 ghµBB (cid:21) (10) The hole spin-flip couples +1 and -2 states as well as -1 and +2 states. For each of these two couples of states we apply the Hamiltonian (10), which results in the following hole contribution to the 4 × 4 exciton Hamiltonian: ex = βhkhe−iφ Hh  2 ghµBB βhkheiφ βhkhe−iφ − 1 2 ghµBB  (11) − 1 2 ghµBB 1 2 ghµBB βhkheiφ 0 0 0 0 0 0 0 0 1 4. In-plane magnetic field (Voight geometry) If the magnetic field is applied in the plane, it splits electron and hole states polarized in the plane of the quantum wells. Suppose that the field is applied in the x-direction. In the electron and hole basis the Zeeman Hamiltonian is in this case: He,h = − 1 2 ge,hµBB σx =(cid:20) 0 − 1 2 ge,hµBB − 1 2 ge,hµBB 0 (cid:21) . (12) Note that the hole g-factor in plane of the quantum well is different from the g-factor in Faraday configuration, in general. This maps into the (+1,−1, +2,−2) exciton basis as a Zeeman Hamiltonian of the form: HZ = − µBB 2 0 0 ge gh 0 gh ge 0 0 ge gh 0 gh ge 0 0     5. Exchange terms (13) Besides the contributions from electron and hole spin orbit interactions and Zeeman splitting, there may be a purely excitonic contribution to the Hamiltonian, which is composed from the Hamiltonian for bright excitons written in the basis (+1,-1): Hb = Eb I − δb σx =(cid:20) Eb −δb −δb Eb (cid:21) (14) and the Hamiltonian for dark excitons written in the ba- sis (+2,-2): Hd = Ed I − δd σx =(cid:20) Ed −δd −δd Ed (cid:21) , (15) where I is the identity matrix. The terms with δb and δd describe the splittings of bright and dark states polar- ized along x and y axes in the plane of the structure due to the long-range exchange interaction. The structural anisotropy is virtually inevitable even in the best qual- ity epitaxially grown quantum wells. It arises from the reduced symmetry of heterointerfaces, from local strains and from islands of quantum well with fluctuations elon- gated in certain crystallographic directions. Eb − Ed is the splitting between bright (+1 and -1) and dark (+2 and -2) exciton states due to the short-range exchange interaction. In microcavities, this splitting is addition- ally enhanced due to the vacuum field Rabi splitting of exciton-polariton modes formed by bright excitons and a confined optical mode of the cavity38. The origin of Eqs. (14) and (15) can be easily seen from the exciton Hamiltonian written in the basis of linear x and y polarizations. For example, for the bright excitons: 4 HXY =(cid:20) Eb − δb Hb = C −1 HXY C, 0 0 Eb + δb (cid:21) where: C = 1 √2(cid:20) 1 1 i −i(cid:21) , C −1 = 1 1 i (cid:21) √2(cid:20) 1 −i (16) (17) (18) are the transformation matrices from the linear to circu- lar polarization basis and vice versa39. The same reason- ing can be applied to the dark excitons as well. The sum of Hamiltonians Hb and Hd, written in the 4 × 4 exciton spin basis is: ex = Hex  0 0 Eb −δb −δb Eb 0 0 0 0 0 Ed −δd 0 −δd Ed   (19) Now, the full exciton Hamiltonian can be written as: ex + Hh Eb + 1 ex ex ex +H 2 (ge − gh) µBB −δb βekeeiφ βhkheiφ ex = He Htot =  −δb Eb − 1 2 (ge − gh) µBB βhkhe−iφ βekee−iφ βekee−iφ βhkheiφ 2 (ge + gh) µBB Ed − 1 −δd βhkhe−iφ βekeeiφ −δd Ed + 1 2 (ge + gh) µBB . (20)   For the translational motion of an exciton as a whole particle the exciton momentum is given by Pex = (me + mh) vex , where me and mh are in-plane effective masses of an electron and of a heavy hole, respectively; vex is the exciton velocity. Having in mind that the ex- citon translational momentum is a sum of electron and hole translational momenta given by Pe,h = me,hve,h, where ve and vh are the electron and hole velocity, re- spectively, one can easily see that vh = ve = vex. Having in mind that Pex = kex and Pe,h = ke,h, we have kex = kh + ke, with ke = kex and kh = mh ex depends on the exciton cen- ter of mass wave-vector kex and on the angle φ between this angle and one of the structure axes (e.g. (100)-axis). kex. Thus, Htot me+mh me+mh me Vishnevsky et al 40 . Their analysis confirms the presence of linear in kex spin orbit terms in the exciton Hamilto- nian introduced above. B. Spin Density Matrix It should be noted that in this consideration the wave- vectors kex, kh, ke are related to the translational motion of the exciton as a whole particle, with hole and electron as its constituents. The wave-vector of relative motion of the electron and hole "inside" the exciton is zero on average but may be important for each given moment of time. Recently, the effect of relative electron-hole motion on the spin-orbit effects of excitons has been analyzed by Having constructed the Hamiltonian for excitons prop- agating with a wavevector kex, we now consider the de- scription of their spin state. We shall use the spin density matrix, ρ = ΨihΨ, where Ψ = (Ψ+1, Ψ−1, Ψ+2, Ψ−2) are the components of the exciton wavefunction projected onto the four spin states, (Ψ+1i ,Ψ−1i ,Ψ+2i ,Ψ−2i). 1. Relation to Stokes' vectors and polarization degrees of light The exciton spin DM is given by: ρ = ΨihΨ =  −1Ψ+1 Ψ∗ −1Ψ−1 Ψ∗ −1Ψ+2 Ψ∗ −1Ψ−2 Ψ∗ +1Ψ+1 Ψ∗ +1Ψ−1 Ψ∗ +1Ψ+2 Ψ∗ +1Ψ−2 Ψ∗ Ψ∗ Ψ∗ Ψ∗ Ψ∗ +2Ψ+1 Ψ∗ +2Ψ−1 Ψ∗ +2Ψ+2 Ψ∗ +2Ψ−2 Ψ∗ −2Ψ+1 −2Ψ−1 −2Ψ+2 −2Ψ−2,   (21) The elements of the upper left quarter of this DM are linked to the intensity of light emitted by bright exciton states, I = Ψ∗ −1Ψ−1, and to the components of the Stokes' vector, Sx, Sy and Sz of the emitted light: +1Ψ+1 + Ψ∗ I 2 ρ11 = + Sz, ρ12 = Sx − iSy, ρ21 = Sx + iSy, ρ22 = I 2 − Sz. (22) (23) (24) (25) These expressions can be summarized more succinctly using the Pauli matrices as: ρ21 ρ22 (cid:21) = (cid:20) ρ11 ρ12 I 2 I + S.σ, (26) where S = (Sx, Sy, Sz) is the Stokes' vector and we recall that I is the identity matrix. Note that the trace of the spin density matrix is a number of particles in the system, which is not conserved because of the finite lifetime, in contrast with the full quantum optical density matrix which has the trace equal to unity. Often when studying the polarization structure of fields with non-uniform intensity, it is useful to compare the polarization degrees of emitted light, which can be given by normalizing the Stokes' vectors to the light in- tensity. The circular polarization degree is: ρc = 2Sz I = ρ11 − ρ22 ρ11 + ρ22 . The horizontal-vertical linear polarization degree is: ρl = 2Sx I = ρ12 + ρ21 ρ11 + ρ22 . (27) (28) The linear polarization degree measured in the diagonal axes (also referred to as a diagonal polarization degree) is given by: ρd = 2Sy I = i ρ12 − ρ21 ρ11 + ρ22 . (29) 2. Liouville equation The dynamics of the DM is given by the quantum Li- ouville equation: i dρ dt =h Htot ex , ρi , (30) 5 where the Hamiltonian is composed from the electron, hole and exciton contributions given by Eqs. (6), (11) and (19) (considering the Faraday magnetic field config- uration). So far, we have neglected all relaxation or scattering processes in the system. The commonly used way to ac- count for these processes is through the introduction of a phenomenological Lindblad superoperator to the Liou- ville equation: i dρ dt =h Htot ex , ρi − L (ρ) , where the Lindblad superoperator is introduced as: (31) . (32) L (ρ) = i  ρ11/τb ρ12/τb ρ13/τc ρ14/τc ρ21/τb ρ22/τb ρ23/τc ρ24/τc ρ31/τc ρ32/τc ρ33/τd ρ34/τd ρ41/τc ρ42/τc ρ43/τd ρ44/τd   τb is the bright exciton decoherence time, τd is the dark exciton decoherence time and τc is the characteristic de- coherence time of processes between dark and bright ex- citons. Note that dissipation may be crucial in the de- scription of exciton spin currents in realistic systems. In particular, within this formalism, in the presence of dis- sipation, the current conservation and flux conservation conditions become valid if completed by exciton genera- tion and decay in the continuity equation. In the rest of this manuscript we shall neglect dissipation to simplify the model system and to clarify the physical mechanisms which govern the characteristics of exciton spin currents. Namely, we shall assume L (ρ) = 0. We stress that, in the experiment, the magnitude of predicted spin currents may be reduced by dissipation. The formalism described so far, in sections II A and this section (II B), has been successfully applied in the description of spin transport in gases of cold excitons in coupled GaAs/AlGaAs quantum wells8. In this work, cold excitons are generated within localized spots and then fly away ballistically in radial directions. The ele- ments of the DM, ρij, are dependent on the distance from the excitation spot r = vext at time t and the polar an- gle, φ. The propagation speed vex = kex/ (me + mh). In this work, when solving the Liouville equation intro- duced above, we shall refer to the experimental config- uration of Ref. 8. In particular, this implies a specific choice of the initial conditions for Eq. (30): we shall as- sume that at zero time excitons are not moving. We shall assume that they populate the eigenstates of the exciton Hamiltonian Htot ex taken with kex = 0 following a thermal distribution with a temperature T . We shall assume that once created in the equilibrium state, the excitons start moving apart in the radial direction. Thus, implicitly, we account for a non-linear effect: dipole-dipole repul- sion of excitons which makes them acquire a certain in- plane velocity vex. This non-linearity is crucial to move the system out of equilibirum. The rest of exciton prop- agation and spin dynamics is modelled using the linear equation (30), where the Hamiltonian Htot ex contains now the off-diagonal terms proportional to kex. C. Electron and hole spin currents Now, the exciton spin DM (21) can be represented in terms of electron and hole wavefunctions as: We shall normalize exciton, electron and heavy-hole functions to unity, namely: 6 Ψ∗ +1Ψ+1 + Ψ∗ −1Ψ−1 + Ψ∗ +2Ψ+2 + Ψ∗ Ψ∗ e,+ 1 2 Ψe,+ 1 2 Ψ∗ h,+ 3 2 Ψh,+ 3 2 −2Ψ−2 = 1 Ψe,− 1 = 1 2 + Ψ∗ e,− 1 2 + Ψ∗ h,− 3 2 Ψh,− 3 2 = 1 (33) (34) (35) e,− 1 2 e,− 1 2 e,− 1 2 Ψ∗ Ψ∗ Ψ∗ Ψ∗ e,− 1 2 2 Ψe,− 1 Ψe,+ 1 Ψe,+ 1 Ψe,− 1 2 2 2 ρ =  h,+ 3 2 h,+ 3 2 h,+ 3 2 Ψ∗ Ψ∗ Ψ∗ Ψ∗ h,+ 3 2 2 Ψh,+ 3 Ψh,− 3 Ψh,+ 3 Ψh,− 3 2 2 2 e,+ 1 2 e,+ 1 2 e,+ 1 2 Ψ∗ Ψ∗ Ψ∗ Ψ∗ e,+ 1 2 2 Ψe,− 1 Ψe,+ 1 Ψe,+ 1 Ψe,− 1 2 2 2 Ψ∗ Ψ∗ Ψ∗ Ψ∗ h,− 3 2 h,− 3 2 h,− 3 2 h,− 3 2 2 Ψh,+ 3 Ψh,− 3 Ψh,+ 3 Ψh,− 3 2 2 2 e,+ 1 2 e,+ 1 2 e,+ 1 2 Ψ∗ Ψ∗ Ψ∗ Ψ∗ e,+ 1 2 2 Ψe,− 1 Ψe,+ 1 Ψe,+ 1 Ψe,− 1 2 2 2 Ψ∗ Ψ∗ Ψ∗ Ψ∗ h,+ 3 2 h,+ 3 2 h,+ 3 2 h,+ 3 2 2 Ψh,+ 3 Ψh,− 3 Ψh,+ 3 Ψh,− 3 2 2 2 e,− 1 2 e,− 1 2 e,− 1 2 Ψ∗ Ψ∗ Ψ∗ Ψ∗ e,− 1 2 2 Ψe,− 1 Ψe,+ 1 Ψe,+ 1 Ψe,− 1 2 2 2 Ψ∗ Ψ∗ Ψ∗ Ψ∗ h,− 3 2 h,− 3 2 h,− 3 2 h,− 3 2 2 Ψh,+ 3 Ψh,− 3 Ψh,+ 3 Ψh,− 3 2 2 2  (36) , This representation allows us to obtain useful links be- tween the elements of exciton, electron and hole density matrices, in particular: The in-plane component of electron and hole spins can be extracted from the off-diagonal elements of the DM. Namely, the x-component of electron spin is given by: ρe = ΨeihΨe =" Ψ∗ Ψ∗ e,+ 1 2 e,+ 1 2 Ψe,+ 1 Ψe,− 1 2 2 Ψ∗ Ψ∗ e,− 1 2 e,− 1 2 Ψe,+ 1 Ψe,− 1 2 2 # =(cid:20) ρ22 + ρ33 ρ24 + ρ31 ρ13 + ρ42 ρ11 + ρ44 (cid:21) ρh = Ψhi hΨh =" Ψ∗ Ψ∗ h,+ 3 2 h,+ 3 2 Ψh,+ 3 Ψh,− 3 2 2 Ψ∗ Ψ∗ h,− 3 2 h,− 3 2 Ψh,+ 3 Ψh,− 3 2 2 # ρ23 + ρ41 ρ22 + ρ44 (cid:21) =(cid:20) ρ11 + ρ33 ρ14 + ρ32 (38) We know that the components of electron and hole den- sity matrices are linked with the projections of electron and hole spins as: ρe =(cid:20) ρh =(cid:20) 1 2 + Se,z Se,x + iSe,y 1 2 + Sh,z Sh,x + iSh,y 1 Se,x − iSe,y 2 − Se,z Sh,x − iSh,y 2 − Sh,z 1 (cid:21) (cid:21) , (39) (40) where, for the heavy hole we have assigned spin +1/2 to the state +3/2 and spin -1/2 to the state -3/2 accounting for the orbital momentum of these states of +1 and -1, respectively. The z-component of the spin polarization carried by electrons can now be expressed as: Se,z = (ρ22 + ρ33 − ρ11 − ρ44) /2. (41) Similarly, the z-component of the spin polarization car- ried by holes can now be expressed as: Sh,z = (ρ11 + ρ33 − ρ22 − ρ44) /2. (42) Bright-dark exciton splitting Eb − Ed Se,x = (ρ13 + ρ31 + ρ24 + ρ42) /2, (43) while the x-component of the hole spin is given by: (37) Sh,x = (ρ14 + ρ23 + ρ32 + ρ41) /2. (44) The y-component of electron spin is given by: Se,y = i (−ρ13 + ρ31 + ρ24 − ρ42) /2, (45) while the y-component of the hole spin is given by: Sh,y = i (ρ14 − ρ23 + ρ32 − ρ41) /2, (46) III. NUMERICAL RESULTS IN THE DENSITY MATRIX FORMALISM Figure 1 shows the numerical results obtained within the DM formalism for a model system with the same pa- rameters as those of coupled double quantum wells stud- ied in Ref. 8. The parameters are summarized in Table I. TABLE I: Parameters for numerical calculations. Electron mass me Heavy hole mass mh βe Electron Dresselhaus coupling Heavy-hole Dresselhaus coupling Bright exciton XY splitting Dark exciton XY splitting βh δb δd 0.07 m0 0.16 m0 2.7 µeVµm 0.92 µeVµm 0.5 µeV −13 µeV 5 µeV tween bright and dark excitons this Hamiltonian can be rewritten as: 7 H = −δd  0 0 ξeiφ 0 0 0 0 ξe−iφ ξe−iφ 0 0 0 1 ξeiφ 1 0   (47) where ξ = −βeke/δd. In other words, here we only take into account linear splitting of dark exciton states which inevitably results from structural anisotropy even in the best quality samples, and the Dresselhaus field acting on the spin of electron, bound to the hole. We will see that at sufficiently low temperature these two ingredients provide the in-plane asymmetry that ultimately results in the formation of linear polarisation patterns. Indeed, the eigenvalues of this Hamiltonian can be obtained an- 2 δdp1 + 4ξ2. This corresponds alytically : E = ± 1 to two dispersion branches at low energy ∝ ±ξ2 and two branches at high energy ∝ ±(1 + ξ2) for ξ ≪ 1. These branches are shown by red squares in Fig. 1(a).The eigenvectors, starting from the lowest energy and taking δd < 0, can be approximated for small ξ by: 2 δd ± 1 ξe−iφ −ξeiφ −1 1     ,  e−iφ eiφ −ξ −ξ   ,  e−iφ −eiφ ξ −ξ ξe−iφ ξeiφ 1 1   ,    . (48) As mentioned already, at low temperatures kBT ≪ δd, the lowest energy state with zero momentum is given by [0, 0,−1, 1] and is a linearly polarized dark exciton. By linearly polarised dark exciton we mean the dark state which has a dipole moment oriented in a certain way, and which has a zero spin projection to the grows axis of the structure. After an acceleration due to dipole repulsion, this initial state is no longer an eigenstate and oscillates as a function of time between the two eigenstates of the Hamiltonian with which the initial state is not orthogo- nal, namely the first and the third eigenstates listed in Eqs. 48. Among these two states, only the third gives a significant contribution to the observed polarization as it is essentially "bright" (has large projections to +1 and −1 exciton states). The linear and diagonal polarizations originating from this state are readily given by − cos(2φ) and − sin(2φ) respectively. This reproduces the essential features of the numerical results for the linear polariza- tion pattern, as one can see comparing the images cal- culated with the reduced Hamiltonian (Fig. 1(b)) and the full Hamiltonian (Fig. 1(c)). In the particular case considered here (initial state formed essentially by lin- early polarized dark excitons) the Dresselhaus spin-orbit term for electrons leads to formation of the linear polar- ization vortex: the polarization plane is always perpen- dicular to the wave vector direction. The linear polar- ization vortex has been observed experimentally by High et al8,9. Rapid oscillations in radial direction are due periodical change in the occupation of mainly dark and FIG. 1: (color online) (a) Dispersion of the excitonic states calculated using the set of parameters from Table I (green lines) and reduced set of parameters with Eb − Ed = 0, δb = 0, βh = 0 (red symbols). Black arrows indicate the values of the wavevectors used in (b)-(d) kex = 15.3µm−1 and (e) kex = 5µm−1. Double-ended arrows indicate the energies of oscillations between the eigenstates which appear in spatial polarization patterns. (b) Linear polarization degree along x- axis calculated with the simplified set of parameters. (c) Same as (b) but for the full set of parameters from the Table I. (d) Circular polarization degree with parameters from Table I. (e) Same but at kex = 5µm−1. The source area was taken circular with a radius of 4µm. The dispersion of bright and dark exciton modes ob- tained by diagonalisation of the Hamiltonian Htot ex (20) is shown in Fig. 1(a) by green solid lines. The mo- mentum has been chosen along the x direction, but the anisotropy of the band structure remains small. The initial splittings of dark and bright states makes these dispersion curves qualitatively different from those pre- sented by Vishnevsky et al 40. Note also, that Ref. (40) accounts for the exciton kinetic energy which we neglect in Htot ex , leaving only the spin-dependent contributions to the energy. The numerically calculated linear and circu- lar polarization degrees are shown in Figs. 1(c) and (d), respectively. To stay close to the experimental conditions of Ref.8, we have chosen as initial condition the circular source area with a radius of 4 µm, where cold excitons are generated within localized spots. In Fig.1(c), four lobes are unambiguously observable in the pattern of the lin- ear and diagonal (not shown) polarizations. This pattern is a consequence of the Dresselhaus spin-orbit coupling for electrons, and it is characteristic of the chosen initial state at the source: four split eigen-states with zero in- plane wave-vector are occupied. Relative occupation of these states corresponds to the Boltzmann distribution at temperature T = 0.1 K. This choice of initial condi- tions leads to a variety of polarization patterns observed in experiment8. In the absence of the damping this pat- tern is periodic and infinite in the radial direction. In order to reveal the mechanism of formation of the polarization patterns it is instructive to consider a sim- plified version of the Hamiltonian 20. In the absence of magnetic field, zero Dresselhaus effect for holes, zero splitting of bright excitons and zero splitting Eb − Ed be-         mainly bright states, indicated by double-ended arrows in Fig. (1(b)). We should emphasize that with a proper reordering of the basis vectors, the simplified Hamiltonian 47 is analyt- ically equivalent to the Hamiltonian of bilayer graphene. In its simplest expression, the Hamiltonian of bilayer graphene can be written as41: 0 vF ke−iφ vF keiφ 0 0 0 t⊥ 0 0 t⊥ 0 0 0 vF ke−iφ vF keiφ 0 . (49) where vF is the velocity, k the amplitude of the mo- mentum, φ the angle between the momentum and the x axis, and t⊥ is the main coupling term between the two graphene layers, which is given by hopping between two carbon atoms that are superimposed. The four coef- ficients of the associated wavevectors correspond to the probability amplitudes on the two independent sublat- tices of the two graphene layers. By permutation of the basis vectors:1 → 1, 2 → 3, 3 → 4, 4 → 2, the Hamilto- nian is rewritten as: 0 0 vF keiφ 0 0 0 0 vF ke−iφ vF ke−iφ 0 0 0 t⊥ vF keiφ t⊥ 0 . (50) which is equivalent to Eq. 47 with t⊥ ≡ −δd and vF k/t⊥ ≡ ξ. The 4 × 4 Hamiltonian of bilayer graphene is often restricted to the subspace of the two low energy bands. From Eq. 48, the corresponding eigenstates are: [eiφ,±e−iφ]. (51) We can then define a pseudospin vector, which represents the relative phase between the two components of the wavevectors. From Eq. 51, the pseudospin rotates two times when the particle wavevector undergoes one full rotation. Adapting the terminology used for graphene, the polar- ization pattern observed in Fig. 1(b,c) would be nothing else than the fingerprint of the "pseudospin" rotation in the exciton system. In the context of the exciton system under consideration, the phase φ corresponds to the angle between the exciton polarization vector and the chosen x-axis of the structure. The build up of circular polarization requires intro- duction in the model of the splitting between linearly polarized bright exciton states δb . This splitting acts as an effective magnetic field applied to the Stokes vector of light emitted by bright excitons, S = (Sx, Sy, Sz). In particular, if the x-polarized exciton state has a lower energy than the y-polarized exciton state, it creates an 8 effective magnetic field in the x-direction which rotates the Stokes vector in the yz plane. This converts the diag- onal polarization to the circular polarization and leads to the appearance of right- and left-circularly polarized sec- tors in the polarization map of exciton emission. Separa- tion of spins due to linear-to-circular polarization conver- sion is known in exciton-polariton systems as the Optical spin Hall effect. This effect was theoretically predicted in Ref. 12 and experimentally observed in polaritonic10 and excitonic systems8. For microcavity polaritons it may be described in terms of beats between TE and TM polarized polariton modes, while in the exciton sys- tem studied here the effect is more complex due to the mixture of four nearly degenerate (dark and bright) ex- citon states. A recent theoretical paper40 predicts the skyrmion formation in this case. Here we concentrate on the circular polarization patterns appearing due to the beats between linearly polarized exciton states mixed by the exchange interaction. Note that the circular polar- ization pattern is strongly sensitive to the chosen exci- ton wave-vector, which governs the energies of the four involved eigen-states. Figure 1(d) is calculated assum- ing kex = 15.3µm−1. This corresponds to the crossing point of the dispersion branches associated to the first and second exciton eigen-states (see Fig 1(a)). For com- parison, Fig. 1(e) shows the circular polarization pattern calculated with kex = 5µm−1. One can see that the four lobes pattern of circular polarization is washed out, because the lowest energy state of the system remains essentially dark. Rapid oscillations that show up, have the period determined by the splitting between the two lowest states. ex 2mex Finally, let us underline that the dispersion curves shown in Fig. 1(a) do not take into account the kinetic energy of excitons. The kinetic energy would shift all curves up by K = 2k2 . Note that this does not affect the splittings between exciton eigenstates and would not affect the spin dynamics of excitons. In the spin density formalism developed above we assign to all excitons the same kinetic energy, K. In a realistic system, the kinetic energy may be spread, in which case averaging of the obtained polarization patterns over kex may be needed. This averaging would smooth the fast oscillations seen on the images Fig. 1 (c,d,e,f). In the next section devoted to the non-linear spin dynamics we will fully take into account the kinetic energy of propagating excitons. The distribution of in-plane projections of electron and hole spins for the same choice of parameters as above is shown in Figs. 2(a-h). The left panels show electron spins and the right panels show the hole spins. The di- rection of in-plane spin component is shown by arrows, while the length of each arrow is proportional to the com- puted value of the transverse spin component. The up- per panels, Figs. 2(a,b), show the spin distributions in the absence of a magnetic field. In this case the electron and hole spins are oriented along the effective Dressel- haus fields which are oriented differently for electrons and heavy holes, as we have discussed in the previous section. be observed directly in optical experiments. However, they can be deduced from fitting the exciton polariza- tion maps, e.g., using the formalism described above. 9 FIG. 2: (color online) Spatial distribution of electron and hole spin components in the plane of the quantum well structure, calculated at zero magnetic field and at B = 5 T at three different orientations. The parameters are given in Table I, at kex = 15.3µm−1 .The source area was taken circular with a radius of 4µm. The decrease of the in-plane spin component upon prop- agation corresponds to the build-up of the z-component of electron and hole spins (Figure not shown), due to ro- tation of the exciton spin around the effective magnetic field. The magnetic field strongly changes the spin distri- bution in real space. The spin textures become strongly anisotropic in the case of in-plane (x- or y-oriented) mag- netic field. Note, that the in-plane isotropy in the system is broken by the splitting between x- and y- polarized exciton states, which is why switching of the magnetic field between x and y axes strongly affects the distribu- tion of electron and hole spins. It should be noted also that electron and hole in-plane spin textures can hardly FIG. 3: (color online) Spatial distribution of exciton linear polarization ((a)-(d), length of the bars maps the intensity in arbitrary units) and color maps of the exciton circular polar- ization degree ((e)-(f)), same color code as in Fig. 1). Mag- netic field B = 0 in (a), (e), and B = 5T along z in (b),(f), x in (c),(g), y in (d),(h). Parameters are given in Table I, kex = 15.3µm−1. The source area was taken circular with a radius of 4µm. Figure 3 shows how the magnetic field affects spatial patterns of linear (a-d) and circular (e-h) polarization. Switching the magnetic field orientation between the x-, y- and z-axes one can dramatically affect the polarization patterns. Having in mind that the exciton polarization patterns can be directly observed in near-field photolu- minescence experiments, fitting of these patterns to the experimental data would allow extracting the Dressel- haus constants and exciton exchange splittings, which, in turn, allow to restore electron and hole spin textures8. Figure 4(a-d) illustrates a peculiar regime where the Dresselhaus fields for electrons and holes are taken to be zero and there is no magnetic field applied, but electrons are subjected to the Rashba field (the Rashba field for heavy holes is zero). Figure 4(a) shows the electron spin distribution in space, where the spins are clearly aligned 10 quantum Liouville equation (31) is a very efficient tool for the description of effects linear in the exciton density. On the other hand, one cannot straightforwardly incorporate non-linear interaction terms in this equation. The treat- ment of non-linear effects in a partially coherent system is a non-trivial task. Much simpler is the treatment of non-linear effects in a perfectly coherent system, such as a condensate at zero temperature. In this case, the ensem- ble of excitons can be described by a single 4-component wave function Ψ = (Ψ+1, Ψ−1, Ψ+2, Ψ−2)T . The linear dynamics of this wave-function for ballistically propagat- ing excitons having a wavevector kex is described by the Schrodinger equation: i d dt Ψi = HΨi , (52) where the Hamiltonian is the same as in Eq. (20). This equation represents a set of four coupled linear differen- tial equations for four exciton spin components. Non- linear effects lead to the condensate evolution in real and reciprocal space. From now on we shall consider the exciton spin dynam- ics in real space (2D), so that the wave function Ψ will be- come coordinate-dependent and will not be restricted to one single value of kex. The non-linear interaction terms for multi-component exciton gases are introduced and discussed in detail in Ref. 42. Here we expand Eq. (52) by introducing the kinetic energy (to describe the real space dynamics) and the interaction terms. On the other hand, we neglect the magnetic field, for simplicity. This results in a system of four non-linear Schrodinger or GP equations42 -- 44: i dΨ+1 dt i dΨ−1 dt i dΨ+2 dt = − βeme Ψ+1 + mex (cid:16)kx − iky(cid:17) Ψ+2 2 ∇2 2mex βhmh mex (cid:16)kx − iky(cid:17) Ψ−2 + α1 Ψ+12 Ψ+1 + + α2 Ψ−12 Ψ+1 + α3 Ψ+22 Ψ+1 + α4 Ψ−22 Ψ+1 + W Ψ∗ −1Ψ+2Ψ−2, (53) = − βeme Ψ−1 + mex (cid:16)kx + iky(cid:17) Ψ−2 2 ∇2 2mex βhmh mex (cid:16)kx + iky(cid:17) Ψ+2 + α1 Ψ−12 Ψ−1 + + α2 Ψ+12 Ψ−1 + α3 Ψ−22 Ψ−1 + α4 Ψ+22 Ψ−1 + W Ψ∗ +1Ψ+2Ψ−2, (54) = − βeme Ψ+2 + mex (cid:16)kx + iky(cid:17) Ψ+1 2 ∇2 2mex βhmh mex (cid:16)kx − iky(cid:17) Ψ−1 + α1 Ψ+22 Ψ+2 + + α2 Ψ−22 Ψ+2 + α3 Ψ+12 Ψ+2 + α4 Ψ−12 Ψ+2 + W Ψ∗ −2Ψ+1Ψ−1, (55) FIG. 4: (color online) Spatial distribution of electron (a) and hole (b) in-plane spin component calculated at B=0 in the absence of the Dresselhaus field βe = 0; βh = 0, but including Rashba field αe = 2.7µeVµm;. Other parameters are given in Table I, kex = 15.3µm−1. The source area was taken circular with a radius of 4µm. Corresponding patterns of linear (c) and circular (d) exciton polarization degree are shown with the same color code as in Fig. 1. along the Rashba field force lines. Interestingly, the hole spins become polarized as well, while no effective field acts on them (Figure 4(b)). This is an illustration of the exciton effect: bound in excitons by Coulomb interaction and subject to the exchange induced exciton effects, the holes acquire in-plane spin polarization. The non-zero spin polarization of heavy holes is possible due to the ex- citon exchange effects. Figure 4(c,d) shows the resulting linear and circular exciton polarization patterns. One can see that the Rashba effect induces polarization patterns strongly different from those induced by the Dresselhaus effect, which is why from the shape of polarization pat- terns one can conclude on the nature of spin-orbit cou- pling in the system. IV. NON-LINEAR SPIN DYNAMICS OF PROPAGATING EXCITONS AND EXCITON-POLARITONS In the previous section we operated with a spin density matrix which is very convenient for the description of par- tially coherent and partially polarized exciton gases. The i dΨ−2 dt = − βeme Ψ−2 + mex (cid:16)kx − iky(cid:17) Ψ−1 2 ∇2 2mex βhmh mex (cid:16)kx + iky(cid:17) Ψ+1 + α1 Ψ−22 Ψ−2 + + α2 Ψ+22 Ψ−2 + α3 Ψ−12 Ψ−2 + α4 Ψ+12 Ψ−2 + W Ψ∗ +2Ψ+1Ψ−1. (56) Here kx,y = −i ∇x,y, mex = me + mhh. To make this system more compact we have omitted the terms de- scribing exchange induced exciton splittings given by the Hamiltonian (20). We do not discuss here the nature and value of the interaction constants α1,2,3,4 and W . In the system of indirect excitons in coupled GaAs/AlGaAs quantum wells, as a zeroth approximation, one can take α1 = α2 = α3 = α4. Note also that in microcavities, where the lower exciton-polariton mode is strongly de- coupled from dark excitons, the dark exciton states may be almost empty at low temperatures. If this is the case, the spin dynamics of the exciton-polariton condensate is given by the first two of the four GP equations (53 and 54) with α3,4 = W = 0. The remaining constants α1,2 have been widely discussed in literature45. The GP equations are widely used for the description of coherent propagation of exciton-polaritons in micro- cavities46. They allow for the studying of interesting topology effects such as: quantum vortices47 -- 52; half- quantum vortices53 -- 55; bright56 and dark51,57 -- 60 solitons. The polarization of light emitted by an exciton or exciton-polariton condensate can be obtained as: ρc = ρl = ρd = 2Sz I 2Sx I 2Sx I = = Ψ+12 − Ψ−12 Ψ+12 + Ψ−12 , 2Re(cid:8)Ψ∗ +1Ψ−1(cid:9) Ψ+12 + Ψ−12 , 2Im(cid:8)Ψ∗ +1Ψ−1(cid:9) Ψ+12 + Ψ−12 . = − (57) (58) (59) These expressions easily follow from the definition of the spin density matrix. A significant limitation of the GP equations as a theo- retical tool is that they assume a coherent state of the sys- tem. If one is interested in the spin structure of the zero temperature ground state of excitons in a Bose-Einstein condensate43, then this assumption is fulfilled by defini- tion. However, in real systems there is an incomplete co- herence that, strictly speaking, requires a description of statistical mixtures, perhaps involving density matrices. Furthermore, Eqs. (53-56) have been written assuming an infinite lifetime for the particles (be they excitons or exciton-polaritons), which is never the case of real non- equilibrium systems. Often pumping and radiative decay terms are introduced into Eqs. (53-56) phenomenologi- cally30,44. While in the case of a resonant coherent pump, one can imagine that the exciton/exciton-polariton dis- tribution inherits coherence directly, it is less obvious how an incoherent pump can be modelled. A phenomenologi- 11 cal model introduced by Wouters and Carusotto61 of in- coherent pumping has allowed the modelling of the first order coherent fraction observed in many experimental configurations based on condensation26,62 -- 65. While the GP model cannot model the phase transition during for- mation of a condensate and/or superfluid, it can offer a suitable description of spin currents once spatial coher- ence has formed. Here we will consider a localized source of the four- component indirect exciton system, as corresponds to the localized bright spot sources generating exciton conden- sates9. We focus our attention on the possible spin po- larization textures of coherent excitons propagating away from the source. We do not attempt to describe the par- tially coherent state within the source, noting that in exciton-polariton systems spin currents have been gen- erated from both coherent66 and incoherent11 tightly fo- cused spots utilizing the optical spin Hall effect12 in a similar way. Since the Gross-Pitaevskii equations are only valid for coherent excitons, we restrict the exciton wavefunction to lie outside of the source area. The ef- fect of the source is then characterized by the chosen boundary condition along the edge of the source area. Given that dark excitons have lower energy than bright excitons [due to the exchange splitting of Eq. (19)], it is reasonable to expect the source to provide linearly po- larized dark excitons. By fixing the values of the exciton wavefunction along the edges of the source area, which is assumed circular, to such a distribution the bound- ary condition acts as an effective source for the exciton wavefunction outside the source area. The indirect excitons are known to have very long life- time, typically in the range of 10ns to 10µs67. This allows them to cover distances of a few hundreds of µm with negligible loss8. Consequently, when we focus on the be- haviour of excitons in a small (10 × 10µm2) area around the source, the main loss of excitons is caused by their escape from the area of interest rather than their decay (recombination). To model the spin currents we thus em- ploy an absorbing boundary condition to allow the solu- tion of Eqs. (53-56) in a finite area. This allows a balance between source and loss to achieve a steady state of the non-equilibrium system (for a continuous pump), where both source and loss appear as boundary conditions.8 The steady state solution of the system is independent of the initial condition. Exciton intensity and polarization distributions calcu- lated within the GP approach are shown in Figs. 5 and 7. Note that all the plotted quantities are spatially av- eraged over 1.5µm to account for the typical resolution of experimental setups8. In principle excitons can dis- play features on the scale of the de Broglie wavelength, λ = 2π/kex. Such features are on the sub-micron scale and are far beyond experimental resolution. We note also that while in the DM approach we could consider exci- tons having a fixed radial velocity, in the GP approach we necessarily cover the whole range of wavevectors and propagation velocities (the dispersion obtained from the GP approach is shown in Fig. 6). Also, exciton-exciton interactions modify the exciton dispersion. For all these reasons, the DM and GP approaches cannot give iden- tical results, in principle. The DM approach is suitable for the description of both coherent and partly coherent exciton gases, while the GP approach better catches the dispersive propagation features and accounts for nonlin- ear effects. Both approaches are complementary, and it is instructive to compare the results, obtained within these two models. The exciton density decreases as excitons travel away from the source (Fig. 5a). This is not due to exciton recombination, which is expected to be very slow, but more simply due to the spreading out of excitons in all directions. The intensity spread need not be perfectly circularly symmetric due to the presence of the spin-orbit (Dresselhaus terms), which can introduce a directional dependence of the exciton velocity. Fig. 5b shows the exciton brightness degree, defined as: ρb = Ψ+12 + Ψ−12 − Ψ+22 − Ψ−22 Ψ+12 + Ψ−12 + Ψ+22 + Ψ−22 . (60) This quantity represents the degree to which the bright exciton density exceeds the dark exciton density. There is a conversion of dark to bright excitons as they spread out from the source, which can be expected from the presence of Dresselhaus coupling terms. FIG. 5: (color online) Spatial distribution of the total exciton density (a) and exciton brightness degree (b) [see Eq. (60)] in the absence of a magnetic field. The parameters were the same as those used for the DM calculations, given in Table I, with: B = 0; W = 0.2α; mex was taken as 0.21 of the free electron mass. The source area was taken circular with a ra- dius of 1µm. The images are presented with spatial averaging over 1.5µm. The scattering parameter α and intensity at the source center were chosen such that the interaction energy, α (cid:0)Ψ+12 + Ψ−12 + Ψ+22 + Ψ−22(cid:1) = 1µeV (being com- parable to the other energy scales in the system, we are in a nonlinear regime). The absorbing boundary condition used in calculations appears outside of the plotted range, at a radius of 15µm from the source center. The polarization distribution is shown in Fig. 7 and can be significantly influenced by non-linearity in the system. The left-hand plots show the results for negligible non- linearity (α = 0; W = 0), which is equivalent to a weak 10-3 0.01 0.1 1 12 40 20 0 V e Μ E - 20 - 20 - 10 0 10 20 kΜm -1 FIG. 6: (color online) Dispersion relation obtained from the Gross-Pitaevskii equations, corresponding to Figs. 5 and 7b,d and f. The dispersion is obtained by Fourier transform of the wavefunctions in space and time, from which the grayscale map of intensity is obtained. The curves show the bare dis- persion, obtained from diagonalization of Htot ex + 2 k2 2m . pump intensity. Here the polarization distributions are qualitatively similar to those calculated in the density matrix formalism. In analogy to the (intrinsic) spin Hall effect2,3 and the optical spin Hall effect10,12 the presence of the spin-orbit coupling terms introduces a directional dependence of the polarization. The patterns of the po- larization degrees divide into quadrants. Some quantita- tive differences with the DM calculations appear due to the presence of different wavevectors. The right-hand plots show the case of a moderate non- linearity, with interaction strength comparable to the other energy scales of the system. The most drastic effect is on the circular polarization degree, which be- comes higher and each quadrant of circular polarization divides further giving an eight-lobed pattern to the po- larization degree. The interaction terms that we have introduced are all spin conserving and it can be noted that we have considered the spin isotropic case. Even the W nonlinear interaction term, which allows the inter- conversion of bright and dark exciton pairs does not ap- pear to directly change the spin polarization, conserving both circular and linear polarizations upon scattering. Still, the nonlinear interaction terms can have a drastic effect on the polarization structure. This is because they are able to shift (renormalize) the dispersion branches in the system. Given that the potential energy of ex- citons is fixed by their interaction energy at the source and that this energy is converted into kinetic energy at distances away from the source, any shifts in the disper- sion branches can change the wavevector of propagat- ing excitons. Even if the nonlinear induced shifts of the dispersion branches were not polarization dependent, a change in the wavevector of an exciton can allow it to experience a different effect from the k-dependent spin- orbit coupling terms. In this way, richer structures can appear in the nonlinear regime. Note, that the build-up of circular polarisation clearly seen in Fig. 7 (e,f) would not yield 100% circularly polarised excitons, as it might be expected in the ideal case of the optical spin Hall effect for exciton-polaritons12. In our system, the preces- sion of electron and hole spins has different frequencies, and the interplay between dark-to-bright and linear-to- circular polarisation conversion prevents formation of a purely circularly polarised state. The current of each of these densities is given by a prod- uct of the exciton speed and the corresponding density: ja = kex mex ρjj (61) 13 µB 2 with j = 1, 2, 3, 4 for a = +1,−1, +2,−2, respectively. Experimentally, one can measure the magnetization cur- rent associated with the exciton density current. The magnetization carried by propagating excitons can be found as: Mz = − [(gh − ge) (ρ11 − ρ22) + (gh + ge) (ρ33 − ρ44)] (62) This expression is obtained having in mind that an elec- tron with a spin projection on the z-axis of ±1/2 con- tributes to the magnetization projection on the z-axis ∓ µB 2 ge, and a heavy hole with the spin projection of ±3/2 contributes to the magnetization ∓ µB 2 gh. Hence, the magnetization (spin) current produced by the exci- tons having a wave-vector kex will be given by: jM (kex) = − [(gh − ge) (ρ11(kex) − ρ22(kex)) µBkex 2mex + (gh + ge) (ρ33(kex) − ρ44(kex))] (63) The total magnetization current in the exciton gas can be obtained by integration over all wave-vectors: jtot M = − A (2π)2 Z jM (kex)dkex (64) Here A is the area of the sample. This current may be detected, for example, by spatially resolved Kerr rotation spectroscopy. FIG. 7: (color online) Spatial distribution of the polarization state of excitons in the absence of a magnetic field: horizontal- linear polarization degree (a, b), diagonal polarization degree (c, d) and circular polarization degree (e, f) for bright ex- citons, which corresponds to the near field emission pattern of light. The left-hand plots show results in the absence of nonlinear interactions (α = 0; W = 0), while the right-hand plots show the case of a moderate nonlinearity. The parame- ters were the same as in Fig. 5 V. EXCITON SPIN CURRENTS Consider an exciton state characterized by a wavevec- tor kex and described by the DM ρ. Let us recall that the elements of this matrix ρ11, ρ22, ρ33, ρ44 are the densities of +1,-1,+2 and -2 spin polarized excitons, respectively. FIG. 8: (color online) Spatial distribution of the exciton spin current calculated in arbitrary units using the parameters summarized in Table I at kex = 15.3µm−1 at B = 0 (a) and B = 5T (b)-(d) along the z, x and y-axis, respectively. The source area was taken circular with a radius of 4µm. Figure 8 shows the spin current density jM / (2πr) cal- culated for the system of indirect excitons which we con- sidered above in the absence of external magnetic field (a) and in the presence of a magnetic field of 5T oriented normally to the plane of the structure (b) and along x- and y-axes (c,d). In all cases the current intensity de- creases as one moves away from the excitation spot, as the exciton density decreases inversely with the radius r. One can see that the total spin (magnetization) of propagating excitons experiences oscillations and has a strong angular dependence. The spin currents are sup- pressed by an in-plane magnetic field, which is not sur- prising: jM describes propagation of the normal-to plane spin component, which is strongly reduced by an in-plane magnetic field. The images show the total spin carried by both bright and dark excitons. They do not directly correspond to the polarized photoluminescence map for two reasons: first, dark excitons do not contribute to the photoluminescence; second, the polarization degree of photoluminescence does not experience the 1/r decay characteristic of the total spin density. On the other hand, the images presented in Figure 8 do correspond to the signal of spatially resolved Kerr or Faraday rotation, which is sensitive to the normal-to-plane magnetization. VI. SPIN CURRENTS IN EXCITON CONDENSATES The approach formulated above can be extended to the description of spin currents in coherent exciton (or exciton-polariton) condensates accounting for particle- particle interactions. In this case we need to replace the momentum kex by a momentum operator p = −i ∇ and the diagonal components of the DM ρ11, ρ22, ρ33, ρ44 by the exciton densities ψ+12,ψ−12,ψ+22,ψ−22, respectively, in the expressions 61 and 64. In this case the density currents become: jα = −i mex Ψ∗ α∇Ψα, (65) and the total magnetization current can be expressed as: iµB jtot M = +1∇Ψ+1 − Ψ∗ 2mex (cid:2)(gh − ge)(cid:0)Ψ∗ + (gh + ge)(cid:0)Ψ∗ [(gh − ge) (j+1 − j−1) + (gh + ge) (j+2 − j−2)] µB 2 = − +2∇Ψ+2 − Ψ∗ −1∇Ψ−1(cid:1) −2∇Ψ−2(cid:1)(cid:3)(66) (67) 14 FIG. 9: (color online) Spatial structure of the spin density current j+1 (a) and the total magnetization current jtot M (b). The arrows show the directional dependence of the vector fields in space, while the colour code illustrates the intensity. The parameters were the same as in Fig. 5. Arbitrary units are used for both the spin density and total magnetization current. One can also introduce the spin conductivity tensor linking the components of the density current (65) with the gradient of potential acting upon each of the exciton spin components: jα,l = σl,m α,β∇Uβ,m, (68) where l = x, y and m = x, y indicate the in-plane projec- tions of the current and potential gradient, respectively. One can see that σl,m α,β is a 64-component tensor in the general 2D case. The origin of the potential gradient ∇Uβ,m needs to be discussed separately. ∇Uβ,m can orig- inate from the gradient of the quantum well width, gra- dient of the barrier height, or it can be induced by exci- tons themselves due to e.g. dipole-dipole repulsion. Indi- rect excitons have built-in dipole moments, the laterally modulated external electric field in the z-direction can create an in-plane potential landscape and, in turn,∇U for them. This was used in studies of transport of in- direct excitons in various electrostatic potential land- scapes including potential energy gradients28,68,69, cir- cuit devices70 -- 72, traps73 lattices74,75, moving lattices- conveyers76, and narrow channels72,77,78. The distribution of the spin density current, j+1, is shown in Fig. 9 (a). The current density propagates out- ward from the source in all directions, decreasing in in- tensity. The apparent rotation of the current density is a nonlinear effect coming from the interactions in the sys- tem. The other spin density currents, j−1, j+2 and j+2, display a similar behaviour. The magnetization current is shown in Fig. 9 (b). The current is stronger closer to the source, where the intensi- ties are stronger. The magnetization current is predicted to rotate around the source. VII. POLARIZATION CURRENTS Spatially resolved measurements of the polarization de- grees ρc, ρl and ρd of light emitted by excitons give access to the exciton polarization currents. In terms of the DM formalism, they can be defined as products of the exciton 15 speed and the corresponding polarization degree: jc (kex) = jl (kex) = jd (kex) = kex mex kex mex kex mex ρc = ρl = ρd = kex mex kex mex kex mex ρ11 − ρ22 ρ11 + ρ22 ρ12 + ρ21 ρ11 + ρ22 ρ12 − ρ21 ρ11 + ρ22 (69) (70) (71) The total polarization currents can be obtained inte- grating the expressions (69-71) over reciprocal space: jc,l,d tot = − A (2π)2 Z dkexjc,l,d (kex) (72) The polarization currents in an exciton condensate can be found from the GP equations (53-56) as: FIG. 10: (color online) Spatial structure of the polarization currents jc (a) and jl (b). The arrows show the directional dependence of the vector fields in space, while the colour code illustrates the intensity. The parameters were the same as in Fig. 5. Arbitrary units are used for the polarization currents. jc = − jl = − jd = − i mex(cid:0)Ψ∗ mex(cid:0)Ψ∗ mex(cid:0)Ψ∗ +1∇Ψ+1 − Ψ∗ Ψ+12 + Ψ−12 +1∇Ψ−1 + Ψ∗ Ψ+12 + Ψ−12 +1∇Ψ−1 − Ψ∗ Ψ+12 + Ψ−12 −1∇Ψ−1(cid:1) −1∇Ψ+1(cid:1) −1∇Ψ+1(cid:1) i (73) (74) (75) The distributions of the polarization currents calcu- lated within GP approach are shown in Fig. 10. A strik- ing nonuniform structure appears due to the presence of the spin-orbit coupling terms. Close to the source spot, there is a strong circularly polarized current that rotates around the source. This can be attributed to the rotating circular polarization degree already observed in Fig. 7. Away from the source, the circular polarization current decays, which can be expected due to the decay of the spin density current observed in Fig. 9. Along the ver- tical axis (x = 0), a strong circular polarization current remains due to the particularly fast change of the circular polarization degree in this region. The linearly polarized current can be stronger further away from the source than at closer distances. This is attributed to an increasing linear polarization degree further from the source. In ad- dition, one can recall that while spin density currents are generally weaker further from the source, there is some compensation due to the conversion between dark and bright excitons (as shown in Fig. 5b, the bright exciton fraction increases further from the source). VIII. CONCLUSIONS Bosonic spin transport is a young and promising area of solid-state physics. The theories of mesoscopic transport of charge carriers and quantum transport are among the most interesting chapters of modern physics. Substitu- tion of fermions by bosons and of a scalar electric charge by a spin vector cannot be formally done in these the- ories. Basically, all mesoscopic and quantum transport effects need to be reconsidered if we speak about electri- cally neutral bosonic spin carriers like excitons or exciton- polaritons. This is why the area of "spin-optronics" essentially remains terra incognita. Experimentally, di- rect measurements of transport of indirect excitons and exciton-polaritons in time-resolved imaging experiments have become possible in recent years. In this work, we have demonstrated that exciton polarization currents are inseparably connected with electron and hole spin cur- rents. The intensity and direction of exciton polarization currents and electron and hole spin currents is governed by an interplay of spin-orbit effects, Zeeman effects and exciton exchange effects. In the non-linear regime, the pattern of spin currents may also be affected by spin- dependent exciton-exciton interactions. We have developed two complementary approaches to the description of exciton spin currents and textures. The DM formalism allows for description of the spin transport effects in both classical exciton gases and condensates of non-interacting excitons, while the GP equations de- scribe propagation of exciton condensates. We predict non-trivial topologies of interacting exciton spin in con- densates, and suggest tools of their control, such as ex- ternal magnetic and electric fields, and source intensity. We have demonstrated, that ballistic propagation of ex- citons may result in a build up of polarization patterns, which may be observed in near-field photoluminescence spectra. This work has been supported by the EU FP7 ITN INDEX, EU FP7 PodiTrodi, the E.P.S.R.C., the EU FP7 Marie Curie EPOQUES, ANR-2011-NANO-004-06, and DOE. J. R. L. acknowledges Chateaubriand Fellow- ship from the Embassy of France in the United States. The authors are deeply grateful to T. Ostatnicky, Y.G. Rubo, M.M. Glazov, I.A. Shelykh, and A. Bramati for many useful discussions on the peculiarities of bosonic spin transport. 16 1 M. I. Dyakonov and V. I. Perel, Phys. Lett. A, 35, 459 (1971). 2 J. Sinova, D. Culcer, Q. Niu, N. A. Sinitsyn, T. Jungwirth, and A. H. MacDonald, Phys. Rev. Lett., 92, 126603 (2004). 3 E. G. Mishchenko, A. V. Shytov and B. I. Halperin, Phys. Rev. Lett., 93, 226602 (2004). 4 Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Aw- shalom, Science, 306, 1910 (2004). 5 J. Wunderlich, B. Kaestner, J. Sinova, and T. Jungwirth, Phys. Rev. Lett., 94, 047204 (2005). 6 S. O. Valenzuela and M. Tinkham, Nature 442, 176 (2006). 7 J. R. Leonard, Y. Y. Kuznetsova, Sen Yang, L. V. Butov, T. Ostatnicky, A. Kavokin, and A. C. Gossard, Nano Lett. 9, 4294 (2009). 8 A.A. High, A.T. Hammack, J.R. Leonard, Sen Yang, L.V. Butov, T. Ostatnicky, M. Vladimirova, A.V. Kavokin, T.C.H. Liew, K.L. Campman, A.C. Gossard, Phys. Rev. Lett., 110, 246403, (2013). 9 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, 483, 584, Nature, (2012). 10 C. Leyder, T. C. H. Liew, A. V. Kavokin, I. A. Shelykh, M. Romanelli, J. Ph. Karr, E. Giacobino and A. Bramati, Nature Phys., 3, 628 (2007). 11 E. Kammann, T. C. H. Liew, H. Ohadi, P. Cilibrizzi, P. Tsotsis, Z. Hatzopoulos, P. G. Savvidis, A. V. Kavokin, and P. G. Lagoudakis, Phys. Rev. Lett., 109, 036404 (2012). 12 A. Kavokin, G. Malpuech, and M. Glazov, Phys. Rev. Lett., 95, 136601 (2005). 13 J. Keeling, Phys. Rev. B, 78, 205316 (2008). 14 J. Keeling, M. J. Bhaseen, and B. D. Simons, 105, 043001 (2010). 15 I. A. Shelykh, G. Pavlovic, D. D. Solnyshkov, and G. Malpuech, Phys. Rev. Lett., 102, 046407 (2009). 16 I. A. Shelykh, A. V. Kavokin, Y. G. Rubo, T. C. H. Liew, and G. Malpuech, Semicond. Sci. Technol., 25, 013001 (2010). 17 T. C. H. Liew, I. A. Shelykh, and G. Malpuech, Physica E, 43, 1543 (2011). 18 A. Amo, T. C. H. Liew, C. Adrados, R. Houdr´e, E. Gia- cobino, A. V. Kavokin, and A. Bramati, Nature Photon., 4, 361 (2010). 19 D. Sanvitto, S. Pigeon, A. Amo, D. Ballarini, M. De Giorgi, I. Carusotto, R. Hivet, F. Pisanello, V. G. Sala, P. S. S. Guimaraes, R. Houdr´e, E. Giacobino, C. Ciuti, A. Bra- mati, and G. Gigli, Nature Photon, 5, 610 (2011). 20 R. Hivet, H. Flayac, D. D. Solnyshkov, D. Tanese, T. Boulier, D. Andreoli, E. Giacobino, J. Bloch, A. Bramati, G. Malpuech, and A. Amo, Nature Phys., 8, 724 (2012). 21 J. Keeling and N. G. Berloff, Contempory Physics, 52, 131 (2011). 22 I. Carusotto and C. Ciuti, Rev. Mod. Phys. 85, 299 (2013). 23 M. H. Szyma´nska, F. M. Marchetti, and D. Sanvitto, Phys. Rev. Lett., 105, 236402 (2010). 24 A. Amo, D. Sanvitto, F. P. Laussy, D. Ballarini, E. del Valle, M. D. Martin, A. Lemaıtre, J. Bloch, D. N. Krizhanovskii, M. S. Skolnick, C Tejedor, L Vina, Nature, 457, 291 (2009). 25 C. Adrados, T. C. H. Liew, A. Amo, M. D. Martin, D. Sanvitto, C. Ant´on, E. Giacobino, A. Kavokin, A. Bramati, and L. Vina, Phys. Rev. Lett., 107, 146402 (2011). 26 E. Wertz, A. Amo, D. D. Solnyshkov, L. Ferrier, T. C. H. Liew, D. Sanvitto, P. Senellart, I. Sagnes, A. Lemaıtre, A. V. Kavokin, G. Malpuech, and J. Bloch, Phys. Rev. Lett., 109, 216404 (2012). 27 Y. Y. Kuznetsova, M. Remeika, A. A. High, A. T. Ham- mack, L. V. Butov, M. Hanson, and A. C. Gossard, Optics Letters, 35, 1587 (2010). 28 J.R. Leonard, M. Remeika, M.K. Chu, Y.Y. Kuznetsova, A.A. High, L.V. Butov, J. Wilkes, M. Hanson and A.C. Gossard, Appl. Phys. Lett. 100, 231106 (2012). 29 I. Carusotto and C. Ciuti, Phys. Rev. Lett., 93, 166401 (2004). 30 I. A. Shelykh, Y. G. Rubo, G. Malpuech, D. D. Solnyshkov, and A. Kavokin, Phys. Rev. Lett., 97, 066402 (2006). 31 See, e.g., E. L. Ivchenko and G. E. Pikus, "Superlattices and other heterostructures", Symmetry and optical phe- nomena, Springer, Berlin, (1997). 32 E. I. Rashba and E. Ya. Sherman, Physics Letters A, 129, 175 (1988). 33 E. L. Ivchenko, "Optical spectroscopy of semiconductor nanostructures", Alpha Science (2005). 34 L.P. Gor'kov, I.E. Dzyaloshinskii, Sov. Phys. JETP 26, 449 (1968). 35 I.V. Lerner, Yu.E. Lozovik, Sov. Phys. JETP 51, 588 (1980). 36 Yu.E. Lozovik, I.V. Ovchinnikov, S.Yu. Volkov, L.V. Bu- tov, D.S. Chemla, Phys. Rev. B 65, 235304, (2002). 37 J.-W. Luo, A. N. Chantis, M. van Schilfgaarde, G. Bester, A. Zunger, Phys. Rev. Lett., 104, 066405 (2010). 38 See, e.g., A. V. Kavokin, J. J. Baumberg, G. Malpuech, and F.P. Laussy, "Microcavities", Oxford University Press, Oxford (2007). 39 See, e.g., M. Born and E. Wolf, "Principles of optics", Perg- amon, London, (1970). 40 D. V. Vishnevsky, H. Flayac, A. V. Nalitov, D. D. Sol- nyshkov, N. A. Gippius, and G. Malpuech, Phys. Rev. Lett. 110, 246404 (2013). 41 E. McKann and M. Koshino, Reports on Progress in Physics, 056503 (2013). 42 Y. G. Rubo and A. V. Kavokin, Phys. Rev. B, 84, 045309 (2011). 43 M. Matuszewski, T. C. H. Liew, Y. G. Rubo, and A. V. Kavokin, Phys. Rev. B, 86, 115321 (2012) 44 O. Kyriienko, E. B. Magnusson, and I. A. Shelykh, Phys. Rev. B, 86, 115324 (2012). 45 M. Vladimirova, S. Cronenberger, D. Scalbert, K. V. Ka- vokin, A. Miard, A. Lemaıtre, J. Bloch, D. Solnyshkov, G. Malpuech, and A. V. Kavokin, Phys. Rev. B, 82, 075301 (2010). 46 See, e.g., M. Wouters and I. Carusotto, Phys. Rev. Lett., 105, 020602 (2010). 47 D. M. Whittaker, Supperlattice. Microstruct., 41, 297 (2007). 48 T. C. H. Liew, A. V. Kavokin, and I. A. Shelykh, Phys. Rev. B, 75, 241301 (2007). 49 T. C. H. Liew, Yu. G. Rubo, and A. V. Kavokin, Phys. Rev. Lett., 101, 187401 (2008). 50 F. M. Marchetti, M. H. Szymanska, C. Tejedor, and D. M. Whittaker, Phys. Rev. Lett., 105, 063902 (2010). 51 S. Pigeon, I. Carusotto, and C. Ciuti, Phys. Rev. B, 83, 17 144513 (2011). 52 K. G. Lagoudakis, F. Manni, B. Pietka, M. Wouters, T. C. H. Liew, V. Savona, A. V. Kavokin, R. Andr´e, and B. Deveaud-Pl´edran, Phys. Rev. Lett., 106, 115301 (2011). 53 Yu. G. Rubo, Phys. Rev. Lett., 99, 106401 (2007). 54 H. Flayac, I. A. Shelykh, D. D. Solnyshkov, and G. Malpuech, Phys. Rev. B, 81, 045318 (2010). 55 F. Manni, K. G. Lagoudakis, T. C. H. Liew, R. Andr´e, V. Savona, and B. Deveaud-Pl´edran, Nature Communica- tions, 3, 1309 (2012). 56 O. A. Egorov, D. V. Skryabin, and F. Lederer, Phys. Rev. B, 82, 165326 (2010). lykh, Phys. Rev. B, 85, 075312 (2012). 66 A. Amo, T. C. H. Liew, C. Adrados, E. Giacobino, A. V. Kavokin, and A. Bramati, Phys. Rev. B, 80, 165325 (2009). 67 A.G. Winbow, A.T. Hammack, L.V. Butov, A.C. Gossard, Nano Lett. 7, 1349 (2007). 68 M. Hagn, A. Zrenner, G. Bohm, and G. Weimann, Appl. Phys. Lett. 67, 232 (1995). 69 A. Gartner, A. W. Holleitner, J. P. Kotthaus, and D. Schuh, Appl. Phys. Lett. 89, 052108 (2006). 70 A. A. High, A. T. Hammack, L. V. Butov, M. Hanson, and A. C. Gossard, Opt. Lett. 32, 2466 (2007). 57 A. V. Yulin, O. A. Egorov, F. Lederer, and D. V. Skryabin, 71 A. A. High, E. E. Novitskaya, L. V. Butov, M. Hanson, Phys. Rev. A, 78, 061801 (2008). 58 H. Flayac, D. D. Solnyshkov, G. Malpuech, Phys. Rev. B, 84, 125314 (2011). 59 E. A. Ostrovskaya, J. Abdullaev, A. S. Desyatnikov, M. D. Fraser, and Y. S. Kivshar, Phys. Rev. A, 86, 013636 (2012). 60 A. Amo, S. Pigeon, D. Sanvitto, V. G. Sala, R. Hivet, I. Carusotto, F. Pisanello, G. Lem´enager, R. Houdr´e, E. Giacobino, C. Ciuti, and A. Bramati, Science, 332, 1167 (2011). 61 M. Wouters and I. Carusotto, Phys. Rev. Lett., 99, 140402 (2007). 62 G. Tosi, G. Christmann, N. G. Berloff, P. Tsotsis, T. Gao, Z. Hatzopoulos, P. G. Savvidis, and J. J. Baumberg, Na- ture Phys., 8, 190 (2012). 63 G. Christmann, G. Tosi, N. G. Berloff, P. Tsotsis, P. S. Eldridge, Z. Hatzopoulos, P. G. Savvidis, and J. J. Baum- berg, Phys. Rev. B, 85, 235303 (2012). 64 V. Kohnle, Y. L´eger, M. Wouters, M. Richard, M. T. Portella-Oberli, and B. Deveaud-Pl´edran, Phys. Rev. Lett., 106, 255302 (2011). 65 H. Flayac, G. Pavlovic, M. A. Kaliteevski, and I. A. She- and A. C. Gossard, Science 321, 229 (2008). 72 G Grosso, J Graves, A T Hammack, A A High, L V Bu- tov, M Hanson, and A C Gossard, Nature Photon., 3, 577 (2009). 73 A. A. High, A. K. Thomas, G. Grosso, M. Remeika, A. T. Hammack, A.D. Meyertholen, M. M. Fogler, L. V. Bu- tov, M. Hanson, and A. C. Gossard, Phys. Rev. Lett. 103, 087403 (2009). 74 M. Remeika, J. C. Graves, A. T. Hammack, A. D. Meyert- holen, M. M. Fogler, L. V. Butov, M. Hanson, and A.C. Gossard, Phys. Rev. Lett. 102, 186803 (2009). 75 M. Remeika, M. M. Fogler, L. V. Butov, M. Hanson, and A. C. Gossard, Appl. Phys. Lett. 100, 061103 (2012). 76 A. G. Winbow, J. R. Leonard, M. Remeika, Y. Y. Kuznetsova, A. A. High, A. T. Hammack, L. V. Butov, J. Wilkes, A. A. Guenther, A. L. Ivanov, M. Hanson, and A. C. Gossard, Phys. Rev. Lett. 106, 196806 (2011). 77 X. P. Vogele, D. Schuh, W. Wegscheider, J. P. Kotthaus, and A. W. Holleitner, Phys. Rev. Lett. 103, 126402 (2009). 78 K. Cohen, R. Rapaport, and P. V. Santos, Phys. Rev. Lett. 106, 126402 (2011).
1709.06936	1	1709	2017-07-19T19:32:54	Energy Transfer Controlled by Dynamical Stark Shift in Two-level Dissipative Systems	[ "cond-mat.mes-hall", "physics.optics", "quant-ph" ]	A strong electromagnetic field interacting with an electron system generates both the Rabi oscillations and the Stark splitting of the electron density. Changing of the electron density gives rise to nonadiabatic effects due to existence of the electron-vibrational interaction in a dissipative system. In this Letter, the mechanism of energy transfer between the electron system and the phonon reservoir is presented. This mechanism is based on establishment of the coupling between the electron states dressed by the electromagnetic field and the forced vibrations of reservoir oscillators under the action of rapid changing of the electron density with the Rabi frequency. The photoinduced vibronic coupling results in appearance of the states that are doubly dressed by interaction, first time due to the electron-photon interaction, and second time due to the electron-vibrational interaction. Moreover, this coupling opens the way to control energy which can be transferred to (heating) or removed from (cooling) the phonon reservoir depending on the parameters of the electromagnetic pulse.	cond-mat.mes-hall	cond-mat	Energy Transfer Controlled by Dynamical Stark Shift in Two-level Dissipative Systems Research Center "Information Optical Technologies", ITMO University, Birzhevaya liniya, 14, Saint Petersburg, 199034, Russia A. V. Ivanov,* A strong electromagnetic field interacting with an electron system generates both the Rabi oscillations and the Stark splitting of the electron density. Changing of the electron density gives rise to nonadiabatic effects due to existence of the electron-vibrational interaction in a dissipative system. In this Letter, the mechanism of energy transfer between the electron system and the phonon reservoir is presented. This mechanism is based on establishment of the coupling between the electron states dressed by the electromagnetic field and the forced vibrations of reservoir oscillators under the action of rapid changing of the electron density with the Rabi frequency. The photoinduced vibronic coupling results in appearance of the states that are doubly dressed by interaction, first time due to the electron-photon interaction, and second time due to the electron-vibrational interaction. Moreover, this coupling opens the way to control energy which can be transferred to (heating) or removed from (cooling) the phonon reservoir depending on the parameters of the electromagnetic pulse. DOI: worth to stress that the considered physical situation can be interpreted as a manifestation of the pseudo-Jahn-Teller effect [14] in the electron-photon system. In this Letter, the effect of the displacement of the equilibrium position (EDEP) of vibrational coordinates under the action of a strong electromagnetic field is taking into account for describing the energy transfer between the two-level system and the phonon reservoir. Keywords: Two-level system, dynamical Stark shift, dressed states, nonadiabatic effect, vibronic coupling PACS: 03.65.Yz, 42.50.Hz, 63.20.kd. The evident example of a two-level dissipative system (TLDS) is a semiconductor quantum dot (QD) which is widely used in various applications of the quantum optics, nanoelectronics, and quantum information. Physical features of QD based setups are ultimately depend on the interaction between the electron system and the phonon reservoir. This interaction is responsible for changing the quantum states of the phonon subsystem and, thus, the temperature of the QD. From a practical point of view, it is important to control the energy transfer between the electron and phonon subsystems of the QD. For instance, the processes of the coherent phonon generation [1] and cooling [2] can be implemented using the optical control. One of the convenient approaches to organize such control is the dynamical Stark shift method. This method has been used to realize the fast quantum gates for trapped cold ions [3], the atomic cooling [4], the precise qubit thermometry of an oscillator [5], the population inversion of a single QD [6], and the coherent quantum interference between the Stark split dressed states (DSs) in a semiconductor quantum well [7]. Consider a two-level electron subsystem interacting with a strong monochromatic electromagnetic field and a phonon reservoir. Let us divide the total Hamiltonian into two parts. First part, H1, contains the electron and photon subsystems the dipole approximation and the rotating wave approximation (RWA) while second part, H2, includes the energy of the phonon 0, and the energy of the electron-phonon subsystem, Hpn interaction, Ve-pn. Thus, the Hamiltonian of the system can be written as H H H interaction their g c * (cid:33) 0 and (cid:14) (cid:86) (cid:16) (cid:86) 0 (cid:14) H 1 c (cid:14) (cid:14) g (cid:33) in (cid:14) 1 , 2 , (cid:32) c c (cid:14) (cid:33) (cid:14) (cid:72) (cid:86) (cid:90) 0 2 z H V (cid:32) e-pn 0 pn (cid:14) (cid:32) H 2 . (1) Interaction of a dissipative system with a strong electromagnetic field leads to changing both the energy spectrum of electron states [8, 9] and the energy spectrum of vibrational modes of the reservoir [10]. Thus, an additional mode appears in the vibrational spectrum due to the periodic displacement of equilibrium positions of reservoir oscillators under the Rabi oscillations of the electron density [11, 12]. Additionally, the optically coupled electron levels become split in the conditions of the ac Stark effect. This fact is known as removing the degeneracy in the electron-photon system. Since the energy separation between the Stark split levels is comparable with the energy of vibrational modes, the nonadiabatic effects must be taken into account [13]. It is the action of Here, (cid:304) is the energy difference between the ground and excited electron levels; (cid:305)z, (cid:305)+, and (cid:305)- are the Pauli operators associated to the TLDS with the ground 1(cid:77) and the 2(cid:77) electron states; c and c+ are the annihilation excited and creation operators of the electromagnetic field of frequency (cid:550)0; g0 is the coupling strength of the electron- photon interaction. The solution of the Schrödinger equation for H1 is well known and can be written in the basis of electron states that are dressed by an electromagnetic field. If initially only the ground electron state is populated, the solution has the following form [15] (cid:60) (cid:114) mU (cid:32) ( ), , t m n (cid:587) U (cid:14) m m 1 1 (cid:167) (cid:114)(cid:168) 2 (cid:169) (cid:32) (cid:71) (cid:183) (cid:184)(cid:58) 2 (cid:185) S , m (cid:91) 1 (cid:32) 0 , (cid:14) ( ), t m n (cid:587) S U (cid:16) m m m g g 0 (cid:16) 2 2 (cid:71)(cid:58) (cid:32) (2) (cid:14) , . 4 (cid:71) 1 (cid:72) (cid:90)(cid:16)(cid:32) 0 (cid:33) (cid:91) 2 ( ), , t m n , vibrational interaction can be represented in the following form [16] 2 g 0 ( m (cid:14) 1) , F 0 (cid:32) (cid:166) k , s i t (cid:16) (cid:58) 2 (cid:33) H (cid:99) (cid:32) ED (cid:16) (cid:14) F e 0 ( f U U C U 2 (cid:170) (cid:33) (cid:172) k s m m m , ) (cid:14) m i * F e (cid:14) (cid:33) 0 k S g , 2 s 12 m 2 t (cid:58) (cid:16) , ( U 2 ) (cid:16) m S g * m k , 21 s (cid:16) (cid:14) 2 m mU U g (cid:16) s k , 22 (cid:186) (cid:188) , f k , s (cid:507) l , (cid:68) (cid:32) R e , l (cid:68) (cid:16) (cid:32) (cid:166) , (cid:68) l R . g , l (cid:68) k R (cid:152) g , l (cid:68) i e , (cid:507) e (cid:68) (cid:152) k , , s l (cid:68) 2 Nm (cid:68) (cid:90) (cid:33) k , s (6) Here, s denotes the phonon mode of frequency (cid:90)k,s and k,s is the coupling strength of the electron- wave vector k; gij phonon interaction; ek,s (cid:68) are the normalized polarization vectors of vibrational coordinates for an (cid:68)-type oscillator of mass m(cid:68); N is the number of reservoir oscillators; (cid:507) is the displacement operator; Rg and Re are the equilibrium positions of oscillations of the normal coordinates related to the ground and excited BSs, respectively. The obtained matrix element is responsible for establishment of the photoinduced vibronic (PIV) coupling in the TLDS. The matrix element describes the vibrations of frequency 2(cid:525) under the action of the Rabi oscillations of the electron density between the BSs. These vibrations have the amplitude which is proportional to the module of the displacement, (cid:507), and the coupling constants of the electron- vibrational and electron-photon (factor (cid:16) ) interactions. Since the frequency of the vibrations is equal to the frequency of the electron transition between the DSs, the electron-vibrational interaction leads to the electron density oscillations between these states and defines the back action of the vibrations on the electron density. Thus, the interaction of Eq. (6) gives rise to appearance of the vibronic coupling [17] in the electron-photon system. To determine the PIV states which can be represented as a linear combination of the DSs, we have to solve the following differential system for coefficients , m mU U(cid:14) i (cid:16) (cid:58) (cid:32) (cid:16) ( ) t ( ) t d 2 t iF e 0 d 1, n 2, n d 2, n (cid:32) (cid:16) iF e * 0 i 2 t (cid:58) d 1, n ( ) t . (7) The solution of the system gives the PIV wave functions for different initial conditions. Moreover, using this solution, we can write the PIV states in three different bases (i) in the BS basis, (ii) in the basis of the states which are dressed by the electron-photon interaction, and (iii) in the basis of the doubly dressed states (DDSs), i.e. the states which are dressed firstly by the electron-photon interaction and, then, by the electron-vibrational interaction (Fig. 1). In the last case, the PIV wave function takes the following form , ( ), , t m n ( ), , t m n ( ), , t m n (cid:81) (cid:32) (cid:14) 1 2 (cid:93) 1 1 2 (cid:93) 2 Here, m and n are the numbers of photons and phonons, respectively; Cm is the initial probability amplitude for the state with m photons; (cid:525) is the off-resonance Rabi frequency; (cid:303) is the detuning of the optical field with respect to the electron transition frequency. It should be stressed that the wave function in Eq. (2) is defined for the instant start of the electron-photon interaction in the so-called diabatic case the zeroth- approximation Hamiltonian of H2, the matrix element of the electron-vibrational interaction can be defined as Including H1 [15]. into V (cid:5) e-pn , ( ), H t m n V (cid:99) (cid:32) (cid:60) (cid:60)(cid:5) e-pn (cid:12) (cid:11) exp i H t V 1 (cid:16) (cid:32) (cid:16) (cid:33) H H H (cid:32) ( ), , t m n (cid:11) exp 1 (cid:16) . (3) , i H t (cid:33) 0 0 (cid:14) e-pn 0 pn 0 1 (cid:12) , The energy difference between the DSs is 2(cid:427)(cid:525). For the dynamical Stark shift experiments, this difference is comparable with the vibrational energy of reservoir modes. Such situation gives rise to existence of the DSs which are mixed by the electron-vibrational interaction. In order to take into account the EDEP of vibrational coordinates, let us consider the non-diagonal matrix elements in the DS basis and transfer from this basis to the basis of the bare states (BSs), using the following relationships (cid:91) 1 , (cid:32) ( ), t m n U e (cid:14) m (cid:71) 2 (cid:167) (cid:58)(cid:14)(cid:168) (cid:169) S U e (cid:16) m m (cid:16) i t (cid:183) (cid:184) (cid:185) i (cid:167) (cid:58)(cid:16)(cid:168) (cid:169) (cid:71) 2 (cid:183) (cid:184) (cid:185) t (cid:77) 1 ( ), t m (cid:14) 1, n , 2 ( ), (cid:77) t , t m n (cid:183) (cid:184) (cid:185) (cid:71) 2 (cid:167) (cid:169) i (cid:16) (cid:58)(cid:14)(cid:168) (cid:77) 1 ( ), t m (cid:14) 1, n (cid:91) 2 ( ), , t m n (cid:32) S U e * (cid:16) m m (cid:167) (cid:169) (cid:71) 2 (cid:183) (cid:184) (cid:185) t i (cid:16) (cid:58)(cid:16)(cid:168) (cid:14) mU e (cid:14) , 2 ( ), t m n (cid:77) . (4) (cid:14) (cid:99) 12 H Thus, the non-diagonal part of the H´ in the DS basis can be written in the form (for details see Ref. [16]) (cid:12) (cid:11) (cid:11) (cid:12) 2 H H V (cid:587) U U V V (cid:5) (cid:5) (cid:5) (cid:99) (cid:99) (cid:16) (cid:16) (cid:39) (cid:32) (cid:32) 11 22 22 nd 21 m m m (cid:11) ) ( ) ( e U V C S U U U 2 2 2 t (cid:5) (cid:186) (cid:170) (cid:16) (cid:14) (cid:16) (cid:16) (cid:16) (cid:39) (cid:14) (cid:188) (cid:172) 12 m m m m m m (cid:11) ) ( ) ( e C S U U U e V U 2 2 2 i (cid:5) (cid:186) (cid:170) (cid:16) (cid:16) (cid:14) (cid:16) (cid:39) (cid:16) (cid:188) (cid:172) 21 m m m m m m ( ), ( ), , . (5) t m n V t m n V (cid:5) (cid:32)(cid:5) (cid:77) e-pn ij j (cid:11) cos 2 (cid:58) (cid:12) V e (cid:71) i t (cid:5) 12 (cid:12) V e (cid:71) i t (cid:5) (cid:16) 21 (cid:16) e 2 i (cid:16) (cid:58) t , (cid:77) i i (cid:16) (cid:58) (cid:16) (cid:12) t (cid:58) , (cid:58) 2 (cid:14) * 2 (cid:14) 2 2 (cid:14) t t i ijV(cid:39) (cid:5) are the terms that describe the contribution of Here, the EDEP to the electron-vibrational interaction. For further consideration, we leave only these terms and determine them up to the first order with respect to the small displacement between two equilibrium positions of oscillations of the normal coordinates related to the ground and excited BSs. In this approximation, the electron- (cid:119) t (cid:119) ( ) t (cid:119) t (cid:119) (cid:93) 1 (cid:93) 2 (cid:11) 2 (cid:32) ( ), , (cid:587) U W e t m n (cid:14) (cid:14) 1 m m (cid:12)1 i (cid:587) S U W e (cid:58)(cid:16)(cid:58) (cid:14) (cid:16) (cid:14) 2 m m m ( ), , (cid:587) U W e t m n (cid:16) (cid:14) (cid:32) 1 m m (cid:12)1 i (cid:587) S U W e (cid:58)(cid:14)(cid:58) (cid:16) (cid:16) (cid:14) 2 m m m U F U (cid:14) (cid:16) (cid:58) 0 m 2 U (cid:14) (cid:58) 1 m (cid:58) (cid:32) (cid:58) (cid:14) W (cid:114) 2 , (cid:114) (cid:16) m 2 2 2 (cid:11) 0F 1 1 2 . t t (cid:11) (cid:11) 2 2 , (cid:91) 1 (cid:12)1 i (cid:16) (cid:58)(cid:14)(cid:58) ( ), , t m n 2( ), , t m n (cid:91) (cid:12)1 i t ( ), , t m n (cid:16) (cid:58)(cid:16)(cid:58) (cid:91) 1 , 2( ), t m n (cid:91) U F U 1 (cid:14) (cid:16) m m 2 (cid:58) 1 * (cid:14) (cid:58) 0 2 U (cid:16) m (cid:114) (cid:32) , t (8) W (cid:114) 1 (cid:32) The initial conditions for the coefficients d1,n and d2,n correspond with the electron population of the DSs in the electron-photon system [see Eq. (2)]. Eqs. (8) show that the electron population in the TLDS oscillates between the DSs with the frequency (cid:525)1 and becomes equally divided between the DDSs. At the same time, the phonon numbers are not changed under the action of the considered interaction contribution. FIG. 1. (Color online) Formation of the DDSs in the TLDS: (i) BSs with an electromagnetic field of frequency (cid:90)0 (magenta, dashed arrows); (ii) DSs with a phonon mode of frequency (cid:90)k (red, doted arrows); (iii) DDSs in the first order of the electron-vibrational interaction; (iv) DDSs in the second order of the electron-vibrational interaction. In order to establish the coupling with the phonon modes, we must take into account the next term of the Taylor's series expansion in displacements for the interaction operator. Using the wave functions of Eqs. (8), the matrix element of the quadratic interaction can be readily obtained (for details see Ref. [16]) H (cid:99)(cid:99) (cid:32) ED (cid:33) 2 mC 2 (cid:170) (cid:172) (cid:94) (cid:11) , s k F 11 (cid:14) (cid:117) (cid:166) k , , s n k F 2 * (cid:14) (cid:16) 2 (cid:14) 2 (cid:14) 1 * (cid:16) 1 (cid:11) (cid:12) W W W W e i W W e (cid:58)(cid:16)(cid:58) (cid:14) k F 11 (cid:16) 1 n e k s (cid:14) 2 (cid:14) i (cid:90) k 2( * (cid:16) s i 1 t , , , s (cid:12) (cid:11) ) t , s 2 t (cid:58) (cid:14) * W W e (cid:58)(cid:14)(cid:58) 2( i 1 ) t (cid:14) 1 (cid:16) 2 (cid:186) (cid:188) (cid:16) , s k F 2 (cid:12) n k , s (cid:14) 1 e i (cid:90) k t , s (cid:96) G (cid:114) k 0 (cid:32) , , s k F 2 2 (cid:32) (cid:33) (cid:14) 2 mC W W W W e (cid:12) (cid:33) 2 * (cid:14) 2 * (cid:16) 2 (cid:14) 1 (cid:16) 1 (cid:170) (cid:172) (cid:14) (cid:11) i t (cid:16) (cid:58) 2 (cid:14) * W W e(cid:16) (cid:58)(cid:14)(cid:58) 2( i 1 (cid:16) 1 (cid:14) 2 ) t (cid:94) (cid:11) , s k F 12 (cid:16) k F 2 , * s (cid:12) (cid:117) (cid:166) k , , s n W W e(cid:16) (cid:58)(cid:16)(cid:58) (cid:14) 2( i 1 (cid:14) 1 ) t i (cid:90) k , s t (cid:14) (cid:11) , s k F 11 (cid:14) * (cid:16) 2 n e k s , (cid:16) (cid:186) (cid:188) k F 2 , * s (cid:12) n k , s (cid:14) 1 e i (cid:90) k t , s (cid:96) , , , s (cid:14) (cid:14) (cid:16) s m m 1 (cid:167) f U U n (cid:168) k k 2 (cid:169) ( f U U 2 (cid:16) (cid:14) (cid:32) k(cid:33) , s m m ( ) f U U 2 (cid:16) k(cid:33) , m (cid:14) s m 2 , s k F 11 k F , s (cid:32) 12 s (cid:14) ( k U S h , * 21 m (cid:16) m s , (cid:186) (cid:188) 2 ) (cid:14) m (cid:183) (cid:170) ( k U S h , (cid:184) (cid:172) 12 m (cid:185) (cid:11) ) k U h 2 , (cid:16) 22 m (cid:11) k U h , s (cid:14) (cid:14) 22 m s k U S h , (cid:14) (cid:16) 12 m m k U S h , s (cid:16) 12 m m (cid:12) s 2 ) (cid:12) . (9) , Here, nk,s is the phonon number of mode s and wave vector k,s is the quadratic coupling strength of the electron- k; hij phonon interaction. In Eq. (9), we have left only the non- diagonal matrix elements in the basis of the DSs. It can be seen from Eq. (9) that there are three frequencies in the PIV state spectrum, 2(cid:525), 2((cid:525) – (cid:525)1), and 2((cid:525) + (cid:525)1). Note that the diagonal matrix elements in the basis of the DSs give additional frequency, 2(cid:525)1 (see also Fig. 1). Such situation is similar to existence of a Mollow triplet in the electron- photon system. In further consideration, we will take into account only the term which defines the transitions of frequency 2(cid:525). Other terms can be accounted in the same manner. Then, we assume that the parameters of the electromagnetic field are chosen so that there is a resonance between the transition frequency 2(cid:525) of the DSs and one reservoir phonon mode (s, k0). This assumption makes it possible to use the RWA for the phonon mode. For each selected mode (s, k0), the Hamiltonian associated with Eq. (9) is the Jaynes-Cummings interaction Hamiltonian. Since this Hamiltonian commute with the where n is the parity operator [18], phonon number operator, the final form of the PIV wave functions in the DSs basis can be written as [19] , , ( ) ( ), ,2 ( ) m n D t m n D t t (cid:16) (cid:16) (cid:14) (cid:91) (cid:91) 1,2 2,2 1 2 1 n (cid:14) ( ) ( ), , , ,2 1 t t m n D m n (cid:16) (cid:14) (cid:91) (cid:32) 1,2 1 1 n (cid:16) ,2 , (10) . nD m n (cid:14)(cid:14) (cid:91) 2,2 2 (cid:32) (cid:81) (cid:14) (cid:51) (cid:32) (cid:16) (cid:12) i n (cid:83) 1 (cid:14) , ,2 m n exp (cid:86) z (cid:81) (cid:16) ( ) t (cid:11) , Here, signs + and – stand for the even and odd parity, respectively. Strictly speaking, the BS wave functions also have the parity according to the number of photons but we omit the RWA, the coefficients of Eqs. (10) can be found from the following differential systems this fact for simplicity. Using n (cid:119) t (cid:119) (cid:119) t (cid:119) mC 2 2 D (cid:114) 1,2 n (cid:16) 0.5 0.5 (cid:35) D (cid:114) 2,2 n (cid:14) 0.5 0.5 (cid:35) ( ) t ( ) t (cid:32) (cid:16) i G e D i t (cid:71) 1 (cid:114) k 0 (cid:114) 2,2 ( ) t , n (cid:14) 0.5 0.5 (cid:35) (cid:32) (cid:16) iG e D (cid:114) 1,2 i t (cid:71)(cid:16) 1 * (cid:114) k 0 ( ) t , n (cid:16) 0.5 0.5 (cid:35) 2 n k 0 (cid:14) (cid:35) 1 1 2 2 2(cid:71) 1 (cid:11) k F 11 0 (cid:14) k F 2 0 (cid:32) (cid:58) (cid:16) k . (cid:90) 0 * * (cid:14) (cid:14) 2 (cid:12)(cid:11) W W W W (cid:16) (cid:14) 2 1 (11) (cid:16) 1 (cid:12) , The solution of Eqs. (11) allows to write the final PIV wave functions of the corresponding parity in three different bases in the same manner as the wave functions of Eqs. (8). In the case of the DDS basis, the energy eigenvalues take the form similar as in the DS case E (cid:114) 1 (cid:32) (cid:90) (cid:33) 0 ( m 1) (cid:14) (cid:14) (cid:90) (cid:33) k 0 E (cid:114) 2 (cid:32) (cid:90) (cid:33) 0 ( m 1) (cid:14) (cid:14) (cid:90) (cid:33) k 0 (cid:58) (cid:32) (cid:114) 2 2 (cid:71) 1 (cid:14) 4 G (cid:114) k 0 (cid:167) (cid:168) (cid:169) (cid:167) (cid:168) (cid:169) 2 2 n k 0 (cid:35) 2 n k 0 (cid:35) . , (cid:114) 2 , (cid:114) 2 (cid:16) (cid:58) (cid:33) (cid:14) (cid:58) (cid:33) 1 2 1 2 (cid:183) (cid:184) (cid:185) (cid:183) (cid:184) (cid:185) (12) For an exact resonance, the difference of module squares of the coefficients can be written as ( ) D t 0.5 0.5 * * iU U S s m n ( ) ( U U (cid:186) (cid:14) (cid:16) (cid:188) m (cid:11) (cid:12) (cid:12) (cid:58) , sin 2 (cid:114) 20 D (cid:114) 0.5 0.5 1,2 n (cid:16) (cid:11) cos 2 (cid:117) (cid:58) (cid:170) (cid:32) (cid:172) S s m n ( ) t t (cid:114) 20 (cid:14) (cid:16) m m (cid:16) m t (cid:114) 2,2 (cid:16) (cid:16) (cid:11) ) 2 2 n (cid:14) 2 2 (cid:35) (cid:35) (cid:16) (cid:12) G k G k S U C D m m m U , (cid:14) m (cid:16) 0 0 s n (cid:32) U (cid:16) m (cid:32) (cid:32) U C D (cid:114) 1,2 (cid:14) m m n (cid:16) 0.5 0.5(0) (cid:35) , (cid:114) 2,2 0.5 0.5(0) (cid:35) . n (cid:14) (13) Here, (cid:525)20 is the resonance frequency of (cid:525)2 with (cid:303)1 = 0. In Eq. (13), we assume that the TLDS state is initially a product state which is composed by its subsystem states. Namely, in the equations above, Cm and D± i,n(0) are the initial probability amplitudes for the photon and phonon states, respectively. Eq. (13) shows that the initially presented population difference between the DSs in the electron-photon system oscillates with the frequency of (cid:21)(cid:525)20. Thus, the DDSs are formed by two steps. At the first step, the basis of levels is created due to the first order electron-vibrational interaction. At the second step, the second order interaction leads to broadening of the levels. Finally, taking into account the dispersion of phonon states, we obtain a quasi-continuous electron spectrum in the vicinity of the DSs (Fig. 1). In order to include the dissipation in our consideration, we have to transfer from the wave function description to the density matrix formalism and specify the relaxation channels. The population of the DSs is affected by both the nonradiative and radiative decays. In both cases, we assume that the Weisskopf-Wigner approximation is valid and the phenomenological damping rates can be introduced. This assumption leads to the complete equivalence of the DS and DDS representations (see Table I). The exact estimations of the nonradiative decay rate must be done using the coupling between the photoinduced vibrations and phonon modes of the reservoir [see Eq. (9)] but it is not the aim of this Letter. Additionally, in the case of the ac Stark effect, we assume that the Rabi frequency is large compared to the radiative decay rate. TABLE I. Equivalence of representations of the dressed and doubly dressed states. Electron-photon interaction Electromagnetic field mode of frequency (cid:90)0 Oscillations between the BSs of frequency (cid:525) Appearance of the DSs Radiative decay due to the interaction with the vacuum field mode continuum Electron-phonon interaction Phonon mode of frequency (cid:90)k Oscillations between the DSs of frequency (cid:525)1 Appearance of the DDSs Nonradiative decay due to the interaction with the phonon mode continuum For our consideration, existence of the nonradiative decay means the uniform distribution of the electron population, oscillating between the DSs. Thus, if initially the lower DS has the larger population than the upper DS, the energy will be removed from the phonon reservoir, in the opposite case, if initially the lower DS has the lower population than the upper DS, the energy will be transferred to the phonon reservoir. In the case of the equal populations at the DSs, the energy is not changed. Transferred energy can be defined as the product of the electron population difference and the energy difference of the DSs and equals 2 (cid:58) (cid:33) (cid:11) ( U (cid:14) m 2 ) (cid:16) ( U 2 ) (cid:16) m (cid:12) (cid:32) (cid:33) , (cid:71) (14) in accordance with the energy conservation law. FIG. 2. (Color online) Energy the dynamical Stark shift in the TLDS: reservoir cooling (cid:11)(cid:507) > 0) and reservoir heating ((cid:507) < 0). transfer using It is important to stress that the considered transference of energy can be observed if the frequency of the photoinduced oscillations are comparable with the frequencies of vibrational modes of the reservoir. To satisfy this condition, we must use the optical detunings of the same order of magnitude as the resonance Rabi frequency. Thus, we can control the energy transfer between the electron and phonon subsystems with use of the dynamical Stark shift, choosing properly the intensity and optical frequency of an electromagnetic pulse. Appling a sequence of pulses with the duration that is shorter compared to the radiative decay time, we can heat (negative optical detuning) or cool (positive optical detuning) the TLDS (Fig. 2). In conclusion, we describe the mechanism of energy transfer between the two-level electron system and the phonon reservoir in the case of the RWA. The use of the RWA means the weak electron-vibrational coupling in the TLDS. Remarkably, there are no limitations for this mechanism to occur in the case of the strong coupling. In the strong coupling case, the generalized RWA may be used [18, 20]. At the same time, the case of low temperatures, when some phonon modes are frozen, has to be considered separately. It should be noted that the vibronic coupling may be created by the interaction *e-mail address: [email protected] [1] T. Henighan, M. Trigo, M. Chollet, J. N. Clark, S. Fahy, J. M. Glownia, M. P. Jiang, M. Kozina, H. Liu, S. Song, D. Zhu, and D. A. Reis, Control of two-phonon correlations and the mechanism of high-wavevector phonon generation by ultrafast light pulses, Phys. Rev. B 94, 020302(R) (2016). [2] I. Wilson-Rae, P. Zoller, and A. Imamoglu, Laser cooling of a nanomechanical resonator mode to its quantum ground state, Phys. Rev. Lett. 92, 075507 (2004). [3] D. Jonathan, M. B. Plenio, and P. L. Knight, Fast quantum gates for cold trapped ions, Phys. Rev. A 62, 042307 (2000). [4] J. A. Black and H. Schmidt, Atomic cooling via ac Stark effect, Opt. Lett. 39, 536 (2014). [5] K. D. B. Higgins, B. W. Lovett, and E. M. Gauger, Quantum thermometry using the ac Stark shift within the Rabi model, Phys. Rev. B 88, 155409 (2013). [6] J. H. Quilter, A. J. Brash, F. Liu, M. Glässl, A.M. Barth, V.M. Axt, A.J. Ramsay, M.S. Skolnick, and A.M. Fox, Phonon-Assisted Population Inversion of a Single InGaAs/GaAs Quantum Dot by Pulsed Laser Excitation, Phys. Rev. Lett. 114, 137401 (2015). [7] J. F. Dynes, M. D. Frogley, M. Beck, J. Faist, and C. C. Phillips, ac Stark Splitting and Quantum Interference with Intersubband Transitions in Quantum Wells, Phys. Rev. Lett. 94, 157403 (2005). [8] A. Mysyrowicz, D. Hulin, A. Antonetti, A. Migus, W. T. Masselink, and H. Morkoç, "Dressed Excitons" in a Multiple-Quantum-Well Structure: Evidence for an Optical Stark Effect with Femtosecond Response Time, Phys. Rev. Lett. 56, 2748 (1986). [9] M. A. Bondarev, E. Yu. Perlin, and A. V. Ivanov, Multiphoton Absorption Controlled by the Resonance Optical Stark Effect in Crystals, Opt. and Spectr. 115, 827 (2013). [10] Y. Kayanuma and K. G. Nakamura, Dynamic Jahn- Teller viewpoint for generation mechanism of asymmetric between optical phonon modes and electron states directly [16] as it occurs in the doped solid state systems [19, 21]. Therefore, the results of this work are also valid in the case of optical phonon modes. Since the exact resonance conditions must be satisfied for the optical phonon frequency and the transition frequency between the DSs, the case of the forced vibrations of reservoir oscillators considered in this Letter is the most general. The author of this Letter has been inspired by the recently published experimental results of Ref. [22]. Financial support of the Ministry of Education and Science of the Russian Federation (grant 074-U01) and of the Russian Foundation for Basic Research (grant 17-02- 00598) are gratefully acknowledged. modes of coherent phonons, Phys. Rev. B 95, 104302 (2017). [11] H. J. Zeiger, J. Vidal, T. K. Cheng, E. P. Ippen, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. B 45, 768 (1992). [12] A. V. Kuznetsov and C. J. Stanton, Theory of Coherent Phonon Oscillations in Semiconductors, Phys. Rev. Lett. 73, 3243 (1994). [13] I. B. Bersuker, V. Z. Polinger, Vibronic Interactions in Molecules and Crystals (Springer, Berlin, 1989). [14] U. Öpik, and M. H. L. Pryce, Studies of the Jahn Teller Effect. 1. A Survey of the Static Problem, Proc. R. Soc. A. 238, 425 (1957). [15] N. B. Delone, V.P. Krainov, Atoms in Strong Light Fields (Springer, Berlin, 1985). [16] A. V. Ivanov, Photoinduced vibronic coupling in two- level dissipative systems, Phys. Rev. B. [17] G. Fischer, Vibronic Coupling (Academic, San Diego, 1984). [18] J. Casanova, G. Romero, I. Lizuain, J. J. Garcia- Ripoll, and E. Solano, Deep Strong Coupling Regime of the Jaynes-Cummings Model, Phys. Rev. Lett. 105, 263603 (2010). [19] T. Vaikjärv, V. Hizhnyakov, Time-dependent pseudo Jahn-Teller long-time nonadiabatic relaxation, J. Chem. Phys. 140, 064105 (2014). [20] E. K. Irish, Rotating-Wave Approximation for Arbitrarily Large Coupling, Phys. Rev. Lett. 99, 173601 (2007). [21] A. Ivanov, Yu. Rozhdestvensky, E. Perlin, Vibronic model of laser cooling with stimulated Raman adiabatic passage pumping for Yb3+-ion-doped crystals, JOSA B 33, 1564 (2016). [22] A. J. Brash, L. M. P. P. Martins, A. M. Barth, F. Liu, J. H. Quilter, M. Glässl, V. M. Axt, A. J. Ramsay, M. S. Skolnick, and A. M. Fox, Dynamic vibronic coupling in InGaAs quantum dots, JOSA B 33, C115 (2016). Phonon-mediated Generalized effect:
1108.2010	2	1108	2011-09-22T16:37:09	Quarter-Filled Honeycomb Lattice with a Quantized Hall Conductance	[ "cond-mat.mes-hall" ]	We study a generic two-dimensional hopping model on a honeycomb lattice with strong spin-orbit coupling, without the requirement that the half-filled lattice be a Topological Insulator. For quarter-(or three-quarter) filling, we show that a state with a quantized Hall conductance generically arises in the presence of a Zeeman field of sufficient strength. We discuss the influence of Hubbard interactions and argue that spontaneous ferromagnetism (which breaks time-reversal) will occur, leading to a quantized anomalous Hall effect.	cond-mat.mes-hall	cond-mat	Quarter-Filled Honeycomb lattice with a quantized Hall conductance Ganpathy Murthy,1 Efrat Shimshoni,2 R. Shankar,3 and H.A. Fertig4 1Department of Physics and Astronomy, University of Kentucky, Lexington KY 40506-0055 2Bar-Ilan University, Ramat Gan, Israel 3Yale University, New Haven CT 06520 4Department of Physics, Indiana University, Bloomington, IN (Dated: July 20, 2018) We study a generic two-dimensional hopping model on a honeycomb lattice with strong spin-orbit coupling, without the requirement that the half-filled lattice be a Topological Insulator. For quarter- (or three-quarter) filling, we show that a state with a quantized Hall conductances generically arises in the presence of a Zeeman field of sufficient strength. We discuss the influence of Hubbard interactions and argue that spontaneous ferromagnetism (which breaks time-reversal) will occur, leading to a quantized anomalous Hall effect. PACS numbers: 73.50.Jt Topological Insulators (TIs) are a new state of mat- ter with a bulk gap but protected edge/surface modes[1]. Quantum Hall states[3] are examples of time-reversal vi- olating TIs. Recently, time-reversal invariant TIs have been predicted[1] and seen in experiments[2]. They are characterized by a nontrivial Z2 index in the bulk, and the presence of chiral edge/surface states which are ro- bust against localization due to static disorder. One of the simplest models of two-dimensional TIs is the Kane-Mele model[4], which is a tight-binding model on a honeycomb lattice with hopping and spin-orbit inter- actions. A more realistic model is given by a monolayer of the 3D TI Alkali Iridate A2IrO3 (where A = N a, Li), where the various tight-binding hopping parameters are known from a fit to ab initio calculations[5]. We will be working with a generic model on a honeycomb lattice which has no symmetries other than lattice translations, 2π/3 rotations, and time-reversal. Our goal is to construct an experimentally realizable system with a quantized Anomalous Hall effect (AHE)[6]. The AHE has historically described the effect of mag- netic order (spontaneous or otherwise) on the Hall con- ductance. We will describe a proposal for observing a quantized Hall conductance in the absence of a perpen- dicular magnetic flux which can be realized, for example, in a monolayer of Alkali Iridate doped to 3 4 -filling, and subject to a Zeeman field perpendicular to the monolayer. Previous proposals in this direction[7] have also either explicitly violated time-reversal, or appealed to sponta- neous magnetic order in interacting TIs. Our proposal does require strong spin-orbit couplings but does not re- quire the material to be a TI at half-filling. Generic models we consider share the following prop- erties: They are defined on a honeycomb lattice with two sites per unit cell. Including spin, there are four bands. In the Brillouin Zone (BZ) shown in Fig. (1) there are two kinds of special points. The Γ and M points are time-reversal invariant (modulo a reciprocal lattice vec- tor), while the K and K ′ points go into each other under time-reversal. Time-reversal symmetry forces the bands to be degenerate at the Γ and Mα points. In a sim- FIG. 1: Top: The Brillouin Zone of the honeycomb lattice. The Γ and three M points are time-reversal invariant, while the K and K ′ points go into each other under time-reversal. The arrows show the path along which the band structure will be plotted. Bottom: The next-nearest neighbor hops in A2IrO3 are shown by dot-dashed red lines that go with σ1, the solid blue that go with with σ2, and the dashed green that go with σ3. The primitive lattice vectors are ~a1 and ~a2 while ~b1, ~b2,~b3 are nearest-neighbour hopping vectors. The auxiliary triad shows the projection of the original crystal axes on to the plane of the honeycomb lattice. Note that these directions match the nearest neighbor hops and the labels of the associated σ matrices. ple spin-independent nearest-neighbor only tight-binding model (e.g. Graphene) there are van Hove singularities at Γ and Mα, whereas there are Dirac cones at K and K ′. Neither of these features is generic under the addi- tion of time-reversal invariant spin-orbit couplings: The Dirac cones at the K and K ′ points can be gapped by either a LzSz type spin-orbit coupling, or by a sublat- tice antisymmetric potential[4], while the van Hove sin- gularities are converted into Dirac cones by a Rashba coupling. Fig. (2) shows such a generic time-reversal invariant band structure. We begin with a noninteracting model with three-fold rotational and time-reversal symmetry, and focus on the generic Dirac crossings at the Γ and Mα points. Now con- sider a Zeeman field in the Z direction (perpendicular to the monolayer). This preserves the three-fold rotational symmetry of the lattice, but breaks time-reversal and will therefore provide a mass gap to the Dirac points at Γ and Mα. We know that if the variation of a parameter in the Hamiltonian leads to the touching of two bands at a Dirac point for some critical value, a Chern number of ±1 must be exchanged between the two bands[8] as the parameter passes through that value[8]. By the lattice rotational symmetry all Mα must have the same exchange of Chern number. Since there are three M points and a single Γ point, it is clear that the Chern number exchange must be either ±2 or ±4. This means that upon reversing the Zeeman coupling, the Chern number of the lowest band must reverse, implying that its Chern number is either ±1 or ±2. Of course, the Zeeman coupling should be strong enough to create a hard gap (the bands should not overlap in energy) in order to form an insulator. We will show that in a simple model of A2IrO3 with Hubbard interactions, such a hard gap may form spontaneously. To demonstrate this behaviour in detail, we start with the noninteracting hamiltonian for a freestanding mono- layer of Sodium/Lithium Iridate in the notation of [5]: HSI = −tXhiji (cid:0)c† i cj + hc(cid:1) + Xhhijii c† i tijcj (1) where the spin indices have been suppressed, hiji is a sum over nearest neighbors, hhijii is a sum over next-nearest neighbors, and the matrix tij in spin-space is tij = t′ 0 + it′σa (2) This hopping term is diagonal in the sublattice. Refer- ring to Fig. (1), the hopping is antisymmetric in the sublattice index (A, B). Further, each hop comes with a σa matrix, the index a corresponding to the projection of the original crystal axes on to the 111 plane. Thus, hopping along the primitive lattice vector ±~a1 carries a σ1, hopping along the vector ±(~a1 −~a2) carries a σ2, and hopping along ±~a2 carries a σ3. Let us denote the Pauli matrices in the sublattice space by τa. We obtain hSI (~k) = t′ 0F0(~k) − t(cid:0)fr(~k)τ1 − fi(~k)τ2(cid:1) − t′τ3(cid:0) ~F (~k) · ~σ(cid:1)(3) where fr + ifi = ei~k·~b1 + ei~k·~b2 + ei~k·~b3 f (~k) = (4) F0(~k) = cos(~k · ~a1) + cos(~k · ~a2) + cos(~k · (~a2 − ~a1))(5) F1(~k) = F2(~k) = F3(~k) = sin(~k · (~a2 − ~a1)) − sin(~k · ~a2) sin(~k · ~a1) (6) (7) (8) 2 are estimated as t = 310K, t′ The parameters for the 111 plane of Ir in N a2IrO3 0 = −130K and t′ = 100K. One first diagonalizes ~F · ~σ, which has eigenvalues ± ~F , and obtains the energies at ~k as E±(~k) = t′ 0F0(~k) ±qt2f 2 + t′2 ~F 2 (9) It is seen that the four bands come in two pairs, de- generate at all momenta. This is the result of inversion symmetry and clearly not generic. For a monolayer on a substrate, a Rashba spin-orbit coupling arising from the substrate ~E field perpendicular to the 111 plane is in- duced. A nearest-neighbor Rashba coupling leads to an additional term hR(~k) = iλR ~G(~k) · στ+ − iλR ~G∗(~k) · στ− (10) where G1(~k) = r 2 3(cid:0)ei~k·~b1 − ei~k·~b2 − 1 2 1 2 ei~k·~b3(cid:1) (11) with cyclic permutations of ~b1, ~b2, ~b3 defining G2, G3. Due to its special symmetries, this model does not show generic behavior for λR < t′. For this parameter range, in addition to the usual Dirac points at Γ, and Mα there are also Dirac points at K, K ′. For t′ > t′ ∗ (where t′ ∗ is a function of λR, but tends to 3t/8 as λR → 0) the M points are unsplit. However, for λR < t′ < t′ ∗, each M point splits into three Dirac points lying on the zone boundary. The central point is still at M and the other two "satellites" are symmetrically distributed about it. At the special point λR = t′, the satellites coalesce with the Dirac points at K, K ′. For λR > t′ the would-be Dirac points at K, K ′ are gapped, and the only Dirac points left are the generic ones. In the following we will focus on this case. A typical band structure including λR = 0.4t (satisfy- ing λR > t′) is shown in Fig. (2). One clearly sees the Dirac-like crossings at the Γ and Mα points. 4 For weak Zeeman coupling, a gap will open at the Dirac points, but the two bands will overlap in energy (at different momenta), and the system will be metallic at 1 or 3 4 -filling. Increasing the Zeeman coupling will lead to a hard gap and thus a quantized Hall conductance. For the noninteracting model, the critical Zeeman field is 0.37t. A calculation shows that the Chern number for these parameters is −1. We note parenthetically, that even for the nongeneric case λR < t′ the Chern number of the highest energy band is nonzero (1), but different from the above. Once a hard gap has been established, the Chern number is robust against arbitrary deformations of the hamiltonian which do not close the gap. In particular, one could softly break the lattice rotational symmetry, add a Kane-Mele-type spin-orbit coupling or a staggered sublattice potential, and also tilt the Zeeman away from the perpendicular to the monolayer. Let us now consider the influence of interactions. It is known that a strong enough Hubard U leads to an 3 FIG. 2: The band structure for a monolayer of A2IrO3 with the addition of a Rashba term with λR = 0.4t. The path in the BZ followed is shown in Fig. (1). The generic Dirac crossings at the Γ and Mα points can be seen. FIG. 3: The Hartree-Fock band structure at U = 10t and three-quarter filling in A2IrO3 with λR = 0.1t. The path followed in the BZ is shown in Fig. (1). Note the hard gap and the flatness of the empty band (blue). antiferromagnet in the half-filled TI[9]. An on-site Hub- bard interaction (typically of size U t ≃ 10) spontaneously breaks time-reversal symmetry at 3 4 filling in the Hartree- Fock approximation. Assuming only that lattice trans- lation symmetry is intact but allowing for the possibility of rotational and sublattice symmetry breaking, we find in Hartree-Fock that the magnetization is in the Z di- rection (perpendicular to the monolayer), and the gap is strongly enhanced. Because there is no spin-rotation symmetry in this model, the Z direction is the natu- ral one, and the order-parameter is Ising-like. This im- plies that the spin-waves are fully gapped, and that there should be a finite-temperature phase transition even in the two-dimensional material. Of course, it could happen that the system chooses to break lattice translation symmetry by forming a larger unit cell, as happens in Graphene at a quarter-doping[10]. However, in Graphene, this is a result of Fermi-surface nesting, which is sensitive to the details of the band struc- ture, and is not generic (there are also other competing states in Graphene at 1 4 -filling). Furthermore, a strong enough Zeeman field in the Z direction will make our lat- tice symmetric solution energetically favorable. We will leave a further discussion of the issue of other possible ground states at quarter (or three-quarter) filling[11] to future work. We now speculate on the effects of long range (1/r) Coulomb interactions in this system, which can coexist with the Hubbard interaction. Long-range Coulomb interactions will be present because the mate- rial is insulating, and will likely suppress Charge Density Wave states, which are one avenue of translation sym- metry breaking. Secondly, since the band has a nonzero Chern number we expect skyrmionic excitations[12] to be the lowest energy charged excitations in some regime of parameters. Finally, because the empty band is relatively flat, it may represent an environment in which analogs of fractional quantum Hall states will be stable when the band is partially occupied[13, 14]. One can ask whether other simple lattices can sup- port similar states under generic conditions, where the only Dirac points are those mandated by time-reversal symmetry. The simplest square lattice two-band model does not support such states. The reason has to do with the fact that the BZ has only two M points. Repeat- ing the argument above Eq. (1), we see that the Chern number exchange is ±1 or ±3, implying that the states have Chern number ± 1 2 , which is impossible for a 2D band insulator. The half-filled triangular lattice with one band per unit cell would have a tendency to become an- tiferromagnetic when interacting. Thus the honeycomb lattice seems to be the simplest one in which the desired properties can be realized. 2 , ± 3 In summary, we have shown that in a 2D honeycomb lattice at 3 4 -filling, one can obtain a band with nonzero Chern number with strong enough Rashba coupling, and an external Zeeman coupling. Since the system does not have to be a TI at half-filling, this increases the range of possible material realizations. One promising material is A2IrO3 with A = N a, Li, with the Chern number of the 3 4 -filled system being ±1. In the bulk it is half-filled and antiferromagnetically ordered[15] below TN = 15K. A monolayer of the 111 Ir plane sandwiched by two layers of N a would induce three-quarter doping[16], where we expect it to become a ferromagnetic quantized anomalous Hall insulator with an Ising-like order parameter in the presence of Hubbard interactions of realistic strength. It may be quite difficult to grow a single monolayer doped in the manner described, but possibly the surface layer of a thin enough film will display the same properties. We intend to investigate this in future work. We are grateful to the Aspen Center for Physics (NSF 1066293) for its hospitality while this work was conceived and carried out. We would also like to thank Greg Fiete, Yong-Baek Kim, Dung-Hai Lee, Karyn Le Hur, Steve Simon, and particularly Xiaoliang Qi for illumi- nating conversations. We are also grateful for partial support from NSF-DMR-0703992 and nsf-phy 0970069 (GM), ISF 599/10 (ES), NSF-DMR-0901903 (RS), NSF- DMR-1005035 (HAF), and the US-Israel Binational Sci- ence Foundation-2008256 (ES and HAF). 4 [1] For reviews see, M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010); X.-L. Qi and S.-C. Zhang, http://arxiv.org/abs/1008.2026 (2010). [2] M. Koenig, S. Wiedmann, C. Bruene, A. Roth, H. Buh- mann, L. W. Molenkamp, X.-L. Qi and S.-C. Zhang, Sci- ence 318, 766 (2007); Y. Xia et al, Nature Phys. 5, 398 (2009). [3] K. v Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). [4] C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005). [5] A. Shitade, H. Katsura, J. Kunes, X.-L. Qi, S.-C. Zhang, and N. Nagaosa, Phys. Rev. Lett. 102, 256403 (2009). [6] E. H. Hall, Philos. Mag. 10, 301 (1880); ibid 12, 157 (1881): For a recent review, see, N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Rev. Mod. Phys. 82, 1539 (2010). [7] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988); M. Onoda and N. Nagaosa, Phys. Rev. Lett. 90, 206601 (2003); X.-L. Qi, Y. S. Wu, and S.-C. Zhang, Phys. Rev. B 74, 085308 (2006); R. Yu, W. Zhang, H.-J. Zhang, S.-C. Zhang, X. Dai, and Z. Fang, Science 329, 61 (2010); G. Xu, H. Weng, Z. Wang, X. Dai, and Z. Fang, arXiv:1106.3125; R. Nandkishore, L. Levitov, and A. Chubukov, arXiv:1107.1903. [8] A. W. W. Ludwig, M. P. A. Fisher, R. Shankar, and G. Grinstein, Phys. Rev. B 50, 7526 (1994). [9] D. A. Pesin, and L. Balents, Nature Phys. 6, 376 (2010): S. Rachel and K. Le Hur, Phys. Rev. B 82, 075106 (2010); A. Ruegg, and G. A. Fiete, arXiv:1106.1559. [10] T. Li, arXiv:1103.2420. [11] E. V. Castro, A. G. Grushin, B. Valenzuela, M. A. H. Vozmediano, A. Cortijo, and F. de Juan, arXiv:1105.3937 [12] S. L. Sondhi, A. Karlhede, S. A. Kivelson, and E. H. Rezayi, Phys. Rev. B 47, 16419 (1993); K. Nomura and N. Nagaosa, Phys. Rev. B 82, 161401 (2010). [13] E. Tang, J.-W. Mei, and X.-G. Wen, Phys. Rev. Lett. 106, 236802 (2011); K. Sun, Z.-C. Gu, H. Katsura, and S. Das Sarma, Phys. Rev. Lett. 106, 236803 (2011); T. Neupert, L. Santos, C. Chamon, and C. Mudry, Phys. Rev. Lett. 106, 236804 (2011). [14] D. N. Sheng, Z.-C. Gu, K. Sun, and L. Sheng, arXiv:1102.2658; X.-L. Qi, arXiv:1105.4298. [15] Y. Singh and P. Gegenwart, arXiv:1006.0437 (2010); X. Liu, T. Berlijn, W.-G. Yin, W. Ku, A. Tsvelik, Young- June Kim, H. Gretarsson, Yogesh Singh, P. Gegenwart, J. P. Hill, Phys. Rev. B 83, 220403 (2011); Y. Singh, S. Manni, and P. Gegenwart, arXiv:1106.0429 (2011). [16] X.-L. Qi, private communication.
1912.11020	1	1912	2019-12-23T18:31:56	Nonlocal effects in plasmonic metasurfaces with almost touching surfaces	[ "cond-mat.mes-hall" ]	Geometrical singularities in plasmonic metasurfaces have recently been proposed for the enhancement of light-matter interactions, owing to their broadband light-harvesting properties and extreme plasmon confinement. However, the large plasmon momenta thus achieved lead to failure of local descriptions of the optical response of metals. Here we study a class of metasurfaces consisting of a periodic metal slab with a smooth modulation of its thickness. When the thinnest part shrinks, the two surfaces almost touch, forming a near-singular point. Using transformation optics, we show analytically how nonlocal effects, such as a blueshift of the resonance peaks and a reduced density of states, become important and cannot be ignored in this singular regime. The method developed in this paper is very general and can be used to model a variety of metasurfaces, providing valuable insight in the current context of ultra-thin plasmonic structures.	cond-mat.mes-hall	cond-mat	Nonlocal effects in plasmonic metasurfaces with almost touching surfaces Fan Yang,1, ∗ Emanuele Galiffi,1 Paloma Arroyo Huidobro,2 and John Pendry1 1The Blackett Laboratory, Department of Physics, Imperial College London, London SW7 2AZ, UK 2Instituto de Telecomunica¸coes, Insituto Superior Tecnico-University of Lisbon, Avenida Rovisco Pais 1,1049-001 Lisboa, Portugal Abstract Geometrical singularities in plasmonic metasurfaces have recently been proposed for the en- hancement of light-matter interactions, owing to their broadband light-harvesting properties and extreme plasmon confinement. However, the large plasmon momenta thus achieved lead to fail- ure of local descriptions of the optical response of metals. Here we study a class of metasurfaces consisting of a periodic metal slab with a smooth modulation of its thickness. When the thinnest part shrinks, the two surfaces almost touch, forming a near-singular point. Using transformation optics, we show analytically how nonlocal effects, such as a blueshift of the resonance peaks and a reduced density of states, become important and cannot be ignored in this singular regime. The method developed in this paper is very general and can be used to model a variety of metasurfaces, providing valuable insight in the current context of ultra-thin plasmonic structures. 9 1 0 2 c e D 3 2 ] l l a h - s e m . t a m - d n o c [ 1 v 0 2 0 1 1 . 2 1 9 1 : v i X r a ∗ [email protected] 1 I. INTRODUCTION The advancement of sophisticated nano-fabrication techniques has recently allowed for the realization of atomically thin films, structures with extremely sharp angles and touching points1 -- 7. The exotic behaviour of these singular structures hinges on small-scale geomet- rical features of nanometer and even sub-nanometer-scale, so that quantum effects, such as repulsion, diffusion and spill-out of electrons, become appreciable2,8 -- 11. In the case of noble metals, such as gold and silver, the nonlocal response is dominated by repulsion in the electron gas, resulting in size-dependent linewidth broadening and resonance shift. Singular metasurfaces constitute a special class of structured surfaces which feature sharp edges or ultra-small gaps. A conventional metasurface12 -- 15 is characterized by two wave vec- tors, such that the selection of k-vectors is discretized. On the other hand, singular meta- surfaces feature three wave vectors, owing to the additional length scale introduced by the singularity, which allows surface modes to exist over a continuum of quantum numbers16 -- 18. However, while a local singular structure has a broadband optical response, nonlocality sets a length scale that prevents the formation of a perfect singularity, thus yielding a discrete resonance spectrum. As recently pointed out in previous studies on singular surfaces, such as surfaces with a knife-edge profile19 and graphene-based gratings20, the degree of nonlocality determines the spacing of the resonance peaks in a far-field spectrum, so that singularities offer a valuable window into the microscopic physics of electrons in plasmonic materials. In the past few years, the analytical technique of transformation optics (TO)21,22 has proven a powerful tool for the modelling of singular plasmonic structures both within the local approximation23,24 and including nonlocal effects as described by the hydrodynamic model19,25. Here we deploy TO to model nonlocal effects of plasmonic metasurfaces in the form of thin gold gratings with almost touching surfaces, and explore their nonlocal response. Similar metasurfaces, in a non-singular regime, have been studied in the past26,27. However, we have developed a different theoretical approach, which allows us to investigate the singular regime while accounting for nonlocal effects, which, as we show in the subsequency sections, become important for a near-singular metasurface, whose minimum thickness approaches the single-nanometre regime. This paper is structured as follow: In section II, we present our analytical method, discussing in detail (1) the transformation of the geometry, source and dielectric function, (2) the dispersion relation and (3) the fields in real space and the metasurface reflection 2 coefficient. In section III, we deploy our method in order to study a set of near-singular metasurfaces in terms of their far-field scattering spectra and field profiles, highlighting the contrast between local and nonlocal models. Finally, we close the Paper with conclusions in Secion IV. II. PRINCIPLES II.1. Transformation of geometry, source and dielectric function FIG. 1. Sketch showing the metasurface geometry and the source field in both frames. In the metasurface frame (a) the sources are plane waves (incident, reflected and transmitted waves), which we take to be generated by an array of magnetic line currents located at infinity. The sources are mapped into an array of magnetic line currents in the slab frame (b), where the structure is a flat slab. The metasurface studied in this paper is shown in Fig. 1(a). It consists of a thin slab with periodic smooth thickness modulations and can be generated from the simple slab geometry in Fig. 1(b) by using the transformation26 (cid:18) (cid:18) (cid:48) z = z = T 2π h 2π ln ln (cid:19) 1 ae2πz/h − w0 − y0 (cid:48) 1 /T + y0) + w0 a a(e2πz (cid:19) (1) where w0 = αae− 2π between line sources in the slab frame, which correspond to a linear array of monopoles. h d, d = d1 + d2 is the slab thickness and h is the separation h d1, y0 = αe− 2π In this transformation, z stands for the complex coordinate in the slab frame while z (cid:48) for 3 the coordinate in the metasurface frame. Also, δmin and δmax stand for the minimum and maximum grating thickness shown in Fig. 1(a). Note that T , which acts as a scaling factor in the transformation affecting both the in-plane and out-of-plane components of the complex coordinate z (cid:48) , directly determines the period of the metasurface. In order to generate the metasurface in Fig. 1(a), we also need 0 < α < 1. Moreover, without loss of generality, we place the two surfaces of the slab in the slab frame at x = −d1 and x = d2 with d1 = d2 = d/2, where d is chosen as 1 throughout this paper. We additionally require that the flat interface of the metasurface lies on the y-axis, such that: a = 1 h d2(1 − α2e− 4π e 2π h d) (2) Note that the wavy surface generated by Eq. 1 is not sinusoidal, but nearly so when the modulation amplitude is not strong relative to the grating thickness. The transformation used also changes the form of the source excitation. In the meta- surface frame, we consider an incident plane wave, as shown in Fig. 1(a). For this near- sinusoidal metasurface, we follow the method introduced in17 to obtain the source represen- tation, whereby incident, reflected and transmitted plane waves in the z(cid:48) frame correspond to monopole arrays in the z frame. Note that incident and reflected waves on the right side of the metasurface correspond to the monopoles on the left side of the slab, and vice versa for transmitted waves, due to the coordinate inversion in Eq. 1 [see Fig. 1(b)]. With a lengthy but straightforward calculation (see Appendix A), the three types of e−kyx−xs+ eikyydky + aysgn(ky) e−kyx−xs+ sgn(x−xs+)ky eikyydky sgn(x−xs+)ky eikyydky raysgn(ky) e−kyx−xs+ taysgn(ky) e−kyx−xs− sgn(x−xs−)ky eikyydky (3) waves in the slab frame are obtained and written as  H inc H ref ax z = ∞(cid:82) z = − ∞(cid:82) ∞(cid:82) −∞ −∞ (cid:48) t k 0x k0x H tra z = −∞ 2π ln w0 ∞(cid:82) ∞(cid:82) eikyydky − ∞(cid:82) −∞ eikyydky + −∞ −∞ ky e−kyx−xs+ ky rax e−kyx−xs− ky ax where xs+ = h and the reflected waves, while xs− = h a 2π ln 1+w0y0 ay0 is the x coordinate of the monopole transformed from the incident is that from the transmitted wave. In the above source representation, we have decomposed the source field into two parts: the mode denoted by ax corresponds to waves with the electric field parallel to the metasurface, while the electric field of the ay is normal to the metasurface. At normal incidence only the ax mode is excited17. The discrete nature of the electron gas determines the screening length (δC ∼ 0.1 nm, for noble metals)28 and prevents the electron density from diverging. This effect has been widely 4 FIG. 2. Diagram for the longitudinal mode where the purple layer stands for the decay length of the longitudinal mode in (a) the metasurface frame and (b) the slab frame. The decay length is uniform in the metasurface frame, while after the transformation, the decay length varies along the slab. (c) The effective nonlocal parameter βef f as a function of y at the two interfaces in the slab frame. (d) Dispersion relation for the near-sinusoidal metasurface, where the nonlocal dispersion relation (β = 1.27 × 106 m/s) is compared with the local case (β = 0). The geometric parameters are T = 50 nm, the maximum gap δmax = 10 nm and the minimum gap δmin = 0.5 nm. The dispersion relation at y = 0.1h is illustrated. studied with the hydrodynamic model2,8,29. According to this model, both transverse and longitudinal modes are present inside the metal. The permittivity of the transverse mode, εT , is conserved under the conformal mapping between slab frame and metasurface frame. However, this symmetry does not hold for the longitudinal contribution to the dielectric function, whose form in the slab frame (z = x + iy) is related to that in the metasurface 5 frame (z (cid:48) (cid:48) = x (cid:48) + iy ) by25 εz L(ω, k, z) = εz (cid:48) L (ω, (cid:12)(cid:12)(cid:12)(cid:12) dz dz (cid:48) (cid:12)(cid:12)(cid:12)(cid:12) k) = ε∞ − ω2 p ω(ω + iΓ) − β2 ef f (z)k2 (4) where in the following calculation we choose ε∞ = 1, ωp = 8.95 eV/ and Γ = 65.8 meV/, which are typical parameters for gold30 and the effective nonlocal parameter, which acquires a spatial dependence in the slab frame, is βef f (z) = h T and β = 1.27 × 106 m/s31. (cid:12)(cid:12)(cid:12)(cid:12) y0(ae2πz/h − w0)2 + w0 ae2πz/h (cid:12)(cid:12)(cid:12)(cid:12) β − 1 (5) The effective nonlocal parameter βef f on the two interfaces of the slab is depicted in Fig. 2(c). Note that the spatial deformation in the slab frame is not symmetric at the two interfaces (x = −d1 and x = d2), such that their effective nonlocal parameter βef f are not exactly the same. The blue line stands for the βef f 1 at the interface x = −d1, while the red line is the βef f 2 at the interface x = d2. However, the difference between these two βef f is much smaller than the variation of βef f along the slab. Therefore, in the following calculation, we use (βef f 1 + βef f 2)/2 as the effective β. In the presence of nonlocality, an additional longitudinal mode exists inside the metal. The decay length of this mode is uniform along the interface of metasurface, as shown in Fig. 2(a). However, upon transformation into the slab frame, the decay length of the longitudinal mode becomes periodically modulated along the slab (Fig. 2(b)). Note that the thinnest part of the metasurface is mapped to the part with largest decay length in the slab frame. In other words, nonlocal effects become more important near the minimum gap region. II.2. Dispersion relation In order to account for both the transverse and longitudinal modes we use two potential functions, the magnetic field Hz and electric potential ϕ, in which Hz corresponds to the transverse mode and ϕ to the longitudinal mode19. Using the source representation in Eq. 3, we can express the total field in k-space as  (1 − r)ax (cid:16) (cid:0)c+ekyx + c−e−kyx(cid:1) eikyy, (cid:18) −kyx−xs+ ky −kyx−xs− e −t k (cid:48) 0x k0x e ax Hz(x, ky) = + (1 + r)aysgn(ky) e −kyx−xs+ sgn(x−xs+)ky + b+ekyx(cid:17) (cid:19) ky − taysgn(ky) e −kyx−xs− sgn(x−xs−)ky + b−e−kyx eikyy, eikyy, x < −d/2 −d/2 < x < d/2 x > d/2 (6) 6 and ϕ(x, ky) = sgn(ky) ωε0ε where b±, c± and d± are the excited mode amplitudes, and κ = (d+eκx − d−e−κx)eikyy (cid:113) (7) ε ε−1 determines y + ω2 k2 p βef f the decay of the longitudinal mode. The electric field components can be obtained as i Ex(kx, y) = ωε Ey(kx, y) = − i ωε ∂Hz ∂y ∂Hz ∂x − ∂ϕ ∂x − ∂ϕ ∂y (8) in which Hz contributes the divergence-free part of electric field, while ϕ includes the curl- free part. Imposing the continuity of Hz and Ey, together with vanishing normal component of current density J at the interface between the metal and the dielectric (Jx = 0)2,8, we can obtain the excited mode amplitude. Furthermore, we take a WKB approximation25,32, valid when the phase changes more rapidly than the amplitude (see Appendix C)19. By looking at the pole of these mode amplitudes, we can obtain the dispersion relation given as Eq. C1 in Appendix C. When the slab system is symmetric, i.e. ε1 = 1 [ε1 being the permittivity of the substrate of the metasurface, see Fig. 1(a)], the dispersion relation can be decomposed into anti-symmetric and symmetric modes. The dispersion relation for the anti-symmetric mode is (ε − 1)ky(ekyd + 1) eκd − 1 κ(eκd + 1) + ((ε + 1)ekyd + (ε − 1)) = 0 and for the symmetric mode (ε − 1)ky(ekyd − 1) eκd + 1 κ(eκd − 1) + ((ε + 1)ekyd − (ε − 1)) = 0. (9) (10) The dispersion relations are plotted in Fig. 2(d) for a metasurface parameterized by T = 50 nm, δmax = 10 nm and δmin = 0.5 nm. From Eqs. 9 and 10, it is clear that the dispersion relation reduces to the local case when κ → ∞ (β = 0)24, which is depicted with dashed lines for comparison with the nonlocal dispersion relation (solid lines). From Fig. 2(d), we see that both local and nonlocal dispersion relations have two bands, the lower band representing the symmetric mode while the upper band the anti-symmetric one. However, the nonlocal dispersion relation differs from the local case in its asymptotic behavior. For the conventional local dispersion relation, the anti-symmetric and symmetric modes asymptotically approach ωsp. On the contrary, in a nonlocal model both bands 19. In the large k asymptotically approach the longitudinal bulk mode ω = limit, ω ≈ βef f k, giving rise to a linear dispersion relation, see the solid line in Fig. 2(d). β2 ef f k2 + ω2 p (cid:113) 7 II.3. Field in real space and reflection coefficient After obtaining the field in k-space, we can derive the field in real space by means of a Fourier transform. The potential functions, Hz and ϕ, can then be expressed as follows. First, the magnetic field reads as, k2 pyx(ei(cid:82) y √ √ pyx(ei(cid:82) y −√ 0 kpy(y(cid:48))dy(cid:48) ± e−i(cid:82) y pyx)(ei(cid:82) y −√ 0 kpy(y(cid:48))dy(cid:48) ± e−i(cid:82) y k2 pyx + Λ−e k2 k2 0 kpy(y(cid:48))dy(cid:48)+iφ0) 0 kpy(y(cid:48))dy(cid:48) ± e−i(cid:82) y 1−eiφ0 , 0 kpy(y(cid:48))dy(cid:48)+iφ0) 1 0 kpy(y(cid:48))dy(cid:48)+iφ0) 1 1−eiφ0 , x < −d/2 1−eiφ0 , −d/2 < x < d/2 1 x > d/2 (11) Hz(x, y) =  where φ0 =(cid:82) h/2 i2πaΓ+e i2πa(Λ+e i2πaΓ−e −h/2 kpy(y(cid:48))dy(cid:48). Second, the electric potential inside the metal is given by, (Ω+eκpx − Ω−e−κpx)(ei(cid:82) y 0 kpy(y(cid:48))dy(cid:48) ∓ e−i(cid:82) y 0 kpy(y(cid:48))dy(cid:48)+iφ0) ϕ(x, y) = i2πasgn(kpy)sgn(y) ωε0ε 1 1 − eiφ0 (12) (13) where coefficients Γ±, Λ± and Ω± are the mode amplitudes in real space. From the potentials, the electric field can be calculated straightforwardly from Maxwell's equations. The energy loss of the system can then be obtained by calculating the power flow at the excitation point y = 0. This power flow is then modeled as an effective surface conductivity as17  σer = σx(cid:48) σmr = σy(cid:48) absσe0 ax mode absσm0 sin2 θin ay mode in which σe0 = Z−1 is the incident angle and we give the expression of σ(x,y)(cid:48) Eq. 13 is just the real part of the surface conductivity, whose imaginary part can be derived and σm0 = Z0 are the free space electric and magnetic conductivity, θin in Appendix B and C. Note that abs 0 using the Kramers-Kronig relations17,28,33,34 ∞(cid:90) −∞ σ(e,m)i = − 1 π P σ(e,m)r(s) s − ω ds = 1 π P ∞(cid:90) −∞ ln(cid:12)(cid:12)s − ω(cid:12)(cid:12) dσ(e,m)r(s) ds ds (14) Using the flat surface model17, the metasurface can be represented as a simple flat sheet with the electric and magnetic surface conductivities (σe, σm), as shown in Fig. 3. In Fig. 3(a), the component of the electric field along the y direction excites the ax mode, whose energy dissipation is modeled as an electric surface conductivity. In contrast, the component of electric field in the x direction excites the ay mode, whose associated energy dissipation is equivalent to that of a magnetic surface conductivity. 8 FIG. 3. Flat surface model. The energy dissipation by (a) ax/ (b) ay mode is modeled as an effective electric/magnetic surface conductivity. With this simple flat surface geometry, the reflection and transmission coefficients can be readily obtained as r = t = (cid:112) (cid:112) (cid:112) (cid:112) (cid:112) (cid:112) −4ε1σm + 4σeZ2 4ε1σm + 4σeZ2 0 cos θin 0 cos θin ε1 − sin2 θin − σeσmZ0 ε1 − sin2 θin + σeσmZ0 ε1 − sin2 θin + ε1σeσmZ0 cos θin − 4Z0 ε1 − sin2 θin + ε1σeσmZ0 cos θin + 4Z0 4ε1σm + 4σeZ2 0 cos θin ε1 − sin2 θin + σeσmZ0 ε1 − sin2 θin + ε1σeσmZ0 cos θin + 4Z0 2ε1Z0 cos θin(4 − σeσm) Under normal incidence (θin = 0), they can be written as √ ε1 − 1 + σe/σe0 √ ε1 + 1 + σe/σe0 r = √ 2 t = ε1 √ ε1 + 1 + σe/σe0 (cid:112) (cid:112) ε1 − sin2 θin + 4ε1Z0 cos θin ε1 − sin2 θin + 4ε1Z0 cos θin (15) (cid:112) ε1 − sin2 θin + 4ε1Z0 cos θin (16) (17) III. OPTICAL RESPONSE OF THE METASURFACE In order to study how nonlocality influences the response of the system, the optical response of the metasurface is compared against the local approximation in Fig. 4. The nonlocal parameter is taken as β = 1.27× 106 m/s as in Fig. 2, the grating period is T = 50 nm and the maximum thickness is δmax = 10 nm. Moreover, metasurfaces with different 9 FIG. 4. The reflection spectrum r2 (top row) and the real and imaginary components of the effective surface conductivity σe (middle and bottom rows respectively) of the metasurface under normal incidence are compared in the near-singular case (left column) and in the non-singular one (right column). The geometric parameters are T = 50 nm, the maximum gap δmax = 10 nm. The minimum gap δmin = 0.5 nm for the near-singular case, while δmin = 5 nm for the non-singular case. minimum gaps are considered in Fig. 4, where the near-singular case on the left column (δmin = 0.5 nm) is compared with a non-singular case on the right column (δmin = 5 nm). Under normal incidence, we only have electric surface conductivity, shown in Figs. 4(c)-(f). From the spectrum of surface conductivity, we can derive the reflection spectrum shown in Figs. 4(a) and 4(b). In order to check the accuracy of our analytic calculation, we carry out a comparison with exact numerical simulations implemented in the finite element solver Comsol Multiphysics. Excellent agreement between the two methods is shown in Fig. 8 in Appendix D. From the comparison between local and nonlocal spectra of the near-singular case in the left column of Fig. 4, we can observe a blue-shift of the peaks associated with the onset of nonlocality. Larger differences are observed in the high frequency regime, where the local 10 calculation gives a discrete spectrum, while the nonlocal calculation has a continuous one. However, these differences reduce in the case of non-singular metasurface [Fig. 4 (b,d,f)], where the nonlocal result nearly coincides with the local solution. Therefore, nonlocal effects are significantly more pronounced when the metasurface is more singular. FIG. 5. Nonlocal effects in the far field spectrum. (a) Reflectance as a functioun of frequency and grating period T for the metasurface under normal incidence. The geometric parameters are h = 23.6161d and α = 0.8040, which fixes the shape of metasurface when changing the grating period; (b) Reflectance as a function of frequency and minimum gap δmin for the metasurface under normal incidence. The geometric parameters are T = 50 nm and δmax = 10 nm, which is fixed when changing the minimum gap δmin. In both panels, the white dashed lines corresponds to the peak position of the local calculation. Nonlocality introduces a size-dependence in the response of the metasurface, otherwise scale invariant in the quasistatic limit. In Fig. 5(a), the dependence of reflection on the grating period T and frequency are plotted for the nonlocal case. Since T is a scaling factor, the shape of grating remains unchanged when T varies. In the local model (white dashed lines), the subwavelength character of the grating implies that the response of the system is predominantly quasistatic, and therefore scale-invariant. However, the introduction of nonlocal effects induces the clear blueshift that appears for smaller grating sizes, and which is more pronounced for higher frequencies. We should point out that our analytic theory makes use of the quasi-static approximation, which typically results in a loss of accuracy 11 for larger sizes due to coupling to radiation. However, radiative effects are not as relevant in singular plasmonic gratings as one may expect. While for localized plasmonic structures radiation damping becomes increasingly important for large particle sizes, which makes the quasistatic approximation inaccurate25, quasistatic calculations of singular plasmonic gratings with long periods yield very accurate results. The reason for this is that when the grating period increases, there are less periods per unit length, such that the total dipole moment does not increase19. A quantitative comparison with numerical results for larger- period gratings is provided in Fig. 8 in Appendix D, showing excellent agreement apart from minor discrepancies near the surface plasma frequency, for periods up to 100 nm. We then investigate the geometric parameter δmin, which quantifies the proximity to the singular limit, and the resulting nonlocal effects. The contour plot in Fig. 5(b) shows reflection as a function of the gap size δmin and frequency. When shrinking the gap size δmin, the number of visible peaks increases. This increase in resonances supported by the grating is a result of the hidden dimension at the singular point, which enables its spectral response to accommodate a higher density of states. In the local case, when δmin → 0 these resonances will merge towards a continuum16. However, nonlocality sets a barrier to this continuous limit, saturating the density of states of the metasurface. By comparing the nonlocal spectrum with the local solution (white dashed lines), a saturation of the spectrum merging is observed for the smaller gap sizes δmin. We now turn our attention to the mode profiles, focusing on how the onset of nonlocality affects the near-field response. By substituting the reflection coefficient into the expressions for the fields in real space (Eqs. 11 and 12), the field profile in the slab system is obtained. The resulting field distribution is subsequently obtained by mapping the field into the meta- surface frame. Fig. 6 shows the the near field response in the local case (panels (a)), and in the nonlocal one (panels (b)). The geometry of the metasurface studied here is parame- terized by T = 50 nm, the maximum gap δmax = 10 nm and the minimum gap δmin = 0.5 nm, which is the same as the near-singular metasurface in Fig. 4. From the left to the right column, the colour plots show the amplitude of the electric field E, and its x and y components, Ex and Ey respectively. In order to compare the local and nonlocal plots, a working frequency ω = 0.65ωp, below the surface plasmon frequency ωsp is chosen, so that only the symmetric mode is excited. From this comparison, we see that the introduction of nonlocality leads to a reduced number of SPP oscillations, since nonlocality plays the role of smearing out the near-singular point, which effectively widens the minimum gap size. 12 FIG. 6. The near field profile of a near-singular metasurface under normal incidence (a) without and (b) with nonlocality. The amplitude of the electric field E (left column) and the x and y components Ex and Ey (center and right column respectively) are shown. The presence of nonlocality reduces the number of SPP oscillations in the field profile of Ex and Ey. Note that we have zoomed in for the field E, whose maximum field enhancement factor is ∼ 45 for the local case, which reduces to ∼ 40 for the nonlocal one. The geometric parameters are T = 50 nm, the maximum gap δmax = 10 nm and the minimum gap δmin = 0.5 nm. Furthermore, in the plot of E we have zoomed in the region of interest, where the density of states is reduced as a result of nonlocality. This discrepancy will be more significant if we make the metasurface more singular by reducing the minimum gap size, effectively leading to saturation of the local density of states near the singular point. IV. CONCLUSION In conclusion, we have studied how the response of a singular metasurface consisting of a smooth modulated metallic slab with nearly-touching surfaces is affected by nonlocality. We showed how transformation optics can be deployed to provide an accurate analytic description of repulsive quantum effects as described by the hydrodynamic model, which was shown to oppose the realisation of a geometrical singularity, thereby blueshifting the 13 plasmonic bands of the metasurface. A convenient route towards this analytic solution was demonstrated, based on the representation of the grating as an effective current sheet, whose electric and magnetic surface conductivities were derived. This very general approach can in principle be applied to a wide range of metasurfaces, including surfaces with different kinds of singularities19. From the comparison between the local and nonlocal calculation, we observed a blueshift of all resonance peaks and a reduced density of states. However, these nonlocal effects are only important for a near-singular metasurface, pointing out that these structures offer a convenient route for accessing nonlocal features of the electron gas via far-field measure- ments. Recent advancements in nanofabrication make ultra-small gaps (∼ 1 nm) possible to realize in practice6,7. In such small scale features, nonlocal effects cannot be ignored, and our theory is able to accurately account for these deviations. V. ACKNOWLEDGEMENTS F.Y. acknowledges a Lee Family Scholarship for financial support. J.B.P. acknowledge funding from the Gordon and Betty Moore Foundation. E.G. acknowledges support through a studentship in the Centre for Doctoral Training on Theory and Simulation of Materials at Imperial College London funded by the EPSRC (EP/L015579/1). P.A.H. acknowledges funding from Funda¸cao para a Ciencia e a Tecnologia and Instituto de Telecomunica¸coes under projects CEECIND/03866/2017 and UID/EEA/50008/2019. Appendix A: Source representation In this appendix, we present the derivation of the incident, reflected and transmitted waves in the slab frame. Starting from plane waves in the metasurface frame, we apply the transformation equation (Eq. 1) and write the source fields in the slab geometry. We assume that the metasurface is subwavelength and that the spatial region of interest satisfies (cid:48) (cid:28) λ. For the incident and reflected waves on the right-hand side of the metasurface T (cid:28) x (cid:48) (cid:28) λ) under this assumption, (T (cid:28) x (cid:19)(cid:19) (cid:18) (cid:18) (cid:18) e− 2π ln T z w0 z = h 2π ≈ h 2π (cid:48) 1 /T + y0) + w0 a (cid:19) a(e2πz (cid:48) w0 a + ln 14 (A1) Hence, z (cid:48) ≈ − T 2π setting xs+ = h 2π ln w0 a , we have ln For the transmitted wave on the left side of singular surface we have T (cid:28) −x (cid:18)2πw0 (cid:18)2πw0 h h (z − h 2π ln w0 a (z − xs+) (cid:19) ) (cid:19) (cid:19) 2π T z e (cid:48) )−1 + 2π T z (cid:48) ) + e (cid:19) − w0 a (cid:19) w0 a e 2π T z (cid:48) z = h 2π ≈ h 2π ≈ h 2π z (cid:48) ≈ − T 2π ln ln ln ay0 (1 + 1 y0 (1 − 1 y0 (cid:18) 1 (cid:18) 1 (cid:32) (cid:18)1 + w0y0 (cid:18)2π ay0 ay0 ln ( h 2π ln h 1 + w0y0 ay0 (cid:18)2π h z (cid:48) ≈ T 2π ln (xs− − z)y0(1 + w0y0) (cid:33) (cid:19) y0(1 + w0y0) − z)y0(1 + w0y0) (cid:19) (A2) (A3) (cid:48) (cid:28) λ, so (A4) (A5) (A6) Hence, z (cid:48) ≈ T 2π ln By setting xs− = h 2π ln 1+w0y0 ay0 , we have (cid:18) (cid:18) (cid:18) (cid:18) (cid:18) With Eqs. A3 and A6, we have the source representation z = H0e−ik0xx H inc +ik0y y (cid:48) (cid:48) − −ik0x+k0y 2 ik0x − k0y ≈ H0e ≈ H0 1 + 1 + i 1 + i 2 k0xT 4π k0xT 4π = H0 = H0 T T 2π 2π ln(cid:0) π ln(cid:0) 2πw0 ln(cid:0) 2πw0 d z(cid:1) × e ln(cid:0) 2πw0 (cid:1)2 − i (cid:19) (cid:1)2 + h h h k0xT 4π ∞(cid:90) −∞ − −ik0x−k0y 2 T 2π ln(cid:0) π d z∗(cid:1) (z − xs+)(cid:1)(cid:19) (cid:18) ∞(cid:90) × 1 + e−kyx−xs+ ik0x + k0y 2 T 2π ky eiky ydky + −∞ e−kyx−xs+ ky ax ∞(cid:90) −∞ ln(cid:0) 2πw0 ∞(cid:90) h k0yT 4π −∞ (z∗ − xs+)(cid:1)(cid:19) sgn(ky) e−kyx−xs+ sgn(x − xs+)ky eiky ydky eiky ydky + aysgn(ky) e−kyx−xs+ sgn(x − xs+)ky eiky ydky (cid:19) (A7) for the incident wave, where ax = −i k0xT only need to change k0x into −k0x and obtain 4π H0 and ay = k0yT 4π H0. For the reflected wave, we H ref z = rH0eik0xx +ik0y y (cid:48) (cid:48) = rH0 = rH0 1 − i 1 − i k0xT 4π k0xT 4π ln(cid:0) 2πw0 ln(cid:0) 2πw0 h h (cid:1)2 + i (cid:19) (cid:1)2 k0xT 4π ∞(cid:90) − −∞ ∞(cid:90) −∞ e−kyx−xs+ ky eiky ydky + k0yT 4π e−kyx−xs+ ky rax eiky ydky + 15 ∞(cid:90) −∞ sgn(ky) e−kyx−xs+ sgn(x − xs+)ky eiky ydky (cid:19) raysgn(ky) e−kyx−xs+ sgn(x − xs+)ky eiky ydky (A8) ∞(cid:90) −∞ For the transmitted wave, we have 1 − ik ≈tH0 ≈tH0e z =tH0e−ik H tra (cid:18) (cid:18) (cid:18) ∞(cid:90) =tH0 =tH0 1 − i 1 − i − −∞ (cid:48) 0xx (cid:48) (cid:48) +ik0y y T 2π h ln (cid:18) 2π y0(1 + w0y0)(cid:1)2 − i (cid:19) y0(1 + w0y0)(cid:1)2 + (cid:48) 0x − k0y 2 (cid:48) 0xT k 4π (cid:48) 0xT k 4π ln(cid:0) 2π ln(cid:0) 2π h h taysgn(ky) e−kyx−xs− sgn(x − xs−)ky eiky ydky eiky ydky (A9) −ik0x+k0y 2 T 2π ln( 2π h (xs−−z)y0(1+w0y0)) × e T 2π ln( 2π h (xs−−z∗)y0(1+w0y0)) −ik0x−k0y 2 (cid:19)(cid:19) (cid:18) (xs− − z)y0(1 + w0y0) × 1 − ik (cid:19)(cid:19) (xs− − z∗)y0(1 + w0y0) (cid:18) 2π (cid:1)(cid:19) ln h T 2π (cid:48) 0x + k0y 2 ln(cid:0) z∗ − xs− z − xs− (cid:48) 0xT k 4π ∞(cid:90) −∞ lnz − xs−2 − k0yT 4π (cid:48) e−kyx−xs− k 0x k0x ky ax t Appendix B: Local calculation Without spatial dispersion in the metal, the field can be represented merely by a divergence-free magnetic field Hz, i.e. Eq. 6. By imposing the continuity condition at the two interfaces of Hz and Ey, the field in k-space is obtained. In the calculation of the excited SPP mode in real space, we have two approaches: the discrete-k method and the continuous-k method. For the continuous-k method, we calculate the field generated by a single monopole and then sum the field of all monopoles together. In the discrete-k method, the periodicity of the monopole array only excites SPPs with a specific value of the k-vector (k = ng where n is an integer and g = 2π h is the reciprocal lattice constant). a. Continuous-k method First, we consider two monopoles on the x-axis in the slab frame, one for incident and reflected waves at z = xs+, and the other for transmitted wave at z = xs−. Applying boundary conditions, we can obtain the mode amplitudes b+,− and c+,− in k-space, from which the field H 0 z generated by the two monopoles in real space can be obtained by inverse 16 Fourier transform. H 0 z (x, y) = = ∞(cid:90) −∞  H 0 z (x, ky)eikyydky k2 √ pyxeikpyy, √ i2πaΓ+e k2 pyx + Λ−e −√ i2πa(Λ+e pyxeikpyy, k2 i2πaΓ−e −√ k2 x < −d/2 pyx)eikpyy, −d/2 < x < d/2 x > d/2 (B1) where the coefficients Γ+,− and Λ+,− are the mode amplitudes in real space. Adding together the fields of all monopole pairs with (x = xs+ and x = xs−) at y = nh, we have Hz(x, y) = = (cid:88) n  z (x, y − nh) H 0 k2 √ pyx(eikpyy ± e−ikpyy+ikpyh) √ −√ k2 pyx + Λ−e pyx(eikpyy ± e−ikpyy+ikpyh) −√ k2 k2 i2πaΓ+e i2πa(Λ+e i2πaΓ−e 1 1−eikpy h , pyx)(eikpyy ± e−ikpyy+ikpyh) x < −d/2 1−eikpy h , −d/2 < x < d/2 1 (B2) 1 1−eikpy h , x > d/2 where the positive sign is chosen for ax, and the negative one for ay. From Maxwell equations, the corresponding electric field can be calculated as i ωε0ε ∂Hz ∂y Ex(x, y) = =  k2 ωε0 Γ+e −isgn(y) 2πakpy −isgn(y) 2πakpy −isgn(y) 2πakpy √ pyx(eikpyy ∓ e−ikpyy+ikpyh) √ ωε0ε (Λ+e −√ k2 pyx + Λ−e pyx(eikpyy ∓ e−ikpyy+ikpyh) −√ k2 k2 Γ−e ωε0ε1 1 1−eikpy h , 1 1−eikpy h , pyx)(eikpyy ∓ e−ikpyy+ikpyh) x < −d/2 1−eikpy h , −d/2 < x < d/2 1 x > d/2 (B3) Ey(x, y) = − i ∂Hz ωε0ε ∂x √ √ √ − 2πa  k2 ωε0ε1 ωε0ε ωε0 2πa 2πa = k2 k2 py py k2 √ pyx(eikpyy ± e−ikpyy+ikpyh) √ pyx − Λ−e −√ pyx(eikpyy ± e−ikpyy+ikpyh) −√ k2 k2 k2 Γ+e (Λ+e py Γ−e 1 1−eikpy h , pyx)(eikpyy ± e−ikpyy+ikpyh) x < −d/2 1−eikpy h , −d/2 < x < d/2 1 1 1−eikpy h , x > d/2 (B4) To calculate the energy dissipated by excited SPP mode, we evaluate the power flow at y = 0+. For the ax mode, we have √ pyx 1+eikpy h k2 √ i2πaxΓx+e 1−eikpy h , k2 pyx + Λx−e −√ i2πax(Λx+e pyx 1+eikpy h k2 1−eikpy h , i2πaxΓx−e Hz(x, y) =  17 −√ x < −d/2 1−eikpy h , −d/2 < x < d/2 pyx) 1+eikpy h k2 x > d/2 (B5) The electric field Ex is Ex(x, y) =  Re[Sy](cid:12)(cid:12)y=0+ (cid:20)(cid:18) 1 + eikpyh dx ∞(cid:90) 1 2 −∞ 4π2ax2 √ k2 pyx, √ Γx+e ωε0 k2 pyx + Λx−e −√ ωε0ε (Λx+e k2 pyx, Γx−e −i 2πaxkpy −i 2πaxkpy −i 2πaxkpy ωε0ε1 −√ k2 x < −d/2 pyx), −d/2 < x < d/2 x > d/2 (B6) Therefore, the total power absorbed by the ax mode is P ax abs = 2 (cid:19)∗(cid:18) kpy 1 − eikpyh √ = + + ωε0 Re Λx+2 (cid:0) 2Re[(cid:112)k2 i2Im[(cid:112)k2 kpy ε Λx+Λ∗ x− py] py] (eRe[ √ (eiIm[ k2 py]d − e −Re[ √ k2 py]d − e −iIm[ k2 py]d) + 2Re[(cid:112)k2 Γx+2 √ k2 py]d) + √ −Re[ e k2 py]d + py] Λx−2 −2Re[(cid:112)k2 −i2Im[(cid:112)k2 Λx−Λ∗ x+ py] kpy ε1 −Re[ √ (e py] −iIm[ (e 2Re[(cid:112)k2 Γx−2 √ py] e −Re[ √ √ k2 py]d k2 py]d − eRe[ √ py]d − eiIm[ k2 py]d) py]d)(cid:1)(cid:19)(cid:21) k2 k2  For the ay mode at y = 0+, we have √ k2 pyx, √ i2πayΓy+e k2 pyx + Λy−e −√ i2πay(Λy+e k2 pyx, Hz(x, y) = i2πayΓy−e  Ex(x, y) = −i 2πaykpy −i 2πaykpy −i 2πaykpy √ pyx 1+eikpy h k2 √ Γy+e 1−eikpy h , ωε0 k2 pyx + Λy−e −√ ωε0ε (Λy+e pyx 1+eikpy h k2 Γy−e 1−eikpy h , (cid:19)(cid:18) ωε0ε1 (cid:20)(cid:18) 1 + eikpyh 4π2ay2 Therefore, the power absorbed by ay mode is Γy+2 √ 2Re[(cid:112)k2 1 − eikpyh abs = P ay √ kpy ωε0 Re Λy+2 (cid:0) 2Re[(cid:112)k2 i2Im[(cid:112)k2 kpy ε Λy+Λ∗ y− py] + + py] (eRe[ √ (eiIm[ k2 py]d − e k2 py]d) + −Re[ √ k2 py]d − e −iIm[ k2 py]d) + −√ k2 x < −d/2 pyx), −d/2 < x < d/2 x > d/2 −√ x < −d/2 1−eikpy h , −d/2 < x < d/2 pyx) 1+eikpy h k2 x > d/2 √ −Re[ e k2 py]d + py] Λy−2 −2Re[(cid:112)k2 −i2Im[(cid:112)k2 Λy−Λ∗ y+ py] 2Re[(cid:112)k2 Γy−2 √ py] √ −Re[ e √ kpy ε1 (e py] −Re[ √ k2 py]d − eRe[ √ py]d − eiIm[ k2 py]d) py]d)(cid:1)(cid:19)(cid:21) k2 k2 −iIm[ (e (B7) (B8) (B9) k2 py]d By using the flat surface model17, the power absorption of the ax and ay modes can be modeled as an effective surface conductivity, see Fig. 3. The component of the electric field along the y direction excites the ax mode. For the near-sinusoidal metasurface in this paper, (B10) 18 the ax mode does not have a well-defined symmetry and becomes a mixture of both anti- symmetric and symmetric modes. This is because the two boundaries of the metasurface are not symmetric in relation to the source, resulting in both the anti-symmetric and the symmetric modes being excited. Likewise, the component of the electric field along the x direction excites ay mode, which is anti-symmetric dominantly. Following the same proce- dure we did for the wedge/groove singular metasurface, the effective surface conductivity is expressed as17  σer = σx(cid:48) σmr = σy(cid:48) absσe0 ax mode absσm0 sin2 θin ay mode (B11) and σm0 = Z0 are the free space electric and magnetic conductivity. The 0 k0T 2 Re f (kpy) kpy σ(x,y)(cid:48) abs = where σe0 = Z−1 primed parameter in the above equation following the power-flow method reads (cid:20) (cid:0) Λ (cid:113) (x,y)+2 k2 py] (cid:113) (cid:113) y−2 Γ (cid:48) k2 2Re[ py] (cid:113) (x,y)−2 (cid:48) k2 py] (cid:113) (cid:48)∗ (x,y)+ (cid:113) (x,y)+2 Γ (cid:48) k2 py] kpy ε1 Λ −2Re[ (cid:48) Λ (x,y)−Λ −i2Im[ 2Re[ (cid:48)∗ (cid:48) (x,y)+Λ (x,y)− k2 py] i2Im[ py]d − e−iIm[ k2 py]d − eRe[ k2 py]d − e−Re[ 2Re[ √ (e−iIm[ k2 py]d − eiIm[ (eiIm[ k2 + kpy ε (e−Re[ √ √ k2 py]d) + k2 py]d) + e−Re[ k2 py]d + √ e−Re[ k2 py]d Λ + (eRe[ √ √ √ k2 py] (cid:18) (cid:48) √ √ k2 py]d) py]d)(cid:1)(cid:19)(cid:21) k2 √ in which the primed coefficients Γ (cid:48) (x,y)± and Λ (cid:48) (x,y)± are related to Γ(x,y)± and Λ(x,y)± by17 (B12) and Γx± = (1 − r)Γ (cid:48) x± Λx± = (1 − r)Λ (cid:48) x± (cid:48) Γy± = (1 + r)Γ y± (cid:48) Λy± = (1 + r)Λ y± f (kpy) =  (cid:16) 1+eikpy h 1−eikpy h (cid:17)∗ 1+eikpy h 1−eikpy h 19 ax mode ay mode (B13) (B14) b. Discrete-k method If the field of one monopole is expressed as H 0 z (x, y), the total field can be written as where n is an integer, h is the period of the monopole source array and ∗ stands for convo- lution. Since Hz(x, y) is a periodic function, it can be written as a Fourier series δ(y − nh) (B15) Hz(x, y) = z (x, y − nh) H 0 = H 0 n (cid:88) z (x, y) ∗(cid:88) (cid:88) n Hz(x, y) = cneingy n Hz(x, y)e−ingydy (B16) (B17) (B18) z (x, y − nh)e−ingydy H 0 z (x, τn)e−ing(τn+nh)dτn H 0 z (x, τn)e−ingτndτn H 0 −h/2+nh z (x, y)e−ingydy H 0 gH 0 z (x, ng)eingy (cid:88) z (x, ky) ×(cid:80) n gH 0 z (x, ky) δ(ky − ng)eikyydky n gδ(ky − ng). where cn = = = = = 1 h 1 h 1 h 1 h 1 h −h/2 (cid:90) h/2 (cid:90) h/2 (cid:88) (cid:90) h/2+nh (cid:88) (cid:90) h/2+nh (cid:88) (cid:90) ∞ −h/2 −h/2+nh n n n Therefore, Hz(x, y) = = = gH 0 z (x, ng) −∞ (cid:88) (cid:90) ∞ n −∞ Hence the field representation in k-space is H 0 Hz(x, ky)eikyydky −∞ ∞(cid:90) a(cid:80) a(cid:80) a(cid:80)  Hz(x, y) = = Solving the above equation system, we can calculate the field distribution of the excited mode in real space by inverse Fourier transform. x < −d/2 n gΓn+engxeingy, n g(Λn+engx + Λn−e−ngx)eingy, −d/2 < x < d/2 n gΓn−e−ngxeingy, x > d/2 (B19) 20 where Γn± = b±(cid:12)(cid:12)ky=ng and Λn± = c±(cid:12)(cid:12)ky=ng. Then the electric field can be calculated as Ex(x, y) = i ∂Hz ∂y ωε0 ωε0ε − a − a − a ωε0ε ωε0ε1 (cid:80) (cid:80) x < −d/2 n ng2Γn+engxeingy, (cid:80) n ng2(Λn+engx + Λn−e−ngx)eingy, −d/2 < x < d/2 n ng2Γn−e−ngxeingy, x > d/2 = Ey(x, y) = − i ωε0ε − ia − ia = ωε0 ∂Hz ∂x (cid:80) (cid:80) n ng2Γn+engxeingy, x < −d/2 (cid:80) n ng2(Λn+engx − Λn−e−ngx)eingy, −d/2 < x < d/2 n ng2Γn−e−ngxeingy, x > d/2 ωε0ε ia ωε0ε1   The power absorption in one period is Pabs = ωε0Im[ε](Ex2 + Ey2)dxdy (cid:90) 1 2 slab a2Im[ε] ωε0ε2 a2Im[ε] ωε0ε2 = = (cid:88) (cid:88) n n d/2(cid:90) (cid:18)Λn+2 −d/2 2ng (cid:32)Λ (B20) (B21) (B22) (cid:19) (cid:33) n2g4h (Λn+2e2ngx + Λn−2e−2ngx)dx n2g4h (engd − e−ngd) + Λn−2 −2ng (e−ngd − engd) where the integration on y cancels the terms ei(n−n(cid:48))gy unless n = n(cid:48). From the power absorption in one period, we can derive an effective surface conductivity as the continuous-k method in Eq. B11 where σ(x,y)(cid:48) abs = k0T Im[ε] 8π2ε2 n2g4h (cid:48) n+2 2ng (engd − e−ngd) + (cid:48) Λ n−2 −2ng (e−ngd − engd) (B23) (cid:88) n The above two methods (discrete-k and continuous-k) both give the spectrum of surface conductivity. They are identical when the k-vector of SPP is large (see Fig. 7), but differ when the k-vector is small (near 0 and ωp frequencies). This difference comes from neglecting the branch cut when applying the residue theorem in the continuous-k method. Therefore, the discrete-k method is more accurate for the local calculation. Appendix C: Nonlocal calculation In the presence of nonlocality, we have two potential functions: the divergence-free mag- netic field Hz and the curl-free scalar potential function ϕ. By imposing the continuity of Hz 21 FIG. 7. Real part of the surface conductivity calculated with the discrete-k and continuous-k methods. and Ey, together with vanishing normal component of current density J at the metal inter- face (Jx = 0), we can obtain the mode amplitude in k-space. In the calculation, the WKB approximation is deployed. Under this approximation, the derivative of potential function on y is only for the phase factor eikyy. By looking at the pole of these mode amplitudes, we can arrive at the following dispersion relation: ky κ2 (ε − 1)((e2kyd − 1)(e2κd − 1)ky(ε − 1)ε1 + 2(1 + e2kyd + e2κd − 4e(ky+κ)d + e2(ky+κ)d)ε1κ + (e2kyd − 1)(e2κd + 1)ε(ε1 + 1)κ) + 4e(ky+κ)d(ε(ε1 + 1) cosh[kyd] + (ε2 + ε1) sinh[kyd]) sinh[κd] = 0 (C1) Then by using a Fourier transform, the potential function Hz and ϕ in real space can be expressed as Eqs. 11 and 12. The continuous-k method allows for a more accurate way to study the variation of k vector along the slab in the nonlocal calculation. However, neglecting the branch cut when using the residue theorem for the continuous-k method leads to some discrepancy when k is small. In comparison, the discrete-k method has avoided the residue theorem such that there is no error in the case of the local calculation, but it is unable to describe the variation of k vector along the slab in the nonlocal calculation. In view of the error of the continuous- k method and negligible nonlocal effect at low frequency, we utilize the local results from discrete-k method for low frequency and the nonlocal result from the continuous-k method for high frequency. Following the scenario in the continuous-k method, we evaluate the field at y = 0+ for 22 the ax mode Hz(x, y) =  √ i2πaxΓax √ + e i2πax(Λax −√ + e i2πaxΓax− e k2 pyx 1+eiφ0 1−eiφ0 , k2 pyx + Λax− e k2 pyx 1+eiφ0 1−eiφ0 , −√ x < −d/2 1−eiφ0 , −d/2 < x < d/2 k2 pyx) 1+eφ0 x > d/2 (C2) The electric field Ex is  ωε0 −i 2πaxkpy −i 2πaxkpy −i 2πaxkpy ωε0ε ωε0ε1 Γax + e √ k2 pyx, √ (Λax + e −√ Γax− e k2 pyx, Ex(x, y) = −√ k2 pyx) − i 2πaxsgn(kpy)κp ωε0ε x < −d/2 + eκpx + Ωax− e−κpx), −d/2 < x < d/2 (Ωax k2 pyx + Λax− e x > d/2 (C3) (C4) For the ay mode at y = 0+, we have  Hz(x, y) = √ i2πayΓay k2 pyx, √ + e pyx + Λay− e i2πay(Λay k2 −√ + e i2πayΓay− e k2 pyx, −√ k2 x < −d/2 pyx), −d/2 < x < d/2 x > d/2  Ex(x, y) = (cid:113) (cid:113) −(cid:113) (Λ ay− e ay + e ay + e k2 py x k2 py x 1+eiφ0 1−eiφ0 + Λ −(cid:113) , ay− e k2 py x 1+eiφ0 1−eiφ0 , ωε0 −i 2πay kpy Γ (−i 2πay kpy ωε0ε −i 2πay kpy ωε01 Γ k2 py x ) − i 2πay sgn(kpy )κp ωε0ε (Ω ay + eκpx + Ω ay− e−κpx)) 1+eiφ0 1−eiφ0 x < −d/2 , −d/2 < x < d/2 x > d/2 (C5) As opposed to the local case, the symmetric band and the anti-symmetric band overlap in frequencies between ωsp and ωp. In this case, multiple SPP modes can be excited at the same frequency. We consider three poles together in the calculation of power flow, which can be written as Pabs = 4π2a2 ωε0 (cid:21) f (kij)P ij abs (cid:88) (cid:20) Re i,j=1,2,3 23 (C6) where i and j stand for the index of three modes considered. The expression for P ij abs is √ −( √ k2 pi+ ∗ k2 pj ∗ e ) d 2 + √ k2 pi+ k2 pj ∗ ) d 2 − e (cid:113) (cid:113) Γ−iΓ∗ −j k2 pi + √ k2 pi+ k2 pj ∗ kpi ε1 √ −( k2 pj ∗ ) d 2 − e √ −( pi−√ k2 ∗ k2 pj ) d 2 ) + P ax i,j (cid:18) = kpi +j (cid:113) (cid:113) (cid:113) (cid:0) (cid:113) pi −(cid:113) Γ+iΓ∗ k2 pi + Λ+iΛ∗ k2 pi + Λ+iΛ∗ −j k2 k2 pj kpi ε (cid:113) sgn(kpi)κpi ε κpi −(cid:113) Ω+iΛ∗ −j k2 pj k2 pj k2 +j k2 pj ∗ (e( √ ∗ (e( pi−√ √ (cid:0) Ω+iΛ∗ (cid:113) ∗ (e(κpi−√ κpi + k2 pj k2 pj +j ∗ + + + + √ −( √ k2 pi+ ∗ ) d 2 k2 pj ∗ e k2 pj ) d 2 ) + −(cid:113) ∗ (e(−√ pi−√ k2 +j k2 pj −(cid:113) pi −(cid:113) Λ−iΛ∗ k2 pi + Λ−iΛ∗ −j k2 (cid:113) ∗ (e(−√ (cid:113) −κpi + ∗ (e(−κpi−√ Ω−iΛ∗ k2 pj k2 pj k2 pj +j ∗ √ k2 pi+ ∗ k2 pj ) d −(−√ √ k2 pi+ 2 − e −(−√ pi−√ k2 ∗ k2 pj ) d 2 ) ∗ k2 pj 2 )(cid:1) ) d ∗ k2 pj ) d 2 − e √ ∗ (e(κpi+ k2 pj ∗ ) d 2 − e √ −(κpi+ ∗ k2 pj ) d 2 ) + √ ∗ (e(−κpi+ ∗ k2 pj ) d 2 − e √ −(−κpi+ ∗ k2 pj ) d 2 ) −(κpi−√ ∗ k2 pj ) d 2 ) + ) d 2 − e Ω−iΛ∗ −j −κpi −(cid:113) k2 pj −(−κpi−√ ∗ k2 pj ) d ) d 2 − e 2 )(cid:1)(cid:19) and (cid:17)∗  (cid:16) 1+eiφ0j 1−eiφ0j 1+eiφ0i 1−eiφ0i ax mode ay mode f (kij) = Then, this power flow can be modeled as an effective surface conductivity, written as  σer = σx(cid:48) σmr = σy(cid:48) absσe0 ax mode absσm0 sin2 θin ay mode in which σ(x,y)(cid:48) abs = k0T 2 (cid:21) f (kij)σij abs (cid:88) (cid:20) Re i,j=1,2,3 24 (C7) (C8) (C9) (C10) where σij abs is written as σij abs (cid:18) = kpi (cid:48)∗ +j (cid:113) (cid:113) (cid:48) Γ +iΓ k2 pi + (cid:48) Λ +iΛ (cid:113) (cid:0) (cid:113) pi −(cid:113) (cid:48) Λ +iΛ k2 pi + (cid:48)∗ −j k2 pj k2 kpi ε (cid:113) sgn(kpi)κpi ε Ω (cid:48) +iΛ κpi −(cid:113) (cid:48)∗ −j k2 pj + + + + k2 pj k2 √ k2 pj (cid:48)∗ +j ∗ (e( √ ∗ (e( pi−√ (cid:0) Ω (cid:113) ∗ (e(κpi−√ (cid:48) +iΛ κpi + (cid:48)∗ +j k2 pj k2 pj ∗ √ −( √ k2 pi+ ∗ ) d 2 k2 pj ∗ e −(cid:113) ∗ (e(−√ pi−√ k2 Λ k2 pj (cid:48)∗ +j (cid:48) −iΛ Λ (cid:48) −iΛ k2 pi + (cid:48)∗ −j −(cid:113) pi −(cid:113) (cid:113) ∗ (e(−√ k2 (cid:113) pj (cid:48) −iΛ −κpi + ∗ (e(−κpi−√ (cid:48)∗ +j k2 pj k2 pj Ω ∗ k2 √ k2 pi+ ∗ k2 pj ) d ∗ k2 pj ) d 2 − e 2 − e −(−√ −(−√ √ k2 pi+ ∗ k2 pj pi−√ k2 ∗ k2 pj 2 )(cid:1) ) d ) d 2 ) √ ∗ (e(−κpi+ ∗ k2 pj ) d 2 − e √ −(−κpi+ ∗ k2 pj ) d 2 ) −(−κpi−√ ∗ k2 pj ) d ) d 2 − e 2 )(cid:1)(cid:19) √ −( √ k2 pi+ ∗ k2 pj ∗ e ) d 2 + √ k2 pi+ k2 pj ∗ ) d 2 − e (cid:113) (cid:48) −iΓ Γ (cid:48)∗ −j (cid:113) k2 pi + √ k2 pi+ k2 pj kpi ε1 √ −( k2 pj ∗ ) d 2 ) + k2 pj ∗ ) d 2 − e √ −( pi−√ k2 ∗ k2 pj ) d 2 ) + √ ∗ (e(κpi+ k2 pj ∗ ) d 2 − e √ −(κpi+ ∗ k2 pj ) d 2 ) + −(κpi−√ ∗ k2 pj ) d 2 ) + ) d 2 − e Ω (cid:48) −iΛ −κpi −(cid:113) (cid:48)∗ −j k2 pj Appendix D: Numerical verification (C11) FIG. 8. Numerical verification of the reflection spectrum for the near-sinusoidal metasurface under normal incidence. Top row: T = 50 nm, the maximum gap δmax = 10 nm and the minimum gap δmin = 0.5 nm; bottom row: T = 100 nm, the maximum gap δmax = 20 nm and the minimum gap δmin = 1 nm. We also verify our analytic calculation with finite element solver Comsol Multiphysics. 25 We have considered two sets of near-sinusoidal metasurfaces with period T = 50 nm and T = 100 nm, shown in Fig. 8. For a given metasurface, both local and nonlocal calculations are provided. The good agreement between analytic and numerical solutions confirms the validity of our theoretical framework. The nonlocal simulations are based on the RF and PDE modules in Comsol2,35, where the hydrodynamic system of equations in the metal are implemented as ∇ × ∇ × E = k2 β2∇(∇ · J) + ω(ω + iΓ)J = iωω2 0E + iωµ0J pε0E (D1) 1 T. Søndergaard, S. M. Novikov, T. Holmgaard, R. L. Eriksen, J. Beermann, Z. Han, K. Pedersen, and S. I. Bozhevolnyi, Nature communications 3, 969 (2012). 2 C. Cirac`ı, R. Hill, J. Mock, Y. Urzhumov, A. Fern´andez-Dom´ınguez, S. Maier, J. Pendry, A. Chilkoti, and D. Smith, Science 337, 1072 (2012). 3 A. Wiener, H. Duan, M. Bosman, A. P. Horsfield, J. B. Pendry, J. K. Yang, S. A. Maier, and A. I. Fernandez-Dominguez, Acs Nano 7, 6287 (2013). 4 R. Chikkaraddy, B. De Nijs, F. Benz, S. J. Barrow, O. A. Scherman, E. Rosta, A. Demetriadou, P. Fox, O. Hess, and J. J. Baumberg, Nature 535, 127 (2016). 5 F. Benz, M. K. Schmidt, A. Dreismann, R. Chikkaraddy, Y. Zhang, A. Demetriadou, C. Carnegie, H. Ohadi, B. De Nijs, R. Esteban, et al., Science 354, 726 (2016). 6 R. A. Maniyara, D. Rodrigo, R. Yu, J. Canet-Ferrer, D. S. Ghosh, R. Yongsunthon, D. E. Baker, A. Rezikyan, F. J. G. de Abajo, and V. Pruneri, Nature Photonics 13, 328 (2019). 7 A. Boltasseva and V. M. Shalaev, "Transdimensional photonics," (2019). 8 S. Raza, G. Toscano, A.-P. Jauho, M. Wubs, and N. A. Mortensen, Physical Review B 84, 121412 (2011). 9 J. A. Scholl, A. L. Koh, and J. A. Dionne, Nature 483, 421 (2012). 10 N. A. Mortensen, S. Raza, M. Wubs, T. Søndergaard, and S. I. Bozhevolnyi, Nature commu- nications 5, 3809 (2014). 11 G. Toscano, J. Straubel, A. Kwiatkowski, C. Rockstuhl, F. Evers, H. Xu, N. A. Mortensen, and M. Wubs, Nature communications 6, 7132 (2015). 12 C. L. Holloway, E. F. Kuester, J. A. Gordon, J. O'Hara, J. Booth, and D. R. Smith, IEEE Antennas and Propagation Magazine 54, 10 (2012). 26 13 A. V. Kildishev, A. Boltasseva, and V. M. Shalaev, Science 339, 1232009 (2013). 14 N. Yu and F. Capasso, Nature materials 13, 139 (2014). 15 H.-T. Chen, A. J. Taylor, and N. Yu, Reports on progress in physics 79, 076401 (2016). 16 J. Pendry, P. A. Huidobro, Y. Luo, and E. Galiffi, Science 358, 915 (2017). 17 F. Yang, P. A. Huidobro, and J. B. Pendry, Physical Review B 98, 125409 (2018). 18 E. Galiffi, J. Pendry, and P. A. Huidobro, EPJ Applied Metamaterials 6, 10 (2019). 19 F. Yang, Y.-T. Wang, P. A. Huidobro, and J. B. Pendry, Physical Review B 99, 165423 (2019). 20 E. Galiffi, P. Huidobro, P. Goncalves, N. Mortensen, and J. Pendry, arXiv preprint arXiv:1908.09320 (2019). 21 A. Ward and J. B. Pendry, Journal of modern optics 43, 773 (1996). 22 U. Leonhardt, science 312, 1777 (2006). 23 A. I. Fernandez-Dominguez, Y. Luo, A. Wiener, J. Pendry, and S. A. Maier, Nano letters 12, 5946 (2012). 24 A. Aubry, D. Y. Lei, A. I. Fern´andez-Dom´ınguez, Y. Sonnefraud, S. A. Maier, and J. B. Pendry, Nano letters 10, 2574 (2010). 25 A. Fern´andez-Dom´ınguez, A. Wiener, F. Garc´ıa-Vidal, S. Maier, and J. Pendry, Physical review letters 108, 106802 (2012). 26 M. Kraft, Y. Luo, S. Maier, and J. Pendry, Physical Review X 5, 031029 (2015). 27 J. Pendry, P. A. Huidobro, and K. Ding, Physical Review B 99, 085408 (2019). 28 C. Kittel, Introduction to solid state physics (8th edition) (Wiley New York). 29 A. D. Boardman, Electromagnetic surface modes (John Wiley & Sons, 1982). 30 L. Novotny and B. Hecht, Principles of nano-optics (Cambridge university press, 2012). 31 A. Moreau, C. Ciraci, and D. R. Smith, Physical Review B 87, 045401 (2013). 32 D. J. Griffiths, Introduction to quantum mechanics (2nd edition) (Cambridge University Press, 2016). 33 J. D. Jackson, Classical electrodynamics (John Wiley & Sons, 2012). 34 M. Dressel and G. Guner, Electrodynamics of Solids: Optical Properties of Electrons in Matter (Cambridge University Press, 2002). 35 G. Toscano, S. Raza, A.-P. Jauho, N. A. Mortensen, and M. Wubs, Optics express 20, 4176 (2012). 27
1404.2619	4	1404	2015-08-29T04:47:38	Generalizing the Fermi velocity of strained graphene from uniform to nonuniform strain	[ "cond-mat.mes-hall" ]	The relevance of the strain-induced Dirac point shift to obtain the appropriate anisotropic Fermi velocity of strained graphene is demonstrated. Then a critical revision of the available effective Dirac Hamiltonians is made by studying in detail the limiting case of a uniform strain. An effective Dirac Hamiltonian for nonuniform strain is thus reported, which takes into account all strain-induced effects: changes in the nearest-neighbor hopping parameters, the reciprocal lattice deformation and the true shift of the Dirac point. Pseudomagnetic fields are thus explained by means of position-dependent Dirac cones, whereas complex gauge fields appear as a consequence of a position-dependent Fermi velocity. Also, position-dependent Fermi velocity effects on the spinor wavefunction are considered for interesting cases of deformations such as flexural modes.	cond-mat.mes-hall	cond-mat	Generalizing the Fermi velocity of strained graphene from uniform to nonuniform strain M. Oliva-Leyva1∗ and Gerardo G. Naumis1, 2† 1. Departamento de F´ısica-Qu´ımica, Instituto de F´ısica, Universidad Nacional Aut´onoma de M´exico (UNAM), Apartado Postal 20-364, 01000 M´exico, Distrito Federal, M´exico and 2. School of Physics Astronomy and Computational Sciences, George Mason University, Fairfax, Virginia 22030, USA The relevance of the strain-induced Dirac point shift to obtain the appropriate anisotropic Fermi velocity of strained graphene is demonstrated. Then a critical revision of the available effective Dirac Hamiltonians is made by studying in detail the limiting case of a uniform strain. An ef- fective Dirac Hamiltonian for nonuniform strain is thus reported, which takes into account all strain-induced effects: changes in the nearest-neighbor hopping parameters, the reciprocal lattice deformation and the true shift of the Dirac point. Pseudomagnetic fields are thus explained by means of position-dependent Dirac cones, whereas complex gauge fields appear as a consequence of a position-dependent Fermi velocity. Also, position-dependent Fermi velocity effects on the spinor wavefunction are considered for interesting cases of deformations such as flexural modes. PACS numbers: 73.22.Pr, 81.05.ue, 77.65.Ly I. INTRODUCTION Since its discovery,1 graphene has been subject of many theoretical and experimental studies due to the unique array of its physical properties.2,3 In particular, the pe- culiar relation between its electronic and its mechanical properties has attracted growing interest.4 -- 7 The unusual long interval of elastic response8 makes it possible observ- able changes in the electronic structure, such as the open- ing of a bandgap9,10 or the merging of Dirac cones11,12. The strategy is to use strain engineering as a possibil- ity to guide the electrical transport in graphene-based devices.13 -- 17 In the literature there are different theoretical ap- proaches for studying the influence of lattice deforma- tions over the electronic properties of graphene. A quan- tum field theoretical approach in curved spaces has been alternatively used to predict electronic implications due to out-of-plane deformations.18 -- 20 Also, methods based solely on symmetry considerations have been applied to several problems of strained graphene.21 -- 24 In particu- lar, using group theory techniques, a symmetry analysis has been performed to construct all the possible terms in the low-energy effective Hamiltonian for graphene in presence of a nonuniform strain.24 More recently, a for- mulation based on concepts from discrete differential ge- ometry has shown how the atomistic structure of two- dimensional crystalline membranes dictates their me- chanical, electronic, and chemical properties.25 -- 28 Some particular analytical results are available for the case of uniaxial strain. For periodic strain, a complex fractal spectrum with gaps, localization transitions and topo- logical states are obtained.29 Nevertheless, the most popular theoretical framework for exploring the concept of strain engineering combines a nearest-neighbor tight-binding (TB) model and linear elasticity theory.4,5 As is well known, this TB-elasticity approach, in the continuum limit, predicts the existence of strain-induced pseudomagnetic fields. These pseudo- magnetic fields are described by means of a pseudovector potential A which is related to the strain tensor ¯ by5 Ax = (¯xx − ¯yy), β 2a Ay = − β 2a (2¯xy), (1) where a is the unstrained carbon-carbon distance and β is the electron Gruneisen parameter. Thus, nonuni- form local deformations of the lattice can be inter- preted as a pseudomagnetic field, given by B = ∇ × A (in units of /e) and perpendicular to the graphene sample.30 -- 34 Scanning tunneling microscopy studies in graphene nanobubbles have reported pseudo Landau lev- els, which are signatures of strain-induced pseudomag- netic fields.35,36 A discussion on the pseudomagnetic field theory was reactivated due to the explicit inclusion of the local lat- tice vectors deformation.37 Initially, this lattice correc- tion was supposed to produce an extra pseudovector po- tential (K0-dependent), but later on it was shown its physical irrelevance.38 -- 40 Particularly, in Ref. [26] the ab- sence of the extra pseudovector potential proposed in the theory was demonstrated in an explicit manner. Also, the consideration of the actual atomic positions in the TB Hamiltonian resulted in a more complete analysis on the position-dependent Fermi velocity for strained graphene.39 More recently, another correction has been identified as important in the derivation of the effec- tive low-energy Hamiltonian for deformed graphene, by pointing out that the effective Hamiltonian should be ex- panded around the true Dirac point and not around the unperturbed one.41 -- 44 The principal motivation of the present work is to determine the implications of the strain-induced Dirac point shift in the derivation of the appropriate anisotropic Fermi velocity. Moreover, we discuss a possi- ble generalization of the effective Dirac Hamiltonian for nonuniform in-plane deformations. For that end, we lay out our discussion on a basic principle: the theory for graphene under nonuniform strain should describe the particular case of a uniform strain. The paper is organized as follows. In Sec. II we dis- cuss the effective Dirac Hamiltonian for graphene under a uniform in-plane strain. By comparing with other ap- proaches available in the literature, we demonstrate the relevance of the expansion around of the true Dirac point. In Sec. III we report a generalized effective Dirac Hamil- tonian for graphene under a nonuniform in-plane strain, which reproduces the case of a uniform strain. In Sec. IV we summarize the results of our work. II. DIRAC EQUATION FOR UNIFORMLY STRAINED GRAPHENE: CRITICAL REVISION To illustrate the derivation of the effective Dirac Hamiltonian in presence of strain, we first consider graphene under a uniform strain. We use this partic- ular case as a benchmark to identify the goodness of any effective Dirac Hamiltonian for strained graphene, even if the deformation is nonuniform. In the case of a uniform strain, if a represents a general vector in the unstrained graphene lattice, its strained counterpart is given by the transformation a(cid:48) = ( ¯I + ¯) · a, (2) where ¯I is the 2× 2 identity matrix and ¯ is the position- independent strain tensor. As an import example, one can quote the deformation of the three nearest-neighbor vectors. Selecting the x axis along the graphene zigzag direction, the unstrained nearest-neighbor vectors are, √ ( δ1 = a 2 3, 1), δ2 = a 2 √ (− 3, 1), δ3 = a(0,−1), (3) whereas the strained nearest-neighbor vectors can be ob- tained from δ(cid:48) n = ( ¯I + ¯) · δn, see Fig. 1(a). On the other hand, a uniform strain distorts the recip- rocal space as well. From Eq. (2) follows that if b repre- sents a vector of the unstrained reciprocal lattice, its de- formed counterpart results b(cid:48) = ( ¯I + ¯)−1 · b (cid:39) ( ¯I − ¯)· b (see Fig. 1(b)). However, the high-symmetry points of the Brillouin zone are modified differently. For exam- ple, the high-symmetry point of the unstrained Bril- √ louin zone K0 = ( 4π , 0) moves to the new position 3a 3 (1 − ¯xx/2 − ¯yy/2,−2¯xy) under a uniform √ K = 4π 3a 3 strain.9 For computing the effective Dirac Hamiltonian we start from the nearest-neighbor TB Hamiltonian, † t(cid:48) x(cid:48)bx(cid:48)+δ(cid:48) na n + H.c., (4) H = −(cid:88) x(cid:48),n where x(cid:48) runs over all sites of the deformed A sublattice, † x(cid:48) is the creation operator for an electron on the A sub- a lattice at site x(cid:48) and bx(cid:48)+δ(cid:48) is the annihilation operator n 2 FIG. 1. (Color online) (a) Uniaxial stretching along the zigzag direction of a graphene sample. The zoom of the honeycomb lattice shows the unstrained δi (black, dashed) and strained δ(cid:48) i (red, solid) three nearest-neighbor vectors. (b) Unstrained (black, dashed) and strained (red, solid) first Brillouin zone for the same uniaxial zigzag strain. Note how the reciprocal lattice is contracted along the direction where the lattice is stretched. for an electron on the B sublattice at site x(cid:48) + δ(cid:48) n. The nearest-neighbor hopping parameters t(cid:48) n are modified due to the changes in intercarbon distance and fulfill an ex- n/a − 1)], where t is ponential decay, t(cid:48) the equilibrium hopping parameter.4,45 n = t exp[−β(δ(cid:48) Replacing the creation/annihilation operators with their Fourier expansions,46 we obtain that the Hamilto- nian in momentum space is given by † t(cid:48) ne−ik·( ¯I+¯)·δn a kbk + H.c., H = −(cid:88) (5) k,n and therefore, the closed dispersion relation for uniformly strained graphene is E(k) = ± ne−ik·( ¯I+¯)·δn t(cid:48) (6) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:88) n (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) . As has been documented in other works,9,42 the posi- tions of the minimum of energy, i.e., the KD Dirac points (E(KD) = 0) obtained from the previous equation, do not coincide with the K (K0) high-symmetry points of the strained (unstrained) Brillouin zone. This is illus- trated in Fig. 2. Eqs. (5) and (6) are the main ingredients of the avail- able effective Dirac Hamiltonians. As we will discuss be- low, the main differences come out from the reciprocal- space points used for the approximations. This is also (b)(a)TensionTensionillustrated in Fig. 2, where the idea is to understand how the Dirac cone moves and deforms as strain is applied. A. Effective Hamiltonian around K0 3 To be more precise, if one considers momenta close to the arbitrary reciprocal-space point G, i.e. k = G + q, the Hamiltonian (5) can be casted as The most popular expansion in the literature is to con- sider momenta close to the high-symmetry points of the unstrained Brillouin zone, G = K0.38 -- 40 In this case, Hamiltonian (8) can be written as (1 + iσzq · ( ¯I + ¯) · δn) (9) (cid:18) HG = − 3(cid:88) t(cid:48) n n=1 0 ei(G+q)·( ¯I+¯)·δn e−i(G+q)·( ¯I+¯)·δn 0 (cid:19) , (7) and expanding to first order in q and ¯, as will be used throughout the rest of the paper, we obtain HG (cid:39) − 3(cid:88) t(cid:48) n (cid:18) 0 eiG·δn (cid:19) e−iG·δn 0 n=1 × (1 + iσzG · ¯ · δn), (1 + iσzq · ( ¯I + ¯) · δn) (8) with σz being the diagonal Pauli matrix. In the liter- ature, one can find two kinds of expansions by making G = K0 or G = KD. This leads to two different effective Dirac Hamiltonians, as will be discussed in the following subsections. It is worth mentioning that such Hamiltonians are in fact trying to describe the deformation and movement of the Dirac cone from different points, as explained in Fig.2. Clearly, if one chooses a point which is not the true Dirac point of the strained system, the Hamiltonian will not display the proper symmetries associated with it. Furthermore, one can not pass from one Hamiltonian to the other by using a simple renormalization of the mo- mentum since the Taylor expansions used around each point are different. The Fermi velocity will be used to test these ideas. FIG. 2. (Color online) Sketch of the Dirac cone (red cone) movement as graphene is stretched along the zigzag direc- tion. Three important points are indicated in the reciprocal space, the original Dirac point K0, the high-symmetry point K of the strained reciprocal lattice and the true Dirac point KD. The gray Dirac cone is the image of the red Dirac cone for unstrained graphene. The unstrained reciprocal lattice is pictured with black dots. HK0 (cid:39) − 3(cid:88) t(cid:48) n (cid:18) 0 eiK0·δn (cid:19) e−iK0·δn 0 n=1 × (1 + iσzK0 · ¯ · δn). Using the following identity20,39 (cid:18) 0 eiK0·δn (cid:19) e−iK0·δn 0 σ · δn a σz, = i (10) (cid:18) n (cid:39) t t(cid:48) (cid:18) (cid:19) (cid:19) where σ = (σx, σy) are the non-diagonal Pauli matrices, and writing the three nearest-neighbor hopping parame- ters t(cid:48) n as 1 − β a2 δn · ¯ · δn , (11) the Hamiltonian (9) becomes (see Appendix A) HK0 (cid:39) vF σ · + vF σ · ( (2¯ + Tr(¯) ¯I) ¯I + ¯ − β 4 K0 · ¯ · σ(cid:48)) · q + vF σ · ¯ · K0, a 2 · q − vF σ · A (12) where vF = 3ta/2 is the Fermi velocity for unstrained graphene and σ(cid:48) = (−σz, σx). Here the A vector is given by Eq. (1) and as mentioned, is interpreted as a pseudo- magnetic vector potential when the strain is nonuniform. It is worth pointing out that the expression (1) was de- rived by taking the x axis parallel to the zigzag direction of the graphene lattice and considering a valley (K0) with index +1. In the following, we assume these conditions, unless stated otherwise. Hamiltonian HK0 contains a problem that is very easy to spot. Let us consider a simple isotropic stretch- ing of the lattice, which can be written as ¯ = ¯I. This strain is just a renormalization of the distance be- tween carbons. As a result, the new carbon-carbon distance under isotropic strain is a(cid:48) = a(1 + ) and the new hopping parameter to first order in strain is t(cid:48) = t(1 − β). Thus, the new Fermi velocity obtained straight away from the nearest-neighbor TB Hamiltonian F = 3t(cid:48)a(cid:48)/2 (cid:39) vF (1 − β + ) and therefore, the ef- is v(cid:48) F σ · q, which can not be fective Dirac Hamiltonian is v(cid:48) obtained from Eq. (12) to an isotropic strain. This triv- ial test confirms that HK0 is not appropriate to describe graphene under a uniform strain. Consequently, expan- sions around the high-symmetry points of the unstrained Brillouin zone lead to unsuitable effective Hamiltonians for strained graphene.44 B. Effective Dirac Hamiltonian around KD A second option is to derive an effective Dirac Hamiltonian by expanding (5) around the true Dirac points.41,42,44 In other words, to make G = KD. As reported in previous work,42 the actual positions of the KD Dirac points to the K0 point is given by KD (cid:39) ( ¯I + ¯)−1 · K0 + A, (13) as shown in Fiq. 2. The previous equation confirms the remark that KD coincides with K only for isotropic strain. Using Eq. (13) it is possible to obtain the proper ef- fective Dirac Hamiltonian by developing Eq. (5) around the Dirac points, k = KD + q. Following this approach one can derive that42 H = vF σ · ( ¯I + ¯ − β ¯) · q, (14) where two strain-induced contributions can be recog- nized. The β-independent term, vF σ · ¯ · q, is purely a geometric consequence due to lattice deformation and does not depend of the material as long as it has the same topology. On the other hand, the β-dependent term, −vF βσ · ¯ · q, is owing to the strain-induced changes in the hopping parameters and its contribution depends of the material since β varies depending on the material. For graphene, both contributions have the same order of magnitude. From Eq. (14) one can identify that the appropriate Fermi velocity tensor is given by ¯v = vF ( ¯I + ¯ − β ¯), (15) which consistently reproduces the anisotropic trans- port for uniformly strained graphene.47,48 For example, Eq. (15) yields the correct result vF (1 − β + ) ¯I when the strain is isotropic, ¯ = ¯I. Also, for the case of a uniaxial stretching, with ν being the Poisson ratio, from Eq. (15) one imme- diately obtains the known result9,49 ¯v = vF 0 1 − (1 − β)ν , (17) for the anisotropic Fermi velocity. This expression has been used to calculate and explain the experimentally observed modulation of the transmittance of strained graphene with respect to the polarization direction of the incoming light.49 -- 51 III. GENERALIZED DIRAC HAMILTONIAN FOR NONUNIFORMLY STRAINED GRAPHENE As mentioned in Sec. I, we base our discussion for nonuniformly strained graphene on the following basic (cid:19) 0 0 −ν ¯ = (cid:18)1 (cid:18)1 + (1 − β) 0 , (16) (cid:19) 4 principle: the theory for graphene under nonuniform strain must describe the particular case of a uniform strain. Therefore, one would expect that the effective Dirac Hamiltonian for nonuniformly strained graphene should reduce to the effective Dirac Hamiltonian for the case of a spatially uniform strain. However, none of the effective Dirac Hamiltonian reported in the literature for the case of a nonuniform strain reduce to Eq. (14). This is an inconsistency in the theory of the strain-induced pseu- domagnetic field, which is owing to expansions around points which are not the true Dirac points for strained graphene. Below, we give a proposal to solve the prob- lem. Unlike the case of uniform strain, a nonuniform strain breaks the crystal periodicity. This delicate issue de- pends upon the physical considered limit. For example, if the strain is periodic but with a wavelength compara- ble to the interatomic displacement, in certain cases one needs an infinite number of reciprocal vetors to build the wavefunction, so the present approach can not be made.29 Here we will assume that the strain modulation wavelength is much bigger than the interatomic distance, as well as the amplitude. Under such approximation, the problem is usually solved by starting from the uniform Hamiltonian and changing ¯ to ¯(r). The problem lies in the fact that now the Fermi veloc- ity ¯v(r) depends upon the position, and thus the term ¯vijqk breaks the hermiticity of resulting Hamiltonian. To assure hermiticity, the procedure made in previous works to generalize the Dirac Hamiltonian around the unstrained Dirac point K0, using the replacement,20,39 -- 41 ∂¯vij(r) . (18) ¯vijqk → ¯vij(r) −i ∂ ∂rk − i 2 ∂rk However, as we discussed previously, the strain- induced Dirac point shift must be considered in the derivation of the appropriate Fermi velocity. This issue can be solved by starting from the uniform Hamiltonian around the true Dirac point in the momentum space, and going to real space by means of the replacement44 (cid:18) (cid:19) ¯vijqk → ¯vij(r) −i ∂ ∂rk − K D k (r) − i 2 ∂¯vij(r) ∂rk , (19) (cid:19) (cid:18) where now we have introduced the explicit position- dependence of the Dirac point by denoting it as K D k (r). This approach corresponds to the general scheme of emergence of gravity and gauge fields in the vicinity of the Weyl, Dirac or Majorana points in the energy spectrum.52 -- 54 Thus, according to Eq. (19) and taking into considera- tion local rotations (see Appendix B), the effective Dirac Hamiltonian for nonuniform in-plane strain can be writ- ten as H = σ · ¯v(r) · (−i∇ − KD(r)) − vF σ · Γ, (20) where the position-dependent Fermi velocity tensor ¯v(r) is given by (cid:0) ¯I + ¯(r) − β ¯(r)(cid:1), ¯v(r) = vF (21) the Dirac point KD(r) by, KD(r) =(cid:0) ¯I − ¯(r) + ¯ω(r)(cid:1) · K0 + A(r), (22) and the vector field Γ as Γi = i 2vF ∂¯vij(r) ∂rj = i(1 − β) 2 ∂¯ij(r) ∂rj , (23) with an implicit sum over repeated indices. Let us make some important remarks about our effec- tive Hamiltonian (20). First of all, one can see that Eq. (20) reproduces the limiting case of a uniform strain in a consistent manner. This is the principal merit of Hamil- tonian (20) with respect to previous effective Hamiltoni- ans. At the same time, one can recognize a new position- dependent Fermi velocity tensor (Eq. (21)) as the main difference. This is a very important result because en- ables a more appropriate prediction of spatially-varying Fermi velocity. Nowadays, such effect of strain has been confirmed by experiments using scanning tunneling mi- croscopy and spectroscopy.55,56 In the approach carried out, the position-dependent Dirac point generates the pseudomagnetic fields. This fact can be seen by taking the rotational of the effective potential that appears in Eq. (20) which leads to the pseudomagnetic field, B = ∇ × KD(r), (24) but since ∇ × ((¯(r) − ¯ω(r)) · K0) = 0, the term (¯(r) − ¯ω(r)) · K0 does not contribute to the pseudomagnetic field. Therefore, the value of the pseudomagnetic field is given by B = ∇ × A(r), (25) which is exactly the same pseudomagnetic field that appears in other derivations.20,39 -- 41 Note that, the in- clusion of the local rotations tensor ¯ω(r) was neces- sary to demonstrate the physical irrelevance of the K0- dependent pseudovector potential. On the other hand, the complex gauge field Γ is owing to a position-dependent Fermi velocity and its presence guarantees the hermiticity of the Hamiltonian (20). Un- like A, Γ is a purely imaginary. Thus Γ can not be interpreted as a gauge field and will not give rise to Lan- dau levels in the density of states.39 However, it may have other physical consequences, such as pseudospin precession, i.e., electronic transitions between the two sublattices.39 At present, the experimental signatures of such complex gauge field Γ are open questions. ij = 1 2 (cid:19) , + ∂h ∂ri ∂h ∂rj (cid:18) ∂ui + ∂h ∂ri ∂rj 1 2 ∂uj ∂ri ∂h ∂rj 5 deformations. Volovik et. al.44 found a similar Hamilto- nian, but they used a parametrization thought for curved graphene, where the in-plane coordinates of atoms are identical to their coordinates in the unperturbed honey- comb lattice. The reason is that they were mainly in- terested in a differential geometry interpretation. Here we used the reference laboratory frame, which is more suitable to compare with experiments, because one must use this frame to describe the interaction with external probes or fields.39 However, once the equations of Volovik and Zubkov are written in the reference laboratory frame, Eq. (20) for in-plane deformations can be recovered. Likewise, one can take advantage of both approaches and to write a generalized effective Dirac Hamiltonian. For this end, in β-dependent terms of Eq. (20) one must replace the strain tensor ¯ with the generalized strain tensor = ¯ij + , (26) where u(r) and h(r) are in- and out-of-plane displace- ments respectively. Thus, finally, the generalized effec- tive Dirac Hamiltonian can be written as H = −iσ · ¯v(r) · ∇ − vF σ · A − vF σ · Γ, (27) where now the generalized position-dependent Fermi ve- locity tensor ¯v(r) results (cid:0) ¯I + ¯(r) − β (r)(cid:1), ¯v(r) = vF (28) with the corresponding complex vector field, Γi = i 2vF ∂¯vij(r) ∂rj = i 2 ∂¯ij(r) ∂rj − iβ 2 ∂ij(r) ∂rj , (29) whereas the pseudovector potential A is given by Ax = (xx − yy), β 2a Ay = − β 2a (2xy). (30) A simple exploration shows that our generalized Hamiltonian (27) reproduces our Hamiltonian for in- plane deformations (Eq. (20)) as well as the equations of Volovik and Zubkov,44 for out-of-plane displacements. Note that, we ignored the term(cid:0) ¯I− ¯(r)+ ¯ω(r)(cid:1)·K0 due to its demonstrated irrelevance. Consequently, the gen- eralized Hamiltonian (27) describes the particular case of a uniform strain which resolves an inconsistency of pre- vious effective Hamiltonians. A. Inclusion of out-of-plane deformations the spinor wavefunction B. Effects of position-dependent Fermi velocity on It is worth mentioning that a second check can be made to Eq. (20) by adapting an independent approach devel- oped by Volovik and Zbukov in Ref.[44] for out-of-plane Finally, let us now consider the effects of a position- dependent Fermi velocity tensor on the spinor wavefunc- tion of charge carriers. For this purpose, we consider the case of a out-of-plane deformation along the x axis given by h(x). Then from Eq. (26) it follows that the generalized strain tensor is, xx(x) = (∂xh(x))2 ≡ f (x)/β, 1 2 yy = xy = 0, (31) thus, one immediately obtains that A = (f (x)/(2a), 0), whereas ¯v(x) = vF , Γ = (−if(cid:48)(x)/2, 0). (32) (cid:18)1 − f (x) 0 (cid:19) 0 1 Taking into consideration that the resulting pseudo- magnetic field is zero (B = ∂xAy − ∂yAx), from Eq. (27) one can write the corresponding stationary Dirac equa- tion for the spinor wavefunction Ψ as (cid:0)−i(1 − f (x))∂x − ∂y + if(cid:48)(x)/2(cid:1)ψ2 = εψ1, (cid:0)−i(1 − f (x))∂x + ∂y + if(cid:48)(x)/2(cid:1)ψ1 = εψ2, (33) where the parameter ε is defined as ε ≡ E/(vF ), and If now one supposes that the spinor E is the energy. wavefunction is of the form Ψ = exp(ikyy)Φ(x) then the following differential equation system is obtained, (cid:0)(1 − f (x))∂x + ky − f(cid:48)(x)/2(cid:1)φ2 = iεφ1, (cid:0)(1 − f (x))∂x − ky − f(cid:48)(x)/2(cid:1)φ1 = iεφ2. (cid:105)(cid:18)c1 (cid:19) (cid:104)(cid:90) x ikx + f(cid:48)(x)/2 1 − f (x) dx , c2 Φ(x) = exp In order to recover the case of flat graphene in the appropriate limit one can cast the following ansatz: (34) (35) IV. CONCLUSIONS 6 In this work we revisited the effective Dirac Hamilto- nian for graphene under a uniform strain, starting from a tight-binding description. We simultaneously consid- ered three fundamental strain-induced contributions: the changes in the nearest-neighbor hopping parameters, the reciprocal lattice deformation and the true shift of the Dirac point. In particular, the Dirac point did not co- incide with the high-symmetry points of the strained re- ciprocal lattice. A detailed discussion about this last strain-induced effect demonstrates its relevance to ob- tain the appropriate Fermi velocity. Finally, we pre- sented a generalized effective Dirac Hamiltonian for the case of a nonuniform deformations. This new Hamilto- nian reproduces the case of uniform strain in the cor- responding limit, which was a missing issue in previ- ous approaches. Within the approach carried out, the strain-induced pseudomagnetic fields were obtained ow- ing to the floating character of the Dirac point KD(r), whereas complex gauge fields appeared as a consequence of a position-dependent Fermi velocity. Our expression (28) for the generalized position-dependent Fermi veloc- ity tensor is the main result in this paper. Also, we found closed analytical solutions for the spinor wavefunctions in cases of practical interest on which the Fermi velocity depends on the position. where c1 and c2 are constants. Consequently, the differ- ential system (34) becomes the algebraic system ACKNOWLEDGMENTS (ikx + ky)c2 = iεc1, (ikx + ky)c1 = iεc2, (36) which has infinite solutions if ε = ±(k2 y)1/2. There- fore, finally we find that the stationary Dirac equation (33) has as solution the spinor wavefunction x + k2 (cid:90) x ikx + f(cid:48)(x)/2 (cid:105)(cid:18) 1 (cid:19) (cid:104) Ψ(r) = A exp ikyy + 1 − f (x) dx , seiθ (37) where eiθ = (kx + iky)/ε, A is a normalization constant and s = ±1 denotes the conduction band and valence bands, respectively. A remarkable result follows from our solution (37): Ψ2 ∼ (1 − f (x))−1 (38) i.e. a position-dependent Fermi velocity induces a in- homogeneity in the carrier probability density. For ex- ample, in the interesting case of a flexural mode given (38) one get Ψ2 ∼ by h(x) = h0 cos(Gx), from Eq. (1 − h sin2(Gx))−1, where h = βh2 0G2/2. So that, the carrier probability density is minimum at the valleys and at the crests of the flexural mode. To end, let us point out that our findings can be easily extended to the case of an in-plane deformation (along the x) replacing β by β − 1. We specially thank M. Zubkov and G. Volovik for pointing out a mistake in a previous version of the manuscript. We also acknowledge conversations with J. E. Barrios and G. Murgu´ıa. This work was supported by UNAM-DGAPA-PAPIIT, project IN-102513. M.O.L ac- knowledges support from CONACYT (Mexico). G. Nau- mis thanks a PASPA schoolarship for a sabatical leave at the George Mason University, where this work has been completed. Appendix A In this section, the details of the calculations to de- rive the effective Hamiltonian around K0 are presented. Substituting Eqs. (10) and (11) into Eq. (9) we get 3(cid:88) n=1 HK0 (cid:39) − t (1 − β a2 δn · ¯ · δn)(i σ · δn a σz) × (1 + iσzq · ( ¯I + ¯) · δn)(1 + iσzK0 · ¯ · δn), HK0 (cid:39) − t and expanding to first order in strain, HK0 can be written as (cid:16) (i σ · δn 3(cid:88) (cid:17) a2 δn · ¯ · δn(iσzq · δn) a2 δn · ¯ · δn − β − β + iσzK0 · ¯ · δn − (K0 · ¯ · δn)(q · δn) 1 + iσzq · ( ¯I + ¯) · δn σz) n=1 a . (A1) Now we collect the contribution of each term of this expression, 3(cid:88) (i n=1 −t σ · δn a σz) = 0, (A2) 3(cid:88) n=1 t 3(cid:88) σ · δn (i − t = vF σ · ( ¯I + ¯) · q, n=1 a σz)(iσzq · ( ¯I + ¯) · δn) σ · δn a (i σz)( β a2 δn · ¯ · δn) = −vF σ · A, σ · δn 3(cid:88) n=1 (i t = −vF β 4 σz)( β a2 δn · ¯ · δn)(iσzq · δn) a σ · (2¯ + Tr(¯) ¯I) · q, (A3) (A4) (A5) 3(cid:88) (i n=1 −t σ · δn a σz)(iσzK0 · ¯ · δn) = vF σ · ¯ · K0, (A6) 3(cid:88) σ · δn t (i n=1 a = vF σ · ( a 2 σz)(K0 · ¯ · δn)(q · δn) K0 · ¯ · σ(cid:48)) · q, (A7) where σ(cid:48) = (−σz, σx) and the A vector is given by Eq. (1) if the x axis is selected parallel to the zigzag direction of the graphene lattice. Finally, taking into account the contribution of each term in Eq. (A1), given by Eqs. (A2)-(A7), the effective Hamiltonian around K0 has the form of our Eq. (12). 7 Appendix B In this section, we include the local rotations in the problem of strained graphene. Note that, under an atomic displacement field u(r), the strained nearest- neighbor vectors are given approximately by38 where ∇u is the displacement gradient tensor: [∇u]ij = n (cid:39) ( ¯I + ∇u) · δn, δ(cid:48) (cid:19) (cid:18) ∂ui (cid:18) ∂ui = + + ∂uj ∂ri 1 2 − ∂uj ∂ri ∂rj ∂ui ∂rj 1 2 ∂rj = ¯ij(r) + ¯ωij(r), (B1) (cid:19) , (B2) with ¯ω(r) being the rotation tensor, which is antisym- metric. A position-dependent rotation tensor ¯ω(r) de- scribes the local rotations associated to the displacement field, while if ¯ω is independent on the position, it repre- sents a lattice global rotation which does not have phys- ical implications. nearest-neighbor hopping parameters t(cid:48) dent on the ¯ω(r) tensor, Unlike the strained nearest-neighbor vectors, the three n do not depen- (cid:18) (cid:18) n (cid:39) t t(cid:48) (cid:39) t 1 − β 1 − β a2 δn · ∇u · δn a2 δn · ¯(r) · δn (cid:19) (cid:19) , , (B3) which is an expected result since the rotations do not affect the module of the nearest-neighbor vectors. Thus, one should expect that the ¯ω(r) tensor only appears in β-independent terms, i.e., in terms of purely geometric origin. For our purpose to include the local rotations, let us start with the Hamiltonian of strained graphene in k- momentum space,40 t(cid:48) n 0 eik·( ¯I+∇u)·δn e−ik·( ¯I+∇u)·δn 0 , (B4) (cid:18) H = − 3(cid:88) n=1 (cid:19) where ¯ and ¯ω are considered position-independent. In order to obtain the effective Dirac Hamiltonian one must consider momentum close to the Dirac point, k = KD + q. In this case KD can be casted as KD =(cid:2)( ¯I + ∇u)(cid:62)(cid:3)−1 · (K0 + A), (cid:39) ( ¯I − ¯ + ¯ω) · K0 + A, (B5) which is a generalization of Eq.(13). Substituting Eq. B5 into Eq. B4 and consistently expanding to first order in strain and q results in, (cid:18) (cid:18) 0 H = − 3(cid:88) = − 3(cid:88) n=1 n=1 t(cid:48) n t(cid:48) n e−i(K0·δn+q·( ¯I+∇u)·δn+A·δn) 0 (cid:19) , 0 ei(K0·δn+q·( ¯I+∇u)·δn+A·δn) (cid:19) eiK0·δn e−iK0·δn 0 (1 + iσzq · ( ¯I + ∇u) · δn)(1 + iσzA · δn). 8 (B6) Using once again the identity (10) and replacing t(cid:48) n with the expression (B3) the Hamiltonian (B6) becomes H = −t 1 + iσzq · ( ¯I + ∇u) · δn (cid:16) . a (i n=1 σz) (cid:17) σ · δn 3(cid:88) a2 δn · ¯ · δn(iσzq · δn) − (A · δn)(q · δn) − β a2 δn · ¯ · δn + iσzA · δn − β 3(cid:88) 3(cid:88) σz)(iσzq · ( ¯I + ∇u) · δn) σz)(iσzq∗ · δn), = −t = vF σ · q∗, with q∗ = ( ¯I + ∇u(cid:62)) · q, = vF σ · ( ¯I + ∇u(cid:62)) · q, a σ · δn σ · δn − t n=1 n=1 (i (i a The contribution of each term in the last equation is given by Eqs. (A2), (A4), (A5) and (B7) (B8) (B9) 3(cid:88) n=1 (i t = −vF β 4 σ · δn σz)(A · δn)(q · δn) a σ · (2¯ − Tr(¯) ¯I) · q, 3(cid:88) (i n=1 −t σ · δn a σz)(iσzA · δn) = vF σ · A. (B10) where it is worth mentioning that the contributions of the last two terms in Eq. (B7), Eqs. (A4) and (B10), cancel. ∗ [email protected] † [email protected] 1 K. S. Novoselov et al., Science 306, 666 (2004) 2 A. K. Geim, Science 324, 1530 (2009) 3 K. S. Novoselov, Rev. Mod. Phys. 83, 837 (Aug 2011), http://link.aps.org/doi/10.1103/RevModPhys.83.837 4 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (Jan (cid:19) (cid:18) After looking the contributions of each term, Eq. (B7) can be written as H = vF σ · ( ¯I + ∇u(cid:62) − β ¯) · q, (B11) where ∇u(cid:62) = ¯ − ¯ω. Hamiltonian (B11) can be consid- ered as the generalization of Eq. (14). Now to extend Eq. (B11) to the case of a nonuni- form strain we assume that ¯(r) and ¯ω(r) are position- dependent and pass to real space by means of the rule44 ¯vijqk → ¯vij(r) In consequence, we obtain that the effective Dirac − K D − i 2 ∂ ∂rk ∂¯vij(r) k (r) (B12) −i ∂rk . Hamiltonian for nonuniform in-plane strain is given by H = σ · ¯v(r) · (−i∇ − KD(r)) − vF σ · Γ, (B13) where (cid:0) ¯I + ¯(r) − ¯ω(r) − β ¯(r)(cid:1), KD(r) =(cid:0) ¯I − ¯(r) + ¯ω(r)(cid:1) · K0 + A(r), ¯v(r) = vF (B14) (B15) and Γi = i 2vF ∂¯vij(r) ∂rj = i(1 − β) 2 ∂¯ij(r) ∂rj − i 2 ∂ ¯ωij(r) ∂rj , (B16) with an implicit sum over repeated indices. Finally, we remove the dependence on ¯ω from Eqs. (B14) and (B16) carrying out the following local rotation of the pseu- dospinor ψ → exp( i 2 ¯ωxyσ3)ψ (cid:39) ψ + i 2 ¯ωxyσ3ψ, (B17) and as a consequence, Eq. (B13) takes the form of our Eq. (20). 2009), http://link.aps.org/doi/10.1103/RevModPhys. 81.109 5 M. A. H. Vozmediano, M. I. Katsnelson, and F. Guinea, Physics Reports 496, 109 ISSN 0370-1573, http://www.sciencedirect.com/science/article/pii/ S0370157310001729 (2010), 6 F. Guinea, Solid State Communications 152, 1437 ISSN 0038-1098, http://www.sciencedirect. (2012), com/science/article/pii/S0038109812002256 7 D. Zhan, J. Yan, L. Lai, Z. Ni, L. Liu, and Z. Shen, Advanced Materials 24, 4055 (2012), ISSN 1521-4095, http://dx.doi.org/10.1002/adma.201200011 8 C. Lee, X. Wei, J. W. Kysar, and J. Hone, Science 321, 385 (2008) 9 V. M. Pereira, A. H. Castro Neto, and N. M. R. Peres, Phys. Rev. B 80, 045401 (Jul 2009), http://link.aps. org/doi/10.1103/PhysRevB.80.045401 10 G. Cocco, E. Cadelano, and L. Colombo, Phys. Rev. B 81, 241412 (Jun 2010), http://link.aps.org/doi/10.1103/ PhysRevB.81.241412 11 M. O. Goerbig, J.-N. Fuchs, G. Montambaux, and F. Pi´echon, Phys. Rev. B 78, 045415 (Jul 2008), http: //link.aps.org/doi/10.1103/PhysRevB.78.045415 12 R. de Gail, J.-N. Fuchs, M. Goerbig, F. Pi´echon, and G. Montambaux, Physica B: Condensed Matter 407, 1948 (2012), ISSN 0921-4526, http://www.sciencedirect. com/science/article/pii/S0921452612000774 13 V. M. Pereira and A. H. Castro Neto, Phys. Rev. Lett. 103, 046801 (Jul 2009), http://link.aps.org/doi/10. 1103/PhysRevLett.103.046801 14 T. Low, Y. Jiang, M. Katsnelson, and F. Guinea, Nano Letters 12, 850 (2012), http://pubs.acs.org/doi/abs/ 10.1021/nl2038985 15 Y. Jiang, T. Low, K. Chang, M. I. Katsnelson, and F. Guinea, Phys. Rev. Lett. 110, 046601 (Jan 2013), http: //link.aps.org/doi/10.1103/PhysRevLett.110.046601 16 D. A. Gradinar, M. Mucha-Kruczy´nski, H. Schomerus, and V. I. Fal'ko, Phys. Rev. Lett. 110, 266801 (Jun 2013), http://link.aps.org/doi/10.1103/PhysRevLett.110. 266801 17 H. Hung Nguyen, V. Viet Nguyen, and P. Dollfus, Nan- otechnology 25, 165201 (2014), http://stacks.iop.org/ 0957-4484/25/i=16/a=165201 18 A. Cortijo and M. A. H. Vozmediano, EPL (Euro- physics Letters) 77, 47002 (2007), http://stacks.iop. org/0295-5075/77/i=4/a=47002 19 M. A. H. Vozmediano, F. de Juan, and A. Cortijo, Journal of Physics: Conference Series 129, 012001 (2008), http: //stacks.iop.org/1742-6596/129/i=1/a=012001 20 F. de Juan, M. Sturla, and M. A. H. Vozmediano, Phys. Rev. Lett. 108, 227205 (May 2012), http://link.aps. org/doi/10.1103/PhysRevLett.108.227205 21 J. L. Manes, Phys. Rev. B 76, 045430 (Jul 2007), http: //link.aps.org/doi/10.1103/PhysRevB.76.045430 22 R. Winkler and U. Zulicke, Phys. Rev. B 82, 245313 (Dec 2010), http://link.aps.org/doi/10.1103/ PhysRevB.82.245313 23 T. L. Linnik, Journal of Physics: Condensed Matter 24, 205302 (2012), http://stacks.iop.org/0953-8984/ 24/i=20/a=205302 24 J. L. Manes, F. de Juan, M. Sturla, and M. A. H. Voz- mediano, Phys. Rev. B 88, 155405 (Oct 2013), http: //link.aps.org/doi/10.1103/PhysRevB.88.155405 25 J. V. Sloan, A. A. P. Sanjuan, Z. Wang, C. Horvath, and S. Barraza-Lopez, Phys. Rev. B 87, 155436 (Apr 2013), http://link.aps.org/doi/10.1103/PhysRevB. 87.155436 26 S. Barraza-Lopez, A. A. Pacheco Sanjuan, Z. Wang, and M. Vanevi´c, Solid State Communications 166, 7075 (Jul 2013) 27 A. A. Pacheco Sanjuan, M. Mehboudi, E. O. Har- riss, H. Terrones, and S. Barraza-Lopez, ACS Nano 9 8, 1136 (2014), http://pubs.acs.org/doi/abs/10.1021/ nn406532z 28 A. A. Pacheco Sanjuan, Z. Wang, H. Pour Imani, M. Vanevi´c, and S. Barraza-Lopez, Phys. Rev. B 89, 121403 (Mar 2014), http://link.aps.org/doi/10.1103/ PhysRevB.89.121403 29 G. G. Naumis and P. Roman-Taboada, Phys. Rev. B 89, 241404 (Jun 2014), http://link.aps.org/doi/10.1103/ PhysRevB.89.241404 30 F. Guinea, M. I. Katsnelson, and A. K. Geim, Nat Phys 6, 30 (2010) 31 Z. Qi, D. A. Bahamon, V. M. Pereira, H. S. Park, D. K. Campbell, and A. H. C. Neto, Nano Let- ters 13, 2692 (2013), http://pubs.acs.org/doi/abs/10. 1021/nl400872q 32 M. Neek-Amal and F. M. Peeters, Phys. Rev. B 85, 195446 (May 2012), http://link.aps.org/doi/10.1103/ PhysRevB.85.195446 33 R. Carrillo-Bastos, D. Faria, A. Latg´e, F. Mireles, and N. Sandler, Phys. Rev. B 90, 041411 (Jul 2014), http: //link.aps.org/doi/10.1103/PhysRevB.90.041411 34 Z. Qi, A. L. Kitt, H. S. Park, V. M. Pereira, D. K. Campbell, and A. H. Castro Neto, Phys. Rev. B 90, 125419 (Sep 2014), http://link.aps.org/doi/10.1103/ PhysRevB.90.125419 35 N. Levy, S. A. Burke, K. L. Meaker, M. Panlasigui, A. Zettl, F. Guinea, A. H. C. Neto, and M. F. Crom- mie, Science 329, 544 (2010), http://www.sciencemag. org/content/329/5991/544.abstract 36 J. Lu, A. C. Neto, and K. P. Loh, Nat Commun 3, 823 (May 2012), http://dx.doi.org/10.1038/ncomms1818 37 A. L. Kitt, V. M. Pereira, A. K. Swan, and B. B. Goldberg, Phys. Rev. B 85, 115432 (Mar 2012), http://link.aps. org/doi/10.1103/PhysRevB.85.115432 38 A. L. Kitt, V. M. Pereira, A. K. Swan, and B. B. Goldberg, Phys. Rev. B 87, 159909(E) (Apr 2013), http://link. aps.org/doi/10.1103/PhysRevB.87.159909 39 F. de Juan, J. L. Manes, and M. A. H. Vozmediano, Phys. Rev. B 87, 165131 (Apr 2013), http://link.aps.org/ doi/10.1103/PhysRevB.87.165131 40 M. R. Masir, D. Moldovan, and F. Peeters, Solid State Communications 175 -- 176, 76 (2013), ISSN 0038-1098, http://www.sciencedirect.com/science/article/pii/ S0038109813001555 41 H. T. Yang, Journal of Physics: Condensed Matter 23, 505502 (2011), http://stacks.iop.org/0953-8984/23/ i=50/a=505502 42 M. Oliva-Leyva and G. G. Naumis, Phys. Rev. B 88, 085430 (Aug 2013), http://link.aps.org/doi/10.1103/ PhysRevB.88.085430 43 M. Zubkov and G. Volovik, arXiv:1308.2249 http:// arxiv.org/abs/1308.2249 44 G. Volovik and M. Zubkov, Annals of Physics 340, 352 (2014), ISSN 0003-4916, http://www.sciencedirect. com/science/article/pii/S0003491613002558 45 R. M. Ribeiro, V. M. Pereira, N. M. R. Peres, P. R. Briddon, and A. H. Castro Neto, New J. Phys. 11, 115002 (2009), http://stacks.iop.org/1367-2630/11/ i=11/a=115002 46 C. Bena and G. Montambaux, New J. Phys. 11, 095003 (2009), http://stacks.iop.org/1367-2630/11/ i=9/a=095003 47 M. Oliva-Leyva and G. G. Naumis, Journal of Physics: Condensed Matter 26, 125302 (2014), http://stacks. 48 M. Oliva-Leyva and G. G. Naumis, Journal of Physics: Condensed Matter 26, 279501 (2014), http://stacks. iop.org/0953-8984/26/i=27/a=279501 49 V. M. Pereira, R. M. Ribeiro, N. M. R. Peres, and A. H. Castro Neto, EPL 92, 67001 (2010), http://stacks.iop. org/0295-5075/92/i=6/a=67001 50 G.-X. Ni, H.-Z. Yang, W. Ji, S.-J. Baeck, C.-T. Toh, J.-H. Ahn, V. M. Pereira, and B. Ozyilmaz, Advanced Materials 26, 1081 (2014), ISSN 1521-4095, http://dx.doi.org/10. 1002/adma.201304156 iop.org/0953-8984/26/i=12/a=125302 51 M. Oliva-Leyva and G. G. Naumis, 2D Materials 2, 025001 (2015), http://stacks.iop.org/2053-1583/2/i= 2/a=025001 52 C. D. Froggatt and H. B. Nielsen, Origin of Symmetry 10 (World Scientific, Singapore, 1991) 53 G. E. Volovik, The Universe in a Helium Droplet (Claren- don Press, Oxford, 2003) 54 P. Horava, Phys. Rev. Lett. 95, 016405 (Jun 2005), http: //link.aps.org/doi/10.1103/PhysRevLett.95.016405 55 H. Yan, Z.-D. Chu, W. Yan, M. Liu, L. Meng, M. Yang, Y. Fan, J. Wang, R.-F. Dou, Y. Zhang, Z. Liu, J.-C. Nie, and L. He, Phys. Rev. B 87, 075405 (Feb 2013), http: //link.aps.org/doi/10.1103/PhysRevB.87.075405 56 W.-J. Jang, H. Kim, Y.-R. Shin, M. Wang, S. K. Jang, M. Kim, S. Lee, S.-W. Kim, Y. J. Song, and S.-J. Kahng, Carbon 74, 139 (2014), ISSN 0008-6223, http://www.sciencedirect.com/science/article/pii/ S0008622314002619
1603.01941	1	1603	2016-03-07T05:25:22	Density Functional Theory based Study of Chlorine Doped WS2-metal Interface	[ "cond-mat.mes-hall" ]	Investigation of a TMD-metal interface is essential for the effective functioning of monolayer TMD based field effect transistors (FETs). In this work, we employ Density Functional Theory (DFT) calculations to analyze the modulation of the electronic structure of monolayer WS2 with chlorine doping and the relative changes in the contact properties when interfaced with gold and palladium. We initially examine the atomic and electronic structures of pure and doped monolayer WS2 supercell and explore the formation of mid gap states with band splitting near the conduction band edge. Further we analyze the contact nature of the pure supercell with Au and Pd. We find that while Au is physiosorped and forms n-type contact, Pd is chemisorped and forms p-type contact with a higher valence electron density. Next, we study the interface formed between the Cl-doped supercell and metals and observe a reduction in the Schottky barrier height (SBH) in comparison to the pure supercell. This reduction found is higher for Pd in comparison to Au which is further validated by examining the charge transfer occurring at the interface. Our study confirms that Cl doping is an efficient mechanism to reduce the n-SBH for both Au and Pd which form different types of contact with WS2.	cond-mat.mes-hall	cond-mat	Density Functional Theory based Study of Chlorine Doped WS2-metal Interface Anuja Chanana and Santanu Mahapatra1, a) Nanoscale Device Research Laboratory, Department Of Electronic Systems Engineering, Indian Institute of Science, Bengaluru, Karnataka 560012, India (Dated: 27 September 2018) Investigation of a TMD-metal interface is essential for the effective functioning of monolayer TMD based field effect transistors (FETs). In this work, we employ Density Functional Theory (DFT) calculations to analyze the modulation of the electronic structure of monolayer WS2 with chlorine doping and the relative changes in the contact properties when interfaced with gold and palladium. We initially examine the atomic and electronic structures of pure and doped monolayer WS2 supercell and explore the formation of mid gap states with band splitting near the conduction band edge. Further we analyze the contact nature of the pure supercell with Au and Pd. We find that while Au is physiosorped and forms n-type contact, Pd is chemisorped and forms p-type contact with a higher valence electron density. Next, we study the interface formed between the Cl-doped supercell and metals and observe a reduction in the Schottky barrier height (SBH) in comparison to the pure supercell. This reduction found is higher for Pd in comparison to Au which is further validated by examining the charge transfer occurring at the interface. Our study confirms that Cl doping is an efficient mechanism to reduce the n-SBH for both Au and Pd which form different types of contact with WS2. 2,3, WSe2 5 and MoTe2 After the fabrication of FET using monolayer MoS2,1 2D layered transition metal dichalcogenides(TMDs) have gar- nered enormous attention in the electron devices commu- 4, nity. Apart from MoS2, other TMDs as such WS2 6 are also explored as channel material for MoSe2 FET's. In the absence of efficient doping techniques, these transistors exhibit SBH at source/drain contact which leads to low ON current. Tremendous efforts are dedicated to reduce the contact resistance at TMD-metal interface by employing different techniques both theoretically and experimentally.7 -- 12 However, most of these efforts are focused towards MoS2 and WSe2 and a minimum study is devoted to other TMD's- metal contact interfaces. A study using ballistic MOSFET model reveals that WS2 outperforms all other TMD's13. Ex- perimental reports of WS2 device fabrication2 and Cl doping technique for reducing WS2-metal contact resistance are also reported14. However, theoretical investigations of WS2 metal contact interface using first principles is still lacking in the lit- erature. Since first principles are extensively used to analyze the graphene-metal15,16, MoS2-metal17 and WSe2-metal18, it is believed that it will efficiently describe the contact nature with other TMD's as well. For WS2, chlorine doping, which is done by replacing sulfur atoms is the first ever method demon- strated experimentally to reduce the WS2-contact resistance, exhibiting both high drain current and field-effect mobility.14 In this work we employ density function theory (DFT) to study the electronic structure of the interface between WS2 and one physiosorped (Au) and one chemisorped (Pd) metal. We then examine the effect of chlorine doping by substitut- ing sulfur atoms to address the SBH mitigation at WS2-metal interface. We start by analyzing the electronic dispersion character- istics of pure and Cl-doped WS2 and explore the shifts in the energy bands with respect to Fermi level (Ef ) and fur- ther analyze the mid gap states formed near the conduction a)[email protected] band (CB) edge. To preserve the stability in the 5x5 WS2 su- percell, the number of chlorine dopants substituting the sulfur atoms is kept one. We perform the formation energy calcula- tions (EF orm) to find the stability of the chlorine doped struc- ture. Next, this supercell is interfaced with <111> cleaved surfaces of Au and Pd respectively and the contact nature is studied. The Schottky barrier height (SBH) is evaluated us- ing the projected bandstructure and the density of states. The charge transfer across the interface is analyzed using valence electron density and charge density difference. Electron local- ization function (ELF) is employed to study the localization of electrons at the WS2-metal interface. A thorough examination of all the above analysis leads to a conclusion that n-type dop- ing of WS2 using chlorine as an effective substitute, results in lowering of n-SBH at the WS2-metal interface. We carry out the DFT simulations employing Atomistix Tool Kit (ATK)19 with Local Density Approximation (LDA)- Perdew Zung (PZ)20 as the exchange correlation to investigate the WS2 (pure and doped) - metal (Au and Pd) interfaces. The pseudopotentials of various elements (Tungsten, Sulfur, Chlorine, Gold and Palladium) as generated by Hartwingster- Goedecker-Hutter21 based on the concept of fully relativistic all electron calculation are used for the present study. The orbital contribution in Tier 4 basis set is found to be appro- priate for generating the band gap of monolayer WS2. It is worth noting that in transition metals (e.g W and Pd) the over- lap between core and valence electrons is substantial. The commoonly used double zeta polarized (DZP) basis func- tion generated using FHI pseudopotential does not consider the semicore electrons and might not yield results of con- siderable accuracy. On the other hand HGH pseudopoten- tials include semi core electrons and thus are expected to pro- duce the electronic structure of the TMD-metal interface pre- cisely. However, their usage significantly enhances the com- putational budget since they require a larger number of basis functions as compared to DZP. The density mesh cut off is 75 Hartree with a Monkhorst Pack k-point22 sampling of 9x9x1 mesh. The convergence criteria of the self consistent loop is set at a value of 10−5 Hartree. We first confirm the band gap of WS2 unit cell and find the value to be 2.07 eV, nearly consistent with the ear- lier reports23,24. A 5x5 supercell is formed using the opti- mized unit cell which is computationally appropriate to study the doping similar to the one reported here25. The inter- face strain between the pure and doped WS2 supercell with <111> cleaved surface of gold and palladium is found to be 1.3% and 2.1% respectively. Doping is achieved by re- placing one sulfur atom with chlorine atom in the supercell which amounts to 2% of the total sulfur atoms. For evaluat- ing EF orm for sulfur substitution, X2 dimers of each Cl2 and S2 are taken and then the total energy of the substitutional atoms and host atoms in the formation energy equation26 are calculated. Four layers of <111> cleaved surface of metal are then interfaced with the WS2 supercell and is found relevant to study the MoS2-metal interface.27 To minimize false interactions between the periodic interfaced geometries, vacuum length of 20 A is considered. A force optimiza- tion of (0.001eV/ A) is performed for both unit cell and su- percell (pure and doped) of the WS2 using limited memory BroydenFletcher-GoldfarbShannon (LBFGS) method. The volume of the structure is allowed to change using a stress optimization of (0.001eV/ A−3) in both unit cell and supercell of WS2. Figure 1 (a) shows the stable atomic structure of chlorine doped 5x5 supercell of WS2. We perform formation energy calculations to find the thermodynamically stable position for sulfur substitution by chlorine atoms with EF orm values as 2.92 eV. This supercell is interfaced with <111> cleaved sur- face of Au and Pd respectively. While interfacing the lattice constant of WS2 is fixed and the metal slab is subjected to match with it. Figure 1 (c) and (d) shows the ball and stick model of the doped 5x5 WS2-Au and WS2-Pd interface. Total energy calculations reveal 2.7 A and 2.2 A as the equilibrium interlayer separation between the WS2 (pure and doped) with Au and Pd metal slabs. The binding energy values obtained per sulfur atom is -0.18 eV (pure) and -0.19 eV (doped) for gold and -0.39 eV (pure) and -0.4 eV (doped) for palladium re- spectively. We calculate the work function of pure and doped WS2 and metals Au and Pd and are shown in Figure 1(f). Substitution by chlorine doping amounts to 2% of total sul- fur atoms which leads to a very high doping concentration in 5x5 supercell of WS2. However, to realize the substitution by considering the experimental doping concentration value of 6 x 1011 cm−2 studied here Ref.(14), a much bigger supercell is required leading to further increase in computational burden. To reduce the computational cost and study the effect of dop- ing on SBH reduction we limit ourselves with a 5x5 supercell. The electronic structures of both pure and doped WS2 along with the density of states are compared in Figure 2 to high- light the effect of doping. The band gap of pure WS2 is 2.07 eV. The positions of CB and Valence Band (VB) is verified with PDOS kept along the side of bandstructure. We see that substitution of single chlorine dopant shifts the energy bands towards the EF and creates midgap states in the vicinity of EF . The number of midgap states is 2 , one of which splits into two bands between the high symmetry points of brillouin 2 FIG. 1. (a) Chlorine substituted 5x5 supercell of WS2 presented us- ing the ball-stick model. Atomic model of the pure WS2 5x5 su- percell (b),(c) and doped supercell (d),(e) interfaced with <111> cleaved surface of face centered cubic (FCC) metals Gold and Pal- ladium. The configuration is stable when the interlayer distance is 2.7 A for gold and 2.2 A for palladium. dint denotes the equilib- rium distance with minimum binding energy along the z-direction. (f) Work function values of pure and doped WS2 along with the met- als Au and Pd calculated using DFT simulations. zone shown in the inset of Figure 2 (c). The two mid gap states have significant contribution in the PDOS as well. The main contributor of these states is chlorine which is shown in inset of Figure 2 (d) with purple color. The energy level is 0.184 eV for the mid gap state near the conduction band min- ima (CBM). The splitting of bands leads to two energy levels 0.028 eV and 0.020 eV respectively. These defect states orig- inate from the hybridization of the Cl 3p and the W 5d states. There exists one unpaired extra electron due to n-type dop- ing by chlorine. As tungsten lies in the 6th period of periodic table, chlorine, which has an extra negative charge, leads to the energy splitting of W d-orbitals and thus contribute to an extra mid gap state. The band gap in doped WS2 is also differ- ent as compared to the pure supercell (by 0.05 eV), which is attributed to the hybridization and change in the atomic struc- ture due to chlorine substitution. We then perform simulations to evaluate the nature of con- tact formed with WS2 (pure and doped) with Au and Pd. The Schottky barrier height is assessed using the electronic struc- ture and DOS of projected WS2. Valence electron density and charge density difference are used to estimate the charge trans- fer across the interface. The localisation of electrons at the WS2-metal interface is studied using the ELF. Both p-type and n-type SBH of projected WS2 (pure and doped) with Au and Pd are shown in Figure 3 (a)-(d). Since the interface involves two different atomic structures, the na- ture band gap is lost and projected band structure comprises 3 FIG. 4. 1D projection of average valence electron density along the z-direction for pure (a) WS2-Au and (b) WS2-Pd interface. The value of valence electron density is evaluated between the nearest sulfur atomic layer of WS2 and the nearest layer of metal formed at the WS2-metal interface geometry and is shown by black horizontal lines. The position of atoms are shown by black dotted lines. energy values quoted earlier confirm this nature. Moreover, the band gap regime also consists of less interface states in Au as compared to Pd (Figure 3 (a,c)). For a pure WS2-Au interface, the amount of complexity in the bandstructure is less, but for a doped interface it is very high. This happens because the atoms in the doped supercell already hybridize with the substitutional impurity. The dif- ference is visible in Figure 3 (b) and (d) with respect to Fig- ure 3 (a) and (c). It further makes the determination of band gap edges difficult. However, density of states along with the electronic structure of projected WS2, can be used to identify the CBM and valence band maxima (VBM) as demonstrated in Ref.8 The difference between the CBM and VBM of all the interfaces differs from the band gap of pure WS2 by 0.04 eV. Strong interface hybridization causes emergence of more mid gap states for both the pure and doped supercells as vis- ible in Figure 3 (c)-(d) for Pd due to its chemisorption nature with WS2. We observe that Au shows a n-type SBH and Pd shows p-type SBH when interfaced with WS2. The values of SBH are 0.78 eV and 0.76 eV respectively. In the complete band structure shown by grey lines, we see that the for Au the band lines are more dispersed in comparison to Pd. This in- dicates strong bonding of WS2 with Pd. Moreover, below the Fermi level, we see a complete metallization for Pd leading to the formation of p-type contact. Here, we see a similarity in the nature of contact between WS2 and WSe2 while form- ing an interface with Au and Pd.18 With the introduction of n-type impurity, the n-SBH exhibits a reduction for both met- als. The n-SBH reduction attained is 0.09 eV for Au and 0.11 eV for Pd. We see the reduction in the n-SBH is higher for Pd in comparison to Au. Pd showing a p-type contact with WS2 also exhibits a reduction in the n-SBH and increase in p-SBH. This validates the observation that the chlorine is ef- fective in reducing n-SBH for both the type of metals forming p-type and n-type SBH with WS2, and is equally competent for chemisorped and physiosorped metal interface with WS2. A slight variance in the band gap value of WS2 2.07 (for pure WS2 bandstructure) - 2.1 and 2.09 (projected WS2 bandstruc- ture) is also observed due to heterogeneous atomic interface. We also evaluate the valence electron density for WS2- metal interface as shown in Figure 4. The value of valence FIG. 2. Projected bandstructure and DOS of (a)-(b) pure WS2 and (c)-(d)chlorine doped WS2. PDOS placed along the side verifies the position of CB and VB. Doping with chlorine leads to appearance of mid gap states in the WS2 band gap and their contribution is stud- ied using the DOS. The number of mid gap states is 2 for doped WS2.The inset in (c) shows the exact position of CBM and the mid gap states. The purple line in (d) shows the chlorine contribution in PDOS and is highlighted in the inset. The contribution of mid gap states is confirmed by the PDOS and peaks in PDOS occurs at the same energy level as in bandstructure. CB, VB, EF , and Eg denotes conduction band edge, valence band edge, Fermi Level and band gap respectively. Fermi level is at zero energy. Dashed brown lines joins the CBM and VBM in bandstructure and DOS. FIG. 3. Projected electronic bandstructure of WS2-metal (Au and Pd) interface for (a)-(c) pure (b)-(d) doped supercells. The projected bandstructure (wine color lines) is superimposed on the total band- structure WS2-metal systems(grey lines). Au forms an n-type SB contact whereas Pd forms a p-type SB contact. φSB,P and φSB,N are the p-type and n-type Schottky barrier heights. Eg is the total band gap measured from the CBM and VBM. EF is the Fermi Level. of interface states. The origin of these states lies in the com- plex electronic hybridization occurring at the WS2-metal in- terface. Au is found to be physiosorped with WS2 where as Pd is chemisorped. Au, being an s-electron metal shows less hy- bridization as compared to Pd which is a d-electron metal and this effect is similar to MoS2-metal interface27. The binding 4 FIG. 5. 1D projection of average charge density difference along the z-direction for pure WS2 with (a) Au (b) Pd interface geometry. To study the charge transfer at the interface thoroughly, we show only single layer of gold atoms out of four. The relative positions of atoms are shown by dotted lines along the y axis. The filled green area indi- cates the accumulation region and the yellow indicates the depletion region. Values on the y-axis show that the charge accumulation in Pd is nearly double as compared to Au. electron density is calculated as the minimum electron den- sity at the metal-semiconductor interface. Higher values of electron density at the interface implies a better electron injec- tion efficiency. The values with Au are for pure 0.10727 A−3 and 0.10816 A−3 for doped supercell. For pure and doped supercell with Pd they are 0.21888 A−3 and 0.2212 A−3 re- spectively. The high values for Pd indicate that more charge transfer has occurred when it is interfaced with WS2. More- over the difference in the density values of doped and pure interface is higher for Pd as compared to Au. To substantiate the observations made in the above analy- sis, we examine the charge density difference of the respec- tive geometries. Figure 5 shows the average charge density difference of WS2-metal interface for (a) Au and (b) Pd. A dipole gets created at the interface since both accumulation and depletion regions exist. Both the regions are shown by dissimilar colors to highlight the difference between them. At the junction between the nearest sulfur atom and metal atom, two peaks exist for depletion region and one peak for accu- mulation region. Moreover the accumulation shows a higher charge difference in Pd as compared to Au. A higher deple- tion regime implies that the charge carriers are repelled back from the surface and evinces a probability of least transmis- sion across the interface. On the other hand, a higher accumu- lation regime implies that more charge is transferred across the interface. Hence we observe more hybridization for Pd rather than Au. The area calculated between the nearest S and metal atom using the methodology exhibited in Ref.8 is 0.03181 x 10−3 for Au and 0.06031 x 10−3 for Pd respec- tively and thus affirms the charge transfer is high for Pd. For pure and doped supercell-metal interface, the charge density difference curve shows a similar nature and the difference in areas for pure and doped case is higher for Pd. Because of this difference, a greater reduction on n-SBH is observed in Pd with respect to Au. To study the localization of electrons at the WS2-metal in- terface, we study ELF which is dimensionless quantity. It is defined as the possibility of finding an electron in the neighborhood of a reference electron. Its value ranges from FIG. 6. 1D projection of average electron localization function along the z-direction of pure WS2 with (a) Au (b) Pd contact. The positions of atoms are shown by black dotted lines along the y-axis. The blue line indicates the value of ELF of the nearest sulfur atom in WS2 monolayer. 0<ELF<1, with the maximum limit ELF=1 corresponding to prefect localization and ELF=0.5 corresponding to electron- gas like pair probability28. Figure 6 shows the value of ELF for WS2-Au and WS2-Pd interface. The value obtained for Pd is less in comparison to Au. This implies that Au is more localized than Pd indicating a better bonding for WS2-Pd in- terface. It is interesting to note that for TMD channel based tran- sistors, in a top contact device structure, two types of inter- faces exist. One interface is between the TMD-metal and the other is between the TMD underneath the metal contact and TMD forming the channel18. Here we find the SBH formed at the first interface. For calculating the SBH at the second interface, transport simulations need to be conducted along the device length. The effective SBH combining both the interfaces can then be determined using LDOS distri- bution. However, carrying transport simulations employing pseudopotentials with basis functions having higher orbital contribution requires powerful computing system, as the num- ber of atoms increase manifold from bulk-configuration to device-configuration. Nevertheless while discussing SBH and the associated contact resistance in the context of ON current of the MOSFET, the gate current is kept high and the channel is inverted. In this condition, the CBM of the TMD channel region (not touching the metal) can be assumed to be lower than CBM of the other TMD regime. Hence, the SBH evalu- ated in this work can be considered as the effective SBH. In the above study we investigate the contact nature of pure and Cl doped monolayer WS2 with Au<111> and Pd<111>. First the electronic structures of pure and doped supercells are investigated. We find that band energies near to the conduc- tion band edge align close to Fermi level with the formation of mid gap states and band splitting. Further, the pure and doped optimized supercell are adsorbed on the metals. Inter- face with pure supercell determines that Au is physiosorped with WS2 and has an n-type SBH while Pd is chemisorped with WS2 having a p-type SBH. Adsorption with Cl doped su- percell shows a reduction of n-type SBH for both the metals. All the results are validated by studying charge redistribution at the WS2-metal interface. The above understandings may further contribute to explore other TMD-metal interfaces, as well as in investigating various possible dopants for chalco- genide substitution. ACKNOWLEDGMENTS The work was supported by Science Engineering and Research Board -Department of Science and Technol- ogy (SERB DST), Govt. India, under Grant no. SR/S3/EECE/0151/2012. The authors would like to thank Quantumwise support staff for their useful discussions. of 1B. Radisavljevic, A. Radenovic, J. Brivio, V. Giacometti, and A. Kis, "Single-layer mos2 transistors," Nature nanotechnology 6, 147 -- 150 (2011). 2W. Sik Hwang, M. Remskar, R. Yan, V. Protasenko, K. Tahy, S. Doo Chae, P. Zhao, A. Konar, H. (Grace) Xing, A. Seabaugh, and D. Jena, "Transis- tors with chemically synthesized layered semiconductor ws2 exhibiting 105 room temperature modulation and ambipolar behavior," Applied Physics Letters 101, 013107 (2012). 3X. Liu, J. Hu, C. Yue, N. D. Fera, Y. Ling, Z. Mao, and J. Wei, "High performance field-effect transistor based on multilayer tungsten disulfide," ACS Nano 8, 10396 -- 10402 (2014). 4S. Das and J. Appenzeller, "Wse2 field effect transistors with enhanced ambipolar characteristics," Applied Physics Letters 103, 103501 (2013). 5S. Larentis, B. Fallahazad, and E. Tutuc, "Field-effect transistors and in- trinsic mobility in ultra-thin mose2 layers," Applied Physics Letters 101, 223104 (2012). 6N. R. Pradhan, D. Rhodes, S. Feng, Y. Xin, S. Memaran, B.-H. Moon, H. Terrones, M. Terrones, and L. Balicas, "Field-effect transistors based on few-layered -mote2," ACS Nano 8, 5911 -- 5920 (2014). 7W. S. Leong, X. Luo, Y. Li, K. H. Khoo, S. Y. Quek, and J. T. Thong, "Low resistance metal contacts to mos2 devices with nickel-etched-graphene electrodes," ACS nano 9, 869 -- 877 (2014). 8A. Chanana and S. Mahapatra, "Theoretical insights to niobium-doped monolayer mos2 gold contact," Electron Devices, IEEE Transactions on 62, 2346 -- 2351 (2015). 9S. Das, H.-Y. Chen, A. V. Penumatcha, and J. Appenzeller, "High perfor- mance multilayer mos2 transistors with scandium contacts," Nano Letters 13, 100 -- 105 (2013). 10J. Kang, W. Liu, and K. Banerjee, "High-performance mos2 transistors with low-resistance molybdenum contacts," Applied Physics Letters 104, 093106 (2014). 11M. R. Laskar, D. N. Nath, L. Ma, E. W. Lee II, C. H. Lee, T. Kent, Z. Yang, R. Mishra, M. A. Roldan, J.-C. Idrobo, et al., "P-type doping of mos2 thin films using nb," Applied Physics Letters 104, 092104 (2014). 12J. Suh, T.-E. Park, D.-Y. Lin, D. Fu, J. Park, H. J. Jung, Y. Chen, C. Ko, C. Jang, Y. Sun, R. Sinclair, J. Chang, S. Tongay, and J. Wu, "Doping 5 against the native propensity of mos2: Degenerate hole doping by cation substitution," Nano Letters 14, 6976 -- 6982 (2014). 13L. Liu, S. Bala Kumar, Y. Ouyang, and J. Guo, "Performance limits of monolayer transition metal dichalcogenide transistors," Electron Devices, IEEE Transactions on 58, 3042 -- 3047 (2011). 14L. Yang, K. Majumdar, H. Liu, Y. Du, H. Wu, M. Hatzistergos, P. Y. Hung, R. Tieckelmann, W. Tsai, C. Hobbs, and P. D. Ye, "Chloride molecular doping technique on 2d materials: Ws2 and mos2," Nano Letters 14, 6275 -- 6280 (2014). 15P. A. Khomyakov, G. Giovannetti, P. C. Rusu, G. Brocks, J. van den Brink, and P. J. Kelly, "First-principles study of the interaction and charge transfer between graphene and metals," Phys. Rev. B 79, 195425 (2009). 16C. Gong, G. Lee, B. Shan, E. M. Vogel, R. M. Wallace, and interfaces," K. Cho, Journal of Applied Physics 108, -- (2010). "First-principles study of metalgraphene 17I. Popov, G. Seifert, and D. Tom´anek, "Designing electrical contacts to mos2 monolayers: A computational study," Phys. Rev. Lett. 108, 156802 (2012). 18J. Kang, W. Liu, D. Sarkar, D. Jena, and K. Banerjee, "Computational study of metal contacts to monolayer transition-metal dichalcogenide semi- conductors," Phys. Rev. X 4, 031005 (2014). v.15.beta 19"Atomistix Quantumwise ToolKit ," http://quantumwise.com/. 20J. P. Perdew and A. Zunger, density-functional approximations Phys. Rev. B 23, 5048 -- 5079 (1981). "Self-interaction for many-electron correction to systems," 21C. Hartwigsen, S. Goedecker, and J. Hutter, "Relativistic separable dual- space gaussian pseudopotentials from h to rn," Phys. Rev. B 58, 3641 -- 3662 (1998). 22H. J. Monkhorst and J. D. Pack, "Special points for brillouin-zone integra- tions," Physical Review B 13, 5188 (1976). 23H. Jiang, "Electronic band structures of molybdenum and tungsten dichalcogenides by the gw approach," The Journal of Physical Chemistry C 116, 7664 -- 7671 (2012). 24A. Kumar and P. Ahluwalia, "Electronic structure of transition metal dichalcogenides monolayers 1h-mx2 (m= mo, w; x= s, se, te) from ab-initio theory: new direct band gap semiconductors," The European Physical Jour- nal B 85, 1 -- 7 (2012). 25K. Dolui, I. Rungger, C. Das Pemmaraju, and S. Sanvito, "Pos- sible doping strategies for mos2 monolayers: An ab initio study," Phys. Rev. B 88, 075420 (2013). 26C. G. Van de Walle and J. Neugebauer, "First-principles calcu- to iii-nitrides," lations Journal of Applied Physics 95, 3851 -- 3879 (2004). and impurities: Applications for defects 27C. Gong, L. Colombo, R. M. Wallace, and K. Cho, "The unusual mecha- nism of partial fermi level pinning at metal -- mos2 interfaces," Nano letters 14, 1714 -- 1720 (2014). 28a. D. Becke and K. E. Edgecombe, "A simple measure of electron localiza- tion in atomic and molecular systems," J. Chem. Phys. 92, 5397 (1990).
1008.4868	1	1008	2010-08-28T13:54:22	A Graphene Quantum Dot with a Single Electron Transistor as Integrated Charge Sensor	[ "cond-mat.mes-hall", "quant-ph" ]	We have developed an etching process to fabricate a quantum dot and a nearby single electron transistor as a charge detector in a single layer graphene. The high charge sensitivity of the detector is used to probe Coulomb diamonds as well as excited spectrum in the dot, even in the regime where the current through the quantum dot is too small to be measured by conventional transport means. The graphene based quantum dot and integrated charge sensor serve as an essential building block to form a solid-state qubit in a nuclear-spin-free quantum world.	cond-mat.mes-hall	cond-mat	A Graphene Quantum Dot with a Single Electron Transistor as Integrated Charge Sensor Ling-Jun Wang§, Gang Cao§, Tao Tu§, Hai-Ou Li§, Cheng Zhou§, Xiao-Jie Hao† , Zhan Su§, Guang- Can Guo§, Guo-Ping Guo§, and Hong-Wen Jiang† §Key Laboratory of Quantum Information, University of Science and Technology of China, Chinese Academy of Sciences, Hefei 230026, People’s Republic of China † Department of Physics and Astronomy, University of California at Los Angeles, California 90095, USA. * Corresponding authors. Email: [email protected], [email protected]. ABSTRACT: We have developed an etching process to fabricate a quantum dot and a nearby single electron transistor as a charge detector in a single layer graphene. The high charge sensitivity of the detector is used to probe Coulomb diamonds as well as excited spectrum in the dot, even in the regime where the current through the quantum dot is too small to be measured by conventional transport means. The graphene based quantum dot and integrated charge sensor serve as an essential building block to form a solid-state qubit in a nuclear-spin-free quantum world. KEYWORDS: graphene, quantum dot, single electron transistor, integrated charge sensor 1 Graphene has attracted a lot of research interest because of its unique electronic properties which make it a promising candidate for future nanoelectronics. 1–3 Because of the weak spin-orbit coupling and largely eliminated hyperfine interaction in graphene, it is highly desirable to coherently control the spin degree of freedom in graphene nanostructures for spin-based quantum computation. Recently, there was a striking advance on experimental production of graphene single or double quantum dots, 4–7 which is an important first step towards such promise. The measurement of individual electrons or its spins in GaAs quantum dots (QDs) have been realized by so-called charge detection via a nearby quantum point contact (QPC) or single electron transistor (SET). 8,9 In particular, the combination of high speed and very high charge sensitivity has made SET for use in studying a wide range of physical phenomena such as discrete electron transport, 8,10,11 qubit readout, 12,13 and nanomechanical oscillators. 14,15 So far most SETs have been using Al/AlOx/Al tunnel junctions, graphene SET reported here is technologically simple, reliable and can operate well above liquid-helium temperature, making it an attractive candidate for use in various charge detector applications. Here we realize an all graphene nanocircuit integration with SET charge read-out for QDs. The QD and the SET in the same material are defined in a single etching step which enables optimized coupling and sensing ability. The SET is placed in close vicinity to the QD giving rise to a strong capacitive coupling between the two systems. Once an additional electron occupies the QD, the potential in the neighboring SET is modified by capacitive interaction that gives rise to a measurable conductance change. Even if charge transport through the QD is too small to be measured by conventional transport means, the SET charge sensor also allows measurements. These devices demonstrated here provide robust building blocks in a practical quantum information processor. The graphene flakes were produced by mechanical cleaving of graphite crystallites by scotch tape and then were transferred to a highly doped Si substrate with a 100 nm SiO2 top layer. Thin flakes were found by optical microscopy and single layer graphene flakes were selected by the Raman spectroscopy 2 measurement. Next, a layer of poly(methyl methacrylate) (PMMA) is exposed by standard electron beam lithography (EBL) to form a designed pattern. The unprotected areas are carved by oxygen reactive ion etching. We used the standard EBL and lift off technique to make the ohmic contact (Ti/Au) on the present graphene devices. One of our defined sample structures with a quantum dot and proximity SET is shown in Figure 1. The quantum dot is an isolated central island of diameter 90 nm, connected by 30 nm wide tunneling barriers to source and drain contacts. Here, Si wafer was used as the back gate and there is also a side gate. The SET has a similar pattern while the conducting island has much larger diameter (180 nm). Electronic transport through both the devices exhibits Coulomb blockade characteristics with back/side gate voltage. The distance between CB peaks is determined by the sum of charging and quantum confinement energies and the former contribution becomes dominant for our devices with diameter > 100 nm. Accordingly, we refer to it as SET rather than QD. The device was first immersed into a liquid helium storage dewar at 4.2 K to test the functionality of the gates. The experiment was carried out in a top-loading dilution refrigerator equipped with filtered wiring and low- noise electronics at the base temperature of 10 mK. In the measurement, we employed the standard AC lock-in technique. Figure 2a shows the conductance through the dot GQD for applied side gate voltage Vsg. Clear Coulomb blockade peaks are observed related to charging of the tunable dot on the graphene. The current through the dot becomes too small to be measurable for side gate voltages in the range of 0.2-0.5 V. Figure 2b shows the conductance through the SET versus side gate voltage Vsg. The SET is close as possible to the QD and in this way charging signals of the dot were detected by tracking the change in the SET current. The addition of one electron to the QD leads to a pronounced change of the conductance of the charge detector by typically 30%. The slope of the SET conductance is the steepest at both sides of its Coulomb blockade resonances giving the best charge read-out signal. To offset the large current background, we used a lock-in detection method developed earlier for GaAs dot. 16 A square shaped pulse was superimposed on the dc bias on side gate voltage Vsg. A lock-in detector in sync with the pulse frequency measured the change of SET current due to the pulse modulation. Figure 2C shows a 3 typical trace of the lock-in signal of the transconductance through the SET dISET /dVsg. These sharp spikes or dips originate from the change of the charge on the dot by one electron. It shows essentially the same features as Figure 2a but is much richer especially in the regime, where the direct dot current is too small to be measured. We also note that the individual charge events measurement has been demonstrated in graphene QD with QPC detector based on graphene nanoribbon. 17 More quantitative information on the system can be obtained from the measurement of the height response of one certain peak in Figure 2c as a function of the modulating pulse frequency on the side gate. The resulting diagram for the SET dISET /dVsg gain magnitude is shown in Figure 3. The green dashed line indicates the gain of 0.707 (-3 dB), corresponding approximately to the bandwidth of 600 Hz of the SET device. By applying a signal of 5 × -210 electrons on the back gate of SET and measuring the signal with a signal-to-noise ratio of 1, we achieved a charge sensitivity of 10 -3 e / Hz which is a nice match to that obtained previously in a GaAs QD and superconducting Al SET detector system. 18 The information contained in the signal goes beyond simple charge counting. For instance, the stability diagram measurement can reveal excited states, which is crucial to get information of the spin state of electrons on a quantum dot. 19 Figure 4a shows Coulomb diamonds for the conductance GQD through the dot versus bias voltage Vsd and side gate voltage Vsg. For comparison, Figure 4b shows the transconductance of the SET dISET /dVsg as a function of the same parameters. A perfect match between the QD transport measurements and the detector signal is observed. Moreover, the discrete energy spectrum of the graphene quantum dot is revealed by the presence of additional lines parallel to the diamond edges. These lines indicate the quantum dot is in the high bias regime where the source-drain bias is so high that multiple dot levels involving the excited states can participate in electron tunneling. 20 The excited states become much more visible in the charge detector signal than the direct measurement. All of these features have been seen in GaAs QD with QPC 20 but here we achieve the goal with an all graphene nanocircuit of QD with SET. In the previous case, the QD and QPC detector are separated by typically 100 nm in width. In the present case, the SET detector is 50 nm from the edge 4 of the QD. Therefore it is expected that the capacitance coupling between QD and SET is enhanced compared to the conventional case realized in semiconductor QD and QPC. This enhanced coupling leads to a larger signal-to-noise ratio of the SET detector signal that can be exploited for time resolved charge measurement or charge/spin qubit readout on the QD. In summary, we have presented a simple fabrication process that facilitates a quantum dot and highly sensitive single electron transistor charge detector with the same material of graphene. Typically the addition of a single electron in QD would result in a change in the SET conductance of about 30%. This way the charging events measured by the charge detector and direct transport through the dot merge perfectly and more information beyond the conventional transport means is also guaranteed. The devices demonstrated here represent a fascinating avenue to realize a more complex and highly controllable electronic nanostructure formed from molecule conductors such as graphene. Acknowledgements: This work was supported by the National Science Foundation (DMR-0804794), the National Basic Research Program of China (Grants No. 2009CB929600, No. 2006CB921900), and the National Natural Science Foundation of China (Grants No. 10804104, No. 10874163, No. 10934006). References (1) Geim, A. K.; Novoselov, K. S. Nature Materials. 2007, 6, 183. (2) Geim, A. K. Science 2009, 324, 1530. (3) Ihn, T. ; Guttinger, J.; Molitor, F.; Schnez, S.; Schurtenberger, E.; Jacobsen, A.; Hellmüller, S.; Frey, T.; Droscher, S.; Stampfer, C.; Ensslin, K. Materials Today 2010, 13, 44. (4) Ponomarenko, L.; Schedin, F.; Katsnelson, M.; Yang, R.; Hill, E.; Novoselov, K.; Geim, A. K. Science 2008, 320, 356. 5 (5) Stampfer, C.; Schurtenberger, E.; Molitor, F.; Guettinger, J.; Ihn, T.; Ensslin, K. Nano Lett. 2008, 8, 2378. (6) Molitor, F.; Knowles, H.; Droscher, S.; Gasser, U.; Choi, T.; Roulleau, P.; Guttinger, J.; Jacobsen, A.; Stampfer, C.; Ensslin, K.; Ihn, T. Europhys. Lett. 2010, 89, 67005. (7) Liu, X. L.; Hug, Dorothee; Vandersypen, L. M. K. Nano Lett. 2010, 10, 1623. (8) Lu, W.; Ji, Z. Q; Pfeiffer, L.; West, K. W; Rimberg, A. J. Nature 2003, 423, 422. (9) Elzerman, J. M.; Hanson, R.; van Beveren, L. H. W.; Witkamp, B.; Vandersypen, L. M. K.; Kouwenhoven. L. P. Nature 2004, 430, 431. (10) Bylander, J.; Duty, T.; Delsing, P. Nature 2005, 434, 361. (11) Gotz, G.; Steele, G. A.; Vos. W. J.; Kouwenhoven, L. P. Nano Lett. 2008, 8, 4039. (12) Lehnert, K. W.; Bladh, K.; Spietz, L. F.; Gunnarsson, D.; Schuster, D. I.; Delsing, P.; Schoelkopf, R. J. Phys. Rev. Lett. 2003, 90, 027002. (13) Duty, T.; Gunnarsson, D.; Bladh, K.; Delsing, P. Phys. Rev. B 2004, 69, 140503(R). (14) Knobel, G.; Cleland, A. N. Nature 2003, 424, 291. (15) LaHaye, D.; Buu, O.; Camarota, B.; Schwab, K. C. Science 2004, 304, 74. (16) Elzerman, J. M.; Hanson, R.; van Beveren, L. H. W.; Vandersypen, L. M. K.; Kouwenhoven. L. P. Appl. Phys. Lett. 2004, 84, 4617. (17) Guttinger, J.; Stampfer, C.; Hellmuller, S.; Molitor, F.; Ihn, T.; Ensslin, K. App. Phys. Lett. 2008, 93, 212102. (18) Berman, D.; Zhitenev, N. B.; Ashoori, R. C.; Shayegan, M. Phys. Rev. Lett. 1999, 82, 161. 6 (19) Hanson, R.; Witkamp, B.; Vandersypen, L. M. K.; van Beveren, L. H. W.; Elzerman, J. M.; Kouwenhoven. L. P. Phys. Rev. Lett. 2004, 91, 196802. (20) Hanson, R. ; Kouwenhoven, L. P.; Petta, J. R.; Tarucha, S.; Vandersypen, L. M. K. Rev. Mod. Phys. 1997, 79, 1217. 7 (a) (b) FIGURE 1. (a) Scanning electron microscope image of the etched sample structure. The bar has a length of 200 nm. The upper quantum dot as main device has a diameter of 90 nm while the bottom single electron transistor as charge sensor has a diameter of 180 nm. The bright lines define barriers and the side gates. (b) Schematic of a representative device. 8 FIGURE 2. (a) Conductance through the quantum dot vs the side gate voltage. (b) The example of conductance through the single electron transistor for the same parameter ranges as in panel a. The steps in conductance are well aligned with the Coulomb blockades in panel a and are about 30% of the total signal. (c) Transconductance of the single electron transistor for the same parameter ranges as in panel a. The spikes and dips indicate the transitions in the charge states by addition of single electron in quantum dot. In particular, the charge detection can allow measurement in the regime (0.2-0.5 V) where the current through the dot is too small to be measurable by direct means. Data in panels a, b and c were recorded simultaneously during a single sweep. The vertical red dashed lines are a guide for the eyes to relate features in these graphs. 9 FIGURE 3. The SET dISET /dVsg gain magnitude with the modulating pulse frequency. The green dashed line illustrates the bandwidth of the SET device about 600 Hz corresponding to a gain of 0.707(-3 dB). Due to the stray capacitances, the response decreases rapidly after 600 Hz. The logarithmically frequency axis is scaled for clarity. 10 (a) (b) FIGURE 4. (a) Plot of the different conductance of the quantum dot vs bias and gate voltage applied on the dot. From the lines parallel to the edges of Coulomb diamonds, we can identify the excited states. (b) Transconductance of the single electron transistor for the same parameters as in panel a. Perfect matching with panel a and more excited spectrum resolved in the signal indicate the single electron transistor can be used as a highly sensitive charge detector. Data in panels a and b were recorded simultaneously during a single sweep. Green dashed lines are a guide for identifying the excited states. 11
1211.0072	2	1211	2013-02-15T19:46:35	Electrical Control of Two-Dimensional Neutral and Charged Excitons in a Monolayer Semiconductor	[ "cond-mat.mes-hall" ]	Monolayer group VI transition metal dichalcogenides have recently emerged as semiconducting alternatives to graphene in which the true two-dimensionality (2D) is expected to illuminate new semiconducting physics. Here we investigate excitons and trions (their singly charged counterparts) which have thus far been challenging to generate and control in the ultimate 2D limit. Utilizing high quality monolayer molybdenum diselenide (MoSe2), we report the unambiguous observation and electrostatic tunability of charging effects in positively charged (X+), neutral (Xo), and negatively charged (X-) excitons in field effect transistors via photoluminescence. The trion charging energy is large (30 meV), enhanced by strong confinement and heavy effective masses, while the linewidth is narrow (5 meV) at temperatures below 55 K. This is greater spectral contrast than in any known quasi-2D system. We also find the charging energies for X+ and X- to be nearly identical implying the same effective mass for electrons and holes.	cond-mat.mes-hall	cond-mat	Electrical Control of Neutral and Charged Excitons in a Monolayer Semiconductor Jason S. Ross1, Sanfeng Wu2, Hongyi Yu3, Nirmal J. Ghimire4,5, Aaron M. Jones2, Grant Aivazian2, Jiaqiang Yan5,6, David G. Mandrus4,5,6, Di Xiao7, Wang Yao3, Xiaodong Xu1,2† 1 Department of Material Science and Engineering, University of Washington, Seattle, Washington 98195, USA 2 Department of Physics, University of Washington, Seattle, Washington 98195, USA 3 Department of Physics and Center of Theoretical and Computational Physics, The University of Hong Kong, Hong Kong, China 4 Department of Physics and Astronomy, University of Tennessee, Knoxville, Tennessee 37996, USA 5 Materials Science and Technology Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee, 37831, USA 6 Department of Materials Science and Engineering, University of Tennessee, Knoxville, Tennessee, 37996, USA 7 Department of Physics, Carnegie Mellon University, Pittsburg, PA 15213, USA * These authors contributed equally to the work. † Email: [email protected] Abstract: Monolayer group VI transition metal dichalcogenides have recently emerged as semiconducting alternatives to graphene in which the true two-dimensionality (2D) is expected to illuminate new semiconducting physics. Here we investigate excitons and trions (their singly charged counterparts) which have thus far been challenging to generate and control in the ultimate 2D limit. Utilizing high quality monolayer molybdenum diselenide (MoSe2), we report the unambiguous observation and electrostatic tunability of charging effects in positively charged (X+), neutral (Xo), and negatively charged (X-) excitons in field effect transistors via photoluminescence. The trion charging energy is large (30 meV), enhanced by strong confinement and heavy effective masses, while the linewidth is narrow (5 meV) at temperatures below 55 K. This is greater spectral contrast than in any known quasi-2D system. We also find the charging energies for X+ and X- to be nearly identical implying the same effective mass for electrons and holes. Introduction Above bandgap photo-excitation creates electrons and holes in the conduction and valence bands respectively. If the screening is weak enough the attractive Coulomb interaction between one electron and one hole creates a bound quasi-particle known as a neutral exciton (Xo) which has an energy structure similar to a neutral hydrogen atom. Excitons can further become charged by 1 binding an additional electron (X-) or hole (X+) to form charged 3-body excitons analogous to H– or H2 + respectively1–3. These exciton species are elementary quasi-particles describing the electronic response to optical excitation in solids and are integral to many optoelectronic applications from solar cells and light emitting diodes (LEDs)4 to optical interconnects5 and quantum logical devices6,7. The main arena for the exploration of excitonic physics has been 3D semiconductors and their heterostructures that form quasi-2D quantum wells where the carrier wavefunction typically occupies a few tens to thousands of atomic layers. Observation and control of excitons in truly 2D systems has been a long pursued goal, largely motivated by the enhancement in exciton and trion binding energies in the strict 2D limit8. In addition, unlike semiconductor heterostructures, the close proximity of external stimuli to the exciton wavefunction can offer unprecedented tunability and diversifies the applications possible in devices made with 2D semiconductors. Here we report the experimental observation and control of the fascinating excitonic physics in a 2D semiconductor by utilizing high quality monolayer molybdenum diselenide (MoSe2). MoSe2 belongs to the group-VI transition metal dichalcogenides which form in layers weakly bound to each other by Van der Waals forces. The monolayers have received much attention recently as they make up a new class of 2D semiconductors with a direct bandgap in the visible frequency range and are predicted to exhibit coupled spin-valley physics9. Recent progresses focus on MoS2, including the demonstration of having a direct bandgap10,11, high mobility electronics12, optical generation of valley polarization13–15, and electrical control of Berry phase properties16. Using a back-gated field effect transistor (FET) device, we demonstrate the reversible electrostatic tunability of the exciton charging effects from positive (X+) to neutral (Xo) and to negative (X-). We observe a large trion binding energy of 30 meV with a narrow emission linewidth of 5 meV. These narrow, well separated features have temperature dependence of typical 2D excitons and exist at high temperature suggesting remarkable stability. Interestingly, the binding energies of X+ and X- are similar implying that low energy electrons and holes in MoSe2 have the same effective mass. Our work demonstrates that monolayer MoSe2 is a true 2D semiconductor opening the door for the investigation of phenomena such as exciton condensation17–19 and the Fermi-edge singularity20,21, as well as for a new generation of optoelectronic devices such as LEDs and excitonic circuits5. 2 Results Crystal structure and spectral features of monolayer MoSe2. In monolayer MoSe2, Mo and Se atoms form a 2D hexagonal lattice with trigonal prismatic coordination (Fig. 1a). First principles calculations show that it has a direct bandgap at the corners (K points) of the first Brillouin zone (Fig. 1b). The curvature of the bands suggests comparable effective mass for low energy electrons and holes at K points (Fig. 1c) 9. These band edge electrons and holes near K points are predominantly from the d-orbitals of Mo atoms. Their wavefunctions are calculated to be strongly confined in the Mo layer within a length scale of ~0.2 nm in the out-of-plane direction. Monolayer MoSe2 is thus an ideal nanomaterial for exploring excitonic physics in the ultimate 2D limit. We use mechanical exfoliation to obtain monolayer MoSe2 on 300 nm SiO2 on n+ doped Si and atomic force microscopy (AFM) to identify the layer thickness22. Figure 1d,e show the optical micrograph and the corresponding AFM image of a representative sample, in which the monolayer thickness of ~0.7 nm is identified12 (Fig. 1f). Standard electron beam lithography (EBL) is used to fabricate monolayer field effect transistors (Fig. 1g). With the contacts simply grounded, the n+ Si functions as a back gate providing uniform electrostatic doping in the MoSe2 (Fig. 1h). The excitonic features of MoSe2 are investigated by differential reflectance and micro- photoluminescence (µ -PL) measurements (Methods). Figure 2 shows the results from an unpatterned MoSe2 sample, S1. At 20 K, we observe two main features associated with the A and B excitons in the differential reflection spectrum10,11,14,23–25 (Fig. 2a). The presence of A and B excitons has been attributed to spin-orbit coupling induced valence band splitting in bulk24. The observed energy difference of ~200 meV agrees well with the calculated splitting (180 meV) in monolayers (Fig. 1c). With the same sample and temperature under 2.33 eV laser excitation, the PL spectrum does not show a measurable feature which can be attributed to the B exciton, likely since it is not the lowest energy transition. Instead, we observe two pronounced peaks at 1.659 and 1.627 eV in the vicinity of the A exciton (Fig. 2b). Note that the PL spectrum lacks the broad low energy peak observed in MoS2 which has been attributed to defect-related, trapped exciton states10,11,14,25. 3 The striking spectral features demonstrate the high quality of our MoSe2 samples (Methods) and provide strong evidence for monolayer MoSe2 being a direct bandgap semiconductor (Supplementary Figure S1) where the two distinct transitions are excitons. The higher energy emission at 1.659 eV is the neutral exciton, Xo, and the lower energy peak is a trion3. In unpatterned samples, we assume the trion to be X- since all measured devices show n-doped characteristics (Supplementary Figure S2). All measured unpatterned samples show a binding energy, which is the energy difference between trion and Xo, of ~30 meV (Fig. 2b inset). This is more than twice typical numbers reported in GaAs quantum wells3,26,27 and similar to a recent mention in MoS2 14. Gate dependence of MoSe2 photoluminescence. In order to confirm the above assignment and control the exciton charging effects, we performed gate dependent PL measurements using monolayer MoSe2 FETs. Here, the excitation laser is at 1.73 eV for better resonance with the luminescent states. Figure 3a shows a color map of the PL spectrum of device D1 at 30 K as a function of back-gate voltage, Vg, in which we clearly observe four spectral features whose intensities strongly depend on Vg. Near zero Vg the spectrum shows a broad low energy feature around 1.57 eV and a narrow high energy peak at 1.647 eV. With large Vg of either sign, these peaks disappear and a single emission peak dominates the spectrum. Both peaks (at negative or positive Vg) have similar energies and intensities with the latter increasing with the magnitude of Vg. This observed gate dependence confirms the assignment of states as labeled in Figure 3a. Since the broad low energy peak does not show up in unpatterned samples before FET fabrication, we attribute it to exciton states trapped to impurities (XI)10,11,14,25 which are likely introduced during EBL processing and are not the focus of this paper. The sharp peak at 1.647 eV is the Xo, slightly red-shifted compared to unpatterned samples. From the gate dependence, we identify the peaks near 1.627 eV as the X- and X+ trions when Vg is largely positive and negative respectively. Remarkably, these two distinct quasi-particles (X+ and X-) exhibit a nearly identical binding energy. The difference is within 1.5 meV over the whole applied Vg range. Since the binding energy of a trion is dependent on its effective mass, this observation implies that the electron and hole have approximately the same effective mass. 4 The gate dependent measurements unambiguously demonstrate the electrical control of exciton species in a truly 2D semiconductor, as illustrated in Figure 3b. The conversion from Xo to trion can be represented as e(h) + Xo → X- (X+), where e and h represent an electron or a hole respectively. By setting Vg to be negative, the sample is p-doped, favoring excitons to form lower energy bound complexes with free holes. As Vg decreases, more holes are injected into the sample and all Xo turn into X+ to form a positively charged hole-trion gas. With positive Vg a similar situation occurs with free electrons to form an electron-trion gas. In the following, we show that a standard mass action model can be used to describe the conversion dynamics. Figure 3c shows the extracted Xo (black) and trion (red) peak intensity as a function of Vg where we have adjusted the negative Vg data due to background signal. The plot shows that the maximum Xo intensity is about equal to the saturated trion PL when Xo vanishes. This observation indicates conservation of the total number of Xo and trion in the applied voltage range and similar radiative decay rates for both quasi-particles. Thus the PL intensity represents the amount of the corresponding exciton species. Since the dynamic equilibrium of free electrons, holes, and excitons are governed by the rate equation and law of mass action27, we calculate the gate dependent Xo and trion abundance (Supplementary Figure S3), shown by the solid lines in Figure 3c, which agrees with the data. In the simulation, we first fit the Xo curve to obtain our two free parameters: the maximum background electron concentration ݊୆ ୫ୟ୶ = 3.6 × 10ଵ଴ cmିଶ when the trion intensity saturates and the photo-excited electron concentration ݊୔ dependence with these parameters held fixed (Supplementary Note 1). The deviation in the trion = 1.5 × 10ଵ଴ cmିଶ . We then fit the trion gate experimental data from the calculated curve near zero Vg is artificial due to the mutual background from Xo and XI. We note that this inferred electron concentration is much smaller than the product of the gate capacitance and Vg. This discrepancy can be attributed to the large contact resistance (Supplementary Figure S4) of the sample, which prevents the carrier concentration from reaching equilibrium on the experimental time scale at 30 K. We expect that future improved contact technologies will eliminate this effect. Temperature dependence of MoSe2 photoluminescence. The observed exciton states also show fine features consistent with 2D excitons, such as temperature dependent line shape, peak energy, and relative weight of Xo and trion, which further supports the excitonic nature of this 5 monolayer system (Fig. 4). Figure 4a shows the evolution of X- and Xo (normalized PL) as a function of temperature in an unpatterned sample, S2, under 1.96 eV laser excitation. At low temperatures, we again observe a binding energy of 30 meV. As the temperature rises we see the X- signal drop significantly at about 55 K which we attribute to electrons escaping their bound trion state due to thermal fluctuations (Supplementary Note 1). Figure 4b is the zoom-in plot at 15K where we observe slightly different line shapes for X- and Xo. The Xo peak is symmetric showing homogenous thermal broadening effects and is well fit by a hyperbolic secant function which yields a full width half maximum of 5 meV27,28. However, the X- peak shows a slightly asymmetric profile with a long low-energy tail consistent with electron recoil effects27: The recombination of a X- with momentum k will emit a photon and leave a free electron with the same momentum k due to momentum conservation. From ℏ௞మ energy conservation, the emitted photon has an energy ℏ߱ = ℏ߱୭ , where ℏ߱୭ is the ∗మ ଶ௠౛ ெ౔ష energy of trions with ݇ = 0, and ܯଡ଼౥ and ܯଡ଼ష are the Xo and X- effective masses, respectively. The lineshape of trion PL will thus be the convolution of a symmetric peak function (hyperbolic ெ౔ − ౥ secant) and an exponential low-energy tail function (Supplementary Figure S5 and Supplementary Note 2). When the temperature is above 70K, we find homogenous broadening dominates over electron recoil and the PL spectrum is fit well by two hyperbolic secant functions (Fig. 4c). From the fits we extract the X- and Xo peak position (Fig. 4d) and the ratio of the integrated intensity of the X- to the Xo (Fig. 4e) where we do not present trion data above 150 K because it becomes negligible. We find that the peak positions are fit well (solid line in Fig. 4d) using a standard semiconductor bandgap dependence29 of ܧ୥ሺܶሻ = ܧ୥ሺ0ሻ − ܵۦℏ߱ۧ ቂcoth ቀۦℏఠۧ − 1ቁቃ ଶ௞் where ܧ୥ሺ0ሻ is the ground state transition energy at 0 K, ܵ is a dimensionless coupling constant, and ۦℏ߱ۧ is an average phonon energy. From the fits we extract for Xo (X-) the ܧ୥ = 1.657 (1.625) eV, ܵ = 1.96 (2.24) and ۦℏ߱ۧ = 15 meV for both. Applying our mass action model (Supplementary Figure S3 and Supplementary Note 1) with a trion binding energy of 30 meV results in a good fit to the X-:Xo intensity ratio (solid line in Fig. 4e). 6 Discussion In summary, we have shown that monolayer MoSe2 is a true 2D excitonic system which exhibits strong electrostatic tuning of exciton charging via a standard back-gated FET. The observed narrow, well separated spectral features are within the Ti-Sapphire laser spectral range and thus provide remarkable opportunities to selectively probe and control specific excitons using current continuous wave and ultrafast Ti-Sapphire laser technologies. Our results further demonstrate that high quality monolayer dichalcogenides can serve as a platform for investigating excitonic physics and photonic applications in the truly 2D limit with the potential to outperform quasi-2D systems. The results represent unique prospects for this burgeoning class of 2D materials in addition to the recent attention received for their valley physics. Our work expands the horizons for using 2D semiconductors for diverse fundamental studies and technical applications. Note: During the review process, we became aware of the independent observation of 30 and investigation of PL intensity of thin film negatively charged exciton in monolayer MoS2 MoSe2 as a function of temperature31. Methods Sample growth. MoSe2 single crystals were grown by vapor transport. First, MoSe2 powder was prepared by heating a stoichiometric mixture of Mo (Alfa, 99.999%) powder and Se (Alfa, 99.999%) pieces sealed in a quartz tube under 1/3 atmosphere of pure argon. The ampoule was slowly heated up to 850oC and kept at this temperature for 24 hours. Then, 3.5g of MoSe2 powder was mixed with 0.5g of iodine and sealed in a quartz tube. The sealed ampoule was loaded into a tube furnace for 10 days with the hot zone kept at 1050oC and the growth zone at 1000◦C. Plate-like crystals with typical dimensions 6 × 10 × 0.1 mm3 were obtained at the cold end. Room temperature X-ray diffraction confirmed the samples are single phase with 2H structure. Elemental analysis was performed using a Hitachi TM-3000 tabletop electron microscope equipped with a Bruker Quantax 70 energy dispersive x-ray (EDX) system. The analysis confirmed the expected stoichiometric Mo:Se ratio in the crystals. Device fabrication. After mechanical exfoliation of MoSe2 monolayers on SiO2 substrates, a JEOL JBX-6300FS electron beam lithography system was used to pattern contacts in a 300nm 7 bilayer PMMA resist. An e-beam evaporator was used to deposit 6/60 nm of Ti/Au followed by liftoff in hot acetone, cleaning in several IPA baths, and drying under N2. Optical measurements. All spectra were collected via a Princeton Instruments Acton 2500i grating spectrometer and an Andor Newton CCD with samples in a helium flow cryostat. For differential reflectance measurements an Ocean Optics tungsten halogen white light source is reflected into a 40X ultra-long working distance objective using a neutral density (ND), achromatic beam splitter and then focused on the sample (~2 um spot size). The reflected signal from the sample is collected by the objective and sent back through the beam splitter into the spectrometer. For PL measurements, a similar setup is used except the ND beam splitter is replaced by a dichroic beam splitter appropriate for the laser wavelength and a laser line notch filter is inserted before the spectrometer. Acknowledgments: The authors thank David Cobden and Ming Gong for helpful discussions. This work is mainly supported by the US DoE, BES, Materials Sciences and Engineering Division (DE-SC0008145). HY and WY were supported by Research Grant Council of Hong Kong (HKU706412P). NG, JY, DM, and DX were supported by US DoE, BES, Materials Sciences and Engineering Division. Device fabrication was performed at the University of Washington Microfabrication Facility and NSF-funded Nanotech User Facility. Author Contribution: X.X. conceived the experiments. J.R. fabricated the devices and performed the measurements, assisted by S.W., A.J. and G.A.. S.W., J.R. and X.X. performed data analysis. H.Y., W.Y. and D.X. contributed to the theoretical explanation. N.G., J.Y. and D.M. synthesized and performed bulk characterization measurements on the MoSe2 crystals. All authors discussed the results and contributed to writing the manuscript. 8 References 1. 2. 3. 4. 5. 6. 7. 8. 9. Lampert, M. Mobile and Immobile Effective-Mass-Particle Complexes in Nonmetallic Solids. Physical Review Letters 1, 450–453 (1958). Finkelstein, G., Shtrikman, H. & Bar-Joseph, I. Negatively and positively charged excitons in GaAs/AlxGa1-xAs quantum wells. Physical Review B 53, R1709–R1712 (1996). Kheng, K., Cox, R. & D’Aubigné, M. Observation of negatively charged excitons X- in semiconductor quantum wells. Physical review letters 71, 1752–1755 (1993). Scholes, G. D. & Rumbles, G. Excitons in nanoscale systems. Nature materials 5, 683– 696 (2006). High, A. A., Novitskaya, E. E., Butov, L. V, Hanson, M. & Gossard, A. C. Control of exciton fluxes in an excitonic integrated circuit. Science 321, 229–231 (2008). Carter, S. G. et al. Quantum coherence in an optical modulator. Science 310, 651–653 (2005). Kira, M., Koch, S. W., Smith, R. P., Hunter, a. E. & Cundiff, S. T. Quantum spectroscopy with Schrödinger-cat states. Nature Physics 7, 799–804 (2011). Stébé, B. & Ainane, A. Ground state energy and optical absorption of excitonic trions in two dimensional semiconductors. Superlattices and Microstructures 5, 545–548 (1989). Xiao, D., Liu, G.-B., Feng, W., Xu, X. & Yao, W. Coupled Spin and Valley Physics in Monolayers of MoS2 and Other Group-VI Dichalcogenides. Physical Review Letters 108, 1–5 (2012). 10. Splendiani, A. et al. Emerging photoluminescence in monolayer MoS2. Nano letters 10, 1271–1275 (2010). 11. Mak, K., Lee, C., Hone, J., Shan, J. & Heinz, T. Atomically Thin MoS2: A New Direct- Gap Semiconductor. Physical Review Letters 105, 2–5 (2010). 12. Radisavljevic, B., Radenovic, A., Brivio, J., Giacometti, V. & Kis, A. Single-layer MoS2 transistors. Nature nanotechnology 6, 147–150 (2011). 13. Cao, T. et al. Valley-selective circular dichroism of monolayer molybdenum disulphide. Nature communications 3, 887 (2012). 14. Mak, K. F., He, K., Shan, J. & Heinz, T. F. Control of valley polarization in monolayer MoS2 by optical helicity. Nature Nanotechnology 7, 494–498 (2012). 9 15. Zeng, H., Dai, J., Yao, W., Xiao, D. & Cui, X. Valley polarization in MoS2 monolayers by optical pumping. Nature nanotechnology 7, 490–493 (2012). 16. Wu, S. et al. Electrical Tuning of Valley Magnetic Moment Through Symmetry Control in Bilayer MoS2. Nature Physics. DOI:10.1038/NPHYS2524 (2013). 17. Eisenstein, J. P. & Macdonald, a H. Bose-Einstein condensation of excitons in bilayer electron systems. Nature 432, 691–694 (2004). 18. Butov, L. V, Gossard, a C. & Chemla, D. S. Macroscopically ordered state in an exciton system. Nature 418, 751–754 (2002). 19. Deng, H. & Yamamoto, Y. Exciton-polariton Bose-Einstein condensation. Reviews of Modern Physics 82, 1489–1537 (2010). 20. Chen, W. et al. Excitonic enhancement of the Fermi-edge singularity in a dense two- dimensional electron gas. Physical Review B 45, 8464–8477 (1992). 21. Finkelstein, G., Shtrikman, H. & Bar-Joseph, I. Optical Spectroscopy of a Two- Dimensional Electron Gas near the Metal-Insulator Transition. Physical Review Letters 74, 976–979 (1995). 22. Novoselov, K. S. et al. Two-dimensional atomic crystals. Proceedings of the National Academy of Sciences of the United States of America 102, 10451–10453 (2005). 23. Evans, B. L. & Hazelwood, R. A. Optical and structural properties of MoSe2. Physica Status Solidi (a) 4, 181–192 (1971). 24. Coehoorn, R., Haas, C. & De Groot, R. Electronic structure of MoSe2, MoS2, and WSe2. II. The nature of the optical band gaps. Physical Review B 35, 6203–6206 (1987). 25. Korn, T., Heydrich, S., Hirmer, M., Schmutzler, J. & Schüller, C. Low-temperature photocarrier dynamics in monolayer MoS2. Applied Physics Letters 99, 102109 (2011). 26. Warburton, R. et al. Optical emission from a charge-tunable quantum ring. Nature 405, 926–929 (2000). 27. Esser, A., Runge, E. & Zimmermann, R. Photoluminescence and radiative lifetime of trions in GaAs quantum wells. Physical Review B 62, 8232–8239 (2000). 28. Esser, a., Runge, E., Zimmermann, R. & Langbein, W. Trions in GaAs Quantum Wells: Photoluminescence Lineshape Analysis. Physica Status Solidi (a) 178, 489–494 (2000). 29. O’Donnell, K. P. & Chen, X. Temperature dependence of semiconductor band gaps. Applied Physics Letters 58, 2924-2926 (1991). 10 30. Mak, K. F. et al. Tightly bound trions in monolayer MoS2. Nature Materials (2012).doi:10.1038/nmat3505 31. Tongay, S. et al. Thermally Driven Crossover from Indirect toward Direct Bandgap in 2D Semiconductors: MoSe2 versus MoS2. Nano letters 12, 5576–5580 (2012). 11 Figures: Figure 1 MoSe2 characteristics and devices. a, Coordination structure and top view of monolayer MoSe2. b, Density function theory calculated band structure. c, band structure at K point shows 180 meV valence band splitting due to spin-orbit coupling. d, Optical micrograph of exfoliated MoSe2 flake on 300nm SiO2. Scale bar: 5 µm. e, atomic force microscope (AFM) image of area highlighted in (d). Scale bar: 1 µm. f, AFM line scan along dashed line in (e). g, optical micrograph of MoSe2 device. Scale bar: 5 µm. h, schematic of back-gated MoSe2 device. 12 a b R / R ) s t n u o c 3 0 1 ( y t i s n e t n I L P 0.6 0.4 0.2 0.0 -0.2 7.5 6.0 4.5 3.0 1.5 0.0 1.50 Energy (eV) 1.75 2.00 2.25 A B ) . u . a ( L P 30 meV X- Xo 1.60 1.64 Energy (eV) 1.68 1.50 1.75 2.00 Energy (eV) 2.25 2 Differential Figure and reflectance photoluminescence spectra of monolayer MoSe2 at 20 K. a, Differential reflectance shows A and B excitons. b, Photoluminescence (PL) excited by 2.33 eV laser shows neutral exciton (Xo) and the lower energy charged exciton (X-). PL from the B exciton has not been observed. Inset: PL of the exciton peaks. The X- shows a charging energy of about 30 meV. 13 Figure 3 Electrostatic control of exciton charge. a, MoSe2 photoluminescence (color scale in counts) is plotted as a function of back-gate voltage. Near zero doping, we observe mostly neutral and impurity-trapped excitons. With large electron (hole) doping, negatively (positively) charged excitons dominate the spectrum. b, Illustration of the gate dependent trion and exciton quasi-particles and transitions. c, trion and exciton peak intensity vs. gate voltage at dashed arrows in (a). Solid lines are fits based on the mass action model. 14 a ) . u . a ( L P d e z i l a m r o N b T(K): 295 270 250 230 210 190 170 150 130 120 110 100 90 80 70 65 60 55 50 45 40 35 30 25 20 15 ) s t n u o c ( L P 10000 5000 0 c 12000 ) s t n u o c ( L P 8000 4000 0 d ) V e ( n o i t i s o P k a e P 1.64 1.60 1.56 1.62 1.64 1.66 0 50 100 150 200 250 300 e ) o X / - X ( o i t a R a e r A 2 1 0 1.52 1.56 1.60 1.64 1.68 1.72 Energy (eV) 1.62 1.65 1.68 0 50 100 150 Energy (eV) Temperature (K) Figure 4 Temperature dependence of photoluminescence spectrum. a, Normalized photoluminescence of monolayer MoSe2 vs. temperature. b, Line shape fitting at 15 K. Black is data. Red and blue curves are fits with and without considering the electron-recoil effect. c, Data and fit at 70 K using two symmetric peaks. d, Neutral exciton (black) and trion (red) peak position vs. temperature with fits (solid lines). e, Integrated area ratio of trion:exciton vs. temperature with mass action model fitting (red). 15 Supplementary Information 240 160 80 0 ) ) s - W u ( / s t n u o c ( L P Single Layer Bilayer 1.60 1.62 1.64 Energy (eV) 1.66 1.68 Supplementary Figure S1: Photoluminescence (PL) of single and bilayer MoSe2. Both spectrums were taken at 20 K with 532 nm excitation laser. Intensity is normalized to laser power and integration time. Under the same experimental conditions, bilayer samples (confirmed by atomic force microscopy) exhibit suppressed photoluminescence intensity and an overall redshift compared to single layer samples. Like MoS2, these PL features suggest the crossover to direct bandgap at the single layer10,11 and offer a method of determining the layer thickness. 16 Supplementary Figure S2: More gate dependence of photoluminescence. a, device D2 under 1.96 eV laser excitation. b, device D3 under 1.73 eV laser excitation. Both show nonzero electron doping at back gate Vg = 0. In the main text we assume that trions observed in unpatterned samples (no device fabrication after mechanical exfoliation) are negatively charged electron-trions, or X-. These figures provide supporting evidence of this where we see, as with all MoSe2 devices we fabricate, a negative gate voltage is required in order to maximize the Xo. Since a maximized Xo corresponds to a minimized carrier density (see Supplemental Figure S3 and Supplementary Note 1), this gate dependence suggests that samples are originally n-doped. This is also consistent with observations12,14 in MoS2. 17 Supplementary Figure S3: Mass action model plots. Calculated densities of a, exciton, b, trion and c, electron 2D MoSe2 as a function of background electron density and temperature with a fixed amount of absorbed photons of 3.2 ൈ 1010 cm-2 during exciton life time. The white and blue lines correspond to the experimental data in the main text for gate and temperature dependence respectively. Details on the model are found in Supplementary Note 1. 18 a 1.6 1.2 0.8 0.4 ) A n ( d s -0.4 I 0.0 -0.8 -1.2 b V g (V): 65 60 50 40 15 -15 -40 -60 30 0 -30 -50 -65 V sd (mV): 500 350 200 100 50 10 4 3 2 1 ) A n ( d s I -1 0 -400 -200 0 200 V (mV) sd 400 -60 -40 -20 0 20 V (V) g -2 40 60 Supplementary Figure S4: Low temperature transport of single layer MoSe2. a. Source- drain bias vs. current for various gate voltages. b. Gate voltage vs. current for various source- drain biases. The gate dependent photoluminescence (PL) presented in the main text was measured at zero source-drain bias. These figures illustrate that at any gate voltage, there is miniscule conductance for this experimental condition. Specifically, we find the resistance near zero bias to be on the order of 100 GΩ for all gate voltages. Only with appreciable biases above 100 mV do devices show significant conduction. With this in mind, it will be interesting to investigate the gate dependent PL with good metal contacts. 19 Supplementary Figure S5: Electron recoil effect on photoluminescence lineshape in trions. Electron recoil effects result in an exponential lineshape on one side the photoluminescence peaks of trions. Details are described in Supplemental Note 2. 20 Supplementary Note 1: Mass action model. To determine the relative intensity of the photoluminescence (PL) signals from the various exciton species we adopt a steady state (dynamical equilibrium) model of all the particles in our system. Here, for simplicity, we consider electron-trion and assume that the free hole in the system is negligible. We denote ݊ଡ଼ , ݊ଡ଼ష , and ݊ୣ for the concentration of Xo, X-, and free electrons. ݊୔ ≡ ݊ଡ଼ + ݊ଡ଼ష denotes the number of photoexcited electrons, and ݊୆ ≡ ݊ୣ + ݊ଡ଼ష denotes the background electrons (doping level) before light excitation. ݊୆ and ݊୔ are the initial conditions controlled by gate voltage and laser intensity, while ݊ଡ଼ , ݊ଡ଼ష , and ݊ୣ are steady state variables. To establish the relationship between these quantities we first write the reaction rate equation for trion formation, Xo + e- → X-. From the law of mass action with trions we have28,32,33: = ܣ ݇୆ܶ exp (− ܧ୘ ݊ଡ଼݊ୣ ݇୆ܶ) ݊ଡ଼ష Here ܶ is the temperature, kB is Boltzmann constant, ܧ୘ is the trion binding energy and ܣ = in MoSe2 in which ܯଡ଼౥ = ݉ୣ + ݉୦ and ܯଡ଼ష = 2݉ୣ + ݉୦ are ସெ౔ ౥௠౛ ଵ = 6.18 × 10ଵଵ ୡ୫మ ୫ୣ୚ πℏమெ౔ష the exciton and trion effective masses respectively. Next, from charge conservation we obtain: ݊ୣ + ݊ଡ଼ + 2݊ଡ଼ష = ݊୔ + ݊୆ . Solving the above equations gives ݊ଡ଼ష = ݊୔ + ݊୆ + ݊୅ − ඥሺ݊୔ + ݊୆ + ݊୅ሻଶ − 4݊୔݊୆ 2 Here, ݊୅ = ܣ ݇୆ܶ exp (− ). By plotting these quantities as a function of T and ݊୆ , we are ா౐ able to model the full range of our experimental data (see Supplemental Figure S3). We note that ௞ా் the entire picture and the values we have here for electron-trion can be applied directly to the hole-trion thanks to the same effective mass of electron and hole in our massive Dirac Fermion system. Therefore we have also explained the data in the hole-domain, which is labeled by the negative value of ݊୆ in Supplemental Figure S3. 21 Supplementary Note 2: Electron recoil effects. Because of energy and momentum conservation, the radiative recombination of an electron- trion ℏమ௞మ with center of mass wave vector k results in a conduction electron with wave vector k and a photon with energy ℏ߱ = ℏ߱௢ + ∗ + ݉௛ , where ܯ் = 2݉௘ ∗ = ℏ߱௢ − ℏమ௞మ ℏ௞మ ∗ is the − ∗మ ெ೅ ଶ௠೐ ଶ௠೐ ଶெ೅ ∗ is the exciton mass, and ℏ߱଴ is the energy of trion with k = 0. The trion mass, ܯ௑ = ݉௘ ∗ + ݉௛ photon emission rate is given by the optical matrix element ܯ(ܓ) and trion distribution ݂(ܓ), which is ெ೉ ெ೉ ܴሺ߱ሻ = න ݀ܓܯሺܓሻଶ݂ሺܓሻߜ(ℏ߱ − ℏ߱଴ + ܧሺ݇ሻ) where ܧ ሺ݇ሻ ≡ . In the low density limit, the distribution function ݂(ܓ) can be ℏ௞మ ∗మ ଶ௠೐ ெ೅ approximated by the Boltzmann distribution ݂ሺܓሻ ∝ exp ቀ− ଶெ೅௞ಳ்ቁ. The optical matrix element ℏమ௞మ ܯሺܓሻଶ has already been numerically evaluated by Stébé et al.34. Although for the 2D case they only give results for ߪ ≡ ∗ ௠೐ ∗ = 0, 0.1, 0.2, and 0.3, we find those curves can all be well ௠೓ approximated by an exponentially decaying function ܯሺܓሻଶ = ܯሺ૙ሻଶ݁ିଶ଻.ଷா(௞)/ாబ . With ܧ଴ ≡ 2ሺ1 + ߪሻܧ௑ ଷ஽ , and ܧ௑ ଷ஽ is the 3D exciton binding energy. We expect this equation applies for any ߪ ∈ ሾ0,1ሿ. The photon emission rate as a function of ߱ is then ܴሺ߱ሻ = න ݀ܓܯሺܓሻଶ݂ሺܓሻߜ൫ℏ߱ − ℏ߱଴ + ܧሺ݇ሻ൯ + ݉௘ ∗ 1 = ܴሺ߱଴ሻ exp ൤− ൬27.3 ݇஻ ܶ൰ ሺℏ߱଴ − ℏ߱ሻ൨ Θሺ߱଴ − ߱ሻ ܧ଴ ܯ௑ Here Θ is the Heaviside step function. In the MoSe2 system, ߪ = 1 and the 3D exciton binding ∗ = 50 meV. We have plotted the line shape ܴሺ߱ሻfor several temperatures in energy is ܧ௑ ଷ஽ = ܴ௬ Supplemental Figure S5. Thus the trion line shape can be fit through convolution of the above contribution with a symmetric broadening effect. 22 Supplemental References 32. Vercik, a., Gobato, Y. G. & Brasil, M. J. S. P. Thermal equilibrium governing the formation of negatively charged excitons in resonant tunneling diodes. Journal of Applied Physics 92, 1888 (2002). Siviniant, J., Scalbert, D., Kavokin, a., Coquillat, D. & Lascaray, J.-P. Chemical equilibrium between excitons, electrons, and negatively charged excitons in semiconductor quantum wells. Physical Review B 59, 1602–1604 (1999). Stébé, B., Feddi, E., Ainane, A. & Dujardin, F. Optical and magneto-optical absorption of negatively charged excitons in three- and two-dimensional semiconductors. Physical Review B 58, 9926–9932 (1998). 23 33. 34.
1907.05065	1	1907	2019-07-11T09:11:20	Room temperature Organic Exciton-Polariton Condensate in a Lattice	[ "cond-mat.mes-hall", "cond-mat.soft", "physics.optics" ]	Interacting Bosons, loaded in artificial lattices, have emerged as a modern platform to explore collective manybody phenomena, quantum phase transitions and exotic phases of matter as well as to enable advanced on chip simulators. Such experiments strongly rely on well-defined shaping the potential landscape of the Bosons, respectively Bosonic quasi-particles, and have been restricted to cryogenic, or even ultra-cold temperatures. On chip, the GaAs-based exciton-polariton platform emerged as a promising system to implement and study bosonic non-linear systems in lattices, yet demanding cryogenic temperatures. In our work, we discuss the first experiment conducted on a polaritonic lattice at ambient conditions: We utilize fluorescent proteins as an excitonic gain material, providing ultra-stable Frenkel excitons. We directly take advantage of their soft nature by mechanically shaping them in the photonic one-dimensional lattice. We demonstrate controlled loading of the condensate in distinct orbital lattice modes of different symmetries, and finally explore, as an illustrative example, the formation of a gap solitonic mode, driven by the interplay of effective interaction and negative effective mass in our lattice. The observed phenomena in our open dissipative system are comprehensively scrutinized by a nonequilibrium model of polariton condensation. We believe, that this work is establishing the organic polariton platform as a serious contender to the well-established GaAs platform for a wide range of applications relying on coherent Bosons in lattices, given its unprecedented flexibility, cost effectiveness and operation temperature.	cond-mat.mes-hall	cond-mat	Room temperature Organic Exciton-Polariton Condensate in a Lattice M. Dusel1†∗, S. Betzold1†, O. A. Egorov2, S. Klembt1, J. Ohmer3, U. Fischer3, S. Hofling1,4 & C. Schneider1,∗ 1Technische Physik, Universitat Wurzburg, Am Hubland, D-97074 Wurzburg, Germany 2Institute of Condensed Matter Theory and Solid State Optics, Abbe Center of Photonics, Friedrich-Schiller-Universitat Jena, Max-Wien-Platz 1, D-07743 Jena, Germany 3Department of Biochemistry, Universitat Wurzburg, Am Hubland, D-97074 Wurzburg, Germany 4SUPA, School of Physics and Astronomy, University of St Andrews, St Andrews KY16 9SS, United Kingdom † These authors contributed equally to this work ∗ Corresponding author. Email: [email protected] (C. S.), [email protected] (M. D.) Interacting Bosons, loaded in artificial lattices, have emerged as a modern platform to ex- plore collective manybody phenomena, quantum phase transitions and exotic phases of mat- ter as well as to enable advanced on chip simulators. Such experiments strongly rely on well- defined shaping the potential landscape of the Bosons, respectively Bosonic quasi-particles, and have been restricted to cryogenic, or even ultra-cold temperatures. On chip, the GaAs- based exciton-polariton platform emerged as a promising system to implement and study bosonic non-linear systems in lattices, yet demanding cryogenic temperatures. In our work, 1 we discuss the first experiment conducted on a polaritonic lattice at ambient conditions: We utilize fluorescent proteins as an excitonic gain material, providing ultra-stable Frenkel ex- citons. We directly take advantage of their soft nature by mechanically shaping them in the photonic one-dimensional lattice. We demonstrate controlled loading of the condensate in distinct orbital lattice modes of different symmetries, and finally explore, as an illustra- tive example, the formation of a gap solitonic mode, driven by the interplay of effective in- teraction and negative effective mass in our lattice. The observed phenomena in our open dissipative system are comprehensively scrutinized by a nonequilibrium model of polariton condensation. We believe, that this work is establishing the organic polariton platform as a serious contender to the well-established GaAs platform for a wide range of applications relying on coherent Bosons in lattices, given its unprecedented flexibility, cost effectiveness and operation temperature. The implementation of well-controllable lattice potentials for interacting quantum particles and Bosonic condensates1 is an engineering task of importance towards the realization of advanced classical2 and quantum simulators3. The possibilities to engineer Hamiltonians in well-defined experimental settings has furthermore served as an inspiration to explore topology in synthetic systems4. While ultra-cold atoms in optical lattices5, trapped ions6, superconducting circuits7, 8 and photonic on-chip architectures9 are considered as the leading platforms for a controlled ex- perimental implementation and manipulation, they intrinsically suffer from a variety of serious drawbacks: The atomic approach relies on ultra-low temperatures, purely photonic approaches suffer from very small non-linearities, and superconducting circuits compose a serious technolog- 2 ical challenge. A hybrid approach, seeking to merge the best of all worlds, involves the implementation of strongly interacting photons on chip, in particular in the form of exciton-polaritons. Such hybrid excitation can inherit the low-loss nature of photons, and still acquire a notable non-linearity aris- ing from the excitonic part10, 11. Since the initial demonstration of Bose-Einstein condensation of microcavity exciton polaritons12, one particular focus was set on engineering the potential land- scape of polaritons, in the spirit of on-chip quantum simulation13. While cavity polaritons and their condensates have been observed in a variety of systems, including II-VI12, III-V semiconductors14, layered materials15, 16, as well as organic materials17 -- 19 due to the technological challenges required to precisely control their energy landscape, the vast majority of approaches towards periodic ar- rangements were conducted on the mature GaAs platform13 . The degree of energy, position, coupling and phase control has now reached such a level, that enabled the first implementation of ultra-fast simulators of the X-Y Hamiltonian20, as well as synthetic topological Chern insulators21 and topological lasers22. Even electrical injection has recently accomplished23. However, these experiments still require cooling by liquid helium, and (in most cases) highly advanced nano- technology for chip processing. In this work, we address these two fundamental complications: We utilize fluorescent pro- teins as active material, hosting ultra-stable Frenkel excitons. Furthermore, we directly take ad- vantage of the soft nature of the material, and mechanical shape it into the photonic lattice envi- ronment, which has the form of a one-dimensional lattice of tightly coupled photonic hemispheric 3 cavities. We make the following, striking observations: The high quality of our device allows us to generate a close-to ideal bandstructure of room-temperature exciton-polaritons, dictated by the photonic lattice in the tight-binding regime. Bosonic condensation is facilitated at elevated pump densities, and by shaping the pump spot, we can load the condensate into distinct lattice modes. The subtle interplay between the repulsive polaritonic interaction, and the negative polariton mass yields the formation of a one-dimensional gap solitonic state, which we clearly identify in the realspace expansion of the condensate. We further develop a broad understanding of these nonequilibrium effects in the framework of a numerical model based on solving the Gross-Pitaevskii equation in presence of gain, loss and noise. Our studied device is composed of two distributed Bragg reflectors (DBRs), which sand- wich an optical spacer layer filled with a thin film of the fluorescent protein "mCherry". The layer of mCherry is coated on the hemispheric micro-lenses, and the planar back-mirror mechanically presses the proteins into the indentations, prior to thermally evaporating the solvents. An artistic sketch of the device is shown in Fig. 1a-c. The DBRs each have 10.5 alternating pairs of SiO2- and TiO2-layers with a reflectivity of 99.9 % between 1.90 eV (653 nm) and 2.13 eV (582 nm). This configuration yields calculated mode quality factors exceeding 20000 and support strong coupling conditions with a Rabi-splitting of 240 meV, as we analyze in the supplementary information S1. Prior to coating the top DBR, we prepared a plateau like area on the glass substrate with depth and diameter of about 500 µm and 4000 µm, respectively, by chemical wet etching. Then, lens- 4 shaped indentations were defined by using ion beam milling. These hemispheric dimples have diameters ranging from 3 µm to 5 µm and depths between 100 nm and 350 nm. The shape of the micro-lenses confines the optical field to a spatial area with effective radii of 1.0 µm to 1.5 µm, depending on the radius of curvature of the lens. Consequently, the dispersion relation of cavity photons, and likewise, exciton-polaritons in such a trap acquires the canonical dispersion relation of a massive particle in a harmonic trap (see Fig. 1d), featuring a ladder of dispersion-less, dis- crete resonances with an approximately equal energy spacing. Recently, we have discussed that the trapping of exciton-polaritons in such a structure promotes the formation of room-temperature polariton condensation with high coherence24. However, the controllable confinement of polari- tonic quasi-particles in a single trap also composes a key building block towards the design of more advanced potential landscapes. In the presence of a neighboring hemispheric trap (see Fig. 1e), the optical modes transform into the frequency-split molecular resonances, with a well-controllable molecular coupling25, 26. Next, we study a one-dimensional periodic arrangement of potential traps under non-resonant excitation (532 nm) in the low-density limit. Fig. 1f shows the far field emission spectra of such an lattice (trap size 5 µm, next neighbor distance 2.5 µm). The observed emission below the condensation threshold strongly differs from an isolated hemispheric lens, where the trapped modes are discrete in energy and momentum24. Evanescent coupling of the modes confined in the polariton traps leads to formation of a distinct bandgap spectrum due to the spatial periodicity of the structure. Up to two Bloch-bands are visible in the spectrum of the lower polariton, as shown in Fig. 1f. Because of the deep confinement, a gap between the ground and the first excited Bloch-band evolves as large as 5 meV. The experimentally 5 determined bandgap structure of the spectrum can be fitted by a spectrum of single-particle eigen- states in a periodic one-dimensional lattice following the methodology introduced in27 (details on the model can be found in the methods section, and as well as supplementary information S2). The high stability of the Frenkel Excitons makes the non-linear regime of bosonic condensa- tion accessible. In the following experiment, we excite the one-dimensional lattice with a Gaussian spot (diameter ≈ 3 µm), focused on a hemispheric trap in the lattice. By increasing the pump power from P = 0.95 nJ/pulse over P = 1.5 nJ/pulse (Fig. 2b), P = 6.0 nJ/pulse (Fig. 2c) to P = 29.9 nJ/pulse (Fig. 2d), we experience a significant qualitative modification of the lattice spectrum. First, one can capture a significant, non-linear increase in the emission intensity, centered around the spec- tral region on top of the Bloch-band formed from photonic s-type wavefunctions. The non-linear threshold, characterized by the typical s-shape of the extracted peak area versus pump energy (see Fig 2e), the strong drop in the linewidth down to the resolution limit of our spectrometer, as well as the persisting blueshift of the mode above threshold puts our system well in the framework of room-temperature bosonic condensation. Once the condensation threshold is reached at P = 1.5 nJ/pulse (see Fig. 2b), in qualitative agreement with experiments conducted at 10 K in significantly more shallow potentials28, as well as in the tight-trapping limit29, 30, the polariton condensate emerges at the vicinity of the BZ edges k = ±π/a and progressively shifts into the gap of the structure, with an approximate energy offset of ≈ 0.56 meV with respect to the maximum of the ground band (Fig. 2c). With increasing power of the pump laser, the energy of the condensate shifts further into the gap (Fig. 2d), while 6 remaining localized at the edges of the BZ in the momentum space. Polaritonic condensation in our lattice is not restricted to the anti-binding orbital s-mode of our lattice. It is well-controllable by tuning the overlap between the pump laser and the op- tical mode, making our system extraordinarily versatile: As shown in Fig. 3a, similarly to the scenario discussed above, the on-site pumping condition loads the polaritonic condensate into the anti-binding s-mode, followed by a subsequent energy shift into the gap. The spatial coherence of this mode can be analyzed by measuring its correlation function g1(r, r (cid:48) ). We use a Michelson interferometer with a variable path length in the mirror-retroreflector configuration, which overlaps the realspace luminescence from the device with its mirror image on a beam-splitter, and combines them on a high resolution CCD camera12. The resulting image (Fig. 3b) shows a distinct interfer- ence pattern, in particular around the area of the pump spot. The spatially resolved visibility of the interference fringes, plotted in Fig. 3c, reflects that the coherent mode expands over various lattice sites. In strong contrast, off-site pumping conditions result in a significantly different phenomeno- logical behavior at any moderate pump power above the threshold of condensation. In this setting, the condensate occurs in the binding ground-state of the lattice (see Fig. 3d). Again, by laterally displacing the pump-laser beam, yielding an improved overlap with the p-type orbital mode of the second dispersive band, the condensation process can even be triggered on top of the bandgap (see Fig. 3e). The observed behavior of the condensed state in Fig. 2 and Fig. 3b (on-site pumping) hints at the presence of a spatially localized polariton gap state, which has previously been reported 7 in29 -- 31 for GaAs-based exciton-polariton condensates in photonic waveguide arrays32 and atomic Bose-Einstein condensates in optical lattices33 -- 35. Solitonic behavior is expected to manifest in the localization- and expansion of the conden- sate in our lattice in the high-density regime. Here, we studied the luminescence from our device under on-site pumping conditions and projected its real-space distribution onto a high-resolution camera. The leakage from the microcavity yields a stochastic decay of the quasiparticles, and re- sults in a characteristic decrease of the luminescence intensity away from the pump spot. Fig. 4a and Fig. 4b depict the energy-resolved realspace distribution of the emitted photoluminescence. One can clearly identify the luminescence maxima from the single sites in the lattice (see also angle-resolved measurements in supplementary information S3). While the pump power in Fig. 4a was chosen slightly above the onset of polaritonic condensation (P = 2.4 nJ/pulse), in Fig. 4b the condensation threshold was exceeded by approximately one order of magnitude (P = 18.9 nJ/pulse). The most dramatic effect of the condensation into the gap state can be seen from the intensity traces in Fig. 4c and Fig. 4d. As the excitation power increases by one order of magnitude, the spatial extension of the condensate becomes substantially narrower. This effect of self-localization, lead- ing to an effective reduction of the condensate expansion, is canonical for the formation of a gap solitonic mode and is an excellent agreement with our numeric model (see supplementary infor- mation S2). We furthermore witness a change in the shape of the condensed mode which transits from a squared hyperbolic secant profile (indicating the role of interactions in our systems) at mod- erate pump power to a dominantly exponential decay, in striking similarity to earlier experiments conducted on GaAs-based cavity polaritons in modulated wire cavities31. 8 A remarkable feature in our system, which was pointed out in an earlier work24, is the na- ture of the interaction in the fluorescent protein system, which is crucial for the formation of a gap-solitonic state36. While inter-particle repulsion in GaAs-based cavities is mostly induced by polariton-polariton interaction and interaction with the exciton reservoir, the density dependent blueshift in Frenkel Excitons mostly arises from the screening of the Rabi-splitting with elevated density24. As our experiments confirm, the nature of interaction is not of relevance for the observed phenomenological behavior for our system in the nonlinear regime. Conclusions We have established the organic platform of exciton-polaritons as a new, attractive player to study bosons, and in particular non-linear bosonic condensates in optical lattices. Our platform is low- cost, flexible and can be operated at ambient conditions. Here, we already observe the full band- structure of polaritonic Bloch-bands emerging in the regime of tight binding, the controlled loading of condensates into Bloch-bands with specific orbital symmetry, and the excitation of localized gap states driven by the intrinsic non-linearity of our system. Our work can be extended straight for- wardly to arbitrary two-dimensional lattice geometries, as well as synthetic time dimensions, for on-chip bosonic annealing, quantum simulation and topological polaritonics. Methods Experimental Setup Our optical setup can be used concurrent in near-field (spatially) and far- field (momentum space) resolved spectroscopy and imaging. We used two lasers to excite our 9 sample. First a 532 nm continuous-wave diode laser and second a wavelength-tunable optical parametric oscillator system with ns-pulses tuned to 532 nm that is resonant with the first Bragg minimum of the top mirror of our sample. For angle resolved measurements we used a Fourier imaging configuration. Here a Fourier lens collecting the angle-dependent information in the back- focal plane of the microscope objective (50x, 0.42 NA). To obtain a spatially resolved signal we used the Fourier configuration with an extra lens. The emission is filtered by a 550 nm longpass filter. Both the near-field and the far-field create an image in the focus plane of the spectrometer system. The slit in entrance plane is closed to exclude a narrow section of the image. The emitted signal is dispersed by an optical grating with 1200 lines/mm with a spectral resolution of about 230 µm for energies around 2 eV. For the Michelson measurements the emitted signal is deflected through a beam splitter and overlaid itself with a retroreflector and a mirror. Numerical Modelling For numerical calculations of the polariton dynamics in organic lattices we consider a two-dimensional mean-field model, which has been widely used for simulation of inorganic semiconductor cavities37, 38 and very recently for organic systems39. This model consists of the open-dissipative Gross-Pitaevskii (GP) equation for the condensate wave-function incorpo- rating stochastic fluctuations and coupled to the rate equation for the excitonic reservoir created by the off-resonant pump30, 38. In particular our theoretical analysis confirms that the gap state bifurcates from the upper edge of the band characterised by the "staggered" phase, and therefore inherits the characteristic π-phase shift between neighbouring density peaks36. For the further details on numerical modelling we refer to the supplementary informations. 10 References 1. I. Bloch, J. Dalibard and W. Zwerger, Rev. Mod. Phys. 80, 885 (2008). 2. R. Ozawa, H. M. Price A. Amo, N. Goldman, M. Hafezi, L. Lu, M. C. Rechtsman, D. Schuster, J. Simon, O. Zilberberg and I. Carusotto, Rev. Mod. Phys. 91, 015006 (2019). 3. N. Na and Y. Yamamoto, New J. Phys. 12, 123001 (2010). 4. F. Dalfovo, L. Pitaevskii and S. Stringari, Phys. Rev. A. 54, 4213 (1996). 5. I. Bloch, J. Dalibard and Sylvain Nascimb`ene, Nature Physics 8, 267 - 276 (2012). 6. R. Blatt and C. F. Roos, Nature Physics 8, 277 - 284 (2012). 7. A. A. Houck, H. E. Tureci and Jens Koch, Nature Physics 8, 292 - 299 (2012). 8. J. Q. You and Franco Nori, Nature 474, 589 - 597 (2011). 9. C. Sparrow, E. Martn-Lpez, N. Marviglia, A. Neville, C. Harrold, J. Carolan, Y. N. Joglekar, T. Hashimoto, N. Matsuda, J. OBrien, D. P. Tew and A. Laing, Nature 557, 660 - 667 (2018). 10. G. Munoz-Matutano, A. Wood, M. Johnsson, X. Vidal, B. Q. Baragiola, A. Reinhard, A. Lemaıtre, J. Bloch, A. Amo, G. Nogues, B. Besga, M. Richard and T. Volz, Nature Materials 18, 213 - 218 (2019). 11. A. Delteil, T. Fink, A. Schade, S. Hofling, C. Schneider and A. Imamoglu, Nature Materials 18, 219 - 222 (2019). 11 12. J. Kasprzak, M. Richard, S. Kundermann, A. Baas, P. Jeambrun, J. M. J. Keeling, F. M. Marchetti, M. H. Szyman´nska, R. Andr´e, J. L. Staehli, V. Savona, P. B. Littlewood, B. De- veaud, and L. S. Dang, Nature 443, 409 (2006). 13. C. Schneider, K. Winkler, M. D. Fraser, M. Kamp, E. A. Ostrovskaya and S. Hofling, Rep. Prog. Phys. 80, 1 (2016). 14. R. Balili, V. Hartwell, D. Snoke, L. Pfeiffer, and K. West, Science 316, 1007 (2007). 15. X. Liu, T. Galfsky, Z. Sun, F. Xia, E. Lin, Y. Lee, S. K`ena-Cohen and V. M Menon, Nature Photonics 9, 30 - 34 (2015). 16. M. Waldherr, N. Lundt, M. Klaas, S. Betzold, M. Wurdack, V. Baumann, E. Estrecho, A. Nalitov, E. Cherotchenko, H. Cai, E. A. Ostrovskaya, A. V. Kavokin, S. Tongay, S. Klembt, S. Hofling and C. Schneider, Nature Communications 9, 3286 (2018). 17. J. D. Plumhof, T. Stferle, L. Mai, U. Scherf and R. F. Mahrt, Nature Materials 13, 247 (2013). 18. K. S. Daskalakis, S. A. Maier, R. Murray and S. K`ena-Cohen, Nature Materials 13, 271 (2014). 19. C. P. Dietrich, A. Steude, L. Tropf, M. Schubert, N. M. Kronenberg, K. Ostermann, S. Hofling and M. C. Gather, Sci, Adv. 2, e1600666 (2016). 20. N. G. Berloff, M. Silva, K. Kalinin, A. Askitopoulos, J. D. Topfer, P. Cilibrizzi, W. Langbein and P. G. Lagoudakis, Nature Materials 16, 1120 - 1126 (2017). 12 21. S. Klembt, T. H. Harder, O. A. Egorov, K. Winkler, R. Ge, M. A. Bandres, M. Emmerling, L. Worschech, T. C. H. Liew, M. Segev, C. Schneider and S. Hofling, Nature 562, 552 - 556 (2018). 22. P. St-Jean, V. Goblot, E. Galopin, A. Lemaitre, T. Ozawa, L. Le Gratiet, I. Sagnes, J. Bloch and A. Amo, Nature Photonics 11, 651 - 656 (2017). 23. H. Suchomel, S. Klembt, T. H. Harder, M. Klaas, O. A. Egorov, K. Winkler, M. Emmerling, R. Thomale, S. Hofling and C. Schneider, Phys. Rev. Lett. 121, 257402 (2018). 24. S. Betzold, M. Dusel, O. Kyriienko, C. P. Dietrich, S. Klembt, J. Ohmer, U. Fischer, I. A. Shelykh, C. Schneider and S. Hofling, Preprint at https://arxiv.org/abs/1906.02509 (2019). 25. S. Dufferwiel, F. Li, A. A. P. Trichet, L. Giriunas, I. Farrer, D. A. Ritchie, J. M. Smith, M. S. Skolnick and D. N. Krizhanovskii, Appl. Phys. Lett. 107, 201106 (2015). 26. D. Urbonas, T. Stoferle, F. Scafirimuto, U. Scherf and R. F. Mahrt, ACS Photonics 3, 1542 (2016). 27. K. Winkler, J. Fischer, A. Schade, M. Amthor, R. Dall, J. Gessler, M. Emmerling, E.A. Ostro- vskaya, M. Kamp, C. Schneider, and S. Hofling, New J. Phys. 17, 023001 (2015). 28. C. W. Lai, N. Y. Kim, S. Utsunomiya, G. Roumpos, H. Deng, M. D. Fraser, T. Byrnes, P. Recher, N. Kumada, T. Fujisawa, and Y. Yamamoto, Nature 450, 529 (2007). 29. T. Gao, O. A. Egorov, K. Winkler, M. Kamp, C. Schneider, S. Hofling, A. G. Truscott and E. A. Ostrovskaya, Phys. Rev. Lett. 121, 225302 (2018). 13 30. K. Winkler, O. A. Egorov, I. G. Savenko, X. Ma, E. Estrecho, T. Gao, S. Muller, M. Kamp, T. C. H. Liew, E. A. Ostrovskaya, S. Hofling and C. Schneider, Phys. Rev. B 93, 121303 (2016). 31. D. Tanese, H. Flayac, D. Solnyshkov, A. Amo, A. Lemaitre, E. Galopin, R. Braive, P. Senellart, I. Sagenes, G. Malpuech, and J. Bloch, Nature Communication 4, 1749 (2013). 32. A. A. Sukhorukov and Y. S. Kivshar, Opt. Lett. 28, 2345 (2003). 33. B. Eiermann, T. Anker, M. Albiez, M. Taglieber, P. Treutlein, K.-P. Marzlin and M. K. Oberthaler, Phys. Rev. Lett. 92, 230401 (2004). 34. P. J. Y. Louis, E. A. Ostrovskaya, C. M. Savage and Y. S. Kivshar, Phys. Rev. A. 67, 013602 (2003). 35. P. Zi`n, E. Infeld, M. Matuszewski, G. Rowland and M. Tippenbach , Phys. Rev. A. 73, 063621 (2006). 36. E. A. Ostrovskaya, J. Abdullaev, M. D. Fraser, A. S. Desyatnikov, and Yu. S. Kivshar, Phys. Rev. Lett. 110, 170407 (2013). 37. M. Wouters and I. Carusotto, Phys. Rev. Lett. 99 140402 (2007). 38. M. Wouters, I. Carusotto, and C. Ciuti, Phys. Rev. B, 77, 115340 (2008). 39. N. Bobrovska, M. Matuszewski, K. L. Daskalakis and S. Kna-Cohen, ACS Photonics 5, 111 (2018). 14 40. See Supplementary Information for further experimental characterisation of the polariton con- densate in the 1D lattice as well as numerical modelling below and above threshold condensa- tion. Acknowledgements The Wurzburg group acknowledges financial support from the state of Bavaria. Assistance by Monika Emmerling during fabrication and patterning of the sample is gratefully ac- knowledged. Competing Interests The authors declare that they have no competing financial interests. Correspondence Correspondence and requests for materials should be addressed to C. Schnei- der (email: [email protected]) and M. Dusel (email: Marco.Dusel@uni- wuerzburg.de) Figure 1: Schematic images and angle-resolved measurements of our device. a - c, Artistic illustration of single trap (a), molecular configuration (b) and one-dimensional lattice (c). The hemispheric indentations are filled with the "mCherry" proteins (red). d - f, Angle-resolved pho- toluminescence spectrum of a single trapping site (d), as well as a molecular configuration (e) and the one-dimensional lattice (f), proving the formation of a bandgap polariton spectrum resulting from evanescent coupling between the sites. White lines show calculated single-particle energy bands in the effective periodic potential of the depth - 160 meV. The measurements are recorded at a detuning between the cavity photon and exciton energy of ∆ = EC − EX = - 100 meV (at k = 0 of the ground Bloch-band). abcfyzxyzxyzxde0.20.40.60.81.0Norm. intensity-4-20241.8551.8801.9051.930Energy (eV)ky (µm-1)0.20.40.60.81.0Norm. intensity-4-20241.8251.8501.8751.900x15x15ky (µm-1)0.20.40.60.81.0Norm. intensity-4-20241.8901.9151.9401.965ky (µm-1)Figure 2: Excitation power-dependent analysis of an one-dimensional lattice. a - d, Far- field photoluminescence spectra recorded at various pump powers. Excitation with a continu- ous wave laser (a), the white line shows calculated single-particle energy s-band in the effec- tive periodic potential. Angled-resolved spectra for pump powers at the condensation thresh- old (b, P = 1.5 nJ/pulse), above the threshold (c, P = 6.0 nJ/pulse) and far past the threshold (d, P = 29.9 nJ/pulse). The condensed mode moves into the gap. e, Integrated emission intensity (black) and linewidth (red) versus excitation energy. At P = 1.5 nJ/pulse the linwidth drops to the resolution limit of the spectrometer whereas the intensity clearly features a superlinear increase. f, Excitation power-dependent blueshift of the mode. -2021.9501.9551.960ky (µm-1)Energy (eV)0.050.100.501.00Log. norm. intensity-202ky (µm-1)2000040000Intensity (counts/s)-202ky (µm-1)20004000Intensity (counts/s)-202ky (µm-1)1020Intensity (counts/s)110100101102103104105106107Excitation energy (nJ/pulse)Integrated PL intensity (arb. u.)200250300350 Linewidth (µeV)1100.00.20.40.60.81.0Relative blueshift (meV)Excitation energy (nJ/pulse)1.5 nJ/pulse6.0 nJ/pulse29.9 nJ/pulseCW-LaseraefbcdFigure 3: Controlled loading of the polariton condensate in the lattice. off-site versus on- site pumping and formation of corresponding gap state condensates. a, d, e, Far-field spectra below (left) and above (right) the condensation threshold, corresponding with the pump-potential alignment inserted with a sketch of the position of the pump laser spot (green) aligned with re- spect to the polariton potential. b, Spatial coherence measurement under on-site excitation above threshold and c, extracted coherent magnitude of the interference pattern. cx (µm)y (µm)005101550.250.500.751.00Norm. Magnitudeb0.250.500.751.00Norm. intensity05x (µm)d1.0-2020.20.40.60.8Norm. intensity-2021.8701.8901.910ky (µm-1)ky (µm-1)Energy (eV)0.20.40.60.81.0Norm. intensityP<PthP>Ptha-202ky (µm-1)0.20.40.60.81.0Norm. intensity-2021.8501.8701.890ky (µm-1)Energy (eV)0.20.40.60.81.0Norm. intensityP<PthP>Pthky (µm)-2021.8651.8851.905ky (µm-1)Energy (eV)0.20.40.60.81.0Norm. intensitye-202-10.20.40.60.81.0Norm. intensityP<PthP>PthFigure 4: Self localization of a gap-solitonic mode. a, b, Energy-resolved realspace images under on-site pumping conditions little above threshold (a, P = 2.4 nJ/pulse), and far above threshold (b, P = 18.9 nJ/pulse). c, d, The intensity traces are fitted by a sech2 and an exponential function. For the lower excitation power the sech2-function fits to the intensity profile while for higher excitation power the exponential function fits better. In the insets, the same profiles are shown in log scale. Spatial narrowing of the extension of the condensate characterize the gap-soliton mode. -2021.941.951.96ky (µm-1)Energy (eV)0.20.40.60.81.0Norm. intensityCW-Laser-202ky (µm-1)0.20.40.60.81.0Norm. intensity2.4 nJ/pulse-202ky (µm-1)0.20.40.60.81.0Norm. intensity18.9 nJ/pulsefedabdc-5051.941.951.96Position (µm)0.20.40.60.81.0Norm. intensity-5051.941.951.96Energy (eV)Position (µm)0.20.40.60.81.0Norm. intensity-10-505100.010.11Log. normalized intensityPosition (µm)-20-100100.00.51.0Position (µm) exp sech2-10-505100.010.11Log. normalized intensityPosition (µm)-20-100100.00.51.0Normalized intensityPosition (µm) exp sech22.4 nJ/pulse18.9 nJ/pulse
1707.00252	2	1707	2017-08-25T12:38:12	Effect of magnetic field on the lasing threshold of a semimagnetic polariton condensate	[ "cond-mat.mes-hall" ]	We evidence magnetic field triggered polariton lasing in a microcavity containing semimagnetic quantum wells. This effect is associated with a decrease of the polariton lasing threshold power in magnetic field. The observed magnetic field dependence of the threshold power systematically exhibits a minimum which only weakly depends on the zero-field photon-exciton detuning. These results are interpreted as a consequence of the polariton giant Zeeman splitting which in magnetic field: leads to a decrease of the number of accessible states in the lowest polariton branch by a factor of two, and substantially changes the photon-exciton detuning.	cond-mat.mes-hall	cond-mat	Effect of magnetic field on the lasing threshold of a semimagnetic polariton condensate J.-G. Rousset,∗ B. Piętka,† M. Król, R. Mirek, K. Lekenta, J. Szczytko, W. Pacuski, and M. Nawrocki Institute of Experimental Physics, Faculty of Physics, University of Warsaw, ul. Pasteura 5, PL-02-093 Warsaw, Poland We evidence magnetic field triggered polariton lasing in a microcavity containing semimagnetic quantum wells. This effect is associated with a decrease of the polariton lasing threshold power in magnetic field. The observed magnetic field dependence of the threshold power systematically exhibits a minimum which only weakly depends on the zero-field photon-exciton detuning. These results are interpreted as a consequence of the polariton giant Zeeman splitting which in magnetic field: leads to a decrease of the number of accessible states in the lowest polariton branch by a factor of two, and substantially changes the photon-exciton detuning. I. INTRODUCTION (cid:54)(cid:8)(cid:55) (cid:12)(cid:10)(cid:13)(cid:14)(cid:15)(cid:25)(cid:17) (cid:16)(cid:2)(cid:14)(cid:18)(cid:8)(cid:19)(cid:20)(cid:6)(cid:21) (cid:22)(cid:17)(cid:14)(cid:23)(cid:24) (cid:12)(cid:10)(cid:13)(cid:14)(cid:15)(cid:16)(cid:17) (cid:10)(cid:3)(cid:11)(cid:11)(cid:9)(cid:7) (cid:10)(cid:3)(cid:11)(cid:11)(cid:9)(cid:7) (cid:2)(cid:3)(cid:4)(cid:5)(cid:6)(cid:7)(cid:8)(cid:6)(cid:9) (cid:25)(cid:33)(cid:34)(cid:34) (cid:25)(cid:33)(cid:34)(cid:34) (cid:25)(cid:33)(cid:25)(cid:34) (cid:25)(cid:33)(cid:35)(cid:16) (cid:25)(cid:33)(cid:35)(cid:38) (cid:25)(cid:33)(cid:35)(cid:16) (cid:25)(cid:33)(cid:25)(cid:35) (cid:25)(cid:33)(cid:35)(cid:16) (cid:25)(cid:33)(cid:25)(cid:34) (cid:25)(cid:33)(cid:34)(cid:34) (cid:25)(cid:33)(cid:22)(cid:16) (cid:25)(cid:33)(cid:37)(cid:16) (cid:25)(cid:33)(cid:35) (cid:25)(cid:33)(cid:36) (cid:25)(cid:33)(cid:35) (cid:26)(cid:27) (cid:28)(cid:29) (cid:30)(cid:31) (cid:32)(cid:9) (cid:26)(cid:27) (cid:28)(cid:29) (cid:30)(cid:31) (cid:32)(cid:9) (cid:26)(cid:27) (cid:28)(cid:29) (cid:30)(cid:31) (cid:32)(cid:9) (cid:26)(cid:27) (cid:28)(cid:29) (cid:30)(cid:29) (cid:32)(cid:9) (cid:26)(cid:27) (cid:28)(cid:29) (cid:30)(cid:31) (cid:32)(cid:9) (cid:26)(cid:27) (cid:28)(cid:29) (cid:30)(cid:31) (cid:32)(cid:9) (cid:26)(cid:27) (cid:28)(cid:29) (cid:30)(cid:31) (cid:32)(cid:9) (cid:26)(cid:27) (cid:28)(cid:29) (cid:32)(cid:9) (cid:26)(cid:27)(cid:32)(cid:9) (cid:39)(cid:8)(cid:40)(cid:5) (cid:25)(cid:33)(cid:34)(cid:34) (cid:25)(cid:33)(cid:35)(cid:16) (cid:25)(cid:33)(cid:22)(cid:16) (cid:25)(cid:33)(cid:35) (cid:25)(cid:33)(cid:36) (cid:25)(cid:33)(cid:35) (cid:25)(cid:33)(cid:37)(cid:22) (cid:25)(cid:33)(cid:35)(cid:38) (cid:38)(cid:35)(cid:33)(cid:34)(cid:14)(cid:29)(cid:42) (cid:38)(cid:43)(cid:33)(cid:37)(cid:14)(cid:29)(cid:42) (cid:15)(cid:25)(cid:14)(cid:29)(cid:42) (cid:38)(cid:43)(cid:33)(cid:37)(cid:14)(cid:29)(cid:42) (cid:38)(cid:35)(cid:33)(cid:34)(cid:14)(cid:29)(cid:42) (cid:34)(cid:37)(cid:25)(cid:14)(cid:29)(cid:42) (cid:15)(cid:14)(cid:41)(cid:42) (cid:54)(cid:4)(cid:55) (cid:50)(cid:20)(cid:52)(cid:3)(cid:20)(cid:27)(cid:14)(cid:53)(cid:9)(cid:14) (cid:18)(cid:7)(cid:21)(cid:46)(cid:5)(cid:6)(cid:8)(cid:6) (cid:2)(cid:49)(cid:22)(cid:14) (cid:48)(cid:44)(cid:8)(cid:6)(cid:9) (cid:50)(cid:20)(cid:29)(cid:9)(cid:8)(cid:7) (cid:48)(cid:46)(cid:44)(cid:8)(cid:7)(cid:20)(cid:51)(cid:9)(cid:7) (cid:30)(cid:8)(cid:31)(cid:29)(cid:9)(cid:6)(cid:20)(cid:18)(cid:14)(cid:11)(cid:20)(cid:9)(cid:44)(cid:27) (cid:45)(cid:17)(cid:18)(cid:20)(cid:6)(cid:8)(cid:6)(cid:20)(cid:46)(cid:29)(cid:14) (cid:44)(cid:8)(cid:5)(cid:9)(cid:7) (cid:2)(cid:47)(cid:5)(cid:48)(cid:8)(cid:18)(cid:9) (cid:20)(cid:42)(cid:8)(cid:31)(cid:20)(cid:29)(cid:31)(cid:14)(cid:44)(cid:9)(cid:29)(cid:5)(cid:9)(cid:5) (cid:12)(cid:9)(cid:6)(cid:9)(cid:18)(cid:6)(cid:20)(cid:46)(cid:29) FIG. 1: (a) Scheme of the sample designed for investigating semimagnetic polaritons. The layers are wedge-shaped with the average layer thicknesses given. The gradient of thickness is of the order of 2.5%/ cm. (b) Simplified experimental setup. of 22 pairs for the bottom mirror and 20 pairs for the top one.17,18 Polaritons are created non resonantly by a Al2O3:Ti pulsed laser which energy is tuned to the energy of the first reflectivity minimum on the high energy side of the microcavity stopband. All the presented results were obtained at a temperature of T = 1.5 K. To observe lasing triggered by magnetic field we apply the following procedure. In a first step, in zero magnetic field, the condensation threshold Pth(B = 0) is deter- mined based on the dependence of the emission on the excitation power, as shown in Fig. 2 (a-f). Then the excitation power is set to Pexc = 0.75 Pth(B = 0) and the magnetic field is increased. For the situation pre- sented in Fig. 2, the photon-exciton detuning (defined as the energy difference between the uncoupled photon and exciton at zero in-plane momentum) in zero mag- netic field is δ0 = −8.5 meV and the threshold power in zero magnetic field is Pth(B = 0) = 25 kW/cm2. The magnetic field is increased, keeping the excitation power constant Pexc = 19 kW/cm2. When the magnetic field increases to B ≈ 2 T, we observe polariton condensation similarly to what is observed when increasing the exci- tation power. The dispersion curves at chosen magnetic 7 1 0 2 g u A 5 2 ] l l a h - s e m . t a m - d n o c [ 2 v 2 5 2 0 0 . 7 0 7 1 : v i X r a In the last years, there has been a growing interest for the magneto-optical and spin polarization proper- ties of cavity polaritons and their Bose-Einstein conden- sates. Theoretical and experimental investigations evi- denced, among others, the suppression of superfluidity in magnetic field and the quenching of the Zeeman ef- fect for polariton Bose-Einstein condensates.1 -- 3 More re- cently, the condensation of polaritons triggered by mag- netic field and the phase transition between polariton lasing and photon lasing in magnetic field were investi- gated and interpreted in the frame of the shrinkage of the exciton wave function in magnetic field, the diamagnetic shift and the influence of magnetic field on free carriers and exciton diffusion.4,5 Another approach to study the magneto-optical properties of photonic structures takes advantage of the enhanced magneto-optical properties of diluted magnetic semiconductors (DMS). It was shown that the magneto-optical Kerr effect can be amplified by placing a DMS layer on a distributed Bragg reflector.6 -- 11 Magneto-optical effects of cavity polaritons can be en- hanced by embedding quantum wells containing Mn ions in the cavity.12 -- 19 The structure of the microcavity sam- ple studied in this work is shown in Fig. 1 with the sim- plified experimental setup. The average layer thicknesses are given. In fact the microcavity is wedge-shaped with a gradient of thickness of the order of 2.5%/ cm. In such a structure, the exchange interaction between the d-shell electrons of Mn ions and the s-shell electrons and p-shell holes of the exciton results in an angle dependent giant Zeeman splitting of semimagnetic cavity polaritons.20 In this work, we show that external magnetic field also af- fects the polariton properties in the non-linear regime of excitation. In particular we investigate the decrease of the polariton condensation threshold in magnetic field, which can lead to magnetic field induced lasing under constant excitation power. II. EXPERIMENTAL RESULTS Our sample consists of a 3λ-cavity embedding four 20 nm thick (Cd,Zn,Mn)Te quantum wells.19 The cavity is surrounded by distributed Bragg reflectors consisting fields for which the system is below or above threshold are presented in Fig. 2 (g-l). We observe a typical nar- rowing of the emission in momentum and energy space when increasing the excitation power in zero magnetic field (Fig. 2 (e), (f)) or the magnetic field at a fixed ex- citation power (Fig. 2 (k), (l)), which is characteristic of cavity polariton condensation and lasing. Investigation of the PL intensity dependence on the excitation power, in various magnetic fields from B = 0 T to B = 3 T was performed. As shown in Fig. 3, the non linear increase of PL intensity is observed and the condensation threshold power decreases with magnetic field. The polariton condensation triggered by magnetic field shown in Fig. 2 and the associated decrease of the las- ing threshold power led us to investigate the dependence of the polariton lasing threshold on magnetic field and photon-exciton detuning. Series of measurements per- formed for four different values of the photon-exciton detuning starting from positive δ0 = +3.0 meV, close to zero δ0 = −1.5 meV, and significantly negative δ0 = −2.4 meV and δ0 = −5.0 meV are presented in Fig.4. For a given value of the detuning, PL intensity measurement vs excitation power was performed several times at vari- ous magnetic fields between 0 T and 10 T. This allowed us to determine the lasing power threshold as a function of magnetic field. For each detuning, the threshold power decreases in magnetic field compared to the B = 0 T case. III. DISCUSSION Previous zero-field studies show that the polariton las- ing threshold power strongly depends on the shape of lower polariton dispersion curve which is determined by the photon-exciton detuning.21,22 Two types of related phenomena are considered: first - kinetic ones, such as the change of the polariton lifetime, relaxation processes and their efficiency, and second - thermodynamic ones, such as the change of the polariton effective mass, number of accessible states at the bottom of the lower polariton dispersion curve and the effective lattice temperature.21 Investigation of the condensation threshold dependence on the photon-exciton detuning show21,22 that when in- creasing the detuning from negative to positive values, the lower polariton branch forms a shallower trap in re- ciprocal space. In a shallower trap, polaritons encounter less scattering events to relax to the bottom of the trap and the threshold power decreases due to a more effi- cient relaxation of polaritons to the lowest energy state. For positive detuning, the lower polariton branch is pre- dominantly excitonic and a further increase of the detun- ing increases the effective mass of the polaritons and the number of accessible states on the lower polariton branch close to k = 0. This makes it harder for polaritons to reach an occupancy close to unity at the bottom of the lower polariton branch which is the condition to trigger the stimulated scattering to the lowest energy state, and 2 consequently increases the threshold power. A trade off between the kinetic limitation of the threshold at nega- tive detuning and thermodynamic limitation at positive detuning, minimizes the threshold power close to zero detuning.21,22 In microcavities with semimagnetic quantum wells pre- sented in this work, the photon-exciton detuning, but also the density of states depend strongly on magnetic field.20 Consequently we consider here five mechanisms which may be involved in the magnetic field dependence of the polariton lasing threshold power: Effect of excitonic shift in magnetic field. (i) The giant Zeeman splitting and related redshift of the low energy excitonic component leads to a more posi- tive photon-exciton detuning and increases the efficiency of polaritons relaxation along the lower polariton disper- sion curve, similarly to what is observed for nonmagnetic polaritons when increasing the detuning from negative to zero.21,22 This effect shall be important for negative photon-exciton detuning. (ii) The increase of the photon- exciton detuning induced by magnetic field increases the effective mass and the number of accessible states for po- laritons close to k = 0, making it harder for polaritons to reach occupancy close to unity and trigger the stim- ulated scattering to the lowest energy state. This effect shall be important for positive photon-exciton detuning. Effect of spin degeneracy and relaxation dy- namics. (iii) The magnetic field induced giant Zeeman splitting of polaritons lifts the spin degeneracy and con- sequently decreases by a factor of two the number of ther- mally accessible states23 in the lowest polariton branch compared to the B = 0 T case, when the σ+ and σ− spin polarizations are degenerate. We propose that this reduced density of states makes it easier for polaritons to reach occupancy close to unity. (iv) The magnetic field accelerates spin-lattice relaxation of magnetic ions24 -- 28 and excitons29, therefore we propose that it could also accelerate the dynamics of polaritons towards the low- est energy state. (v) The shrinkage of the exciton wave function in magnetic field results in an increase of the os- cillator strength and consequently a higher Rabi energy, which lowers the condensation threshold. Such a depen- dence on the Rabi energy of the condensation thresh- old has been studied e.g. for wide gap semiconductor microcavities.30,31 In order to discuss the importance of all these mecha- nisms, we analyze the dependence of the polariton con- densation threshold power on magnetic field for vari- ous values of zero magnetic field detuning δ0 (Fig. 4). The excitonic giant Zeeman effect leads to a symmetrical splitting which shifts the excitonic components to lower (σ+ component) and higher (σ− component) energy, but owing to the low temperature at which the experiments are conducted, only the lower component is significantly occupied. Therefore the giant Zeeman effect leads to an increase of the photon-exciton detuning for the occupied branch of polaritons. This means that for negative δ0, the absolute value of the detuning decreases with magnetic 3 FIG. 2: Comparison of two effects inducing polariton lasing: the increase of excitation power and the increase of magnetic field. (a)-(d) Angle resolved PL measurement of the polariton condensation in zero magnetic field when increasing the excitation power; the linear color scale is different for each panel. Panels (e) and (f) present the characteristic nonlinear intensity increase and linewidth narrowing at threshold. (g)-(j) Angle resolved PL measurements of the polariton condensation triggered by magnetic field under fixed excitation power; the same linear color scale is used for panels (g)-(j). Panels (k) and (l) present the characteristic nonlinear intensity increase and linewidth narrowing at threshold. In panel (l), for 1 T≤ B ≤ 1.6 T, the emission at zero angle results from the contribution of two lines with comparable intensity: a narrow one from condensed polaritons and a broad one from uncondensed polaritons, which intensity relative to the narrow line becomes negligibly small above B ≈ 2 T. field, whereas it increases with magnetic field for posi- tive δ0. From these considerations, if one assumes that the magnetic field induced changes of detuning (mecha- nism (i) and (ii)) are the most important factor govern- ing the dependence of the threshold power on magnetic field, one would expect that for negative δ0 the lasing threshold should decrease with magnetic field, whereas it should increase with magnetic field for positive δ0. How- ever, this is not what we observe in the data presented in Fig. 4(a-d), where the lasing threshold power decreases in magnetic field, independently of the zero field detun- ing δ0. Moreover, plotting our data as a function of the magnetic field induced detuning (including the excitonic Zeeman effect), we observe that zero-detuning seems not to be the optimal condition for condensation in magnetic field (4e-h), which differs from what is found for nonmag- netic polaritons, for which the condensation threshold power is minimized close to zero detuning. This clearly means that other effects should be seriously taken in to account to explain the observed decrease of the lasing threshold power in magnetic field. We notice that the minimum threshold power is roughly half of the threshold power in zero magnetic field (Fig. 4). This is a strong argument in favor of mecha- nism (iii) related to the lifting of spin degeneracy in mag- netic field. Indeed, for B ≈ 4 T, the spin polarization of the lower polariton near k = 0 is close to saturation (as shown in Fig. 5). Then, for B ≈ 4 T the number of accessible states at the bottom of the lower polariton branch (in the vicinity of k = 0) is roughly half the one in zero magnetic field. As a consequence, the number of polaritons needed for the onset of stimulated scattering to the lowest energy state, is also roughly decreased by a factor of two. th Comparing the four series of measurements shown in Fig. 4, we observe that the minimum value of the thresh- old power P min (normalized with respect to the thresh- old in zero magnetic field) is lower for negative zero-field detuning δ0 than for positive detuning δ0. This depen- dence can be understood in terms of an interplay between mechanisms (i), (ii), and (iii): when δ0 is negative, the magnetic field induced increase of detuning makes the 3T 2T 1T 0T 1010 109 108 107 ) t i n u L P . b r a ( y t i s n e n t i 10 Excitation power (kW/cm2) 100 FIG. 3: PL intensity vs excitation power for various magnetic fields. When the magnetic field increases from B = 0 T to B = 3 T, the polariton lasing threshold power decreases. 1.0 0.5 1.0 0.5 T 0 = B @ h t P / h t P 1.0 0.5 1.0 0.5 (a) = + 3.0 meV (b) = - 1.5 meV (c) = - 2.4 meV (d) = - 5.0 meV (e) 1.0 0.5 (f) 1.0 0.5 (g) 1.0 T 0 = B @ h t P / h t P 0.5 (h) 1.0 0.5 0 2 4 6 B (T) 8 10 0 -5 B induced detuning (meV) 10 5 4 80000 60000 ) t i n u . b r a ( y t i s n e n t 40000 20000 circular polarization circular polarization I 0 2 4 6 B (T) 8 10 FIG. 5: Magnetic field dependence of the PL intensity near k = 0 collected in σ+ and σ− circular polarization. The excitation power is above the lasing threshold in zero magnetic field. hanced polariton relaxation in magnetic field, similarly to spin-lattice relaxation (mechanism (iv)), then it would be even more efficient for high magnetic fields,28 which is not confirmed by the results presented in Fig. 4. There- fore, such a mechanism cannot be of major importance in our experiment. The same argument applies for other effects involving the direct influence of magnetic field on the exciton motion.4 Concerning the effect of the shrinkage of the exciton wave function in magnetic field leading to an increase of the coupling strength (mechanism (v)), or to a decrease of the exciton lifetime,4 owing to the small Bohr radius of excitons in CdTe QWs, such an effect is expected32 to be non negligible for magnetic fields higher than 25 T which is far above the range of magnetic fields applied in these investigations. FIG. 4: (a)-(d) Magnetic field dependence of the threshold power normalized to the value in zero magnetic field for var- ious values of the photon-exciton detuning in zero magnetic field δ0. (e)-(f) The same data presented as a function of the photon-exciton detuning taking into account the excitonic Zeeman effect. polaritons relaxation kinetics more efficient which addi- tionally decreases the threshold power (mechanism (i)). When δ0 is positive, the magnetic field induced increase of detuning increases the effective mass and density of states in the vicinity of k = 0 which tends to increase the threshold power (mechanism (ii)), but the influence of this latter mechanism is weaker than effect of the re- duced density of states (mechanism (iii)) resulting in a decrease of the threshold power. This interplay between mechanisms (i), (ii), and (iii) can also explain the increase of the threshold power for higher magnetic fields. If the decrease of the lasing threshold is due to en- IV. CONCLUSIONS The effect of polariton lasing induced by external mag- netic field for microcavity containing semimagnetic quan- tum wells is observed. We show that this effect is due to the decrease of the polariton lasing threshold power in magnetic field. From the analysis of the threshold power dependence on magnetic field for various values of the photon-exciton detuning we can infer that the mecha- nism leading to a decrease by roughly a factor two of the threshold power in magnetic field is the reduction by a factor of two of the number of accessible states in the vicinity of the lowest energy state of the σ+ polarized lower polariton dispersion curve. Since the efficiency of the threshold power decrease by magnetic field is higher for negative detuning than for positive detuning, lasing threshold is governed by an interplay between magnetic field induced changes of the density of states and photon- exciton detuning. The ability to control the operation of a polariton laser by a magnetic field opens new possibil- ities for combining fields of spintronics and polaritonics. Acknowledgments The authors acknowledge Jan Suffczyński and Michał for helpful discussions. The work was Matuszewski supported by the National Science Centre, Poland, under Projects No. 2014/13/N/ST3/03763, No. 2015/16/T/ST3/00506, No. 2015/18/E/ST3/00559, No. 2015/18/E/ST3/00558, and by the Polish Ministry of Science and Higher Education as a research project Di- amentowy Grant No. 0010/DIA/2016/45 in years 2016- 2020 and No. 0109/DIA/2015/44 in years 2015-2019. Research was carried out with the use of CePT, CeZa- Mat and NLTK infrastructures financed by the Euro- pean Union - the European Regional Development Fund within the Operational Programme "Innovative econ- omy" for 2007-2013. 5 ∗ Electronic address: [email protected] † Electronic address: [email protected] 1 Y. G. Rubo, A. Kavokin, and I. Shelykh, Phys. Lett. A 358, 227 (2006). 2 V. D. Kulakovskii, A. S. Brichkin, S. V. Novikov, C. Schneider, S. Höfling, M. Kamp, A. Forchel, and N. A. Gippius, Phys. Rev. B 85, 155322 (2012). 3 J. Fischer, S. Brodbeck, A. V. Chernenko, I. Led- erer, A. Rahimi-Iman, M. Amthor, V. D. Kulakovskii, L. Worschech, M. Kamp, M. Durnev, et al., Phys. Rev. Lett. 112, 093902 (2014). 4 V. P. Kochereshko, M. V. Durnev, L. Besombes, H. Mari- ette, V. F. Sapega, A. Askitopoulos, I. G. Savenko, T. C. H. Liew, I. A. Shelykh, A. V. Platonov, et al., Sci. Rep. 6, 20091 (2016). 5 V. P. Kochereshko, D. V. Avdoshina, P. Savvidis, S. I. Tsintzos, Z. Hatzopoulos, A. V. Kavokin, L. Besombes, and H. Mariette, Semiconductors 50, 1506 (2016). 6 D. P. Cubian, M. Haddad, R. André, R. Frey, G. Roosen, J. L. A. Diego, and C. Flytzanis, Phys. Rev. B 67, 045308 (2003). 7 J. Sadowski, H. Mariette, A. Wasiela, R. André, Y. Merle d'Aubigné, and T. Dietl, Phys. Rev. B 56, R1664 (1997). 8 M. Haddad, R. André, R. Frey, and C. Flytzanis, Solid State Commun. 111, 61 (1999). 9 M. Koba and J. Suffczyński, J. Electromagnet. Wave 27, 700 (2013). 10 M. Koba and J. Suffczyński, Europhys. Lett. 108, 27004 (2014). 11 W. Pacuski, J.-G Rousset, V. Delmonte, T. Jakubczyk, K. Sobczak, J. Borysiuk, K. Sawicki, E. Janik, and J. Kasprzak, Cryst. Growth Des. 17, 2987 (2017). 12 H. Ulmer-Tuffigo, F. Kany, G. Feuillet, R. Langer, J. Bleuse, and J. Pautrat, J. Cryst. Growth 159, 605 (1996). 13 H. Ulmer-Tuffigo, J. Bleuse, F. Kany, R. André, and L. S. Dang, Superlattice Microst. 22, 383 (1997). 14 A. Brunetti, M. Vladimirova, D. Scalbert, and R. André, Phys. Status Solidi C 2, 3876 (2005). 15 A. Brunetti, M. Vladimirova, D. Scalbert, M. Nawrocki, A. V. Kavokin, I. A. Shelykh, and J. Bloch, Phys. Rev. B 74, 241101 (2006). 16 A. Brunetti, M. Vladimirova, D. Scalbert, R. André, D. Solnyshkov, G. Malpuech, I. A. Shelykh, and A. V. Kavokin, Phys. Rev. B 73, 205337 (2006). 17 J.-G. Rousset, J. Kobak, T. Slupinski, T. Jakubczyk, P. Stawicki, E. Janik, M. Tokarczyk, G. Kowalski, M. Nawrocki, and W. Pacuski, J. Cryst. Growth 378, 266 (2013). 18 J.-G. Rousset, J. Kobak, E. Janik, T. Jakubczyk, R. Rud- niewski, P. Piotrowski, M. Ściesiek, J. Borysiuk, T. Slupin- ski, A. Golnik, et al., J. Cryst. Growth 401, 499 (2014). 20 R. Mirek, M. Król, K. Lekenta, 19 J.-G. Rousset, B. Pi¸etka, M. Król, R. Mirek, K. Lekenta, J. Szczytko, J. Borysiuk, J. Suffczyński, T. Kazimierczuk, M. Goryca, et al., Appl. Phys. Lett. 107, 201109 (2015). J.-G. Rousset, M. Nawrocki, M. Kulczykowski, M. Matuszewski, J. Szczytko, W. Pacuski, and B. Piętka, Phys. Rev. B 95, 085429 (2017). 21 J. Kasprzak, D. D. Solnyshkov, R. André, L. S. Dang, and G. Malpuech, Phys. Rev. Lett. 101, 146404 (2008). 22 E. Wertz, L. Ferrier, D. D. Solnyshkov, P. Senellart, D. Ba- joni, A. Miard, A. Lemaître, G. Malpuech, and J. Bloch, Appl. Phys. Lett. 95, 051108 (2009). 23 A. J. Leggett, Rev. Mod. Phys. 73, 307 (2001). 24 T. Strutz, A. M. Witowski, and P. Wyder, Phys. Rev. Lett. 68, 3912 (1992). 25 T. Dietl, P. Peyla, W. Grieshaber, and Y. M. d'Aubigné, Phys. Rev. Lett. 74, 474 (1995). 26 A. V. Scherbakov, A. V. Akimov, D. R. Yakovlev, W. Os- sau, G. Landwehr, T. Wojtowicz, G. Karczewski, and J. Kossut, Phys. Rev. B 62, R10641 (2000). 27 M. Goryca, D. Ferrand, P. Kossacki, M. Nawrocki, W. Pa- cuski, W. Maślana, J. A. Gaj, S. Tatarenko, J. Cibert, T. Wojtowicz, et al., Phys. Rev. Lett. 102, 046408 (2009). 28 M. Goryca, M. Koperski, T. Smoleński, L. Cywiński, P. Wojnar, P. Plochocka, M. Potemski, and P. Kossacki, Phys. Rev. B 92, 045412 (2015). 29 T. Smoleński, W. Pacuski, M. Goryca, M. Nawrocki, A. Golnik, and P. Kossacki, Phys. Rev. B 91, 045306 (2015). 30 R. Johne, D. D. Solnyshkov, and G. Malpuech, Appl. Phys. Lett. 93, 211105 (2008). 31 O. Jamadi, F. Réveret, E. Mallet, P. Disseix, F. Médard, M. Mihailovic, D. Solnyshkov, G. Malpuech, J. Leymarie, X. Lafosse, et al., Phys. Rev. B 93, 115205 (2016). 32 A. V. Kavokin, J. J. Baumberg, G. Malpuech, and F. P. Laussy, Microcavities, Series on Semiconductor Science and Technology (Oxford University press, New York, 2007).
1712.09512	1	1712	2017-12-27T07:03:32	Finite temperature magnon spectra in yttrium iron garnet from mean field approach in tight-binding model	[ "cond-mat.mes-hall" ]	We study magnon spectra at finite temperature in yttrium iron garnet from tight-binding model with nearest neighboring exchange interaction. The spin reduction due to thermal magnon excitation are taken into account via the mean field approximation to the local spin and found to be different at two sets of iron atoms. The resulting temperature dependence of the spin wave gap shows good agreement with experiment. We find only two magnon modes are relevant to ferromagnetic resonance.	cond-mat.mes-hall	cond-mat	Finite temperature magnon spectra in yttrium iron garnet from mean field approach in tight-binding model Ka Shen The Center for Advanced Quantum Studies and Department of Physics, Beijing Normal University, Beijing 100875, China (Dated: August 9, 2021) We study magnon spectra at finite temperature in yttrium iron garnet from tight-binding model with nearest neighboring exchange interaction. The spin reduction due to thermal magnon excitation are taken into account via the mean field approximation to the local spin and found to be different at two sets of iron atoms. The resulting temperature dependence of the spin wave gap shows good agreement with experiment. We find only two magnon modes are relevant to ferromagnetic resonance. I. INTRODUCTION Since its discovery decades ago [1], yttrium iron garnet (YIG) has been regarded as one of the most important magnetic materials due to its extremely low magnetic damping and other intriguing properties [2]. As an insu- lator, YIG is free from Joule heating, revealing its promis- ing applications in future low energy consumption de- vices, which makes it a popular material for recent study on magnonics [3, 4] and spin caloritronics [5]. While the applications in magnonics mainly rely on the coher- ent transport properties of spin waves with relatively long wavelength (∼ µm) and low frequency (∼GHz), in spin caloritronic devices, the short wavelength spin waves, namely, magnons, dominate because of their sig- nificant population from thermal excitation. In the latter case, recent experiments show that the diffusion length of thermal magnons can reach tens of microns [6, 7] and present very interesting behaviors with respect to vari- ation of external magnetic field and temperature [8 -- 10] which have not been fully understood so far. Moreover, the magnon diffusion length is also found to be sensitive to the method of measurement [6, 7]. The full under- standing on those observation obviously requires a com- prehensive study of dissipation process beyond the long wavelength limit into a wide range of magnon frequency up to THz. In the literature, great efforts [11 -- 13] have been de- voted to explain the origin of magnetic damping in YIG measured by ferromagnetic resonant [14 -- 19] or para- metric pumping [20, 21]. Several mechanisms are be- lieved to be relevant, such as magnon-impurity scatter- ing, magnon-phonon scattering, and magnon-magnon in- teraction as well as spin-lattice relaxation. However, both the uniform mode in ferromagnetic resonant exper- iment and finite wave vector magnon from parametric pumping lie in GHz regime, therefore, the applicability of the conclusions from GHz magnons to thermal magnons of THz is questionable. In some recent works, the life- time of the thermal magnons in YIG were estimated with a single parabolic dispersion [8], which however is valid only for GHz magnons and could depart far from the real spectra. Moreover, at room temperature, several magnon bands could be excited in YIG [22, 23], which again re- veals the limitation of the single band model. The full magnon band structure of YIG were mod- eled by Harris [24] within nearest neighboring exchange interaction and measured from neutron scattering by Plant [22], from which the exchange interaction coeffi- cients were fitted out [2]. The next nearest neighboring exchanges were recently also taken into account with the exchange integrals calculated from first principle calcu- lation [25] or fitted to neutron scattering data [26, 27]. Atomistic dynamic simulation shows that the nearest neighboring exchange model with stochastic thermal fluc- tuation is sufficient to quantitatively reproduce the de- crease of the spin wave gap, minimal frequency of the antiferromagnetic-like mode, with increasing tempera- ture [23]. The reason for such a red shift behavior is that the effective magnetic field a local spin felt is suppressed by the reduction of the expectation value of neighboring spins due to magnon excitation. In the low temperature regime, the reduction of average spin is too small to be relevant, resulting in saturate spin wave gap and magne- tization, observed in experiment. The gap and magneti- zation from the atomistic dynamic simulation of classical spin however keep linearly increasing and fails to saturate at very low temperature [23], in contrast to the results from experiment and Harris's quantum model with the same parameters. In the present work, we develop a mean field approach based on nearest neighboring exchange model, where the magnon excitation at finite temperature is taken into ac- count. In our approach, the temperature dependence of the entire magnon spectra can be easily obtained by self- consistently solving the magnon population and its influ- ence on dispersion. With temperature independent ex- change constants, the resulting spin wave gap and mag- netization show good agreement with experiment from zero up to room temperature. In addition to the proper description of the entire spectra beyond the widely used long wavelength approximation, another apparent advan- tage of our approach is that the output tight-binding type wave functions contain only twenty components, which is therefore convenient for quantitative computation of var- ious spin wave properties, e.g., spin wave dissipation due to different mechanisms, as well as the responses to dif- ferent sources, e.g., light or microwave. Moreover, the information of the polarization, responsible to spin See- beck effect [23], is also naturally included in the wave functions and easy to read out. Therefore our approach should be useful for study various physics in YIG at fi- nite temperature. As an application, we discuss ferro- magnetic resonance and find only two of twenty modes are relevant. Our approach can also be extended to nano structure. II. EXCHANGE MODEL In one unit cell in YIG, the collinearly aligning iron atoms at ground state are separated into two sets, eight a atoms and twelve d atoms, according to the polariza- √ tion or the configuration of neighboring oxygen atoms [2]. √ The closest distances are raa = ( 3/4)a0 between two a- √ site atoms, rdd = ( 6/8)a0 between two d-atoms, and rad = ( 5/8)a0 between a and d atoms with a0 being the lattice constant of face-centered cubic lattice. Fol- lowing Harris [24], we start from the nearest neighboring exchange Hamiltonian H = − N(cid:88) (cid:88) rij=raa i 8(cid:88) (cid:88) (cid:88) rij=rad rij=rdd Jaa n=1 (cid:104) 8(cid:88) 20(cid:88) 8(cid:88) i=9 i=1 +2Jad +Jdd Sa(Rin)· Sa(Rin + rij) Sa(Rin)· Sd(Rin + rij) Sd(Rin)· Sd(Rin + rij) 20(cid:88) (cid:105) ,(1) + gµBB· Sa(Rin) + gµBB· Sd(Rin) i=1 i=9 where Jaa, Jdd, and Jad represent the corresponding ex- change coupling constants. N is the number of unit cell. The last two terms are Zeeman energy. In YIG, all these coupling constants are negative [2]. The coordinates of atoms can be found in Ref. [24]. Without loss of generality, we assume that the equilib- rium magnetization follows the small external magnetic field along z axis. By defining transverse components S± = Sx ± iSy, the Holstein-Primakoff transformation of two antiferromagnetically coupled sublattices can be written as [28, 29] a = Sa − a†a, S+ Sz d = −Sd + d†d, S+ Sz 2Sa − a†aa, 2Sd − d†d, a,d = (S+ (2) and S− a,d)† where we have taken into account the antiferromagnetic alignment of two sublattices. Thus, a† and d† are the creation operators of clockwise rota- (cid:80) tion in a-site and anticlockwise rotation in d-site, respec- tively. Applying Eq. (2) into Eq. (1) and Fourier trans- k ai(k)eik· Rin , one obtains the tight- form ain = 1√ N (cid:112) d = d†(cid:112) a = 2 binding type Bogoliubov-de Gennes Hamiltonian for an arbitrary wave vector k (the quadratic order) [2, 24] Hk = † Aij(k)a i (k)aj(k) + † Dij(k)d i (k)dj(k) 20(cid:88) † [Bij(k)a i (k)d i,j=9 † j(−k) + h.c.], (3) 8(cid:88) 8(cid:88) i,j=1 + 20(cid:88) i=1 j=9 where Aij(k) = (−12JadSd + 16JaaSa + gµBB)δij Dij(k) = (−8JadSa + 8JddSd − gµBB)δij −2JaaSaγaa ij (k), −2JddSdγdd (cid:112) ij (k), SaSdγad Bij(k) = −2Jad ij (k) = (cid:80)rij=rηη(cid:48) eik· rij . The last term in with γηη(cid:48) Eq. (3) leads to angular momentum exchange between the two sublattices preserving the chirality. ij (k), (4) III. RESULTS In this section, we present our results from the diag- onalization of the above Bogoliubov-de Gennes Hamil- tonian. The exchange constants are specified to be Jad = −39.8 K, Jdd = −13.4 K, and Jaa = −3.8 K [2]. Both a- and d-sites are approximately of Sa = Sd = 5/2. Note that, instead of unitary transform in fermionic sys- tem, a paraunitary transform is needed in the diagonal- ization procedure of bosonic Hamiltonian matrix under † † basis (ak, dk, a −k, d −k) [30, 31]. The resulting eigenstates contains two sets † 1,kai(cid:48)k + cij(cid:48) αik = cii(cid:48) 2,kd j(cid:48)−k, † i(cid:48)−k + cjj(cid:48) βjk = cji(cid:48) 3,ka 4,kdj(cid:48)k, which satisfy the Boson commutation rule † i(cid:48)k(cid:48)] = δii(cid:48)δkk(cid:48), [αik, α † j(cid:48)k(cid:48)] = δjj(cid:48)δkk(cid:48), [βjk, β (5) (6) (7) (8) guaranteed by the paraunitary transform. Einstein sum- mation convention has been applied with i, i(cid:48) = 1, ..., 8 and j, j(cid:48) = 1, ..., 12. Since the creation of an anticlock- wise rotation of d-site (d†) is equivalent to the annihila- tion of a clockwise rotation, αik are annihilation opera- tors of pure clockwise rotation. In the same sense, βjk are about purely anticlockwise rotation. The dynamics of the transverse components of local spin for two types of modes read (Sx (Sx (Sx (Sx i(cid:48)a, Sy j(cid:48)d, Sy i(cid:48)a, Sy j(cid:48)d, Sy i(cid:48)a)αik(t) = ((cid:60)cii(cid:48) j(cid:48)d)αik(t) = ((cid:60)cij(cid:48) i(cid:48)a)βjk(t) = ((cid:60)cji(cid:48) j(cid:48)d)βjk(t) = ((cid:60)cij(cid:48) 1,k(t),(cid:61)cii(cid:48) 2,k(t),(cid:61)cij(cid:48) 3,k(t),−(cid:61)cji(cid:48) 4,k(t),−(cid:61)cij(cid:48) 1,k(t)) 2,k(t)) 3,k(t)) 4,k(t)) (9) (10) (11) (12) where the time evolution of the coefficients are cii(cid:48) 1,k(t) = eik· ri(cid:48)−iωαik tcii(cid:48) 2,k, cji(cid:48) 3,k(t) = −ik· ri(cid:48)−iωβjk tcji(cid:48) e 4,k. The in- verse transform gives the original site-operators in form of a combination of magnon operators of eigenstates 2,k(t) = e−ik· rj(cid:48)−iωαik tcij(cid:48) 4,k(t) = eik· rj(cid:48)−iωβjk tcij(cid:48) 1,k, cij(cid:48) 3,k, cij(cid:48) 3 ai(cid:48)k = C i(cid:48)i dj(cid:48)k = C j(cid:48)i 1,kαik + C i(cid:48)j 3,−kα 2,−kβ † i−k + C j(cid:48)j 4,kβjk. † j−k, The fluctuation of local spins is then given by † (cid:104)a i(cid:48)ai(cid:48)(cid:105) = 2,k2((cid:104)β † ikαik(cid:105) + C i(cid:48)j 1,k2(cid:104)α C i(cid:48)i 1 N (13) (14) † jkβjk(cid:105) + 1), (cid:88) (cid:88) k k (cid:104)d † j(cid:48)dj(cid:48)(cid:105) = 1 N C j(cid:48)i 3,k2((cid:104)α † ikαik(cid:105) + 1) + C j(cid:48)j 4,k2(cid:104)β (15) † jkβjk(cid:105), (16) which shows that at very low temperature, where † ikαik(cid:105) ≈ the thermal excitation is negligible, † jkβjk(cid:105) ≈ 0, there is still non-zero fluctuation due to the (cid:104)β coupling between a- and d-site rotations via Bij terms in Eq. (3) [32, 33]. i.e., (cid:104)α A. Zero temperature spectra Fig. 1(a) shows all twenty magnon branches in [110] and [100] directions carried out from zero temperature in the absence of external magnetic field. The characteris- tics of polarization αik and βik are distinct from the color red and blue, respectively. Typically, the eight clockwise modes have higher energy than the twelve anticlockwise modes, since the former correspond to the a-site spins ro- tating under the exchange field of six nearby d-site spins, which is stronger (∼ 12JadSd) than the one felt by the d-site spin from the four nearby a-site spins (∼ 8JadSa). This is confirmed by Fig. 1(b) from the calculation with the same parameters but setting all Bij = 0, therefore, no coupling between clockwise and anticlockwise rota- tions. From comparison between the results with and without Bij terms, we can see that the angular momen- tum exchange between the sublattices partially lifts the degeneracy at high symmetry points, and more impor- tantly leads to the correct dispersion of acoustic mode and the antiferromagnetic-like mode, labeled as β1 and α1, separately. The β1 mode is almost isotropic up to 5 THz and starts to become anisotropic approaching the boundary of the Brillouin zone. In Fig. 1(c) we plot the instantaneous spin configura- tion of each mode at the center of Brillouin zone (k = 0). The transverse spins in both α1 and β1 are parallel within the sublattices and antiparallel between sublattices. The magnitudes of spins in β1 mode are all the same, as a result, this tilted configuration is a ground state of the system and stable, explaining its zero frequency. The a FIG. 1: Zero temperature spin wave spectra (a) with and (b) without inter-sublattice coupling Bij. The blue and red colors separate spin waves into two group according to polarizations. (c) Instantaneous spin configuration of all twenty modes in (a) near k = (0, 0, 0). spins in α1 contain larger transverse spin components and therefore dominate the rotation direction of this mode. Without Bij, the two sublattices are decoupled, as a result, the parallel configuration of each sublattice (α(cid:48) 1 and β(cid:48) 1) is of highest energy among the separately clock- wise and anticlockwise modes due to negative exchange coupling constant between the same type of atoms [see Fig. 1(b)]. B. Finite temperature spectra At low temperature, the density of the thermally ex- cited magnon is low enough so that the magnon disper- sion calculated at zero temperature still works well. How- ever as more magnons are excited at high temperatures, the enhanced magnon-magnon interaction can strongly affect the magnon dispersion. Physically, for example, the large magnon density reduces the magnitude of the expectation value of all local spins, hence the exchange field felt by the neighboring spins. The effect of magnon- magnon interaction can be taken into account by includ- 0 5 10 15 20 25-1-0.5 0 0.5 1 1.5 2ω (THz)k (π/a)α1β1(a)[110][100] 0 5 10 15 20 25-1-0.5 0 0.5 1 1.5 2ω (THz)k (π/a)α'1β'1(b)[110][100]increasing frequency(c)a sited sitemode4 FIG. 2: Spin wave spectra at (a) 100 K and (b) 300 K. (c) Effective local spin at a and d as function of temperature. (d) Minimal frequency of antiferromagnetic-like mode as function of temperature. The inset gives its dependence on magnetization. ing the higher order terms beyond quadratic approxima- † † † † † † tion, e.g., those in form of a i aid i aidj, a i a i aiaj, a i a jdj, etc. The modification in magnon spectra then can be carried out by diagrammatic calculation on the real part of the magnon self-energy due to these interaction terms. Here, alternatively, we perform a mean field approxima- tion with a self-consistent procedure, whose advantage has been discussed in the Introduction. Specifically, we keep all magnon-magnon interactions with up to three magnon operators at the same site and apply † † i ai → (cid:104)a i ai(cid:105), a † † jdj → (cid:104)d jdj(cid:105), d † † † † i ai → 2(cid:104)a i ai(cid:105)a i a a i † † † † jdj → 2(cid:104)d jdj(cid:105)d d jd j (17) (18) for interactions between a-site spins to reduce the in- teraction terms back to quadratic order. The resulting effective quadratic Hamiltonian is exactly of the same expression as Eq. (3) after a replacement of Sa(d) by a(d) = Sa(d) − ∆Sa(d). Seff (19) where ∆Sa(d) stands for the average of thermal excitation over all a(d) site spins C i(cid:48)i 1,k2nB(ωαik ) + C i(cid:48)j 2,k2nB(ωβi−k ), ∆Si(cid:48)a = (cid:88) (cid:88) k 1 N ∆Sj(cid:48)d = 1 N C j(cid:48)i 3,k2nB(ωαi−k) + C j(cid:48)j 4,k2nB(ωβik ). k (20) Here, nB corresponds to Plank distribution at thermal equilibrium. By diagonalizing Eq. (3) with Seff a(d), we ob- tain modified spin wave spectra, from which we compute updated Seff a(d) and repeat such process until reaching a self-consistent solution. In Figs. 2(a) and (b), we plot the convergent magnon spectra at 100 K and 300 K, respectively. As we can see, the shape of each band roughly remain the same at different temperatures. However, as the temperature increases, the clockwise rotating mode (in red) move towards to lower frequency regime, which is consistent with experimental observation [22] and atomistic simula- tion [23]. The anticlockwise rotating modes are relatively insensitive to the temperature change. In Fig. 2(c), the effective magnitudes of two sets spins, i.e., Eq. (19), are plotted as function of temperature, which shows weak dependence in the low temperature regime, and an approximately linear decrease in the tem- perature regime. Interestingly, the reduction at d-site is much larger than that at a-site. The reason for such be- havior is that the d site has larger components than a site in the low energy magnon bands, who are efficiently excited. At room temperature, the effective spin reduces to 2.21 and 2.03 respectively. These distinct reductions are actually the origin of the relative shift between an- ticlockwise and clockwise sets in Figs. 2(a) and (b), be- cause the frequencies of the clockwise modes rely on the magnitude of d spins. If the two effective magnitudes are equal Seff d , all frequencies would vary simultane- ously by a global prefactor according to Eq. (3), which is proportional to Seff d at vanishing magnetic field. a = Seff a = Seff We summarize in Fig. 2(d) the frequency at the but- ton of the α1 mode (i.e., the spin wave gap mentioned in the Introduction) at different temperatures as black solid squares. For comparison, the experimental data [22] and results from atomistic simulation [23] are plotted in the same figure as red bullet and open squares, respectively. As we can see that our results show good agreement with experiment especially at low temperature, where limited magnon are excited hence approaching to the zero tem- perature case. Our approach therefore avoids the low 0 5 10 15 20 25-1-0.5 0 0.5 1 1.5 2ω (THz)k (π/a)(a) 100 K[110][100] 0 5 10 15 20 25-1-0.5 0 0.5 1 1.5 2ω (THz)k (π/a)(b) 300 K[110][100] 4 6 8 10 0 100 200 300 400 500 600ω (THz)Temperature (K)(d)This workExp.Ref.23 2 4 6 8 0.4 0.8ω (THz)Ms /M0 1.6 2.0 2.4 Seffa(d)(c)SeffaSeffdtemperature deviation in atomistic simulation based on classical spin. In the inset, we use the magnetization from our calculation at finite temperature and plot the frequency as function of the corresponding magnetiza- tion of the same temperature, which presents good linear dependence. The magnetization at room temperature is around µ0Ms(300 K) = 0.163 T , being reasonable agree- ment with experimetal data 0.175 T [34]. The deviation from experimental value is because of the fact that the temperature dependence of the magnon spectra was not taken into account in the procedure to determine the ex- change constants from temperature dependence of mag- netization [2]. The determination of the correction in exchange constants due to the change in magnon spectra is beyond the scope of the present work. IV. FERROMAGNETIC RESONANCE As discussed in the Introduction, with the knowledge of wave functions, one can calculate various properties. Here, we take ferromagnetic resonance for an example. Assume spatially uniform ac magnetic field transverse to the magnetization, we write out the time dependent perturbation Hamiltonian to describe its coupling with magnetic atoms (cid:88) (cid:88) H(cid:48)(t) = gµBB(t)· [ Sa(Rin) + Sd(Rjn)]. (21) i,n j,n For linear polarization magnetic field along x-direction, j βj0) + h.c. i αi0 + Gβ j(cid:48) C j(cid:48)i∗ 3,0 , Gβ j = (cid:80) i(cid:48) C i(cid:48)j∗ (22) √ H(cid:48)(t) = N gµBBx(t)(Gα i(cid:48) C i(cid:48)i 1,0 + (cid:80) i = (cid:80) 2 (gµBB0)2N(cid:8)Gα π (cid:80) with Gα j(cid:48) C j(cid:48)j Γ(ω) = 2,0 + 4,0, and Bx(t) = B0 cos(ωt). Here, only k = 0 survives in the space integration. The absorption rates are then given by Fermi Golden rule j 2[NB(ωβj0) + 1]δ(ω − ωβj0 )(cid:9), i 2[NB(ωαi0 ) + 1]δ(ω − ωαi0) (23) +Gβ which is proportional to the number of unit cells in mag- netic sample and the input power (∝ B2 0 ). By using the 5 1 and Gβ coefficients in wave function C1,2,3,4, we find only two of the prefactors Gα 1 are non-zero. This actually can be easily understood from the spin configuration in Fig. 1(c), where only α1 and β1 modes have net spin. The vanishing of net spin in other modes makes them transparent to driving field. V. CONCLUSION AND DISCUSSION In conclusion, we calculated the entire magnon spectra at finite temperature within mean field approximation on the magnon-magnon interaction. The temperature de- pendence of magnon spectra shows good agreement with experiment. We find the reduction in magnitude of d spin is larger than that of a atoms. By using the wave func- tions of eigenstates, we analyze the ferromagnetic reso- nance and find that only two modes can be drived by spatially uniform ac magnetic field. We suppose our ap- proach which provides proper magnon spectra and corre- sponding wave functions can be usefully to study various properties in yttrium iron garnet at finite temperature as discussed in the Introduction. One may notice that in Fig. 2 the residual magnetiza- tion (effective local spin) near Curie temperature (560 K) is still finite, which reveals the limitation of the spin wave approximation in the high temperature regime [2]. In the literature, the mean field theory based on sublattices' magnetizations was found to perform well up to Curie temperature [35]. How to link this theory to our tight- binding type mean field approach and how to extend the present spin-wave-based model up to Curie temperature in ferrimagnetic systems are still under study. Acknowledgments The author thanks G. E. W. Bauer and J. L. Cheng for valuable discussions. This work is supported by the Recruitment Program of Global Youth Experts. 1 S. Geller and M. A. Gilleo, Acta Crystallographica 10, 239 6 B. L. Giles, Z. Yang, J. S. Jamison, and R. C. Myers, Phys. (1957). Rev. B 92, 224415 (2015). 2 Y. Cherepanov, I. Kolokolov, and V. L'vov, Phys. Rep. 7 L. J. Cornelissen, J. Liu, R. A. Duine, J. B. Youssef, and 229, 81 (1993). 3 A. A. Serga, A. V. Chumak, and B. Hillebrands, J. Phys. D: Appl. Phys. 43, 264002 (2010). 4 A. V. Chumak, W. I. Vasyuchka, A. A. Serga, and B. Hille- brands, Nature Phys. 453 (2015). B. V. Wees, Nature Phys. 11, 1022 (2015). 8 L. J. Cornelissen, K. J. H. Peters, G. E. W. Bauer, R. A. Duine, and B. J. van Wees, Phys. Rev. B 94, 014412 (2016). 9 L. J. Cornelissen and B. J. van Wees, Phys. Rev. B 93, 5 G. E. W. Bauer, E. Saitoh, and B. J. van Wees, Nature 020403 (2016). Mater. 391 (2012). 10 L. J. Cornelissen, J. Shan, and B. J. van Wees, Phys. Rev. 6 B 94, 180402 (2016). 217201 (2016). 11 E. Schlomann, Phys. Rev. 121, 1312 (1961). 12 M. Sparks, Ferromagnetic Relaxation Theory (MacGraw- 24 A. B. Harris, Phys. Rev. 132, 2398 (1963). 25 L.-S. Xie, G.-X. Jin, L. He, G. E. W. Bauer, J. Barker, Hill, New York, 1964). 13 A. Akhiezer, V. Baryakhtar, and S. Peletminskii, Spin waves (North Holland, Amsterdam, 1968). 14 E. G. Spencer, R. C. LeCraw, and A. M. Clogston, Phys. Rev. Lett. 3, 32 (1959). 15 E. G. Spencer and R. C. LeCraw, Phys. Rev. Lett. 4, 130 (1960). 16 R. C. Fletcher, R. C. LeCraw, and E. G. Spencer, Phys. Rev. 117, 955 (1960). 17 J. Nemarich, Phys. Rev. 136, A1657 (1964). 18 P. Roschmann, IEEE Transactions on Magnetics 11, 1247 (1975). 19 S. Klingler, H. Maier-Flaig, C. Dubs, O. Surzhenko, R. Gross, H. Huebl, S. T. B. Goennenwein, and M. Weiler, Appl. Phys. Lett. 110, 092409 (2017). 20 T. Kasuya and R. C. LeCraw, Phys. Rev. Lett. 6, 223 (1961). and K. Xia, Phys. Rev. B 95, 014423 (2017). 26 J. S. Plant, J. Phys. C 16, 7037 (1983). 27 A. J. Princep, R. A. Ewings, S. Ward, S. Toth, C. Dubs, D. Prabhakaran, and A. T. Boothroyd, njp Quantum Ma- terials 2, 63 (2017). 28 T. Holstein and H. Primakoff, Phys. Rev. 58, 1098 (1940). 29 M. Takahashi, Phys. Rev. B 40, 2494 (1989). 30 J. Colpa, Physica A: Statistical Mechanics and its Appli- cations 93, 327 (1978). 31 B. Flebus, K. Shen, T. Kikkawa, K.-i. Uchida, Z. Qiu, E. Saitoh, R. A. Duine, and G. E. W. Bauer, Phys. Rev. B 95, 144420 (2017). 32 A. Kamra and W. Belzig, Phys. Rev. Lett. 119, 197201 (2017). 33 A. Kamra, U. Agrawal, and W. Belzig, Phys. Rev. B 96, 020411 (2017). 34 S. A. Manuilov, S. I. Khartsev, and A. M. Grishin, J. Appl. 21 A. G. Gurevich and A. N. Anisimov, Zh. Eksp. Teor. Fiz Phys. 106, (2009). 68, 677 (1975). 35 T. B. Mitchell and P. E. Wigen, Journal of Applied Physics 22 J. S. Plant, J. Phys. C 10, 4805 (1977). 23 J. Barker and G. E. W. Bauer, Phys. Rev. Lett. 117, 61, 3259 (1987).
1209.2694	1	1209	2012-09-12T19:23:22	Fano resonance resulting from a tunable interaction between molecular vibrational modes and a double-continuum of a plasmonic metamolecule	[ "cond-mat.mes-hall", "physics.optics" ]	Coupling between tuneable broadband modes of an array of plasmonic metamolecules and a vibrational mode of carbonyl bond of poly(methyl methacrylate) is shown experimentally to produce a Fano resonance, which can be tuned in situ by varying the polarization of incident light. The interaction between the plasmon modes and the molecular resonance is investigated using both rigorous electromagnetic calculations and a quantum mechanical model describing the quantum interference between a discrete state and two continua. The predictions of the quantum mechanical model are in good agreement with the experimental data and provide an intuitive interpretation, at the quantum level, of the plasmon-molecule coupling.	cond-mat.mes-hall	cond-mat	Fano resonance resulting from a tunable interaction between molecular vibrational modes and a double-continuum of a plasmonic metamolecule E. J. Osley,1, 2 C. G. Biris,2 P. G. Thompson,1, 2 R. R. F. Jahromi,1, 2 P. A. Warburton,1, 2 and N. C. Panoiu2 1London Centre for Nanotechnology, University College London, 17-19 Gordon Street, London WC1H 0AH, UK 2Department of Electronic and Electrical Engineering, University College London, Torrington Place, London WC1E 7JE, UK (Dated: October 11, 2018) Coupling between tuneable broadband modes of an array of plasmonic metamolecules and a vibrational mode of carbonyl bond of poly(methyl methacrylate) is shown experimentally to produce a Fano resonance, which can be tuned in situ by varying the polarization of incident light. The interaction between the plasmon modes and the molecular resonance is investigated using both rigorous electromagnetic calculations and a quantum mechanical model describing the quantum interference between a discrete state and two continua. The predictions of the quantum mechanical model are in good agreement with the experimental data and provide an intuitive interpretation, at the quantum level, of the plasmon-molecule coupling. PACS numbers: 73.20.Mf, 78.67.Pt, 78.68.+m, 42.82.Et, 81.07.-b The spectral control of optical absorption in metallic nanostructures has generated rapidly growing interest, both as a means for studying light-matter interaction at the deep-subwavelength scale and to develop new func- tional nanomaterials. Plasmonic materials provide an ideal platform for achieving enhanced optical interaction at the nanoscale as their primary building blocks consist of highly resonant particles, which strongly confine and enhance the optical field [1 -- 4]. These properties are com- monly employed to increase the optical coupling between plasmonic structures and an optically active medium, enabling many applications, including surface-enhanced Raman spectroscopy [5, 6], surface-enhanced infrared ab- sorption spectroscopy [7, 8], enhancement of the excita- tion rate and fluorescence of quantum dots (QDs) and molecules [9 -- 11], chemical sensing [12, 13], and biode- tection [14, 15]. One effective approach for tailoring the shape of plasmonic resonances is the Fano interference between a narrow (discrete) resonance and a broad (con- tinuous) one [16 -- 19]. A defining feature of Fano resonances is their pro- nounced spectral asymmetry. This can be exploited in applications that require high sensitivity to changes in the environment. Fano resonances in plasmonic systems can also provide physical insights into light-matter inter- action at the nanoscale, as they are primarily the result of short-range, near-field interactions. Although Fano resonances in plasmonic systems have been chiefly inves- tigated in a classical context, they can also be observed in quantum systems consisting of QDs, atoms, or molecules coupled to metallic nanoparticles [14, 20 -- 22]. In this case the Fano resonance results from the interaction between a quantum discrete state and a classical continuum (lo- calized or extended) plasmon mode. In many applica- tions it is desirable to tune the strength of the inter- action between the discrete and continuous state, and consequently dynamically change the shape of Fano res- onances. However, since the materials and geometrical parameters of a plasmonic structure are difficult to con- trol at the nanoscale, a convenient way to achieve optical tunability has not yet been attained. In this Letter we demonstrate experimentally and the- oretically an efficient approach to achieve in situ tun- ing of the strength of the interaction between optically active vibrational modes of molecules -- here the car- bonyl bond of poly(methyl methacrylate) (PMMA) -- and the localized surface plasmon (LSP) modes of plasmonic metamolecules consisting of asymmetric cross apertures in a thin metallic film (see Fig. 1). These nanostruc- tures have been studied extensively [23, 24], one of their defining properties being that, analogous to anisotropic molecules, their polarizability strongly depends on the polarization of the incident wave. These plasmonic meta- molecules can therefore be tailored to different molecular resonances not only because the resonance frequencies of LSPs are strongly dependent upon their size and shape but, more importantly, because of their intrinsic optical anisotropy. Sample fabrication. -- A CaF2 substrate of thickness 1 mm was partially coated with a 5 nm chromium adhe- sion layer and a 35 nm gold film by thermal evaporation. The array shown in Figs. 1(a) and 1(b) was patterned using electron-beam lithography and argon-ion milling. The aperture geometry, and the orientation of the elec- tric field, E, of incident polarized light relative to the apertures is shown in Fig. 1(c). The apertures have arm lengths Lx= 1900 nm and Ly = 1340 nm and arm widths Gx= 580 nm and Gy= 360 nm. The surface was then spin-coated with a 85 nm PMMA film, filling the apertures. Figure 1(d) shows the final structure in cross section. Optical characterization. -- The transmission and re- flection spectra of the array were measured using Fourier transform infrared spectroscopy (FTIR) microscopy. The 2 (color online). FIG. 1. (a) Scanning-electron micrograph showing part of the cruciform aperture array before PMMA coating, (b) magnified detail of four apertures, (c) aperture geometry, showing the in-plane electric field polarization an- gle, θ, (d) schematic cross section of the structure following PMMA coating. source is a broad-band mid-IR globar which is focused onto the sample. The light is collimated and passed through an aperture which selects radiation being trans- mitted or reflected from a circular sample area 79 µm in diameter. The signal was detected using a HgCdTe detec- tor. The reflection spectra R and transmission spectra T of the samples were normalized to spectra obtained from uncoated areas of gold film and CaF2 substrate, re- spectively. The incident radiation was linearly polarized using a wire grid polarizer with an extinction ratio of 168∶ 1 at 5 µm. Figures 2(a) and 2(c) show, respectively, the measured transmission and reflection spectra of the aperture ar- ray coated with PMMA. In the transmission spectra two global maxima are present; one, at 4.2 µm occurs when the polarization angle θ= 0 and another, at 5.5 µm, when θ= 90○. These global maxima are due to the excitation of two LSP modes, which are supported by the aper- tures [23, 24]. In the reflection spectra the excitation of the plasmon modes produces global minima. The spec- tral signature of the PMMA carbonyl bond is located at 5.79 µm. The shape of the carbonyl resonance peak changes with θ, showing a more pronounced, asymmetric Fano-like line shape as θ approaches zero. The interac- tion of the two broad plasmon resonances with the nar- row molecular resonance forms the Fano resonance. The inset to Fig. 2(a) also shows that, for small values of θ, at wavelengths somewhat longer than the carbonyl bond resonant wavelength, the transmission can be larger than the baseline value in spite of the fact that PMMA is a lossy dielectric. This is an example of absorption-induced FIG. 2. (color online). (a) Transmission and (c) reflection of the structure measured using FTIR; (b) transmission and (d) reflection simulated using the RCWA method. The incident (pur- ple). The resonance of the PMMA carbonyl bond is visible polarization angle θ varies from zero (brown) to θ= 90 ○ at λ= 5.79 µm. The inset to (a) shows the measured spec- carbonyl bond resonance for θ= 0 trum (solid line) and the fitted base-line (broken line) at the ○ . transparency [25]. Furthermore, as shown in Fig. 2(c) the reflectivity of the aperture array decreases with re- spect to the baseline for small θ, whereas it increases for large θ, by as much as ∼100% for θ = 90○. This phe- nomenon we refer to as absorption-induced reflectivity. Numerical simulations. -- To model the optical re- sponse of our PMMA-coated device we used commer- cially available software [26] based on the rigorous cou- pled wave analysis (RCWA). We used a constant refrac- tive index for the CaF2 substrate, a Lorentz-Drude model for the Au and Cr layers [27], and a complex refractive index for the PMMA layer (found by fitting the numer- ically determined transmission to an experimental mea- surement of the transmission of a 85 nm thick PMMA film). Figures 2(b) and 2(d) show the transmission and reflection spectra, respectively, found by numerical sim- ulations; there is good agreement with the experimental data. The simulations also allow spatial field profiles to be calculated; Fig. 3 shows the spatial profile of the field enhancement corresponding to the two LSPs. Theoretical model for molecule-plasmon metamolecule interaction. -- In what follows we introduce a simple quan- tum mechanical model describing the interaction between the molecular vibrational mode and the LSP resonances. Figure 3 shows the interacting quantum mechanical sys- tem schematically. a Hamiltonian, H= H0+ Va+ Vb, where H0= Hm+ Ha+ Hb We assume that the interacting system is described by is the Hamiltonian of the noninteracting molecule ( Hm) 3 πγ (3) ; the interaction with the two continua, ΦE− cos ∆ψE Following the approach of Fano [16], the total Hamil- tonian H can be orthogonalized in a new basis of eigenstates which form two orthogonal continua, i.e., E; 1 HE′; 1 = E; 2 HE′; 2 = Eδ(E − E′), where E; 1E′; 2= 0. These eigenstates can be written as [16] E; 2=ψE, E; 1= sin ∆√ whereΦE is the "dressed" excited statee modified by πγ(E′)(E− E′)E′; i, ΦE=e+ Q i=a,b (here means the Cauchy principal value of the integral) andψE are two continuum states defined as: andψE = π~γ(E)(Va,EE; a+ Vb,EE; b) , ψE ψE= π~γ(E)V∗ E; b . tum interference between the discrete state e and the In Eq. (3) ∆ determines the phase shift induced by quan- E; a+ V∗ dE′ Vi,E′ (5b) (5a) (4) a,E b,E two continua, and γ describes the width of the modified excited state. They are given by: , γ(E)= πVa,E2+Vb,E2≡ Γm γ(E) ∆(E)=− arctan E− Ee− Σ(E) , γ(E′) Σ(E)= π(E− E′) dE′. 2 where the energy shift of the excited state is (6a) (6b) (7) The interaction between the coupled molecule- metamolecule system and a plane wave normally incident tem, in which the molecule is in the ground state and no plasmons are excited. The matrix elements of T are: onto the array can be described by a Hamiltonian T = Tm+ Ta+ Tb, with T= te0e0+∑i=a,b∫ Ti,EE; i0dE+ h. c.. Here,0 is the ground state of the interacting sys- ρa(E) sin θ, and te0=e Tm0, Ta,E =E; a Ta0≡ µa ρb(E) cos θ, where µa,b quantify Tb,E =E; b Tb0≡ µb frequency ω is W(ω) = ∑i=1,2∫ dEE; i T02δ(ŏhω− E). Inserting Eqs. (3) 2πŏh A(ω)=N ωψŏhω whereN = 2πns~I0 is a constant, with ns and I0 being incoming beam, respectively, and F(q, )=(q+)2~(2+1) the coupling strength between the incoming plane wave and the two LSPs and we have used the fact that the plasmon A (B ) is excited by a y(x )-polarized wave. T02F(q, )+ψŏhω T02(cid:6) , in this equation one obtains the absorption of the system, the number of molecules per area and the intensity of the Using Fermi's golden rule, the rate at which the system absorbs photons of (8) FIG. 3. (color online). System schematic showing the ground state g, the two LSP continua, E; a and E; b, and the discrete molecular resonance e. Simulated spatial profiles (taken in the x− y plane at the center of the aperture) show large field enhancement for θ= 0 at λ= 4.2 µm and θ= 90 ○ at λ= 5.5 µm. Aperture boundaries are indicated by dashed corresponds to the excited statee. lines. The atomic structure of the PMMA monomer is shown with the carbonyl bond in pink; the resonance of this bond assume that the eigenstates corresponding to the energy and the two LSPs ( Ha and Hb) and Va and Vb describe the interaction between the molecule and the two LSPs. The molecule is modeled as a two-level system withg ande the ground and excited states, respectively. We E, E; a (E; b), of Ha ( Hb) form two continua. Note continua are orthogonal, too: E; aE′; b= δabδ(E− E′). H0= Eggg+ Eeee+ Q i=a,b S EE; iE; idE, Under these conditions we can write H0 as that the two plasmon modes are orthogonal and thus the (1) of Vi and are proportional to the local electric field of the LSP and h. c. means Hermitian conjugate. Introducing where Eg and Ee are the energies of the ground and ex- cited state of the molecule, respectively. The interac- tion between the molecule and the two LSPs (continua) can be expressed as Vi=∫ Vi,EE; iedE+ h. c., i= a, b, where Vi,E =E; i Vie, i= a, b, are the matrix elements the density of states of the two continua, ρa,b(E), these ρi(E), i= a, b. matrix elements are written as Vi,E= vi i= a, b. We assume that the spectra of the plasmon A (B ) is de- scribed by a Lorentzian centered at the plasmon energy Ea (Eb) with width Γa (Γb), ρi(E)= (2) 1 1+ 2(E−Ei) 2 , Γi is the Fano function. The asymmetry parameter, q, and the reduced energy, , are given by Φŏhω T0 πγ(ŏhω)ψŏhω T0 ; = ŏhω− Ee− Σ(ŏhω) γ(ŏhω) q= . (9) The absorption spectrum in Eq. (8) shows that the coupling of the molecular vibrational mode with the two continua of the plasmonic metamolecule leads to a res- onant response of the optical system around the energy of the vibrational mode,ŏhωࣃ Ee, which is superimposed frequency for which F(q, )= 0, i.e., when q=−. on a broad background. Hence, unlike the case of only one continuum, the absorption does not cancel at the The interaction between the incoming light and the plasmons can be tuned by varying the polarization an- gle; this allows one to modify the mixture of the two LSPs and, consequently, one can easily tune the strength of the interaction between the vibrational mode and the plas- monic metamolecule. In order to provide a more quanti- tative description we assume that the normalized densi- ties of state of the two LSPs are given by Eq. (2). Then, the matrix elements in Eq. (8) and the parameters of the Fano function can be calculated explicitly. The absorp- tion can be cast as A(ω)= N πω γ(ŏhω)µav∗ aρa sin θ+ µbv∗ ρaρbµavb sin θ+ µbva cos θ2(cid:6) , aρa sin θ+ ŏhω−Eb + ŏhω−Ea µav∗ aρa sin θ+ µbv∗ µav∗ ŏhω− Ea ŏhω− Ee− πρava2 µbv∗ + ρbvb2 Γb b ρb cos θ q= 2 = 2 where the Fano parameters are te0 2π Γa b ρb cos θ b ρb cos θ2F(q, )+ , ŏhω− Eb (11a) . Γb (10) Γm Γa (11b) Comparison between theory and experiment. -- To fit the model to the experimental system the absorption A of the structure was extracted from the measured transmis- sion and reflection data using the relation A= 1− T− R, Γa,b can be directly extracted; Ea and Γa from the θ= 0 data and Eb and Γb from the θ= 90○ data. Ee and Γm as shown in Fig. 4(a). From this the parameters Ea,b and (the carbonyl bond parameters) are taken from trans- mission measurements of a 100 nm PMMA film. The parameters va,b, µa,b, and te0 are then extracted for each value of θ using an unconstrained least squares method. Figure 4(b) shows the predictions of the model follow- ing fitting. There is good agreement between the mea- sured spectra and those predicted by the model at all val- ues of θ. In particular, all the main features of the Fano resonance are reproduced by our theoretical model, e.g. the increase of the optical absorption as θ is varied from 4 FIG. 4. (color online). (a) Absorption of the structure, from FTIR measurements of transmission and reflection. (b) Best fit of experimental data using Eq. (10). zero to 90○. Our model also predicts that te0âva,b2 for all angles θ, which means that, as expected, the molecules interact primarily with the optical near-field of the SPPs. This proves that our model can be used to gain insights into the interaction between quantum and plasmonic sys- tems as it allows one to retrieve the values of basic quan- tities, e.g. va,b, µa,b, and te0, which characterize these interactions at the nano-scale. In conclusion, we have demonstrated that plasmonic structures allow in situ tuning of the optical coupling between a quantum and a classical system simply by se- lecting the polarization of incident light. This tuning produces a dramatic change in the spectral line shape about the optical resonance of the quantum system, a consequence of the Fano resonance produced by the cou- pling of broad and narrow resonances. The system has been described using an intuitive quantum mechanical model, which allows a deeper understanding of the un- derlying physics. Our study has the potential to foster exciting new developments in sub-wavelength optics and nanophotonics and in the emerging field of quantum plas- monics. Equally important, the system investigated here has the potential to improve the range and versatility of a series of nanodevices, including chemical and biological plasmonic sensors. This work was supported by EPSRC. [1] W. L. Barnes, A. Dereux, and T. W. Ebbesen, Nature (London) 424, 824 (2003). 5 [2] A. V. Zayats, I. I. Smolyaninov, and A. A. Maradulin, Phys. Rep. 408, 131 (2005). [16] U. Fano, Phys. Rev. 124, 1866 (1961). [17] S. Zhang, D. A. Genov, Y. Wang, M. Liu, and X. Zhang, [3] F. Garcia-Vidal, L. Martin-Moreno, T. Ebbesen, and L. Phys. Rev. Lett. 101, 047401 (2008). Kuipers, Rev. Mod. Phys. 82, 729 (2010). [18] B. Gallinet and O. J. F. Martin, Phys. Rev. B 83, 235427 [4] S. A. Maier, Plasmonics: Fundamentals and Applications (2011). (Springer, New York, 2007). [5] S. M. Nie and S. R. Emery, Science 275, 1102 (1997). [6] K. Kneipp, Y. Wang, H. Kneipp, L. T. Perelman, I. Itzkan, R. R. Dasari, and M. S. Feld, Phys. Rev. Lett. 78, 1667 (1997). [7] D. Enders and A. Pucci, Appl. Phys. Lett. 88, 184104 (2006). [19] B. Lukyanchuk, N. I. Zheludev, S. A. Maier, N. J. Ha- las, P. Nordlander, H. Giessen, and C. T. Chong, Nature Mater. 9, 707 (2010). [20] F. Neubrech, A. Pucci, T. W. Cornelius, S. Karim, A. G. Etxarri, and J. Aizpurua, Phys. Rev. Lett. 101, 157403 (2008). [21] I. M. Pryce, K. Aydin, Y. A. Kelaita, R. M. Briggs, and [8] J. Kundu, F. Le, P. Nordlander, and N. J. Halas, Chem. H. A. Atwater, Nano Lett. 10, 4222 (2010). Phys. Lett. 452, 115 (2008). [22] V. Giannini, Y. Francescato, H. Amrania, C. C. Phillips, [9] J. H. Song, T. Atay, S. Shi, H. Urabe, A. Nurmikko, Nano and S. A. Maier, Nano Lett. 11, 2835 (2011). Lett. 5, 1557 (2005). [10] P. P. Pompa, L. Martiradonna, A. Della Torre, F. Della Sala, L. Manna, M. De Vittorio, F. Calabi, R. Cingolani, and R. Rinaldi, Nature Nanotechnol. 1, 126 (2006). [11] P. Anger, P. Bharadwaj, and L. Novotny, Phys. Rev. Lett. 96, 113002 (2006). [12] C. Soennichsen and A. P. Alivisatos, Nano Lett. 5, 301 (2005). [23] R. M. Roth, N. C. Panoiu, M. M. Adams, J. I. Dadap, and R. M. Osgood, Opt. Lett. 32, 3414 (2007). [24] P. G. Thompson, C. G. Biris, E. J. Osley, O. Gaathon, R. M. Osgood, N. C. Panoiu, and P. A. Warburton, Opt. Express 19, 25035 (2011); E. J. Osley, P. G. Thompson, C. G. Biris, N. C. Panoiu, and P. A. Warburton, Bull. Am. Phys. Soc. 56, APS March Meeting 2011. [25] J. A. Hutchison, D. M. O'Carroll, T. Schwartz, C. Genet, [13] N. Liu, M. Mesch, T. Weiss, M. Hentschel, and H. and T. W. Ebbesen, Angew. Chem. 123, 2133 (2011). Giessen, Nano Lett. 10, 2342 (2010). [26] DiffractMOD, RSoft Design Group. [14] J. N. Anker, W. P. Hall, O. Lyandres, N. C. Shah, J. Zhao, and R. P. Van Duyne, Nature Mater. 7, 442 (2008). [15] C. Wu, A. B. Khanikaev, R. Adato, N. Arju, A. A. Yanik, H. Altug, and G. Shvets, Nature Mater. 11, 69 (2011). http://www.rsoftdesign.com. [27] M. A. Ordal, R. J. Bell, R. W. Alexander, L. L. Long, and M. R. Querry, Appl. Opt. 24, 4493 (1985).
1804.06589	2	1804	2018-04-26T15:26:39	Universal Majorana thermoelectric noise	[ "cond-mat.mes-hall", "cond-mat.supr-con" ]	Thermoelectric phenomena resulting from an interplay between particle flows induced by electric fields and temperature inhomogeneities are extremely insightful as a tool providing substantial knowledge about the microscopic structure of a given system. Tuning, e.g., parameters of a nanoscopic system coupled via tunneling mechanisms to two contacts one may achieve various situations where the electric current induced by an external bias voltage competes with the electric current excited by the temperature difference of the two contacts. Even more exciting physics emerges when the system's electronic degrees freedom split to form Majorana fermions which make the thermoelectric dynamics universal. Here we propose revealing this unique universal signatures of Majorana fermions in strongly nonequilibrium quantum dots via noise of the thermoelectric transport beyond linear response. It is demonstrated that whereas mean thermoelectric quantities are only universal at large bias voltages, the noise of the electric current excited by an external bias voltage and the temperature difference of the contacts is universal at any bias voltage. We provide truly universal, i.e. independent of the system's parameters, thermoelectric ratios between nonlinear response coefficients of the noise and mean current at large bias voltages where experiments may easily be performed to uniquely detect these truly universal Majorana thermoelectric signatures.	cond-mat.mes-hall	cond-mat	Universal Majorana thermoelectric noise P. N. Lebedev Physical Institute of the Russian Academy of Sciences, 119991 Moscow, Russia∗ Sergey Smirnov (Dated: July 8, 2021) Thermoelectric phenomena resulting from an interplay between particle flows induced by elec- tric fields and temperature inhomogeneities are extremely insightful as a tool providing substan- tial knowledge about the microscopic structure of a given system. Tuning, e.g., parameters of a nanoscopic system coupled via tunneling mechanisms to two contacts one may achieve various sit- uations where the electric current induced by an external bias voltage competes with the electric current excited by the temperature difference of the two contacts. Even more exciting physics emerges when the system's electronic degrees freedom split to form Majorana fermions which make the thermoelectric dynamics universal. Here we propose revealing this unique universal signatures of Majorana fermions in strongly nonequilibrium quantum dots via noise of the thermoelectric trans- port beyond linear response. It is demonstrated that whereas mean thermoelectric quantities are only universal at large bias voltages, the noise of the electric current excited by an external bias voltage and the temperature difference of the contacts is universal at any bias voltage. We provide truly universal, i.e. independent of the system's parameters, thermoelectric ratios between nonlinear response coefficients of the noise and mean current at large bias voltages where experiments may easily be performed to uniquely detect these truly universal Majorana thermoelectric signatures. I. INTRODUCTION Originally proposed [1] in the late 1930s the Majorana representation [2], making the Dirac equation real, has since then always been a challenge for experiments on el- ementary particles to reveal, e.g. via neutrinoless double beta decay, a fundamental particle representing on equal footing its own antiparticle. Referred to as Majorana fermions, these are hypothetical spin-1/2 neutral funda- mental particles with neutrinos as possible candidates for massive Majorana particles. In parallel to Majorana challenges within the particle physics another way to implement Majorana fermions is to construct condensed matter systems with quasiparti- cles being identical to their own antiquasiparticles. This is indeed possible when a system acquires a finite super- conducting order parameter. Here Dirac fermions may become highly nonlocal in real space, i.e. they may split, or fractionalize, into two Majorana fermions localized at the system's edges. Although these zero energy Majorana bound states are not fundamental particles as in the case of the particle physics, they are effectively quasiparti- cles which are their own antiquasiparticles. Despite their non-abelian statistics these zero energy Majorana bound states are still referred to as Majorana fermions. The Kitaev model [3] is an example of a one-dimensional con- densed matter system hosting two Majorana fermions at its ends. The topological superconducting phase realized in the Kitaev model has various [4 -- 6] physical implemen- tations among which systems based on topological insu- lators [7, 8] and spin-orbit coupled semiconductors [9, 10] are of particular interest. Once implemented in a condensed matter setup Majo- rana fermions need an experimental proof of their exis- ∗ [email protected], [email protected] tence via unique signatures characteristic of exclusively these fractionalized quasiparticles. To this end transport experiments offer a relatively simple way to detect unique Majorana signatures in a given condensed matter system. The majority of transport proposals focus on mean electric current often providing low-energy behavior of the system's electrical conductance. Examples are given by superconductor-Luttinger liquid junctions [11], Kondo effect in quantum dots side coupled to a topological su- perconductor supporting Majorana fermions at its ends [12], driven topological superconductors [13], Josephson junctions on surfaces of three-dimensional topological in- sulators [14], Kondo effect in topological superconductor- quantum dot-normal lead junction [15], normal metal- superconducting semiconductor-normal metal structures [16], spinon-antispinon systems [17], Coulomb blockaded systems [18], disordered Josephson junctions in tilted magnetic fields [19] and many others. Transport experiments [20, 21] oriented on measure- ments of the mean current, in particular, on its low- energy behavior are steadily improving their analysis of the zero bias anomaly present in the electrical con- ductance from which stronger signatures of Majoranas may be extracted [22]. Although it is often necessary to perform additional analysis to disentangle the Majorana physics from possible interaction induced effects, such as the Kondo effect, or from effects related to partially- separated Andreev bound states [23, 24], recent works [25, 26] suggest an alternative proof of the nonlocal na- ture and topological protection of Majorana zero modes via local measurements of the zero bias conductance. To avoid uncertainty and make transport experiments provide more unique signatures of Majorana fermions one might, in conjunction with measurements of zero bias conductance, resort to measurements of the electric cur- rent fluctuations. Here there are not as many proposals [27 -- 30] for Majorana noise as for the mean electric cur- l = e/2 and e∗ rent. Again the most attention is paid to the low-energy behavior of the electric current noise. Because of the fluctuation-dissipation theorem [31 -- 33] this equilibrium noise, accessed within the system's linear response, is not independent of the mean electric current. Beyond linear response the electric current fluctuates independently of its mean value and thus this nonequilibrium noise is able to provide alternative Majorana fingerprints which, what is remarkable, turn out to be of universal nature [34, 35]. In particular, in Ref. [35] it has been demonstrated that this universal Majorana noise is characterized by two ef- fective charges e∗ h = 3e/2 at low and high It has also been shown that the energies, respectively. low-energy Majorana effective charge e∗ l = e/2, whose Majorana nature is additionally confirmed via Majorana tunneling entropy calculations [36], might be sensitive to thermal fluctuations while the high-energy Majorana ef- fective charge e∗ h = 3e/2 is accessed at high bias voltages which protect this effective charge from thermal noise and, therefore, its measurement may be performed at relatively high temperatures easily reachable in modern laboratories. It is important to note that modern experi- ments have already reached such a high level at which one can get the effective charge with very high accuracy as for example in experiments on quantum dots where the effective charge has been measured in the Kondo regime [37]. Another realm of transport experiments is provided by systems hosting Majorana fermions and possessing tem- perature inhomogeneities. Here examples include heat conduction in a Majorana metal, where the heat con- duction happens via Majorana fermions bound to de- fects [38], and thermal quantum Hall effect [39], where a net heat current results from a temperature difference between the edges which support Majorana chiral edge modes arising for example in a chiral p-wave supercon- ductor. The latter case of dispersive Majorana fermions is particularly attractive for future research on thermo- electric phenomena in systems where Majorana disper- sion relations may undergo qualitative changes such as, for example, the evolution of a Majorana conic disper- sion relation into a Majorana arc shaped dispersion re- lation [40]. In the context of nanoscopic systems one is usually interested in setups where a nanoscopic sys- tem interacts via tunneling mechanisms with two normal metals playing the role of the contacts. An external bias voltage is applied to these contacts which may also have different temperatures. Here one may easily tune the sys- tem's parameters making the electric current induced by the bias voltage and the electric current excited by the temperature difference interfere with one another leading to a highly nontrivial thermoelectric transport [41, 42]. When a nanoscopic setup is designed to host Majorana degrees of freedom, the system's thermoelectric response undergoes essential changes which might be used to iden- tify Majorana fermions in, e.g., normal metal-quantum dot-Majorana bound states junction [43], Majorana-side- coupled quantum dots [44, 45], Majorana bound states 2 system coupled to two normal leads [46]. Up to now Majorana thermoelectric transport has been investigated via analysis of the mean electric current of- ten restricted to its low-energy behavior to obtain lin- ear conductances. Fluctuations of the electric current excited by both the bias voltage and temperature dif- ference have not been addressed. At the same time, as discussed above, fluctuations provide much more unique fingerprints of a given system. Therefore, thermoelec- tric noise is an important and highly conclusive tool to uniquely reveal whether Majorana fermions are present in a given nanoscopic system. In this work we explore fluctuations of thermoelectric transport through quantum dots interacting via tunnel- ing mechanisms with two contacts and one end of a one- dimensional Kitaev's chain, supporting two Majorana fermions at its ends. The contacts are normal metals to which an external bias voltage is applied and in gen- eral the temperatures of the two contacts are not iden- tical. We obtain for this system both the mean electric current and thermoelectric noise as well as their nonlin- ear response coefficients in the transport regime governed by Majorana degrees of freedom. We find 1) that mean thermoelectric quantities become universal only for large bias voltages while having non-universal behavior below the energy scale characterizing the overlap of the Ma- jorana modes; 2) that in contrast to the mean quanti- ties thermoelectric noise is universal in the whole voltage range when the dynamics is essentially governed by Ma- jorana bound states; 3) that the differential thermoelec- tric noise has a universal two plateau structure with the truly universal values of the plateaus (e3/h)[1 + ln(2)] and (e3/h)[1 + ln(21/2)] depending on whether the ra- tio between the bias voltage and the thermal voltage is less than or greater than one; 4) analytical universal high-energy asymptotics of the nonlinear response coef- ficients of the thermoelectric noise; 5) universal ratios between nonlinear response coefficients of the thermo- electric noise and mean current; 6) that at large bias voltages these thermoelectric ratios saturate to truly uni- versal constants independent of the bias voltage; 7) an- alytical values of these constants; 8) that these values are protected by high bias voltages, making them robust against thermal noise and, therefore, these values repre- sent unique truly universal Majorana thermoelectric sig- natures which could be measured in modern experiments even at relatively high temperatures. The paper is organized as follows. In Section II we describe in detail an example of a system where thermo- electric transport in presence of Majorana fermions may be analyzed on the level of both mean quantities and fluc- tuations. The description of the system is given in terms of the Keldysh field integral most convenient to explore different correlation functions in stationary nonequilib- rium which is the case here. In Section III thermoelec- tric transport is analyzed in terms of nonlinear response coefficients: first on the level of mean quantities, in Sub- section III A, and afterwards on the level of fluctuations, 3 quantum dot is Hd = dd†d, (1) where d and d† are, respectively, the quantum dot Dirac fermion annihilation and creation operators with the usual anticommutation relations, {d, d} = 0, {d, d†} = 1. If, as mentioned above, a Zeeman field is applied to the quantum dot, it will destroy the correlation effects due to the Kondo effect [47] so that the latter is not impor- tant for the Majorana physics discussed below. More- over, as is well known, the Kondo effect in the present setup would require the quantum dot to be filled (d < 0). However, for the case of the empty quantum dot (d > 0) the Kondo effect does not arise. On the other side, as will be shown below (see also Ref. [35] for pure electric Majorana noise), the thermoelectric Majorana noise is universal and it does not depend on whether d < 0 or d > 0. Therefore, in order to completely exclude the Kondo effect from the physical setup, we will below al- ways consider the case d > 0. The setup also contains two contacts labeled as left (L) and right (R). They represent noninteracting normal metals which are assumed to be in equilibrium character- ized by the Fermi-Dirac distributions: fL,R() = , (2) exp(cid:0) −µL,R 1 TL,R (cid:1)+1 where µL,R and TL,R are, respectively, the chemical po- tentials and temperatures of the contacts. An external bias voltage V may be applied to the contacts so that µL−µR = eV , where e is the electron charge, and we will assume the symmetric bias, µL,R = ±eV /2, with eV < 0. The temperatures of the contacts are TL = T + ∆T , TR = T with T (cid:62) 0, ∆T (cid:62) 0, i.e. the left contact is hot and the right contact is cold. We consider a typical setup where the two contacts are characterized by the same set of quantum numbers k and energy spectrum k. In this case the Hamiltonian of the contacts is † kc lkclk, (3) (cid:88) (cid:88) l={L,R} k Hc = respectively, † where clk and c the contacts lk are, Dirac fermion annihilation and creation operators with the usual anticommutation relations, {clk, cl(cid:48)k(cid:48)} = 0, † {clk, c l(cid:48)k(cid:48)} = δll(cid:48)δkk(cid:48). If the quantum dot energy level d is spin polarized, i.e. it has a definite spin σ realized via a Zeeman splitting, as mentioned above, then Eq. (3) de- scribes the electrons in the normal metals with the same spin σ. Additionally, we assume massive contacts so that their energy spectrum is continuous and their density of states per spin may be characterized by a constant νc/2 within the energy range relevant for transport. The quantum dot interacts with the contacts via tun- neling, Hd−c = (cid:88) (cid:88) l={L,R} k † lkd + H.c., Tlkc (4) FIG. 1. The physical system represents a quantum dot whose single-particle energy level d may be tuned by an ex- ternal gate voltage to make the quantum dot filled (d < 0) or empty (d > 0). The left (L) and right (R) contacts are normal metals assumed to be in equilibrium characterized by the Fermi-Dirac distributions fL,R(ε) specified by the chem- ical potentials µL,R and the temperatures TL,R. They are coupled via tunneling interactions (with the strengths ΓL,R) to the quantum dot. An external bias voltage V is applied to the contacts, µL − µR = eV < 0. The left contact is hot (shown as red) while the right one is cold (shown as blue), TL = T + ∆T , TR = T , ∆T (cid:62) 0. The grounded topolog- ical superconductor, implemented, e.g, by means of the Ki- taev's chain, supports two Majorana modes γ1,2 at its ends. One of these ends, namely the one supporting the Majorana mode γ1, is coupled via another tunneling interaction (with the strength η) to the quantum dot. The blue and red arrows illustrate schematically the electric current flows excited by the external bias voltage V and the temperature difference ∆T , respectively. The noise of the electric current S>(t) and the mean electric current I are measured in the hot (left) contact, S>(t) = (cid:104)δIL(t)δIL(0)(cid:105), I = (cid:104)IL(cid:105). in Subsection III B. In Section IV we summarize the main results of the paper, draw conclusions and discuss some open issues. Finally, the appendix provides the main technical steps in the calculation of the current-current correlator. II. NONEQUILIBRIUM FIELD-THEORETIC FRAMEWORK: KELDYSH ACTION To explore Majorana thermoelectric transport in a nanoscopic setup we focus on a particular example of a quantum dot system shown in Fig. 1. It includes a noninteracting quantum dot with a single-particle energy level d which is spin nondegenerate. Physical realization may represent a quantum dot subjected to a Zeeman field leading to a spin polarized energy level d. Additionally, the position of this energy level with respect to the chem- ical potential may be controlled by an external gate volt- age so that the quantum dot is filled when d < 0 and empty when d > 0. The Hamiltonian of the isolated where the dependence of the tunneling matrix elements on the contacts quantum numbers is usually neglected, Tlk = Tl. The strength of the tunneling between the quantum dot and the left or right contact is characterized by the quantity Γl ≡ πνcTl2 while the total tunneling strength is defined as Γ ≡(cid:80) l={L,R} Γl = ΓL + ΓR. The final constituent of the setup is a grounded topo- logical superconductor implementing the Kitaev's one- dimensional chain supporting two Majorana zero modes localized at its edges. Due to a finite length of the topo- logical superconductor these Majorana bound states may have a finite overlap with a characteristic energy scale ξ. For large values of ξ the two Majorana fermions merge and behave as a single Dirac fermion while for small val- ues of ξ the low-energy physics is essentially governed by Majorana degrees of freedom. The effective low-energy Hamiltonian of the topological superconductor is Htsc = i 2 ξγ2γ1, (5) where γ1,2 are the Majorana fermion annihilation oper- ators identical to the corresponding creation operators, † γ 1,2 = γ1,2. These annihilation operators satisfy the anti- commutation relations {γi, γj} = 2δij implying that the associative algebra generated by the set {1, γ1, γ2} is the Clifford algebra [48]. The quantum dot interacts with the topological su- perconductor via tunneling involving only the Majorana mode γ1 as has been suggested in numerous literature (see, e.g., Refs. [15, 27, 28, 44]), Hd−tsc = η∗d†γ1 + H.c., (6) where the absolute value η of the tunneling matrix ele- ment characterizes the strength of the tunneling between the quantum dot and the topological superconductor. To explore nonlinear thermoelectric response of the system specified by the Hamiltonian H = Hd + Hc + Hd−c + Htsc + Hd−tsc (7) (cid:90) it is convenient to resort to the Keldysh field integral framework [49, 50] as a general tool providing various correlation functions in a simple systematic way. To this end we write down the source dependent Keldysh gener- ating functional: D[¯θ(t), θ(t)]e i SK [¯θ(t),θ(t);Jl(t)], Z[Jl(t)] = (8) where {¯θ(t), θ(t)} = { ¯ψ(t), ψ(t); ¯φlk(t), φlk(t); ¯ζ(t), ζ(t)} is the set of the Grassmann fields of, respectively, the quantum dot, contacts and topological superconductor while Jl(t) is the source field. Note, the fundamental normalization Z[Jl(t) = 0] = 1. (8) the time argument t runs over the Keldysh closed time contour CK, t ∈ CK. The total Keldysh action SK[¯θ(t), θ(t); Jl(t)] consists of the Keldysh actions of the isolated quantum dot, con- tacts and topological superconductor, the tunneling ac- tions describing the interaction of the quantum dot with In Eq. 4 the contacts and topological superconductor as well as the source action. The Keldysh actions of the isolated quantum dot, Sd[ ¯ψ(t), ψ(t)], contacts, Sc[ ¯φlk(t), φlk(t)], and topologi- cal superconductor, Stsc[¯ζ(t), ζ(t)], have the conventional form of upper triangular 2 × 2 matrices in the retarded- advanced space with the upper/lower diagonal elements representing the inverse retarded/advanced Green's func- tions of the corresponding isolated system while the up- per off-diagonal elements of these matrices have the form i δ (1 − 2f ) where δ → 0+ and f is the corresponding Fermi-Dirac distribution which, e.g., for the contacts is given by Eq. (2). The tunneling actions describing the interactions of the quantum dot with the contacts and topological supercon- ductor are, respectively, given as k dt −∞ l={L,R} (cid:88) Sd−c[ ¯ψ(t), ψ(t); ¯φlk(t), φlk(t)] = = − (cid:90) ∞ (cid:88) (cid:8)Tl[ ¯φlk+(t)ψ+(t)− − ¯φlk−(t)ψ−(t)] + G.c.(cid:9), (cid:90) ∞ − ¯ψ−(t)ζ−(t) − ¯ψ−(t)¯ζ−(t)] + G.c.(cid:9), dt(cid:8)η∗[ ¯ψ+(t)ζ+(t) + ¯ψ+(t)¯ζ+(t)− Sd−tsc[ ¯ψ(t), ψ(t); ¯ζ(t), ζ(t)] = = − −∞ (9) (10) (11) where in the right hand side the time argument t runs over the real axis and the subindex +/− denotes the forward/backward branches of the Keldysh closed time contour. The abbreviation G.c. stands for the Grass- mann conjugation, the generalization of the Hermitian conjugation for the case when Grassmann variables are involved in a mathematical expression. Finally, the source action is (cid:88) (cid:88) Sscr[ ¯ψ(t), ψ(t); ¯φlk(t), φlk(t); Jl(t)] = = − dt Jlq(t)Ilq(t), l={L,R} q={+,−} (cid:90) ∞ −∞ where Ilq(t) is the field representing the electric current in the left (l = L) or right (l = R) contact on the forward (q = +) or backward (q = −) branch of the Keldysh closed time contour: Ilq(t) ≡ ie ¯φlkq(t)ψq(t) − G.c.]. (cid:88) (12) [Tl k Using the total Keldysh action, SK[¯θ(t), θ(t); Jl(t)] = = Sd[ ¯ψ(t), ψ(t)] + Sc[ ¯φlk(t), φlk(t)]+ + Stsc[¯ζ(t), ζ(t)] + Sd−c[ ¯ψ(t), ψ(t); ¯φlk(t), φlk(t)]+ + Sd−tsc[ ¯ψ(t), ψ(t); ¯ζ(t), ζ(t)]+ + Sscr[ ¯ψ(t), ψ(t); ¯φlk(t), φlk(t); Jl(t)], (13) one can easily obtain the mean electric current in the left (l = L) or right (l = R) contact by taking the first derivative of Z[Jl(t)] with respect to the corresponding source field: (cid:104)Il(cid:105) ≡ (cid:104)Ilq(t)(cid:105)SK = i δZ[Jl(t)] δJlq(t) where the angular brackets (cid:104)···(cid:105)SK denote the average of a functional of Grassmann fields with respect to the total Keldysh action (13) taken at zero source field, (cid:12)(cid:12)(cid:12)(cid:12)Jlq(t)=0 (cid:90) (cid:104)F[¯θ(t), θ(t)](cid:105)SK ≡ ≡ D[¯θ(t), θ(t)]e i SK [¯θ(t),θ(t)]F[¯θ(t), θ(t)], (15) SK[¯θ(t), θ(t)] ≡ SK[¯θ(t), θ(t); Jl(t) = 0] = = Sd[ ¯ψ(t), ψ(t)] + Sc[ ¯φlk(t), φlk(t)]+ + Stsc[¯ζ(t), ζ(t)] + Sd−c[ ¯ψ(t), ψ(t); ¯φlk(t), φlk(t)]+ + Sd−tsc[ ¯ψ(t), ψ(t); ¯ζ(t), ζ(t)]. (16) The values of q and t in Eq. (14) are arbitrary since the final result does not depend on the choice of a branch of the Keldysh closed time contour as well as on the choice of an instant of time on that branch because we consider only stationary nonequilibrium states. It is straightfor- ward to verify that Eq. (14) leads to the Meir-Wingreen result [51]. In a similar way one can obtain the current-current correlator by taking the second derivative of Z[Jl(t)] with respect to the corresponding source fields: (cid:104)Il(t)Il(cid:48)(t(cid:48))(cid:105) ≡ (cid:104)Il−(t)Il(cid:48)+(t(cid:48))(cid:105)SK = = (i)2 δ2Z[Jl(t)] δJl−(t)δJl(cid:48)+(t(cid:48)) (cid:12)(cid:12)(cid:12)(cid:12)Jlq(t)=0 . (17) The technical steps one makes in the calculation of the second derivative in Eq. (17) are mainly straightforward. One should only take proper care of the fact that due to the presence of the topological superconductor there will appear anomalous contributions when averaging prod- ucts of four Grassmann fields of the quantum dot. To facilitate one's derivation of the anomalous terms and to avoid technicalities in the main text we provide the main details relevant for the calculation of the second deriva- tive in Eq. (17) in the appendix. The first and second derivatives of the Keldysh gener- ating functional are enough to explore nonlinear Majo- rana thermoelectric response of the system in terms of the mean value and fluctuations of the electric current. III. UNIVERSAL MAJORANA THERMOELECTRIC TRANSPORT For definiteness below we focus on transport measure- ments in the left (L) contact which is assumed to be , (14) where kB is the Boltzmann constant. 5 hot. The temperature difference ∆T between the con- tacts may be parameterized by a thermal voltage VT , VT ≡ kB∆T e , (18) Additionally, we will consider the situation when the quantum dot couples symmetrically to the left and right contacts, ΓL = ΓR = Γ/2. To solely focus on universal Majorana physics we ex- plore the regime dominated by the Majorana tunneling, η > max{d,eV , eVT , kBT, Γ, ξ}. (19) Here d may be tuned by an external gate voltage to satisfy (19). Similarly, the bias voltage V , thermal volt- age VT and temperature T may be externally adjusted to satisfy (19). The Majorana overlap ξ will be small if the topological superconductor in the setup is chosen to be long enough so that the two Majorana fermions do not merge into a single Dirac fermion during the tunnel- ing into the quantum dot. In modern experiments [52] the values of η and Γ are readily controlled via exter- nal gates whose potentials can increase or decrease the height of the potential barriers between the quantum dot and contacts in order to vary Γ as well as between the quantum dot and topological superconductor in order to vary η. In this way in modern laboratories η may be increased while Γ may be decreased in order to reach the condition in (19). Thus, besides its theoretical impor- tance, the Majorana transport regime, specified by (19), is of particular experimental interest. Before we start to discuss our results it is important to mention how we obtain various nonlinear response coeffi- cients. In what follows we use the formalism presented in Section II and the appendix and perform numerical cal- culations of corresponding integrals in the energy domain to obtain the mean electric current and current-current correlations. This formalism allows one to compute not only the mean electric current and current-current cor- relations but also various derivatives of these quantities. The calculations of the derivatives may be performed in two different ways. On one side, one can calculate deriva- tives using finite differences. In this case one computes the mean electric current and current-current correla- tions on a fine grid of the bias voltage and thermal volt- age and after that applies conventional finite difference schemes for the first and second derivatives. On the other side, one notices (see the appendix) that the dependence of both the mean electric current and current-current cor- relations on the bias voltage and thermal voltage enters through the Keldysh components of the Green's func- tions, more specifically, through the Fermi-Dirac distri- butions (2). Therefore, one can first calculate various derivatives of corresponding integrands. This simply re- duces to analytical differentiations of the Fermi-Dirac dis- tributions (2). After that one performs numerical inte- gration of these differentiated integrands which now in- volve various analytical derivatives of the Fermi-Dirac 6 presented below are analytical expressions to which nu- merical results converge when one gradually increases the numerical precision and no further approximation to the theoretical model in Section II is assumed. Finally, we would like to specify what we understand under universality of our results which will be presented below. We will call a quantity universal if this quantity is independent of the parameters characterizing exclu- sively the quantum dot. In our case such a parameter is d which is tuned by an external gate voltage. This is experimentally relevant because in a realistic experi- ment one may easily vary the gate voltage and observe which transport quantities do not change in response to this variation of the gate voltage. Similar universality happens in the Kondo effect [47] where universality is understood as independence of response coefficients on d which enters only through the Kondo temperature TK. The difference between the Majorana universality and Kondo universality is in the scaling. For the present case of the Majorana universality the scaling is given by Γ whereas for the Kondo universality it is given by the Kondo temperature TK which is also a function of Γ. In all of our calculations presented below we use d > 0. More specifically, to perform concrete calculations we put d = 8Γ but universal results do not change if one uses other values for d satisfying the condition in (19). Us- ing positive values for d one puts the quantum dot in the empty orbital regime [47]. As a result, in a realis- tic experiment the Kondo effect is switched off and one observes only the Majorana universality. Note, that in the absence of the Majorana bound states one would ob- serve strong dependence of response coefficients on d for bias voltages eV ∼ d. Therefore, observing indepen- dence of transport quantities on the gate voltage for any bias voltage already provides information about signa- tures of unpaired Majorana fermions. Of course, such a quantity, which is universal in the above sense, may de- pend on the tunneling coupling to the contacts, Γ, or on the tunneling coupling to the topological superconduc- tor, η, or on both. Thus the universality understood in the above sense is only a qualitative Majorana signature but not quantitative. For example, a response coefficient may have a universal asymptotic behavior at high bias voltages. Let us assume that this universal asymptotic behavior is given by a certain universal (independent of d) function of eV /Γ. Although the universality of this function is a qualitative signature of unpaired Majorana fermions, the values of the coefficients in the expansion of this universal function in powers of eV /Γ will depend on the definition of Γ and cannot serve as a quantitative signature of unpaired Majorana fermions. To provide quantitative signatures of unpaired Majorana fermions we define truly universal quantities. We will call a quan- tity truly universal if this quantity does not depend on both the parameters of the quantum dot, such as d, and the tunneling couplings to the contacts, Γ, as well as to the topological superconductor, η. For example, if there are two universal quantities, the ratio between these two The magnitude of the first derivative of the FIG. 2. mean electric current with respect to the thermal voltage, ∂I(V, VT )/∂VT, as a function of the bias voltage V at T = 0, eVT /Γ = 10−5 and for different gate voltages, specified by the quantum dot single-particle energy level d. The Majorana fermions overlap weakly, ξ/Γ = 10−4. The strength of the Majorana tunneling is η/Γ = 103. Here the dips in the non- universal (eV (cid:46) ξ) behavior of the curves with d/Γ = 4.0 and d/Γ = 8.0 correspond to those values of the bias volt- age where ∂I(V, VT )/∂VT = 0, that is to those values of V where ∂I(V, VT )/∂VT changes its sign. The inset shows the robustness of the universal (independent of d) behavior of the second derivative of the mean electric current with respect to the thermal voltage at large bias voltages when the tempera- ture increases. At higher temperatures there are values of the bias voltage where ∂2I(V, VT )/∂V 2 T = 0, that is points where ∂2I(V, VT )/∂V 2 T changes its sign. Since we use the logarith- mic scale, we plot the magnitude of the second derivative, T , which displays dips at its zeroes. ∂2I(V, VT )/∂V 2 distributions (2). We find that in all of our calculations these two ways of computing derivatives give the same results within a good numerical precision. This is a very good test that our numerical results are reliable. How- ever, in order to reach higher precision to obtain analyti- cal expressions, which will be discussed below, we prefer to use the second way involving analytical differentiations of the Fermi-Dirac distributions (2). Another important aspect concerns the analytical asymptotic limits presented below. They are obtained by inspection of our numerical results. We will use the term asymptotic limit for an analytical expression if our nu- merical results reproduce this analytical expression with any desired numerical precision by adjusting the physi- cal parameters of the system to satisfy the inequalities, specifying the regime of applicability of this analytical expression, with any desired degree of accuracy. In more mathematical terms, referring to an analytical expression as an asymptotic limit means that the stronger the in- equalities, specifying the regime of applicability of this analytical expression, are fulfilled the more digits after the decimal point in our numerical results reproduce this analytical expression. In other words, for the theoreti- cal model presented in Section II the asymptotic limits 10-1010-910-810-710-610-510-410-310-210-1100101102103eV[Γ]10-1410-1210-1010-810-610-410-2100dI/dVT[e2/h]εd = 0.0εd = 2.0 Γεd = 4.0 Γεd = 8.0 Γ10-1100101102103eV[Γ]10-1010-810-610-410-2100102d2I/dV2T[e3/Γ h]kBT = 0.0kBT = 1.0 ΓkBT = 5.0 ΓkBT = 10.0 Γ4π23eV ΓeVTΓ4π23eV Γ( )-3eVTΓ34π2( )eV Γ-37 mean electric current provides only non-universal Ma- jorana signatures and, therefore, it does not represent any interest as a unique universal signature of Majorana fermions. To access the universal Majorana physics in the mean thermoelectric quantities one has to resort to the nonlin- ear (eV > ξ) thermoelectric response. In this universal regime we find the following asymptotic limits for the thermoelectric coefficient ∂I(V, VT )/∂VT : ∂I(V, VT ) ∂VT = − e2 h 4π2 3 eV Γ eVT Γ η (cid:29) Γ (cid:29) (eV , eVT ), eV (cid:29) ξ ∂I(V, VT ) ∂VT = − e2 h 4π2 3 (cid:18)eV (cid:19)−3 eVT Γ Γ η (cid:29) eV (cid:29) Γ (cid:29) eVT , Γ (cid:29) ξ. (20) (21) (22) (23) for and for The universal laws given by Eqs. (20) and (22) are depicted by dashed lines in Fig. 2. The inset in Fig. 2 shows the universal (indepen- dent of d) Majorana behavior of the thermoelectric co- efficient representing the second derivative of the mean electric current with respect to the thermal voltage, ∂2I(V, VT )/∂V 2 T , as a function of the bias voltage V in the high-energy regime specified by the inequality in (23). This universal Majorana high-energy behavior is inde- pendent of the thermal voltage VT . As one can see, the second derivative ∂2I(V, VT )/∂V 2 T is very robust with re- spect to high temperatures. At high bias voltages it is given by the asymptotic limit, (cid:18)eV (cid:19)−3 , (24) ∂2I(V, VT ) ∂V 2 T = − e3 Γh 4π2 3 Γ shown as the dashed line in the inset in Fig. 2. The expression in Eq. (24) follows from Eq. (22) if the in- equality in (23) is fulfilled. The universal behavior of ∂2I(V, VT )/∂V 2 T given by Eq. (24) is almost unchanged for eV (cid:29) kBT when the temperature increases up to kBT ∼ 10 Γ, or in units of η, kBT ∼ 0.01η. As it follows from the above analysis, to focus on uni- versal thermoelectric Majorana signatures in the mean electric current one has to apply bias voltages eV (cid:29) ξ. In Fig. 3 we show the universal Majorana behavior of the thermoelectric coefficient representing the first derivative of the mean electric current with respect to the thermal voltage, ∂I(V, VT )/∂VT , as a function of the thermal volt- age VT at zero temperature and at a fixed value of the bias voltage Γ (cid:29) eV (cid:29) ξ. For the particular example shown in Fig. 3 we put eV /ξ = 100, eV /Γ = 0.01. As a result, the curve in Fig. 3 is universal in the whole FIG. 3. The first derivative of the mean current with respect to the thermal voltage, ∂I(V, VT )/∂VT , as a universal (inde- pendent of d) function of the thermal voltage VT at T = 0, eV /Γ = 10−2. The other parameters have the same values as in Fig. 2. The inset shows the robustness of the universal (independent of d) behavior of the second mixed derivative ∂2I(V, VT )/∂V ∂VT at large thermal voltages when the tem- perature increases. quantities may be truly universal because Γ and η may cancel out in this ratio which therefore becomes a unique quantitative signature of unpaired Majorana fermions. Below in the discussion of our results we will always dis- tinguish between universal and truly universal results. A. Nonequilibrium response of the mean thermoelectric current Let us first examine the universal Majorana signatures one can detect in the thermoelectric response of the mean electric current I(V, VT ) as a function of the bias voltage and thermal voltage. It is obtained from Eq. (14) with l = L, that is I(V, VT ) ≡ (cid:104)IL(cid:105). In Fig. 2 we show the magnitude of the thermoelec- tric coefficient representing the first derivative of the mean electric current with respect to the thermal volt- age, ∂I(V, VT )/∂VT, as a function of the bias voltage V at zero temperature for the case eVT (cid:28) Γ and for differ- ent gate voltages parameterized by the values of d. One can see that despite the fact that the Majorana fermions are well separated (ξ/Γ = 10−4) and the Majorana tun- neling is strong (η/Γ = 103), the curves do not collapse on a single universal (independent of d) curve. We find that only for eV (cid:29) ξ the curves do collapse on a single universal curve whose shape does not depend on the gate voltage which regulates the value of d. As mentioned in Section II, we show only the case d > 0 since universal Majorana signatures do not depend on both the magni- tude and sign of d. At the same time for d > 0 the quantum dot is in the empty orbital regime which ex- cludes the Kondo resonance [47] from the physical setup. Thus we conclude that the linear thermoelectric response, i.e. response at small bias voltages (eV (cid:28) ξ), of the 10-1010-910-810-710-610-510-410-310-210-1100101102103eVT[Γ]10-1010-810-610-410-2100102-dI/dVT[e2/h]10-1100101102103eVT[Γ]10-610-410-2100-d2I/dVdVT[e3/Γ h]kBT = 0.0kBT = 1.0 ΓkBT = 5.0 ΓkBT = 10.0 Γ4π23eV ΓeVTΓπ16eV Γ( )-2eVTΓ16π( )ΓeVT-28 FIG. 4. The first derivative of the greater current- current correlator with respect to the thermal voltage, ∂S>(V, VT )/∂VT , as a function of the gate voltage, speci- fied by the quantum dot single-particle energy level d, for different values of the Majorana overlap energy ξ. Here the parameters are as follows T = 0, V = 0, eVT /Γ = 10−6 and η/Γ = 103. FIG. 5. The first derivative of the greater current- current correlator with respect to the thermal voltage, ∂S>(V, VT )/∂VT , as a universal (independent of d) function of the bias voltage V at T = 0, eVT /Γ = 10−5. The Majorana overlap energy and the strength of the Majorana tunneling are ξ/Γ = 10−4 and η/Γ = 103, respectively. range of the thermal voltage. The universal behavior of ∂I(V, VT )/∂VT as a function of the thermal voltage VT in the regime specified by the inequality (21) is given by the asymptotic limit, Eq. (20), shown in Fig. 3 as the dashed line with a positive slope. However, for η (cid:29) eVT (cid:29) Γ (cid:29) eV (cid:29) ξ we find the following universal asymptotic limit: ∂I(V, VT ) ∂VT = − e2 h π 16 eV Γ (cid:18) eVT (cid:19)−2 Γ (25) (26) which is shown in Fig. 3 as the dashed line with a nega- tive slope. The inset in Fig. 3 shows the universal Majorana behavior of the thermoelectric coefficient representing the second mixed derivative of the mean electric current with respect to the thermal voltage and the bias voltage ∂2I(V, VT )/∂V ∂VT , as a function of the thermal voltage VT in the high-energy regime specified by the inequality in (25). The universal Majorana high-energy behavior of ∂2I(V, VT )/∂V ∂VT is independent of the bias voltage V . Like the second derivative ∂2I(V, VT )/∂V 2 T as a func- tion of the bias voltage V , the second mixed derivative ∂2I(V, VT )/∂V ∂VT as a function of the thermal voltage VT is robust with respect to high temperatures. At high thermal voltages it is given by the asymptotic limit ob- tained from Eq. (26) if the inequality in (25) is fulfilled. Its universal behavior is shown by the dashed line in the inset in Fig. 3. It is protected by high thermal volt- ages eVT (cid:29) kBT from thermal destruction effects as shown in the inset for high temperatures, kBT ∼ 10 Γ (kBT ∼ 0.01η). B. Thermoelectric fluctuations beyond linear response Now we address the universal Majorana signatures present in the thermoelectric response of the fluctuations of the electric current. To this end we use Eq. (17) with l = l(cid:48) = L to obtain the current-current correla- tor (cid:104)IL(t)IL(t(cid:48))(cid:105). Finite deviations of the electric current from its mean value, δIL(t) = IL(t) − I(V, VT ), may be characterized via the greater current-current correlator defined as S>(t, t(cid:48); V, VT ) ≡ (cid:104)δIL(t)δIL(t(cid:48))(cid:105). It can be expressed through the correlator (cid:104)IL(t)IL(t(cid:48))(cid:105) as follows: S>(t, t(cid:48); V, VT ) = (cid:104)IL(t)IL(t(cid:48))(cid:105) − I 2(V, VT ). (27) Due to the stationary nonequilibrium S>(t, t(cid:48); V, VT ) = S>(t − t(cid:48); V, VT ) and the physical quantity measured ex- perimentally is the Fourier transform S>(ω; V, VT ) = dt eiωtS>(t; V, VT ). (28) (cid:90) ∞ −∞ Below we focus on the zero frequency noise as a function of the bias voltage and thermal voltage, S>(V, VT ) ≡ S>(ω = 0; V, VT ). (29) As we know from the previous subsection the mean Majorana thermoelectric response becomes universal (in- dependent of d) only if the bias voltage is large enough, eV (cid:29) ξ. The first question is thus whether the same happens with the Majorana thermoelectric response of the fluctuations of the electric current. We find that, in contrast to the mean Majorana thermoelectric response, the fluctuations of the electric current are universal in the whole range of the bias voltage. This means that the Majorana thermoelectric response of the electric current 10-310-210-1100101102εd[Γ]0.250.50.7511.251.51.75dS>/dVT[e3/h]1+ln(2)ξ = 0.1 Γξ = 4.0.10-2Γξ = 2.0.10-2Γξ = 1.0.10-2Γξ = 1.0.10-4Γξ = 50.0 Γξ = 10.0 Γξ = 7.0 Γξ = 5.0 Γξ = 1.0 Γ10-1110-1010-910-810-710-610-510-410-310-210-1100101102eV[Γ]0.250.50.7511.251.51.75dS>/dVT[e3/h]1+ln(2)1 +ln(2)2eV = eVTeV = Γ9 FIG. 7. The universal (independent of d) behavior of the second mixed derivative of the greater current-current correlator with respect to the thermal voltage and the bias voltage, ∂2S>(V, VT )/∂V ∂VT , at large bias voltages V . The thermal voltage is eVT /Γ = 10−2. The Majorana over- lap energy and the strength of the Majorana tunneling are ξ/Γ = 10−4 and η/Γ = 103, respectively. The curves demon- strate robustness of ∂2S>(V, VT )/∂V ∂VT at eV (cid:29) kBT when the temperature increases. The inset shows the uni- versal nonlinear response thermoelectric ratio of the sec- ond mixed derivative of the greater current-current correla- tor with respect to the thermal voltage and the bias volt- age to the second derivative of the mean electric current times the electron charge with respect to the thermal volt- age,(cid:8)∂2S>(V, VT )/∂V ∂VT (cid:9)/(cid:8)∂2[eI(V, VT )]/∂V 2 (cid:9), at T = 0. T of this truly universal constant is ∂S>(V, VT ) ∂VT = e3 h [1 + ln(2)]. (31) For η (cid:29) Γ (cid:29) eV (cid:29) eVT , Γ (cid:29) ξ (32) fluctuation the coefficient ∂S>(V, VT )/∂VT does not depend on V either and is equal to a different truly universal constant. The asymptotic limit in this case is thermoelectric (cid:20) (cid:21) ∂S>(V, VT ) ∂VT = e3 h 1 2 1 + ln(2) . (33) These two plateaus are explicitly visible in Fig. 5. In the high-energy regime specified in (23) the uni- versal asymptotic limit of the fluctuation thermoelectric coefficient ∂S>(V, VT )/∂VT is (cid:18)eV (cid:19)−2 ∂S>(V, VT ) ∂VT = e3 h ln(2) 2 Γ . (34) As a function of the thermal voltage VT the fluctuation thermoelectric coefficient ∂S>(V, VT )/∂VT is shown in Fig. 6. In the regimes specified in (30) and (32) this dependence has, respectively, the plateaus (31) and (33) with the only difference in the order of these plateaus. FIG. 6. The first derivative of the greater current- current correlator with respect to the thermal voltage, ∂S>(V, VT )/∂VT , as a universal (independent of d) function of the thermal voltage VT at T = 0, eV /Γ = 10−5. The Majorana overlap energy and the strength of the Majorana tunneling are ξ/Γ = 10−4 and η/Γ = 103, respectively. noise is independent of d at any bias voltage V . In Fig. 4 we show the thermoelectric coefficient representing the first derivative of the greater noise with respect to the thermal voltage, ∂S>(V, VT )/∂VT , as a function of the gate voltage, parameterized by the value of d, at zero temperature, zero bias voltage, eVT (cid:28) Γ and different values of the Majorana overlap energy ξ in the regime specified in (19) where the Majorana tunneling domi- nates. As can be seen, for large values of the Majorana overlap energy there is a very strong dependence on the gate voltage with the unit plateau at small gate voltages. When ξ decreases the value of the plateau grows and the plateau becomes wider. Finally, when ξ is very small (the same as in the previous subsection on the mean elec- tric current, ξ/Γ = 10−4) the Majorana fermions overlap very weakly. In this situation the value of the plateau saturates at the asymptotic limit (e3/h)[1 + ln(2)] and the width of the plateau gets extremely wide. There- fore, when the Majorana tunneling is strong enough, so that (19) is satisfied, the fluctuation thermoelectric co- efficient ∂S>(V, VT )/∂VT becomes independent of d, Γ and η, that is it becomes truly universal with the value (e3/h)[1 + ln(2)] (the black horizontal line in Fig. 4). The universal Majorana behavior of the fluctuation thermoelectric coefficient ∂S>(V, VT )/∂VT as a function of the bias voltage V is shown in Fig. 5 at zero tempera- ture and eVT (cid:28) Γ. The values of ξ and η are the same as for the mean electric current from the previous sub- section so that the Majorana overlap is very weak and the Majorana tunneling is very strong. For η (cid:29) Γ (cid:29) eVT (cid:29) eV , Γ (cid:29) ξ (30) it turns out that the fluctuation thermoelectric coefficient ∂S>(V, VT )/∂VT does not depend on V and is truly uni- versal (independent of d, Γ and η). The asymptotic limit 10-1110-1010-910-810-710-610-510-410-310-210-1100101102eVT[Γ]0.250.50.7511.251.51.75dS>/dVT[e3/h]1+ln(2)1 +ln(2)2eVT = eVeVT = Γ10-1100101102103eV[Γ]10-1210-1010-810-610-410-2100102-d2S>/dVdVT[e4/Γ h]kBT = 0.0kBT = 1.0.10-2ΓkBT = 5.0.10-2ΓkBT = 8.0.10-2ΓkBT = 0.1 ΓkBT = 1.0 Γ100101102103104eV[Γ]00.10.150.2d2S>/dVdVTd2(eI)/dV2Tln(2)eV-33ln(2)4π2Γ( )10 age, (cid:8)∂2S>(V, VT )/∂V 2(cid:9)/(cid:8)∂2[eI(V, VT )]/∂V 2 FIG. 9. The universal nonlinear response thermoelec- tric ratio of the second derivative of the greater current- current correlator with respect to the bias voltage to the second derivative of the mean electric current times the electron charge with respect to the thermal volt- mal voltage is eVT /Γ = 10−5. The Majorana overlap energy and the strength of the Majorana tunneling are ξ/Γ = 10−4 and η/Γ = 103, respectively. The curves are shown for dif- ferent temperatures and demonstrate that at eV (cid:29) kBT this universal nonlinear response thermoelectric ratio is re- markably robust against thermal fluctuations whose ampli- tude grows when the temperature increases. (cid:9). The ther- T however, we were unable to identify the universal asymp- totic limit of the fluctuation thermoelectric coefficient ∂S>(V, VT )/∂VT . This high-energy universal law is defi- nitely not a power dependence and it will be a challenge for future research. In Fig. 7 we show the high-energy universal behav- ior of the fluctuation thermoelectric coefficient repre- senting the second mixed derivative of the greater noise with respect to the thermal voltage and the bias voltage, ∂2S>(V, VT )/∂V ∂VT , as a function of the bias voltage at eVT (cid:28) Γ and for different temperatures. As one can see, at eV (cid:29) kBT the high-energy behavior of the fluctu- ation thermoelectric coefficient ∂2S>(V, VT )/∂V ∂VT is almost independent of the temperature and obeys the universal asymptotic limit, (cid:18)eV (cid:19)−3 Γ ∂2S>(V, VT ) ∂V ∂VT = − e4 Γh ln(2) , (36) shown in Fig. 7 as the dashed line. The high-energy universal behavior (36) follows from the high-energy uni- versal behavior (34) of the fluctuation thermoelectric co- efficient ∂S>(V, VT )/∂VT . Comparison of the high-energy universal behavior of the mean thermoelectric coefficient ∂2I(V, VT )/∂V 2 T in Eq. (24) with the high-energy universal be- havior of the fluctuation thermoelectric coefficient ∂2S>(V, VT )/∂V ∂VT in Eq. (36) shows that the ratio be- tween these two nonlinear response coefficients is a truly The universal (independent of d) behavior of FIG. 8. the second derivative of the greater current-current correlator with respect to the bias voltage, ∂2S>(V, VT )/∂V 2, at large bias voltages V . The thermal voltage is eVT /Γ = 10−5. The Majorana overlap energy and the strength of the Majorana tunneling are ξ/Γ = 10−4 and η/Γ = 103, respectively. The curves are shown for different temperatures and demonstrate remarkable robustness of ∂2S>(V, VT )/∂V 2 at eV (cid:29) kBT when the temperature increases. Here the lowest plateau (33) is followed by the highest one (31). The physical mechanism underlying the fact that the truly universal plateau (31) in the regime (30) is larger than the truly universal plateau (33) in the regime (32) can be understood as follows. As schematically shown in Fig. 1, at T = 0 the Fermi-Dirac distribution in the left (hot) contact is smeared over the energy range eVT around the Fermi energy of this contact. When we have eVT (cid:29) eV , there exist two electron flows. The first flow, from the right (cold) contact to the left contact, is induced by electrons coming from the filled states be- low the Fermi energy of the right contact to the partially filled states below and above the Fermi energy of the left contact. The second flow, from the left contact to the right contact, is induced by electrons coming from the partially filled states above the Fermi energy of the left contact to the empty states above the Fermi energy of the right contact. Both of these electron flows fluctuate and contribute to the total noise S>(V, VT ) which grows linearly as a function of VT with the slope given by Eq. (31). However, when we decrease the thermal voltage up to the bias voltage, eVT ∼ eV , the second flow sig- nificantly decays and becomes negligible for eVT (cid:28) eV because the population of the states in the left contact above its Fermi energy rapidly goes to zero. In this sit- uation the noise contribution from the second flow van- ishes and only fluctuations of the first flow essentially contribute to the total noise S>(V, VT ) which still grows linearly as a function of VT but with the reduced slope given by Eq. (33). In the high-energy regime, η (cid:29) eVT (cid:29) Γ (cid:29) eV , Γ (cid:29) ξ, (35) 10-1100101102103104eV[Γ]10-1410-1210-1010-810-610-410-2100102-d2S>/dV2[e4/Γ h]kBT = 0.0kBT = 1.0 ΓkBT = 5.0 ΓkBT = 10.0 ΓeV3Γ2( )-3100101102103104eV[Γ]10-410-310-210-1100101d2S>/dV2d2(eI)/dV2TkBT = 0.0kBT = 1.0 ΓkBT = 5.0 ΓkBT = 10.0 Γ98π2universal constant with the asymptotic limit, ∂2S>(V,VT ) ∂V ∂VT ∂2[eI(V,VT )] ∂V 2 T = 3 ln(2) 4π2 , (37) shown in the inset in Fig. 7 as the dashed horizon- tal line. As can be seen in the inset, indeed, at zero temperature and large bias voltages, eV (cid:29) Γ, in the regime specified in (19) the thermoelectric ratio between ∂2S>(V, VT )/∂V ∂VT and ∂2[eI(V, VT )]/∂V 2 T is indepen- dent of d, Γ, η and the bias voltage V . However, out- side the high-energy regime, e.g., eV ∼ Γ, or when the Majorana tunneling condition (19) is not satisfied, e.g., eV > η, one can see deviations from the truly universal value (37). The inset in Fig. 7 demonstrates that at low bias voltages there are strong deviations from the truly universal value (37). Since these deviations de- pend on Γ, the behavior at low bias voltages is not truly universal. Nevertheless the thermoelectric ratio between ∂2S>(V, VT )/∂V ∂VT and ∂2[eI(V, VT )]/∂V 2 T is still uni- versal even at low bias voltages because it does not de- pend on d in this voltage range. As expected, we find that outside the validity of the Majorana tunneling con- dition (19), e.g., eV > η, the thermoelectric ratio be- tween ∂2S>(V, VT )/∂V ∂VT and ∂2[eI(V, VT )]/∂V 2 T is not universal (i.e. it depends on d) because the Majorana tunneling is not effective at such large bias voltages where the physics is not governed anymore by the Majorana fermions of the topological superconductor. In Fig. 8 we show the high-energy universal behav- ior of the fluctuation nonlinear response coefficient rep- resenting the second derivative of the greater noise with respect to the bias voltage, ∂2S>(V, VT )/∂V 2, as a func- tion of the bias voltage at eVT (cid:28) Γ and for different temperatures. In the regime specified in (23) the fluctu- ation nonlinear response coefficient ∂2S>(V, VT )/∂V 2 is pretty much insensitive at eV (cid:29) kBT to the tempera- ture increase and its universal asymptotic limit is (cid:18)eV (cid:19)−3 . (38) ∂2S>(V, VT ) ∂V 2 = − e4 Γh 3 2 Γ (24) with Eq. Comparing Eq. (38) one obtains that the thermoelectric ratio between ∂2S>(V, VT )/∂V 2 and ∂2[eI(V, VT )]/∂V 2 T is a truly universal constant with the asymptotic limit ∂2S>(V,VT ) ∂V 2 ∂2[eI(V,VT )] ∂V 2 T = 9 8π2 (39) in the regime specified in (23). This truly universal con- stant is shown in Fig. 9 as the dashed horizontal line. As demonstrated in Fig. 9, in the regime specified in (23) the universal Majorana thermoelectric ratio between the fluc- tuation nonlinear response coefficient ∂2S>(V, VT )/∂V 2 and the mean nonlinear coefficient ∂2[eI(V, VT )]/∂V 2 T is remarkably robust at eV (cid:29) kBT with respect to ther- mal noise excited at high temperatures. 11 At this point it is important to notice that the truly universal ratios (37) and (39), which are valid in the Ma- jorana tunneling regime (19) and at high bias voltages eV (cid:29) Γ, might result from symmetries of the full count- ing statistics as has been discussed in Refs. [53 -- 56]. In particular, in Refs. [55, 56] symmetries of the full count- ing statistics were used to derive relations between non- linear response coefficients beyond the Onsager-Casimir relation. Generalization of those relations between non- linear response coefficients and derivation of (37) and (39) using symmetries of the full counting statistics in the presence of topological superconductors supporting Majorana bound states is an important fundamental task which is, to our knowledge, still unexplored and definitely represents a special topic for future research. Let us estimate at which temperatures one can ex- perimentally observe the truly universal high-energy Majorana plateau (39) in the thermoelectric ratio be- tween the fluctuation nonlinear response coefficient ∂2S>(V, VT )/∂V 2 and the mean nonlinear response co- efficient ∂2I(V, VT )/∂V 2 T . We take the highest temper- ature shown in Fig. 9, that is kBT = 10 Γ. In units of η we have kBT = 0.01η. Since η is the largest energy scale, it should not exceed the induced super- conducting energy gap ∆ so as not to excite the bulk quasiparticles. Therefore, we assume η ∼ ∆. In Ref. [20] the induced superconducting energy gap is estimated as ∆ ≈ 250 µeV. One then obtains the temperature T ≈ 2.5 µeV/kB ≈ 0.03 K = 30 mK. In Ref. [57] a higher value of ∆ is reported, ∆ ≈ 15 meV. In this case one has T ≈ 1.8 K. Such temperatures are already high enough and thus may easily be achieved in modern laboratories. As a consequence, the truly universal thermoelectric con- stant (39) represents a unique and highly conclusive Ma- jorana signature to be detected in a realistic experiment. IV. CONCLUSION In this work we have explored thermoelectric Majo- rana response of both the mean value and the fluctua- tions of the electric current. The research has been fo- cused on unique universal Majorana thermoelectric sig- natures which may be detected in nanoscopic systems such as quantum dots. It has been shown that mean thermoelectric quantities become universal only at bias voltages high enough to exceed the overlap energy of the two Majorana bound states. For bias voltages below the Majorana overlap energy mean thermoelectric quan- tities are not universal even if the Majorana fermions are well separated. As a consequence, in order to ob- tain universal Majorana signatures from measurements of mean thermoelectric quantities one has to resort to essentially nonlinear response. In contrast, the thermo- electric response of the fluctuations of the electric current turns out to be universal at any bias voltage if the Ma- jorana fermions are well separated. We have obtained various fluctuation thermoelectric coefficients in differ- ent transport regimes. In particular, it has been shown that the differential thermoelectric noise has a universal two plateau structure with the truly universal values of the plateaus (e3/h)[1 + ln(2)] and (e3/h)[1 + ln(21/2)] de- pending on whether the ratio between the bias voltage and the thermal voltage (characterizing the temperature difference between the contacts) is less than or greater than one. Further, universal high-energy behavior of the fluctuation nonlinear response coefficients has been pre- sented and universal ratios between nonlinear response coefficients of the thermoelectric noise and the mean cur- rent have been obtained. Finally, we have demonstrated that at large bias voltages these thermoelectric ratios sat- urate to truly universal constants independent of the bias voltage and found the asymptotic limits of these truly universal constants. Importantly, these truly universal constants are protected by high bias voltages, making them robust against thermal noise. This robustness is crucial for realistic measurements. Therefore, the truly universal constants characterizing thermoelectric ratios at high bias voltages represent unique truly universal Majorana signatures challenging modern experiments on thermoelectric noise in quantum dots. It is fair to mention that many issues within Majo- rana noise still remain to be addressed. In particular, the above results have been demonstrated for a simple model where only one single-particle energy level of the quan- tum dot is involved in transport. However, in many re- 12 alistic systems several single-particle energy levels of the quantum dot may often contribute to transport. It is, therefore, interesting whether the results demonstrated here will change and, if so, what kind of Majorana fluc- tuation fingerprints one might expect in a multilevel sys- tem. This question as well as many other interesting issues will be a challenge for our future research on Ma- jorana noise. ACKNOWLEDGMENTS The author thanks Milena Grifoni, Wataru Izumida and Meydi Ferrier for important discussions. Appendix: Basic steps in the calculation of the current-current correlator Taking the second derivative in Eq. (17) one easily finds that the current-current correlator is the sum of a one-particle, S1ll(cid:48) (t, t(cid:48)), and a two-particle, S2ll(cid:48) (t, t(cid:48)), terms, (cid:104)Il(t)Il(cid:48)(t(cid:48))(cid:105) = S1ll(cid:48) (t, t(cid:48)) + S2ll(cid:48) (t, t(cid:48)). (A.1) The one-particle contribution has the form: (cid:18) e (cid:19)2(cid:88) (cid:8)(cid:10)[ψ1(t(cid:48)) + ψ2(t(cid:48))][− ¯ψ1(t) + ¯ψ2(t)](cid:11) (cid:2)GR SK S1ll(cid:48) (t, t(cid:48)) = δll(cid:48)Tl2(−i) − GA lk(t − t(cid:48))(cid:3)+(cid:10)[ψ1(t) − ψ2(t)][ ¯ψ1(t(cid:48)) + ¯ψ2(t(cid:48))](cid:11) 2 k (cid:2)−GR SK lk(t(cid:48) − t) + GK lk(t(cid:48) − t) + GA lk(t − t(cid:48)) + GK lk(t − t(cid:48))− lk(t(cid:48) − t)(cid:3)(cid:9), (A.2) where the angular brackets are defined in Eq. (15), (t − t(cid:48)) are, respectively, the retarded, advanced GR,A,K and Keldysh Green's functions of the isolated contacts lk and we have performed the Keldysh rotation of the Grass- mann fields of the quantum dot, (cid:2)ψ1(t) + qψ2(t)(cid:3), ψq(t) = 1√ 2 (cid:2) ¯ψ2(t) + q ¯ψ1(t)(cid:3). ¯ψq(t) = 1√ 2 (A.3) The one-particle contribution in Eq. (A.2) involves av- erages of products of only two Grassmann fields of the quantum dot. This contribution contains only normal terms, that is terms of the form (cid:104)ψs(t) ¯ψs(cid:48)(t(cid:48))(cid:105)SK , where s, s(cid:48) = 1, 2. It is obvious that anomalous terms, that is terms of the form (cid:104)ψs(t)ψs(cid:48)(t(cid:48))(cid:105)SK or (cid:104) ¯ψs(t) ¯ψs(cid:48)(t(cid:48))(cid:105)SK , cannot appear in this case. However, such terms do arise when one averages products of four Grassmann fields in the two-particle contribution S2ll(cid:48) (t, t(cid:48)). The two-particle contribution has the form: (cid:18) e (cid:19)2(cid:90) ∞ (cid:90) ∞ (t2 − t)(cid:3)−(cid:2)GR −∞ dt1 2 (t2 − t) − ¯ψ2(t2)GA S2ll(cid:48) (t, t(cid:48)) = Tl2Tl(cid:48)2 (−1) 2 − ¯ψ1(t2)GK (t1 − t(cid:48))(cid:3)−(cid:2)GR − GA + ¯ψ2(t1)GA (t − t2)ψ2(t2)(cid:3)(cid:2)− ¯ψ1(t) + ¯ψ2(t)(cid:3)(cid:9)(cid:8)(cid:2)ψ1(t(cid:48)) + ψ2(t(cid:48))(cid:3)(cid:2) ¯ψ1(t1)GR (t(cid:48) − t1)ψ2(t1)(cid:3)(cid:2) ¯ψ1(t(cid:48)) + ¯ψ2(t(cid:48))(cid:3)(cid:9)(cid:29) dt2 (t − t2)ψ1(t2) + GK lk2 l(cid:48)k1 (t(cid:48) − t1)ψ2(t1)+ (t(cid:48) − t1)ψ1(t1) + GK l(cid:48)k1 + GA −∞ l(cid:48)k1 l(cid:48)k1 k1,k2 lk2 lk2 lk2 lk2 . l(cid:48)k1 SK (cid:88) (cid:28)(cid:8)(cid:2)−ψ1(t) + ψ2(t)(cid:3)(cid:2) ¯ψ1(t2)GR lk2 13 (t2 − t)− (t − t2)ψ2(t2)− (t1 − t(cid:48)) + ¯ψ1(t1)GK l(cid:48)k1 (t1 − t(cid:48))+ (A.4) Since the total Keldysh action in Eq. (16) is quadratic, the average of four Grassmann fields is given as the sum of products of the averages of two Grassmann fields, ¯ψ2ψ3 (cid:104)ψ1 − (cid:104)ψ1 ¯ψ4(cid:105)SK = (cid:104)ψ1 ¯ψ2(cid:105)SK(cid:104)ψ3 ¯ψ4(cid:105)SK− ¯ψ2(cid:105)SK − (cid:104)ψ1ψ3(cid:105)SK(cid:104) ¯ψ2 ¯ψ4(cid:105)SK(cid:104)ψ3 ¯ψ4(cid:105)SK . (A.5) In Eq. (A.5) ψ1, ¯ψ2, ψ3, ¯ψ4 schematically denote those Grassmann fields which are taken from a given product of four square brackets in Eq. (A.4) in the same order as these Grassmann fields appear in these four square brackets. which is subtracted from the current-current correlator (cid:104)IL(t)IL(t(cid:48))(cid:105) in Eq. (27). All those terms in Eq. (A.4) which correspond to the second and third terms in the right hand side of Eq. (A.5) give, respectively, the normal and anomalous two-particle contributions to S>(t, t(cid:48); V, VT ). To treat the normal and anomalous contributions to S>(t, t(cid:48); V, VT ) from both S1ll(cid:48) (t, t(cid:48)) and S2ll(cid:48) (t, t(cid:48)) it is convenient to introduce a particle-hole space via the Grassmann fields ψis (i = p, h; s = 1, 2), ψis(t) ≡ ¯ψs(t), ψis(t) ≡ ψs(t), i = p, i = h. (A.6) All those terms in Eq. (A.4) which correspond to the first term in the right hand side of Eq. (A.5) do not rep- resent interest for the calculation of the greater current- current correlator S>(t, t(cid:48); V, VT ) in Eq. (27) because they give just the square of the mean electric current The averages in Eqs. (A.2) and (A.4) are then expressed in terms of the hole-particle, hole-hole and particle-particle retarded, advanced and Keldysh Green's functions of the quantum dot via, respectively, the follow- ing relations: (cid:104)ψhs(t)ψps(cid:48)(t(cid:48))(cid:105)SK = (cid:104)ψhs(t)ψhs(cid:48)(t(cid:48))(cid:105)SK = (cid:104)ψps(t)ψps(cid:48)(t(cid:48))(cid:105)SK = (cid:18)i GR (cid:18)i GK (cid:18) i GA 0 hp(t − t(cid:48)) i GK i GA hh(t − t(cid:48)) i GR hh(t − t(cid:48)) 0 i GA pp(t − t(cid:48)) i GK i GR (cid:19) hp(t − t(cid:48)) hp(t − t(cid:48)) (cid:19) hh(t − t(cid:48)) (cid:19) pp(t − t(cid:48)) pp(t − t(cid:48)) 0 , , . (A.7) (A.8) (A.9) Since the total Keldysh action in Eq. (16) is quadratic, the calculation of the Green's functions in the right hand sides of Eqs. (A.7)-(A.9) is reduced to calculations of the corresponding elements of the inverse kernel of this ac- tion. This is a quite simple, although somewhat lengthy, mathematical step which one may easily perform. After this step one finds the expressions for the hole-particle, hole-hole and particle-particle retarded, advanced and Keldysh Green's functions of the quantum dot. All the retarded and advanced Green's functions have already been found in Ref. [36] in the energy domain. For com- pleteness we give them here: −8(η∗)2 N R GR hp() = hh() = hp() , GR f () hp()]∗, GA hp() = [GR GA f () = 44 − 2(Γ2 + 42 hp() = 2{−4η2 − (ξ2 − 2)[iΓ + 2(d + )]}. N R hh() = [GR d + 4ξ2 + 16η2) + ξ2(Γ2 + 42 , GR f () pp()]∗, GA pp() = pp() = [GR −8η2 , f () hh()]∗, d) + i 4Γ[3 − (ξ2 + 2η2)], (A.10) Finally, for the Keldysh components of the hole-particle, hole-hole and particle-particle Green's functions one obtains the following expressions: 14 GK hp() = hh() = N K hp() f ()2 , GK N K hh() f ()2 , GK hp() = −2 i Γ(cid:8)(FL() + FR())(ξ2 − 2)2[Γ2 + 4(d + )2]+ + 16η2[(FL() + FR())(ξ2 − 2)(d + ) + η2(FL() − FL(−) + FR() − FR(−))](cid:9), hh() = 8Γ(η∗)2(cid:8)(ξ2 − 2)[(FL() + FL(−) + FR() + FR(−))(Γ − 2 i d)+ + 2 i (FL(−) − FL() + FR(−) − FR())] + 4 iη2(FL(−) − FL() + FR(−) − FR())(cid:9), N K pp() f ()2 , pp() = N K N K pp() = −(N K N K hh())∗, (A.11) where FL,R() ≡ 1 − 2fL,R(). Using Eqs. (A.2), (A.4), (A.5) and (A.7)-(A.11) one may without much effort express the zero frequency noise in Eq. (29) as well as its various derivatives as integrals in the energy domain. These integrals may be calculated numerically to obtain the quantities discussed in the main text as functions of the bias voltage and thermal voltage. [1] E. Majorana, "Teoria simmetrica dell'elettrone e del positrone," Nuovo Cimento 14, 171 (1937). [2] C. Itzykson and J.-B. Zuber, Quantum Field Theory (McGraw-Hill, New York, St. Louis, San Francisco, 1980). [3] A. Yu. Kitaev, "Unpaired Majorana fermions in quantum wires," Phys.-Usp. 44, 131 (2001). [4] J. Alicea, "New directions in the pursuit of Majorana fermions in solid state systems," Rep. Prog. Phys. 75, 076501 (2012). [5] M. Leijnse and K. Flensberg, "Introduction to topological superconductivity and Majorana fermions," Semicond. Sci. Technol. 27, 124003 (2012). [6] M. Sato and S. Fujimoto, "Majorana fermions and topol- ogy in superconductors," J. Phys. Soc. Japan 85, 072001 (2016). [7] L. Fu and C. L. Kane, "Superconducting proximity effect and Majorana fermions at the surface of a topological insulator," Phys. Rev. Lett. 100, 096407 (2008). [8] L. Fu and C. L. Kane, "Josephson current and superconductor/quantum-spin-Hall- junction," Phys. Rev. B noise insulator/superconductor 79, 161408(R) (2009). at a [9] R. M. Lutchyn, J. D. Sau, and S. Das Sarma, "Ma- jorana fermions and a topological phase transition in semiconductor-superconductor heterostructures," Phys. Rev. Lett. 105, 077001 (2010). [10] Y. Oreg, G. Refael, and F. von Oppen, "Helical liquids and Majorana bound states in quantum wires," Phys. Rev. Lett. 105, 177002 (2010). [11] L. Fidkowski, J. Alicea, N. H. Lindner, R. M. Lutchyn, and M. P. A. Fisher, "Universal transport signatures of Majorana fermions in superconductor-Luttinger liquid junctions," Phys. Rev. B 85, 245121 (2012). [12] M. Lee, J. S. Lim, and R. L´opez, "Kondo effect in a quantum dot side-coupled to a topological superconduc- tor," Phys. Rev. B 87, 241402(R) (2013). [13] A. Kundu and B. Seradjeh, "Transport signatures of Flo- quet Majorana fermions in driven topological supercon- ductors," Phys. Rev. Lett. 111, 136402 (2013). [14] R. Ilan, J. H. Bardarson, H.-S. Sim, and J. E. Moore, "Detecting perfect transmission in Josephson junctions on the surface of three dimensional topological insula- tors," New J. Phys. 16, 053007 (2014). [15] M. Cheng, M. Becker, B. Bauer, and R. M. Lutchyn, "Interplay between Kondo and Majorana interactions in quantum dots," Phys. Rev. X 4, 031051 (2014). [16] A. M. Lobos and S. Das Sarma, "Tunneling transport in NSN Majorana junctions across the topological quantum phase transition," New J. Phys. 17, 065010 (2015). [17] R. Wang, H.-Y. Lu, B. Wang, and C. S. Ting, "Spinon Majorana fermions," Phys. Rev. B 94, 125146 (2016). [18] R. M. Lutchyn and L. I. Glazman, "Transport through a Majorana island in the strong tunneling regime," Phys. Rev. Lett. 119, 057002 (2017). [19] H. Huang, Q.-F. Liang, D.-X. Yao, and Z. Wang, "Ma- jorana φ0-junction in a disordered spin-orbit coupling nanowire with tilted magnetic field," Physica C: Super- conductivity and its Applications 543, 22 (2017). [20] V. Mourik, K. Zuo, S. M. Frolov, S. R. Plissard, E. P. A. M. Bakkers, and L. P. Kouwenhoven, "Sig- natures of Majorana fermions in hybrid superconductor- semiconductor nanowire devices," Science 336, 1003 (2012). [21] S. M. Albrecht, A. P. Higginbotham, M. Madsen, F. Kuemmeth, T. S. Jespersen, J. Nygard, P. Krogstrup, and C. M. Marcus, "Exponential protection of zero modes in Majorana islands," Nature 531, 206 (2016). [22] H. Zhang, C.-X. Liu, S. Gazibegovic, D. Xu, J. A. Lo- gan, G. Wang, N. van Loo, J. D. S. Bommer, M. W. A. de Moor, D. Car, R. L. M. O. het Veld, P. J. van Veld- hoven, S. Koelling, M. A. Verheijen, M. Pendharkar, D. J. Pennachio, B. Shojaei, J. S. Lee, C. J. Palmstrøm, E. P. A. M. Bakkers, S. D. Sarma, and L. P. Kouwen- hoven, "Quantized Majorana conductance," Nature 556, 74 (2018). [23] G. Kells, D. Meidan, and P. W. Brouwer, "Near-zero- energy end states in topologically trivial spin-orbit cou- pled superconducting nanowires with a smooth confine- ment," Phys. Rev. B 86, 100503(R) (2012). [24] C.-X. Liu, J. D. Sau, T. D. Stanescu, and S. D. Sarma, "Andreev bound states versus Majorana bound states in quantum dot-nanowire-superconductor hybrid struc- tures: Trivial versus topological zero-bias conductance peaks," Phys. Rev. B 96, 075161 (2017). [25] E. Prada, R. Aguado, and P. San-Jose, "Measuring Ma- jorana nonlocality and spin structure with a quantum dot," Phys. Rev. B 96, 085418 (2017). [26] D. J. Clarke, "Experimentally accessible topological qual- ity factor for wires with zero energy modes," Phys. Rev. B 96, 201109(R) (2017). [27] D. E. Liu, M. Cheng, and R. M. Lutchyn, "Probing Ma- jorana physics in quantum-dot shot-noise experiments," Phys. Rev. B 91, 081405(R) (2015). [28] D. E. Liu, A. Levchenko, and R. M. Lutchyn, "Majo- rana zero modes choose Euler numbers as revealed by full counting statistics," Phys. Rev. B 92, 205422 (2015). [29] C. W. J. Beenakker, "Random-matrix theory of Majo- rana fermions and topological superconductors," Rev. Mod. Phys. 87, 1037 (2015). [30] S. Valentini, M. Governale, R. Fazio, and F. Taddei, "Finite-frequency noise in a topological superconducting wire," Physica E 75, 15 -- 21 (2016). [31] H. Nyquist, "Thermal agitation of electric charge in con- ductors," Phys. Rev. 32, 110 (1928). [32] H. B. Callen and T. A. Welton, "Irreversibility and gen- eralized noise," Phys. Rev. 83, 34 (1951). [33] L. D. Landau and E. M. Lifshitz, Statistical Physics. Part 1: Course of Theoretical Physics, 3rd ed., Vol. 5 (Perga- mon Press, Oxford, 1980). [34] A. Haim, E. Berg, F. von Oppen, and Y. Oreg, "Current correlations in a Majorana beam splitter," Phys. Rev. B 92, 245112 (2015). [35] S. Smirnov, "Non-equilibrium Majorana fluctuations," New J. Phys. 19, 063020 (2017). [36] S. Smirnov, "Majorana tunneling entropy," Phys. Rev. B 92, 195312 (2015). [37] M. Ferrier, T. Arakawa, T. Hata, R. Fujiwara, R. Dela- grange, R. Weil, R. Deblock, R. Sakano, A. Oguri, and K. Kobayashi, "Universality of non-equilibrium fluctu- ations in strongly correlated quantum liquids," Nature Physics 12, 230 (2016). [38] T. Senthil and M. P. A. Fisher, "Quasiparticle local- ization in superconductors with spin-orbit scattering," Phys. Rev. B 61, 9690 (2000). [39] N. Read and D. 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). [40] T. Mizushima and M. Nitta, "Topology and symmetry of surface Majorana arcs in cyclic superconductors," Phys. Rev. B 97, 024506 (2018). [41] A. Dorda, M. Ganahl, S. Andergassen, W. von der Lin- 15 den, and E. Arrigoni, "Thermoelectric response of a cor- related impurity in the nonequilibrium Kondo regime," Phys. Rev. B 94, 245125 (2016). [42] I. Titvinidze, A. Dorda, W. von der Linden, and E. Ar- rigoni, "Thermoelectric properties of a strongly corre- lated layer," Phys. Rev. B 96, 115104 (2017). [43] M. Leijnse, "Thermoelectric signatures of a Majorana bound state coupled to a quantum dot," New J. Phys. 16, 015029 (2014). [44] R. L´opez, M. Lee, L. Serra, and J. S. Lim, "Thermoelec- trical detection of Majorana states," Phys. Rev. B 89, 205418 (2014). [45] H. Khim, R. L´opez, J. S. Lim, and M. Lee, "Thermoelec- tric effect in the Kondo dot side-coupled to a Majorana mode," Eur. Phys. J. B 88, 151 (2015). [46] J. P. Ramos-Andrade, O. ´Avalos-Ovando, P. A. Orellana, and S. E. Ulloa, "Thermoelectric transport through Ma- jorana bound states and violation of Wiedemann-Franz law," Phys. Rev. B 94, 155436 (2016). [47] A. C. Hewson, The Kondo Problem to Heavy Fermions (Cambridge University Press, Cambridge, 1997). [48] J. Fuchs and C. Schweigert, Symmetries, Lie Algebras and Representations, a Graduate Course for Physicists (Cambridge University Press, Cambridge, 1997). [49] A. Kamenev and A. Andreev, "Electron-electron inter- actions in disordered metals: Keldysh formalism," Phys. Rev. B 60, 2218 (1999). [50] A. Altland and B. Simons, Condensed Matter Field The- ory, 2nd ed. (Cambridge University Press, Cambridge, 2010). [51] Y. Meir and N. S. Wingreen, "Landauer formula for the current through an interacting electron region," Phys. Rev. Lett. 68, 2512 (1992). [52] D. Goldhaber-Gordon, H. Shtrikman, D. Mahalu, D. Abusch-Magder, U. Meirav, and M. A. Kastner, "Kondo effect in a single-electron transistor," Nature 391, 156 (1998). [53] J. Tobiska and Yu. V. Nazarov, "Inelastic interaction cor- rections and universal relations for full counting statistics in a quantum contact," Phys. Rev. B 72, 235328 (2005). [54] D. Andrieux and P. Gaspard, "Fluctuation theorem for transport in mesoscopic systems," J. Stat. Mech. 2006, P01011 (2006). [55] K. Saito and Y. Utsumi, "Symmetry in full counting statistics, fluctuation theorem, and relations among non- linear transport coefficients in the presence of a magnetic field," Phys. Rev. B 78, 115429 (2008). [56] E. Iyoda, Y. Utsumi, and T. Kato, "Nonequilibrium ex- tension of Onsager relations for thermoelectric effects in mesoscopic conductors," J. Phys. Soc. Japan 79, 045003 (2010). [57] E. Wang, H. Ding, A. V. Fedorov, W. Yao, Z. Li, Y.- F. Lv, K. Zhao, L.-G. Zhang, Z. Xu, J. Schneeloch, R. Zhong, S.-H. Ji, L. Wang, K. He, X. Ma, G. Gu, H. Yao, Q.-K. Xue, X. Chen, and S. Zhou, "Fully gapped topological surface states in Bi2Se3 films induced by a d- wave high-temperature superconductor," Nature Physics 9, 621 (2013).
1005.5121	2	1005	2010-10-21T12:45:52	Atypical Fractional Quantum Hall Effect in Graphene at Filling Factor 1/3	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	We study the recently observed graphene fractional quantum Hall state at a filling factor $\nu_G=1/3$ using a four-component trial wave function and exact diagonalization calculations. Although it is adiabatically connected to a 1/3 Laughlin state in the upper spin branch, with SU(2) valley-isospin ferromagnetic ordering and a completely filled lower spin branch, it reveals physical properties beyond such a state that is the natural ground state for a large Zeeman effect. Most saliently, it possesses at experimentally relevant values of the Zeeman gap low-energy spin-flip excitations that may be unveiled in inelastic light-scattering experiments.	cond-mat.mes-hall	cond-mat	Atypical Fractional Quantum Hall Effect in Graphene at Filling Factor νG = 1/3 1Laboratoire de Physique des Solides, CNRS UMR 8502, Univ. Paris-Sud, F-91405 Orsay cedex, France Z. Papi´c1,2,3, M. O. Goerbig1, and N. Regnault2 2Laboratoire Pierre Aigrain, D´epartement de Physique, ENS, CNRS, 24 Rue Lhomond, F-75005 Paris, France 3Institute of Physics, University of Belgrade, P. O. Box 68, 11 000 Belgrade, Serbia We study, with the help of exact diagonalization calculations, a four-component trial wave function that may be relevant for the recently observed graphene fractional quantum Hall state at a filling factor νG = 1/3. Although it is adiabatically connected to a 1/3 Laughlin state in the upper spin branch, with SU(2) valley-isospin ferromagnetic ordering and a completely filled lower spin branch, it reveals physical properties beyond such a state that is the natural ground state for a large Zeeman effect. Most saliently, it possesses at experimentally relevant values of the Zeeman gap low-energy spin-flip excitations that may be unveiled in inelastic light-scattering experiments. PACS numbers: 73.43.Nq, 71.10.Pm, 73.20.Qt The recent observation of the fractional quantum Hall effect (FQHE) in graphene [1, 2] has proven the relevance of Coulomb interactions in this novel two-dimensional (2D) electron system, in agreement with theoretical ex- pectations [3 -- 8]. The most pronounced state is the one found when the ratio νG = nel/nB between the electronic density nel and that of the flux quanta nB = eB/h is νG = 1/3. Although this state is reminiscent, at first sight, of the prominent 1/3 state observed in semicon- ductor heterostructures [9], which is described to great accuracy by the Laughlin state [10], several questions arise when taking fully into account the four-component strucure of graphene, due to its four-fold spin-valley de- generacy. Whereas first numerical approaches [5] con- sidered the physical spin to be frozen by the Zeeman effect and concentrated on the valley-isospin degree of freedom in a two-component system, a four-component approach [8] seems to be more appropriate in view of the rather small energy scale associated with the Zeeman ef- fect ∆Z , when compared to the leading energy scale of the Coulomb interaction, e2/ǫlB at the magnetic length lB = p¯hc/eB. Indeed for a g-factor of 2 [11], one obtains ∆Z /(e2/ǫlB) ∼ 0.002pB[T] × ǫ, where ǫ is the relative dielectric constant. A further complication arises in graphene, as compared to the 2D electron gas in semiconductor heterostructures, when one considers the definition of the filling factor νG, which is proportional to the carrier density nel. In graphene, the carrier density vanishes at the Dirac point, where the spectrum is particle-hole symmetric. In the presence of a magnetic field, a four-fold degenerate zero- energy Landau level (LL) is formed that happens to be half-filled when nel = 0 and thus νG = 0 -- the situa- tion at νG = 0 is therefore more reminiscent of a fill- ing factor of ν = 2 in a usual four-component quantum Hall system [8], and the observed FQHE at νG = 1/3 corresponds to a situation where two of the four spin- valley subbranches are completely filled and a third one 1/3-filled (ν = 2 + 1/3). As a consequence the ob- served FQHE is not a simple Laughlin state with an SU(4)-spin-valley ferromagnetic ordering, which would arise at νG = −2 + 1/3 (or by particle-hole symme- try, at νG = 2 − 1/3) [6, 8]. A natural candidate for large values of the Zeeman gap would then be a valley- SU(2)-ferromagnetic Laughlin state Ψv−SU(2) in the spin- ↓ branch of the zero-energy LL similar to the usual 1/3 physics. In this scenario, both states K and K ′ are com- pletely filled in the spin-↑ branch. The small relative value of the Zeeman gap, however, casts doubts on such a scenario of complete spin polarization induced by an external field, without considering a cooperative effect mediated by the Coulomb interaction. ↓,L Here, we analyse the system in the zero-energy LL with the help of exact-diagonalization (ED) calculations for relativistic electrons in the spherical geometry with SU(4) symmetry that interact via the Coulomb interac- tion [4, 5, 8]. We show that already for a very small Zeeman effect, one may obtain a FQHE at νG = 1/3 in graphene. This state may be described in terms of a four- component Halperin wave function ΨSU(4) 2+1/3 which is adi- abatically connected to the valley-SU(2)-ferromagnetic state in the upper spin branch. The latter is the natural ground state for a large Zeeman splitting. Most saliently, in spite of this adiabatic connection, the low-energy ex- citations in an intermediate range of the Zeeman split- ting are different from those of the simple SU(2) Laugh- lin state. In addition to the charge-density-wave mode with its characteristic magneto-roton minimum and the valley-isospin wave, which is the Goldstone mode asso- ciated with the spontaneous valley-isospin breaking in the spin-↓ branch, we find a low-energy spin-flip mode with a gap that depends linearly on the Zeeman cou- pling. These modes may be experimentally accessible in inelastic light-scattering measurements that have re- vealed similar modes in conventional quantum Hall sys- tems in GaAs heterostructures [12, 13]. That electrons in graphene reside at the sample surface makes this novel 2D electron system even better adapted to optical mea- surements than the latter. In order to describe the FQHE state at νG = 1/3, which corresponds to a filling factor of ν = 2 + 1/3 when counted from the bottom of the central n = 0 LL, we investigate the trial four-component wave function ΨSU(4) 2+1/3 = Y ξ=K,K ′ Y i<j (cid:16)z↓,K i − z↓,K ′ j Y i,j (cid:17)3 (1) (cid:16)z↓,ξ j (cid:17)3 i − z↓,ξ (cid:16)z↑,ξ i − z↑,ξ × Y ξ=K,K ′ Y i<j j (cid:17) , j where zσ,ξ denotes the complex coordinate of the j-th electron in the spin-valley subbranch σ, ξ (σ =↑ or ↓ and ξ = K or K ′). We have omitted an ubiqitous Gaussian factor in the expression. Notice that, in the absence of a symmetry-breaking field, the wave function (1) is not a good trial state because the Coulomb interaction po- tential respects the SU(4)-spin-valley symmetry [4, 14], whereas the wave function (1) is not an eigenstate of the SU(4)-Casimir operators. This is indeed corroborated by our ED calculations for N = 17 particles with NB = 6 flux quanta thread- ing the sphere, the relation between N and NB being NB = (3/7)N − 9/7 for the state (1) [15], which yields the required filling ν = 7/3 in the thermodynamic limit. The ground state is then found in spin sectors differ- ent from that, 2Sz = 11, expected for the state (1). A simple manner to stabilize the state (1) is to use appro- priate pseudo-potentials [16] that break the SU(4)-spin- valley symmetry in the interaction potential. However, surprisingly, this trial state becomes the ground state also when the SU(4) symmetry is broken by an exter- nal field -- e.g. a very small value of the Zeeman effect turns out to be sufficient, ∆1 Z ≃ 0.01e2/ǫlB, which is a tiny fraction of the leading Coulomb interaction energy scale e2/ǫlB ∼ 625(pB[T]/ǫ) K. In a typical experimen- tal situation, the 1/3 state has been found at a magnetic field of roughly B ∼ 9...12 T [1, 2], which corresponds to a ratio of ∆Z /(e2/ǫlB) ∼ 0.006...0.008 × ǫ if one uses g ∼ 2 [11]. This value is slightly smaller than our the- oretical estimate if one considers the smallest possible value of the dielectric constant (ǫ = 1 for free-standing graphene). However, virtual interband excitations lower the dielectric constant that becomes ǫ∞ ≃ 4 for free- standing graphene in the large-wave-vector limit [17], also in a strong magnetic field [18], and the precise value of the dielectric constant in graphene remains an open issue. Notice that it is even still under debate whether the Zee- man splitting is indeed the dominant SU(4)-symmetry- breaking perturbation or whether the valley splitting is more relevant. Our theory and the conclusions of this paper, however, remain valid also in the latter case if one interchanges the role of spin and valley isospin and if one replaces ∆Z by a "valley Zeeman effect" [19]. The energy spectrum, which we have obtained in ED calculations, is shown in Fig. 1 as a function of the Zee- ] B l ε / 2 e [ E -10.15 -10.2 -10.25 -10.3 -10.35 -10.4 -10.45 2 2Sz<7 2Sz=7 2Sz=9 2Sz=11 -10.16 -10.18 -10.2 -10.22 -10.24 0 0.005 0.01 0.015 0.02 0 0.01 0.02 0.03 ∆ z [e2/εlB] 0.04 0.05 0.06 FIG. 1: (Color online) Energy spectrum for N = 17 electrons at a filling factor νG = 1/3 (ν = 2 + 1/3), as a function of ∆Z , obtained from ED calculations of the Coulomb interaction on the sphere (NB = 6) with implemented SU(4) symmetry. Above Z ≃ 0.01e2/ǫlB, the ground state is found in the maximally spin- ∆1 polarized sector (2Sz = 11, red diamonds). The inset shows a zoom on the region for small values of ∆Z . K' K K' K C ISW SF FIG. 2: Classification of the excitations of the ΨSU(4) state. The excitations of a one-component Laughlin state are found in the same spin-valley sector (C), whereas the Goldstone mode due to the broken SU(2) valley-isospin symmetry in the spin-↓ branch is an insospin-wave mode (ISW). In addition to these conventional modes, the four-component state (1) possesses a spin-flip (SF) mode. 1/3 man gap ∆Z . Above the critical value ∆1 Z , the ground state is found in the maximally-polarized spin sector that corresponds to the state (1), whereas the excited state with the lowest energy is in the same spin sector, 2Sz = 11, only above a second value ∆2 Z ≃ 0.03e2/ǫlB. For values of the Zeeman gap ∆1 Z, the excited state with lowest energy is found in the spin sec- tor 2SZ = 9. Above ∆2 Z , however, the energy cost of this spin-flip excitation (SF, see Fig. 2) is larger than the lowest-lying excitation in the fully polarized sector 2Sz = 11 (C in Fig. 2). Z ≤ ∆Z ≤ ∆2 These results suggest that the state (1) may have phys- ical properties beyond the simple 1/3-Laughlin state in the spin-↓ branch, in the form of coherent spin-flip excita- tions in an intermediate range of Zeeman gaps. In order to test this scenario in more detail, we have investigated the two-component wave function ΨSU(2) 1+1/3 = Y i<j (cid:16)z↓ j(cid:17)3 i − z↓ Y i<j (cid:16)z↑ i − z↑ j(cid:17) , (2) which would be a candidate in a two-component quantum Hall system, such as a conventional 2D electron gas in a GaAs heterostructure, at a filling factor ν = 1+1/3. It is insofar related to the four-component wave function (1) as it describes the same physical situation if the valley- isospin degree of freedom for spin-↓ electrons is neglected. The novel wave function (2) therefore does not reveal any valley-isospin-wave mode (ISW, see Fig. 2) that is the Goldstone mode of the spontaneously broken valley- SU(2) symmetry in the spin-↓ branch and that may even- tually become gapped if one takes into account a possible valley splitting. In contrast to its four-component ana- logue (1), the two-component wave function (2) allows for a more detailed study of different system sizes in ED with an implemented SU(2) symmetry. Indeed, our ED cal- culations with an implemented SU(4) symmetry allowed only for one single system size (N = 17 particles, only N↓ = 3 populate the upper spin branch), in which case the subspace with maximal spin polarization (2Sz = 11) is of dimension one such that the overlap with the wave function (1) is trivially 1. Figure 3(a) shows the energy spectrum for N = 22 particles and NB = 15 flux quanta, in the different spin sectors, obtained by ED of the SU(2) Coulomb inter- action potential in the lowest LL. The spectrum is in qualitative agreement with that obtained for the four- component system at ν = 2 + 1/3 (Fig. 1) -- because the wave function (2) is not an eigenstate of the SU(2)- symmetric Coulomb potential, it does not describe the ground state at ∆Z = 0, where one obtains a three-fold degenerate state (with 2Sz = 0, ±2), but in a compress- ible sector (L 6= 0). As for the four-component case, a Z ≃ 0.01e2/ǫlB small symmetry-breaking Zeeman gap ∆1 suffices to stabilize a state with maximal spin polariza- tion (2Sz = 10 and N↓ = 6), which has an overlap of 99% with the wave function (2) [23], and the lowest-lying ex- cited state in an intemediate range of the Zeeman gap, ∆1 Z ≃ 0.08e2/ǫlB, involves a spin flip as it Z ≤ ∆Z ≤ ∆2 is found in the spin sector 2Sz = 8. It has been argued that, for vanishing Zeeman split- ting, the state at ν = 1 + 1/3 should be a spin-singlet composite-fermion (CF) state with reversed flux attach- ment [20]. Hund's rule, according to which the system chooses a maximally polarized spin inside each energy level, would then predict an unpolarized state because ν = 2/3 corresponds to a completely filled lowest CF- LL [20], but the same rule favors a completely polar- ized state if applied to the original electron coordinates. Our results indicate that already for a very small Zee- man splitting, a completely polarized state is favored that satisfies the electronic instead of the CF version of Hund's rule. Notice, however, that a direct numeri- cal comparison between both states, CF spin singlet and state (2), is problematic in the spherical geometry be- cause the spin-singlet state has a different flux-particle- number relation, NB = (3/4)N − 1, than the polarized state (2), NB = (3/4)N − 3/2. We find for N = 20 and (a) 3 ] B l ε / 2 e [ E (b) ] B l ε / 2 e [ E ∆ -12 -12.4 -12.8 -13.2 -12.48 -12.52 -12.56 -12.6 0 2Sz=10 2Sz=8 2Sz=6 2Sz=4 2Sz=2 2Sz=0 0.02 0.04 0.06 0.08 z [e2/εlB] ∆ 0 0.005 0.01 0.1 0.12 0.14 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 ~ B 0 1 2 3 4 L ~B1/2 2Sz=10 2Sz=8 2Sz=6 6 5 7 8 FIG. 3: (Color online) (a) Energy spectrum for N = 22 electrons at a filling factor ν = 1 + 1/3, as a function of ∆Z in the differ- ent spin sectors 2Sz, obtained from ED calculations of the Coulomb interaction on the sphere (NB = 15) with implemented SU(2) sym- metry. The inset shows a zoom on the region for small values of ∆Z . (b) Excitation spectrum at ∆Z = 0.05e2/ǫlB , as a function of the angular momentum L. The energy is measured with respect to the ground state (at L = 0). The spin-flip mode is found in the 2Sz = 8 sector (blue squares) and scales linearly in B, whereas the 2Sz = 10 sector reveals the usual magneto-roton branch (red diamonds), which scales as √B with the B-field. NB = 14 (results not shown) that the ground state is indeed a singlet at low Zeeman splittings, but it is max- imally polarized above a value of ∆Z/(e2/ǫlB) ∼ 0.03, which is on the same order of magnitude as ∆1 Z . In order to gain further insight into the nature of the low-lying excitations, we have calculated the spec- trum [Fig. 3(b)] at an intermediate value of the Zee- man gap, ∆Z = 0.05e2/ǫlB, where spin-flip excitations are expected to be relevant. The spectrum is now plot- ted as a function of the angular momentum in order to make apparent possible low-energy collective excitations of the incompressible state (2). Within the charge sec- tor with no change in the spin polarization, one observes in Fig. 3(b) the usual magneto-roton branch (red dia- monds) [21] which arises from gapped density-wave exci- tations [22] and which is a prominent feature of Laughlin- type physics. However, another mode is apparent in Fig. 3(b) that indicates the presence of collective excitations beyond the usual one-component Laughlin state and that is precisely a spin-flip excitation (blue squares). This mode, which is the lowest-energy excitation in the low- L regime, is well separated from the high-energy part of the excitation spectrum, such that it is likely to be a true collective mode. Notice that in the large-L limit, the magneto-roton branch has a lower energy, and one may thus conjecture that the activation gap, i.e. the energy to create a well-separated quasiparticle-quasihole pair at large values of L, does not involve a spin-flip excitation, but is governed by one-component Laughlin physics. The relevance of collective spin-flip excitations in an intermediate Zeeman-gap range may eventually be tested experimentally in inelastic light-scattering experiments that are capable of probing collective excitations at finite wave vectors [12, 13]. Indeed, these experiments probe characteristic parts of the dispersion relation that show an enhanced density of states (such as at its minima and maxima). Because the spin-flip mode in Fig. 3(b) is almost flat at low angular momenta L that correspond to small wave vectors, one may expect an enhanced peak in such inelastic light-scattering measurements, at energies in the 0.1e2/ǫlB range (roughly half of the energy of the magneto-roton minimum, for the particular choice ∆Z = 0.05e2/ǫlB). As one may see in Fig. 3, the spin-flip excitation scales linearly with the Zeeman gap, such that the associated peak is expected to scale linearly with the magnetic field as well, whereas that of the usual magneto- roton would scale as √B [see Fig. 3(b)]. The observation of such a linear B-field dependence of the light-scattering peak would be clear evidence for the relevance of spin- flip, beyond the properties of the Laughlin liquid, of the νG = 1/3 state in graphene. In conclusion, we have shown, within ED calculations for a four- and a two-component system on the sphere, how a FQHE can arise in graphene at νG = 1/3 even at very small values of a spin-valley symmetry-breaking Zee- man field. Although the leading energy scale is given by the SU(4)-invariant Coulomb interaction, a small Zeeman gap ∆Z /(e2/ǫlB) ∼ 0.01 is sufficient to fully polarize the electronic spin and thus to stabilize the state (1) which we have identified as being responsible for the observed graphene FQHE [1, 2]. In spite of its reminiscence with the Laughlin state, novel collective excitations that are inherent to the four-component character of graphene de- termine the low-energy spectrum at intermediate values of the Zeeman gap, ∆1 Z, that correspond to the experimental situation in which the FQHE has been observed. In order to gain further insight into the nature of these spin-flip excitations, which may be visible in in- elastic light-scattering experiments, we have performed ED calculations in an analogous two-component quan- tum Hall system at a filling factor ν = 1 + 1/3 that corresponds to a completely filled spin-↑ and a one-third filled spin-↓ branch. The spin-flip excitation is well sep- arated from the high-energy part of the energy spectrum thus indicating its collective nature, in addition to the Z ≤ ∆Z ≤ ∆2 4 usual magneto-roton branch that determines the low- energy spectrum in the large angular-momentum regime. This work was supported by Agence Nationale de la Recherche under Grant No. ANR-JCJC-0003-01 and by grants from R´egion Ile-de-France. Z.P. was further- more supported by the Serbian Ministry of Science under Grant No. 141035. We acknowledge fruitful discussions with Ch. Glattli and M. Milovanovi´c. [1] X. Du, I. Skachko, F. Duerr, A. Luican, and E. Y. Andrei, Nature 462, 192 (2009). [2] K. I. Bolotin, F. Ghahari, M. D. Shulman, H. L. Stormer, and P. Kim, Nature 462, 196 (2009). [3] N. M. R. Peres, F. Guinea, and A. H. Castro Neto, Phys. Rev. B 73, 125411 (2006). [4] M. O. Goerbig, R. Moessner, and B. Dou¸cot, Phys. Rev. B 74, 161407 (2006). [5] V. M. Apalkov and T. Chakraborty, Phys. Rev. Lett. 97, 126801 (2006); C. Toke, P. E. Lammert, V. H. Crespi, and J. K. Jain, Phys. Rev. B 74, 235417 (2006). [6] K. Yang, S. Das Sarma, and A. H. MacDonald, Phys. Rev. B 74, 075423 (2006). [7] D. V. Khveshchenko, Phys. Rev. B 75, 153405 (2007). [8] C. Toke and J. K. Jain, Phys. Rev. B 75, 245440 (2007); M. O. Goerbig and N. Regnault, Phys. Rev. B 75, 241405 (2007); for a review on the theory of the FQHE in graphene, see Z. Papi´c, M. O. Goerbig, and N. Regnault, Solid State Comm. 149, 1056 (2009). [9] D. C. Tsui, H. Stormer, and A. C. Gossard, Phys. Rev. Lett. 48, 1559 (1982). [10] R. B. Laughlin, Phys. Rev. Lett. 50, 1395 (1983). [11] Y. Zhang, Z. Jiang, J. P. Small, M. S. Purewal, Y.-W. Tan, M. Fazlollahi, J. D. Chudow, J. A. Jaszczak, H. L. Stormer, and P. Kim, Phys. Rev. Lett, 96, 136806 (2006). [12] M. A. Eriksson, A. Pinczuk, B. S. Dennis, S. H. Simon, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 82, 2163 (1999); Y. Gallais, T. H. Kirschenmann, I. Dujovne, C. F. Hirjibehedin, A. Pinczuk, B. S. Dennis, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 97, 036804 (2006). [13] For a review, see A. Pinczuk in Perspectives in Quantum Hall Effects, edited by S. Das Sarma and A. Pinczuk (John Wiley, New York, 1997). [14] J. Alicea and M. P. A. Fisher, Phys. Rev. B 74, 075422 (2006). [15] R. de Gail, N. Regnault, and M. O. Goerbig, Phys. Rev. B 77, 165310 (2008). [16] F. D. M. Haldane, Phys. Rev. Lett. 51, 605 (1983). [17] J. Gonz´alez, F. Guinea, and M. A. H. Vozmediano, Phys. Rev. B 59, R2474 (1999). [18] K. Shizuya, Phys. Rev. B 75, 245417 (2007); R. Rold´an, J.-N. Fuchs, and M. O. Goerbig, Phys. Rev. B 80, 085408 (2009). [19] For a recent review, see M. O. Goerbig, arXiv:1004.3396, and references therein. [20] X. G. Wu, G. Dev, and J. K. Jain, Phys. Rev. Lett. 71, 153 (1993); J. K. Jain and X. G. Wu, Phys. Rev. B 49, 5085 (1994). [21] F. D. M. Haldane et E. H. Rezayi, Phys. Rev. Lett. 54, 237 (1985); G. Fano, F. Ortolani et E. Colombo, Phys. Rev. B 34, 2670 (1986). [22] S. M. Girvin, A. H. MacDonald et P. M. Platzman, Phys. Rev. B 33, 2481 (1986). [23] Notice that the subspace 2Sz = 10 is of dimension 6, instead of 1 for the subspace with maximal spin polar- ization in the SU(4) calculation with N = 17 electrons. 5
1607.08801	1	1607	2016-07-29T13:30:11	Dispersive readout of valley splittings in cavity-coupled silicon quantum dots	[ "cond-mat.mes-hall" ]	The bandstructure of bulk silicon has a six-fold valley degeneracy. Strain in the Si/SiGe quantum well system partially lifts the valley degeneracy, but the materials factors that set the splitting of the two lowest lying valleys are still under intense investigation. We propose a method for accurately determining the valley splitting in Si/SiGe double quantum dots embedded into a superconducting microwave resonator. We show that low lying valley states in the double quantum dot energy level spectrum lead to readily observable features in the cavity transmission. These features generate a "fingerprint" of the microscopic energy level structure of a semiconductor double quantum dot, providing useful information on valley splittings and intervalley coupling rates.	cond-mat.mes-hall	cond-mat	Dispersive readout of valley splittings in cavity-coupled silicon quantum dots Guido Burkard1 and J. R. Petta2 1Department of Physics, University of Konstanz, D-78457 Konstanz, Germany 2Department of Physics, Princeton University, Princeton, New Jersey 08544, USA The bandstructure of bulk silicon has a six-fold valley degeneracy. Strain in the Si/SiGe quantum well system partially lifts the valley degeneracy, but the materials factors that set the splitting of the two lowest lying valleys are still under intense investigation. We propose a method for accurately determining the valley splitting in Si/SiGe double quantum dots embedded into a superconducting microwave resonator. We show that low lying valley states in the double quantum dot energy level spectrum lead to readily observable features in the cavity transmission. These features generate a "fingerprint" of the microscopic energy level structure of a semiconductor double quantum dot, providing useful information on valley splittings and intervalley coupling rates. PACS numbers: 03.67.Lx, 73.63.Kv, 85.35.Gv I. INTRODUCTION Silicon is a promising material system for spin-based quantum information processing due to weak spin-orbit and hyperfine couplings [1]. Electron spin lifetimes as long as T1 = 3000 seconds were measured as early as 1959 in phosphorous doped silicon [2]. In natural abundance silicon, electron spin coherence times T2 = 60 ms have been reported [3]. Isotopic enrichment has extended the quantum coherence time to T2 = 10 seconds [4]. More- over, the ability to dope silicon with a wide range of donors and acceptors is particularly exciting, as heavy elements such as 209Bi (with nuclear spin quantum num- ber I = 9/2) have a complicated energy level structure that results in so-called "clock transitions" that are first- order-insensitive to magnetic field fluctuations [5]. Cou- pling to the nuclear spin of a single phosphorous donor also allows access to an additional quantum degree of freedom that can be used as a long-lived quantum mem- ory [6, 7]. In terms of its ability to support quantum coherence, the trajectory of the silicon material system is quite im- pressive [8]. On the other hand, silicon presents severe materials challenges in quantum devices, where control at the level of single electrons is desired. Electrons con- fined in Si/SiGe quantum wells have an effective mass m∗ = 0.19 me (roughly three times larger than the GaAs/AlGaAs quantum well system), where me is the free electron mass [9, 10]. As a result, Si quantum de- vices must be significantly smaller than their GaAs coun- terparts to achieve similar orbital excited state energies. Over the past several years, the effective mass challenge has been effectively solved through the development of novel overlapping gate architectures, and the isolation of single electrons in accumulation mode Si/SiGe quantum dots (QD) is becoming routine [11–15]. A major remaining challenge is to understand the fac- tors that limit the valley splitting in silicon [16]. The bulk electronic bandstructure of Si has six equivalent minima (termed valleys) that are located 0.85 of the way from the Brillouin zone edge [9]. In Si/SiGe quantum well systems, the 4% larger lattice constant of Ge strains the Si quantum well, raising in energy the four in-plane ∆4 valleys and lowering in energy the two perpendicular-to- the-plane ∆2 valleys [10]. In view of the interplay be- tween the spin and valley degrees of freedom [17], the ability to probe and, ultimately, control the splitting be- tween the remaining quasi-degenerate valleys in Si/SiGe quantum well systems represents an urgent challenge on the way towards scalable spin qubits in Si/SiGe QDs. Theory suggests that the vertical electric field sets the overall scale of the valley splitting [18]; a prediction that has been experimentally verified in Si MOS (metal-oxide- semiconductor) QDs [19]. However, in the Si/SiGe sys- tem, the valley splitting is known to substantially vary in QD devices fabricated on the same heterostructure. Work by Borselli et al. reports valley splittings in the range of 120 to 270 µeV [20]. In recent work by Zajac et al., valley splittings in the range of 35 – 70 µeV were extracted in the same multiple QD device [14]. Mea- surements by Shi et al. show that the valley splitting can be tuned by using gate voltages to laterally shift the position of the electronic wave function in the two- dimensional electron gas [21]. These experiments sug- gest that the microscopic structure of the QD system (interface roughness, step edges, etc.) plays a strong role in determining the valley splitting [22]. Unfortunately magnetospectroscopy measurements are time consuming to perform and the data can often be ambiguous, espe- cially when the valley splitting is of the order of kBT , where kB is Boltzmann's constant and T is the electron temperature. Therefore the development of new probes of valley splitting will benefit the QD community. In this paper we propose a cavity-based measurement of the low lying energy level structure of few-electron semiconductor double quantum dots (DQD) in the cir- cuit quantum electrodynamics (cQED) architecture. Hy- brid DQD-cQED systems have been used to demonstrate electric dipole couplings g0/2π ranging from 10 to 100 MHz [23–26], quantum control and readout of spin-orbit qubits [25], and spin-photon coupling [27]. In essence, these experiments probe the electric susceptibility χ of a mesoscopic system with a sensitivity well-beyond that of 6 1 0 2 l u J 9 2 ] l l a h - s e m . t a m - d n o c [ 1 v 1 0 8 8 0 . 7 0 6 1 : v i X r a 2 EL(cid:48)(cid:105) − EL(cid:105) is often different than the right dot valley splitting ER. The energy difference between the left dot ground state L(cid:105) and the right dot ground state R(cid:105) is set by the detuning . In general, this system could be used to measure a variety of low lying excited states, such as orbital excited states, valley states, and Zeeman split states [27]. We focus on the Si/SiGe QD system, where valley splittings are typically <200 µeV in energy [14, 20]. The cavity field is sensitive to charge dynamics in the DQD due to the large electric dipole coupling that is achieved in cQED systems [24]. In typical experiments, the cavity is probed by driving it with an input field ain with frequency ωR and detecting the transmitted field aout. Measurements of the cavity response provide useful information about the mesoscopic systems (e.g. a quan- tum dot) embedded in the cavity [23, 29]. Both the amplitude and phase of the transmitted signal provide useful information. As an example, sequential tunnel- ing through a voltage biased DQD was recently shown to result in microwave frequency amplification, such that aout/ain > 1 [30–32]. B. Hamiltonian We model the single electron Si/SiGe DQD using the Hamiltonian, /2 + EL 0 t t(cid:48) H0 = t −t(cid:48) 0 /2 −t(cid:48) −/2 + ER t 0 t(cid:48) t 0 −/2  , (1) which is expressed in the local valley eigenbasis L(cid:48)(cid:105), L(cid:105), R(cid:48)(cid:105), R(cid:105). The valley eigenstates L(cid:105) and L(cid:48)(cid:105) (R(cid:105) and R(cid:48)(cid:105)) of the left (right) QD are split by the valley split- ting EL (ER). In general, the excited states may have a different projection onto the ±z valley basis states (see Appendix A). Therefore, the matrix elements that cou- ple the four levels are distinct. The states L(cid:105) and R(cid:105) (L(cid:48)(cid:105) and R(cid:48)(cid:105)) are hybridized by the (intravalley) inter- dot tunnel coupling t near = 0 ( = ER − EL). The valley state L(cid:48)(cid:105) (R(cid:48)(cid:105)) is coupled to R(cid:105) (L(cid:105)) by the in- tervalley matrix element t(cid:48) leading to avoided crossings near −EL (ER). The DQD energy levels are plotted as a function of the detuning parameter in Fig. 1(b). The left (right) dot energy levels increase (decrease) in energy with in- creasing . In Fig. 1(b) we take EL = 76 µeV, ER = 58 µeV, t = 25 µeV, and t(cid:48) = 13 µeV. For reference, the cavity frequency f0 = 7.8 GHz corresponds to an energy of 32 µeV. C. Electric dipole coupling We assume that the DQD is irradiated with a classi- cal probe field with angular frequency ωR. The probe Figure 1. (a) A cavity-coupled DQD is probed using a small input field ain with frequency ωR. Charge dynamics in the DQD result in changes in the transmitted field aout. The DQD is coupled to source (S) and drain (D) electrodes. (b) DQD energy levels plotted as a function of energy level detuning . In general, the left dot valley splitting EL is different than the right dot valley splitting ER. For this plot, EL = 76 µeV, ER = 58 µeV, t = 25 µeV, and t(cid:48) = 13 µeV. a single electron [28]. The susceptibility is the largest at a DQD interdot charge transition, where a single electron can tunnel from the left dot to the right dot, resulting in an electric dipole moment that is roughly 1000 times larger than in atomic systems [24]. Here we show that the low lying valley structure of a few electron Si/SiGe QD is directly accessible in a hybrid cQED system. The amplitude response of the cavity generates a fingerprint of the DQD energy level structure, providing not only access to the energy level splittings, but also the inter- dot and intervalley coupling rates. We model the system response using realistic parameters that should be acces- sible in future experiments. II. MODEL A. Cavity-coupled double quantum dot Figure 1(a) illustrates the proposed experimental sys- tem. A DQD containing a single excess electron is electric-dipole coupled to a high quality factor super- conducting resonator. We model the DQD as a four- level system consisting of the left dot ground state L(cid:105) = (1, 0)(cid:105), left dot excited state L(cid:48)(cid:105) = (1(cid:48), 0)(cid:105), right dot ground state R(cid:105) = (0, 1)(cid:105), and right dot excited state R(cid:48)(cid:105) = (0, 1(cid:48))(cid:105). The left dot valley splitting EL = (a)(b)EREL-ELERER-ELELERϵSaoutainD��〉�〉��〉�〉-200-1000100200-1000100ϵ(μeV)energy(μeV)3 the choice of a rotating frame is not unique; here, we choose a rotating frame that allows us to describe tran- sitions between levels adjacent in energy. The transition to the rotating frame can be described using the unitary (cid:34) (cid:32) 3(cid:88) (cid:33)(cid:35) UR(t) = exp −it ωRa†a + nωRσnn . (7) n=0 In the rotating frame, we have H = UR ¯HU † R + i URU † R = H0 + HP + HC + HI , (8) with 3(cid:88) n=0 H0 = (En − nωR) σnn, dn+1,nσn+1,n, 2(cid:88) (cid:32) 2(cid:88) n=0 δ HP (cid:39) 1 4 HC = ∆0a†a, HI (cid:39) 2g0 a (9) (10) (11) (cid:33) field generates an oscillating voltage inside the super- conducting resonator. This voltage is directly coupled to the DQD detuning parameter, making it time depen- dent: (t) = 0 + δ cos(ωRt). Here 0 is a static energy level detuning, which can be slowly varied in experiments using dc gate voltages. The parameter δ describes the magnitude of the detuning modulation. The interaction with the probe field thus gives rise to a term in the Hamil- tonian, HP = 1 2 δ cos(ωRt)σz, (2) where, in the same basis used in Eq. (1) above, 1 0 0 0 0 1 0 0 0 0 −1 0 0 0 0 −1  . σz = (3) (cid:0)a + a†(cid:1) σz, The coupling of the DQD to a single quantized mode of a microwave cavity with resonance frequency f0 = ω0/2π can be described as HI = 2g0 (4) where a† and a are the bosonic creation and annihilation operators for the cavity photons and a+a† is proportional to the electric field of the cavity mode. The cavity mode evolves according to the Hamiltoninan HC = ω0a†a in units where = 1. We take g0/2π = 30 MHz in what follows. In order to describe the dissipative dynamics of the DQD-cavity system, including its steady state, it is con- venient to work in the eigenbasis of H0. Writing U0 for the unitary operator that diagonalizes H0, we have ¯H0 = U0 H0U † 0 = Enσnn, (5) where E0 ≤ E1 ≤ E2 ≤ E3 are the ordered eigenvalues of H0, σmn = m(cid:105)(cid:104)n, and n(cid:105) denotes an eigenstate of H0 with eigenvalue En. The dipole operator σz then needs to be transformed into the eigenbasis of H0, 3(cid:88) n=0 3(cid:88) dn+1,nσn+1,n + h.c. , (12) n=0 where ∆0 = ω0 − ωR is the detuning between the cavity resonance frequency and probe frequency. Here, En and dn+1,n = d∗ n,n+1 have to be obtained from the (numeri- cal) diagonalization of H0. The Hamiltoninan described by Eqs. (8)–(12) forms the basis of the following the- oretical analysis of the dispersive readout of the valley splitting in a DQD. III. INPUT-OUTPUT THEORY The response of the DQD system to a microwave probe field can be determined using input-output the- ory [33]. We begin by finding the stationary solution for the equations of motion of the operators a and σn,n+1 in the Heisenberg picture, a = i[H, a] and σn,n+1 = i[H, σn,n+1], including the relevant dissipative terms, a = −i∆0a − κ 2 √ √ a + κ1ain,1 + κ2ain,2 2(cid:88) n=0 D = U0σzU † 0 = dmnσmn, (6) −2ig0 dn,n+1σn,n+1, (13) m,n=0 where the matrix elements dmn = d∗ nm determine the dipole transition matrix elements between energy eigen- states. When transforming the full Hamiltonian H = H0 + HP + HC + HI into the eigenbasis of H0, we have † ¯H = U0 HU 0 where, in ¯HP and ¯HI the operator σz is replaced by D. To remove the time-dependence from our description, we transform the Hamiltonian ¯H into a frame rotating at the frequency ωR and make a rotating wave approxima- tion. Note that in a system with more than two levels, σn,n+1 = −i(En+1 − En − ωR)σn,n+1 − γ 2 γF − 2ig0dn+1,n(pn − pn+1)a, √ + σn,n+1 (14) where κ = κ1 +κ2 +κi is the total cavity decay rate, with κ1,2 the decay rates through the input and output ports, and κi the internal decay rate. ain,1 and ain,2 denote the incoming parts of the external field at the two ends of the cavity, and γ and F are the decay rate and quantum noise within the DQD. For simplicity we have assumed γ to be equal for all transitions. In the following we assume a cavity quality factor Q = f0/κ = 2500 and an electronic dephasing rate of γ = 2.4 GHz. 4 A previous work considered the cavity-coupled dynam- ics with the DQD restricted to the ground state energy level [25]. Thermal population of low lying excited states may be important in the Si/SiGe system due to small valley splittings. To account for finite temperature ef- fects, we have replaced the operator σn,n by the occupa- tion probability pn = (cid:104)σn,n(cid:105) of the nth DQD level. We assume a thermal population of the DQD levels with (cid:80) e−En/kB T n e−En/kB T pn = . (15) The stationary solution is found by setting σn,n+1 = 0 in Eq. (14), neglecting the quantum noise F, and solving for σn,n+1, with the result σn,n+1 = −2g0dn+1,n(pn − pn+1) En+1 − En − ωR − iγ/2 a ≡ χn,n+1a, (16) where we have introduced the electric susceptibility χn,n+1 pertaining to the n → n + 1 transition. Solv- ing for a in the stationary limit ( a = 0) and calculating the outgoing field aout = √ A = aout ain = κ2a, we find: √ −i ∆0 − iκ/2 + 2g0 (cid:80)2 κ1κ2 n=0 dn,n+1χn,n+1 , (17) with the real-valued microwave transmission probability A2 and phase shift ∆φ = − arg(A), which represents the main analytical result of this paper. In general, the cav- ity input port is driven with a weak coherent microwave tone, i.e., ain = α with the coherent-state amplitude α. IV. RESULTS Our goal is to extract information about the valley splittings ER and EL, as well as the valley-dependent tunneling matrix elements t and t(cid:48), from measurements of the cavity transmission. We expect that the elec- tric dipole matrix elements at the avoided crossings in Fig. 1(b) will lead to features in the amplitude and phase of the microwave field transmitted through the DQD. The distances between these four features are determined by ER and EL, thus potentially allowing for the extraction of those two parameters from the analysis of the spec- trum. For brevity, we restrict our discussion to the cavity amplitude response. The phase response provides similar information [25]. We first demonstrate that the cavity transmission is sensitive to low lying valley states by evaluating the mi- crowave transmission probability A2 as a function of and T by numerically diagonalizing H0 for every value of and filling the states according to Eq. (15). A two- dimensional plot of A(, T )2 is shown in Fig. 2(a). Fig- ures 2(b,c) show cuts through this plot at temperatures of T = 1 K and T = 250 mK. At low temperatures, most (a) Microwave transmission coefficient A2 as a Figure 2. function of the DQD detuning and temperature T . The vertical red dotted lines indicate the anticrossings in the DQD spectrum (cf. Fig. 1). The horizontal yellow lines indicate two cuts at temperatures T = 1 K and T = 250 mK for which the transmission coefficient is plotted separately in (b) and (c). The tunneling matrix elements between the QDs are assumed to be t = 25 µeV, while those between opposite valleys are t(cid:48) = 13 µeV. The microwave resonator frequency is f0 = ω0/2π = 7.8 GHz = 32 µeV and the probe field is on resonance with the cavity, ωR = ω0. The valley splittings are chosen as EL = 76 µeV, and ER = 58 µeV. of the population is in the DQD ground-state and corre- spondingly, only the lowest avoided crossing near = 0 is visible. This avoided crossing results in a reduction in the cavity transmission, as has been observed in GaAs and InAs DQDs [24, 25]. As the temperature is increased the population of the higher-lying states increases following Eq. (15) and these states start contributing to the cav- ity response. The L(cid:48)(cid:105)-R(cid:105) avoided crossing appears as a smaller dip around = -80 µeV. Due to the smaller left dot valley splitting, the L(cid:105)-R(cid:48)(cid:105) avoided crossing has a larger contribution to the cavity response, resulting in a deeper dip in the cavity transmission around = 60 µeV. These simulations demonstrate that valley states can be observed in the cavity response. We next show that the cavity response is sensitive to the magnitude of the valley splitting. Figure 3 shows the cavity transmission as a function of detuning and right dot valley splitting. In these simulations the left dot valley splitting is EL = 70 µeV and T = 250 mK. With ER = 0, the cavity transmission is dominated by the L(cid:105)- R(cid:105) ground state anticrossing and the L(cid:48)(cid:105)-R(cid:105) anticross- ing. Here the cavity response is asymmetric with respect to = 0. As the right dot valley splitting increases, a dip in cavity transmission is observed, which is associated with the L(cid:105)-R(cid:48)(cid:105) anticrossing. The competition between valley splitting and thermal excitation becomes apparent as the valley splitting is further increased because the cavity response is only sensitive to states that are occu- pied. As a result, the dip in cavity transmission that is associated with the L(cid:105)-R(cid:48)(cid:105) anticrossing becomes less pronounced with valley splittings beyond ≈ 150 µeV. A second dip in the cavity response emerges for > 0 when ER > 80 µeV. This feature is associated with the higher lying L(cid:48)(cid:105)-R(cid:48)(cid:105)avoided crossing. In semiconductor DQD charge qubit experiments the interdot tunnel coupling can be tuned using electrostatic gate voltages. Tunability of the tunnel coupling is ob- served in charge sensing and photon assisted tunneling measurements [34–36]. In contrast, little is known about the experimental tunability of the intervalley coupling. We now show that the cavity response is sensitive to changes in the intervalley matrix element t(cid:48). The cavity transmission is plotted as a function of and t(cid:48) in Fig. 4. Here the valley splittings are fixed at ER = EL = 60 µeV, t = 25 µeV, and T = 250 mK. For small values of t(cid:48) the L(cid:105)-R(cid:105) ground state anticrossing dominates the cavity response leading to a significant reduction in the cavity transmission near = 0. As t(cid:48) is increased, the dispersive features associated with the L(cid:48)(cid:105)-R(cid:105) and L(cid:105)-R(cid:48)(cid:105) anti- crossings broaden and become more pronounced [see Fig. 4(b)]. The avoided crossings in the energy level diagram begin to merge and are not well defined for t(cid:48) > 20 µeV. As a result, a broad dip is observed in the cavity trans- mission, centered around = 0. These theoretical predic- tions show that measurements of the cavity transmission may lead to useful characterization of the intervalley cou- pling rate. V. CONCLUSIONS cQED-based approaches to quantum information sci- ence have been very productive [37]. They have allowed long-range coupling of qubits, high fidelity readout of cavity coupled quantum devices, and investigations of mesoscopic physics. In this paper, we have demonstrated that the cQED architecture can be used as a sensitive probe of low-lying valley states. For realistic device pa- rameters, the cavity transmission exhibits dips that are 5 Figure 3. (a) Cavity transmission as a function of the inter- dot bias and the right-dot valley splitting ER, with EL = 76 µeV, f0 = ω0/2π = 7.8 GHz = 32 µeV, ωR = ω0, t = 25 µeV, t(cid:48) = 13 µeV, and T = 250 mK. The horizontal yellow lines indicate two cuts at valley splittings: (b) ER = 100 µeV and (c) ER = 10 µeV. The red vertical lines indicate = −EL and = ER; the other avoided crossings are not shown since they are inaccurate due to their vicinity to = 0. associated with energy level anticrossings with low ly- ing valley states. The position of these dips in cavity transmission yields the valley splittings. The tempera- ture dependence of the cavity response also gives infor- mation on the magnitude of the valley splitting. Since the cavity probes the susceptibility of the DQD, it is also sensitive to the curvature of the energy levels, and can be used to extract the intervalley matrix elements. Due to the high energy resolution of narrow-band microwave spectroscopy, coupling Si/SiGe QDs to microwave cavi- ties may allow for efficient measurements of valley split- tings in an approach that is complementary to existing approaches, such as magnetospectroscopy [20, 38] or the quantum Hall effect [39, 40]. The method presented in 6 Appendix A: Valley-dependent model of a double quantum dot Here, we derive our model Eq. (1) for a single elec- tron in a DQD with a single valley-degenerate orbital in each QD. We start from a description of the DQD in a common valley basis for both QDs because it allows us to formulate a model for valley-preserving tunneling through the smooth electrostatic barrier between the two QDs. We then obtain Eq. (1) by changing into the local valley eigenbasis in each QD. The state of the electron on the left (right) QD is denoted l(cid:105) (r(cid:105)), and we introduce the Pauli operators in this left-right orbital Hilbert space as σzl(cid:105) = +l(cid:105) and σzr(cid:105) = −r(cid:105), and σxl(cid:105) = r(cid:105), etc. The valley- independent part of the DQD Hamiltonian can be written as Hd = 2 σz + tcσx, (A1) where represents the DQD energy detuning (bias) en- ergy and tc the inter-dot tunneling matrix element which we can choose to be real. z± z(cid:105) = ±± z(cid:105), τ(cid:48) The two low-energy valley states in the extended two- dimensional electron system in a Si/SiGe quantum well are denoted ± z(cid:105), and we introduce the corresponding x + z(cid:105) = − z(cid:105), valley Pauli operators τ(cid:48) etc. The most general two-level valley Hamiltonian for each of the two individual QDs can then be written as δi·τ (cid:48)/2 where i = l, r and δi is an arbitrary vector whose length determines the bare valley splitting in QD i. Using the sum and difference δ± = (δl ± δr)/2, we can write the valley Hamiltonian of the DQD as H(cid:48) v = 1 2 i(cid:105)(cid:104)i δi · τ (cid:48) = 1 2 (δ+ + σzδ−) · τ (cid:48). (A2) (cid:88) i=l,r Rotating the valley basis such that the common τz valley quantization axis is parallel to δ+ and the τx axis along the projection of δ− into the plane perpendicular to δ+, we find H(cid:48) v = δ 2 τz + 1 2 (δzτz + δxτx) σz. (A3) Combining Hd and H(cid:48) v, we obtain where 11 denotes the 2x2 identity matrix and H(cid:48) 0 = Hd + H(cid:48) v = + HL = 2 HR = − 2 + EL 2 ER 2 EL 2 ER 2 + + , (cid:19) tc11 HR (cid:18)HL tc11 (cid:18)cos θL (cid:19) (cid:18)cos θR sin θR sin θL sin θL − cos θL sin θR − cos θR (A4) (A5) , (A6) , (cid:19) Figure 4. (a) Cavity transmission as a function of and the intervalley matrix element t(cid:48). The horizontal yellow lines indi- cate two cuts at intervalley tunnel couplings: (b) t(cid:48) = 16 µeV and (c) t(cid:48) = 4.7 µeV. Here, we use a fixed t = 25 µeV, EL = ER = 60 µeV, and f0 = ω0/2π = 7.8 GHz = 32 µeV, ωR = ω0, T = 250 mK. Red vertical lines as in Fig. 2. this paper can potentially be applied to probe the energy level structure of different types of quantum dots, e.g. Zeeman energies for spin sublevels in a gradient field [41] or spatially varying g-factors in strong spin-orbit systems [42]. ACKNOWLEDGMENTS Funded by the Army Research Office through grant No. W911NF-15-1-0149 with partial support from the Gordon and Betty Moore Foundation's EPiQS Initiative through Grant GBMF4535, the National Science Foun- dation (DMR-1409556 and DMR-1420541), and the Ger- man Research Foundation (SFB 767). with the valley splittings and angles with ∆L = δ + δz, ∆R = δ − δz, ∆2 EL,R = (cid:113) L,R + δ2 x, δx (A7) (A8) (A9) t = tc cos t(cid:48) = tc sin 7 (A13) (A14) (cid:19) (cid:19) , . (cid:18) θL + θR (cid:18) θL + θR 2 2 . ∆L,R tan θL,R = omitted an irrelevant constant energy shift by(cid:112)δ2 + δ2 (A10) We have shifted the definition of by (EL − ER)/2 and x, in order to center the level crossing of the lower valley eigenstates in the left and right QD at = 0 and zero en- ergy. In order to obtain Eq. (1), we rotate the valley basis about the y-axis by θi in QD i, using the transformation  cos θL − sin θL 2 2 0 0 U =  , sin θL 2 cos θL 2 0 0 0 0 cos θR − sin θR 2 2 0 0 sin θR 2 cos θR 2 and obtain H0 = U H(cid:48) 0U† = /2 + EL 0 t t(cid:48) t −t(cid:48) 0 /2 −t(cid:48) −/2 + ER t 0 Here, since tc was chosen real, both t and t(cid:48) will be real. The Hamiltonian Eq. (A12) is expressed in the local val- ley eigenbasis, (cid:18) (cid:18) (cid:18) (cid:18) L(cid:48)(cid:105) = l(cid:105) L(cid:105) = l(cid:105) R(cid:48)(cid:105) = r(cid:105) R(cid:105) = r(cid:105) (A11)  , t(cid:48) t 0 −/2 (A12) (cid:19) − z(cid:105) cos θL 2 − sin + z(cid:105) + sin θL 2 + z(cid:105) + cos − z(cid:105) θL 2 + z(cid:105) + sin θR 2 − sin θL 2 − z(cid:105) θR 2 + z(cid:105) + cos cos − z(cid:105) θR 2 , (cid:19) (cid:19) , (cid:19) θR 2 (A15) , (A16) (A17) . (A18) [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] G. Feher and E. A. Gere, Phys. Rev. 114, 1245 (1959). [3] A. M. Tyryshkin, S. A. Lyon, A. V. Astashkin, and A. M. Raitsimring, Phys. Rev. B 68, 193207 (2003). [4] A. M. Tyryshkin, S. Tojo, J. J. L. Morton, H. Riemann, N. V. Abrosimov, P. Becker, H. J. Pohl, T. Schenkel, M. L. W. Thewalt, K. M. Itoh, and S. A. Lyon, Nat. Mater. 11, 143 (2012). [5] G. Wolfowicz, A. M. Tyryshkin, R. E. George, H. Rie- mann, N. V. Abrosimov, P. Becker, H. J. Pohl, M. L. W. Thewalt, S. A. Lyon, and J. J. L. Morton, Nat. Nan- otechnol. 8, 561 (2013). [6] J. J. Pla, K. Y. Tan, J. P. Dehollain, W. H. Lim, J. J. L. Morton, F. A. Zwanenburg, D. N. Jamieson, A. S. Dzu- rak, and A. Morello, Nature (London) 496, 334 (2013). [7] K. Saeedi, S. Simmons, J. Z. Salvail, P. Dluhy, H. Rie- mann, N. V. Abrosimov, P. Becker, H. J. Pohl, J. J. L. Morton, and M. L. W. Thewalt, Science 342, 830 (2013). [8] D. D. Awschalom, L. C. Bassett, A. S. Dzurak, E. L. Hu, and J. R. Petta, Science 339, 1174 (2013). [9] T. Ando, A. B. Fowler, and F. Stern, Rev. Mod. Phys. 54, 437 (1982). [10] F. Schaffler, Semi. Sci. Tech. 12, 1515 (1997). [11] S. J. Angus, A. J. Ferguson, A. S. Dzurak, and R. G. Clark, Nano Lett. 7, 2051 (2007). [12] M. G. Borselli, K. Eng, R. S. Ross, T. M. Hazard, K. S. Holabird, B. Huang, A. A. Kiselev, P. W. Deelman, L. D. Warren, I. Milosavljevic, A. E. Schmitz, M. Sokolich, M. F. Gyure, and A. T. Hunter, Nanotech. 26, 375202 (2015). [13] M. Veldhorst, C. H. Yang, J. C. C. Hwang, W. Huang, J. P. Dehollain, J. T. Muhonen, S. Simmons, A. Laucht, F. E. Hudson, K. M. Itoh, A. Morello, and A. S. Dzurak, Nature (London) 526, 410 (2015). [14] D. M. Zajac, T. M. Hazard, X. Mi, K. Wang, and J. R. Petta, Appl. Phys. Lett. 106, 132410 (2015). [15] D. M. Zajac, T. M. Hazard, X. Mi, E. Nielsen, and J. R. Petta, arXiv:1607.07025. [16] M. Friesen and S. N. Coppersmith, Phys. Rev. B 81, 115324 (2010). [17] N. Rohling and G. Burkard, New J. Phys. 14, 083008 (2012). [18] F. J. Ohkawa and Y. Uemura, J. Phys. Soc. Jpn. 43, 917 (1977). [19] C. H. Yang, A. Rossi, R. Ruskov, N. S. Lai, F. A. Mo- hiyaddin, S. Lee, C. Tahan, G. Klimeck, A. Morello, and A. S. Dzurak, Nature Comm. 4, 2069 (2013). [20] M. G. Borselli, R. S. Ross, A. A. Kiselev, E. T. Croke, K. S. Holabird, P. W. Deelman, L. D. Warren, I. Alvarado-Rodriguez, I. Milosavljevic, F. C. Ku, W. S. Wong, A. E. Schmitz, M. Sokolich, M. F. Gyure, and A. T. Hunter, Appl. Phys. Lett. 98, 3569717 (2011). [21] Z. Shi, C. B. Simmons, J. R. Prance, J. K. Gamble, M. Friesen, D. E. Savage, M. G. Lagally, S. N. Copper- smith, and M. A. Eriksson, Appl. Phys. Lett. 99, 233108 (2011). [22] M. Friesen, M. A. Eriksson, and S. N. Coppersmith, 8 Appl. Phys. Lett. 89, 202106 (2006). (1984). [23] M. R. Delbecq, V. Schmitt, F. D. Parmentier, N. Roch, J. J. Viennot, G. Feve, B. Huard, C. Mora, A. Cottet, and T. Kontos, Phys. Rev. Lett. 107, 256804 (2011). [24] T. Frey, P. J. Leek, M. Beck, A. Blais, T. Ihn, K. Ensslin, and A. Wallraff, Phys. Rev. Lett. 108, 046807 (2012). [25] K. D. Petersson, L. W. McFaul, M. D. Schroer, M. Jung, J. M. Taylor, A. A. Houck, and J. R. Petta, Nature (London) 490, 380 (2012). [26] G. W. Deng, D. Wei, J. R. Johansson, M. L. Zhang, S. X. Li, H. O. Li, G. Cao, M. Xiao, T. Tu, G. C. Guo, H. W. Jiang, F. Nori, and G. P. Guo, Phys. Rev. Lett. 115, 126804 (2015). [34] L. DiCarlo, H. J. Lynch, A. C. Johnson, L. I. Childress, K. Crockett, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 92, 226801 (2004). [35] J. R. Petta, A. C. Johnson, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 93, 186802 (2004). [36] C. B. Simmons, M. Thalakulam, B. M. Rosemeyer, B. J. V. Bael, E. K. Sackmann, D. E. Savage, M. G. Lagally, R. Joynt, M. Friesen, S. N. Coppersmith, and M. A. Eriksson, Nano Lett. 9, 3234 (2009). [37] M. H. Devoret and R. J. Schoelkopf, Science 339, 1169 (2013). [38] C. H. Yang, W. H. Lim, N. S. Lai, A. Rossi, A. Morello, [27] J. J. Viennot, M. C. Dartiailh, A. Cottet, and T. Kontos, and A. S. Dzurak, Phys. Rev. B 86, 115319 (2012). Science 349, 408 (2015). [28] J. Stehlik, Y. Y. Liu, C. M. Quintana, C. Eichler, T. R. Hartke, and J. R. Petta, Phys. Rev. Appl. 4, 014018 (2015). [29] O. Dmytruk, M. Trif, C. Mora, and P. Simon, Phys. Rev. B 93, 075425 (2016). [30] Y. Y. Liu, K. D. Petersson, J. Stehlik, J. M. Taylor, and J. R. Petta, Phys. Rev. Lett. 113, 036801 (2014). [31] Y. Y. Liu, J. Stehlik, C. Eichler, M. J. Gullans, J. M. Taylor, and J. R. Petta, Science 347, 285 (2015). [32] A. Stockklauser, V. F. Maisi, J. Basset, K. Cujia, C. Re- ichl, W. Wegscheider, T. Ihn, A. Wallraff, and K. En- sslin, Phys. Rev. Lett. 115, 046802 (2015). [33] M. J. Collett and C. W. Gardiner, Phys. Rev. A 30, 1386 [39] S. Goswami, K. A. Slinker, M. Friesen, L. M. McGuire, J. L. Truitt, C. Tahan, L. J. Klein, J. O. Chu, P. M. Mooney, D. W. van der Weide, R. Joynt, S. N. Copper- smith, and M. A. Eriksson, Nat. Phys. 3, 41 (2007). [40] X. Mi, T. M. Hazard, C. Payette, K. Wang, D. M. Zajac, J. V. Cady, and J. R. Petta, Phys. Rev. B 92, 035304 (2015). [41] M. Pioro-Ladriere, T. Obata, Y. Tokura, Y. S. Shin, T. Kubo, K. Yoshida, T. Taniyama, and S. Tarucha, Nat. Phys. 4, 776 (2008). [42] S. Nadj-Perge, V. S. Pribiag, J. W. G. van den Berg, K. Zuo, S. R. Plissard, E. Bakkers, S. M. Frolov, and L. P. Kouwenhoven, Phys. Rev. Lett. 108, 166801 (2012).
1905.10076	1	1905	2019-05-24T07:54:52	Upconverted electroluminescence via Auger scattering of interlayer excitons in van der Waals heterostructures	[ "cond-mat.mes-hall" ]	The intriguing physics of carrier-carrier interactions, which likewise affect the operation of light emitting devices, stimulate the research on semiconductor structures at high densities of excited carriers, a limit reachable at large pumping rates or in systems with long-lived electron-hole pairs. By electrically injecting carriers into WSe$_2$/MoS$_2$ type-II heterostructures which are indirect in real and k-space, we establish a large population of typical optically silent interlayer excitons. Here, we reveal their emission spectra and show that the emission energy is tunable by an applied electric field. When the population is further increased by suppressing the radiative recombination rate with the introduction of an hBN spacer between WSe$_2$ and MoS$_2$, Auger-type and exciton-exciton annihilation processes become important. These processes are traced by the observation of an up-converted emission demonstrating that excitons gaining energy in non-radiative Auger processes can be recovered and recombine radiatively.	cond-mat.mes-hall	cond-mat	Upconverted electroluminescence via Auger scattering of interlayer excitons in van der Waals heterostructures J. Binder,1, 2 J. Howarth,3, 4 F. Withers,5 M. R. Molas,1, 2 T. Taniguchi,6 K. Watanabe,6 C. Faugeras,1 A. Wysmolek,2 M. Danovich,3, 4 V. I. Fal'ko,3, 4, 7 A. K. Geim,3, 4 K. S. Novoselov,3, 4 M. Potemski,1, 2, ∗ and A. Kozikov3, 4, † 1Laboratoire National des Champs Magnetiques Intenses, CNRS-UGA-UPS-INSA-EMFL, 25 Rue des Martyrs, 38042 Grenoble, France 2Faculty of Physics, University of Warsaw, ul. Pasteura 5, 02-093 Warsaw, Poland 3School of Physics and Astronomy, University of Manchester, Oxford Road, Manchester M13 9PL, UK 4National Graphene Institute, University of Manchester, Oxford Road, Manchester, M13 9PL, UK 5Centre for Graphene Science, College of Engineering, Mathematics and Physical Sciences, University of Exeter, Exeter EX4 4QF, UK 6National Institute for Materials Science, 1-1 Namiki, Tsukuba 305-0044, Japan 7Henry Royce Institute for Advanced Materials, M13 9PL, Manchester, UK arXiv:1905.10076v1 [cond-mat.mes-hall] 24 May 2019 1 Abstract The intriguing physics of carrier-carrier interactions, which likewise affect the operation of light emitting devices, stimulate the research on semiconductor structures at high densities of excited carriers, a limit reachable at large pumping rates or in systems with long-lived electron-hole pairs. By electrically injecting carriers into WSe2/MoS2 type-II heterostructures which are indirect in real and k-space, we establish a large population of typical optically silent interlayer excitons. Here, we reveal their emission spectra and show that the emission energy is tunable by an applied electric field. When the population is further increased by suppressing the radiative recombination rate with the introduction of an hBN spacer between WSe2 and MoS2, Auger-type and exciton-exciton annihilation processes become important. These processes are traced by the observation of an up- converted emission demonstrating that excitons gaining energy in non-radiative Auger processes can be recovered and recombine radiatively. ∗ corresponding author; [email protected] † corresponding author; [email protected] 2 INTRODUCTION Type-II heterostructures made of semiconducting monolayer transition metal dichalco- genides (TMDs) are excellent systems to study indirect interlayer excitons and their prop- erties, in particular at high charge carrier densities. In the regime of large carrier injec- tion many phenomena and processes become important that are negligible at low injection rates1 -- 3. One of those processes is Auger recombination, a non-radiative decay of non- equilibrium carriers, inherent to any semiconducting material. Since for the Auger process to proceed at least three charge carriers are required, the process gains importance with increasing carrier densities. Notably, Auger-type processes become also a factor in limiting the performance of light-emitting diodes (LEDs) at high output power4. The importance of Auger-type exciton-exciton scattering processes was evidenced for direct intralayer excitons in semiconducting TMD monolayers by optical pump probe and time-resolved photolumi- nescence experiments5 -- 10. Here, we study indirect interlayer excitons in electroluminescent devices with a particular focus on the high carrier density regime. In this work we employ constant electric carrier injection, which is the common operating condition for LEDs, to study the Auger recombination and exciton-exciton interactions. For this approach, without strong pumping, one has to render the radiative recombination inefficient to achieve large charge carrier densities. To this end we chose to study van der Waals (vdW) heterostructures with MoS2/WSe2, which are stable semiconducting TMDs that feature a type-II band alignment and a large lattice mismatch (∼ 4%), see Figure 1 (a) and (b). Such heterostructures allow us to study long-lived interlayer excitons (IX) in the energy range of 1− 1.3 eV, important for silicon photonics and telecommunication, with an emission wavelength that can be tuned by an applied electric field. To even further suppress the radiative recombination across the indirect bandgap we produced devices with a monolayer hBN spacer between the TMDs. In such devices the Auger processes are strongly enhanced, dominating the exciton dynamics. Comparison of the behavior of samples with different spacers allows us to establish the qualitative characteristics of exciton-exciton interaction. Besides Auger processes, there is additional interest in the regime of high carrier densities motivated by recent progress in TMD based excitonic devices11 and theoretical predictions that vdW heterostructures could allow observing effects like the condensation of excitons12 or high-temperature super-fluidity13. 3 FIG. 1. Sample structure and selective charge carrier injection. (a) Brillouin zones of WSe2 (red) and MoS2 (blue) illustrating the momentum Q' arising due to lattice mismatch and misorientation angle. (b) Schematic illustration of the type-II band alignment for the MoS2/WSe2 heterostructures with a middle monolayer hBN spacer. The conduction and valence band of WSe2 (MoS2) are represented by red (blue) lines. The hBN layers are represented by grey-shaded rectangles. The black lines depict the quasi Fermi levels in the bottom and top graphene electrodes for an applied voltage above the threshold for hole tunneling into WSe2 and electron tunneling into MoS2. The dashed ellipse indicates the formation of an interlayer exciton consisting of an electron in the conduction band of MoS2 and a hole in the valence band of WSe2. (c) Schematic drawing of the heterostructure shown in (b). (d) Optical microscope image of the active area of device B1. Another ingredient besides the reduction of radiative recombination needed to establish large IX populations is the possibility to selectively inject electrons only into one and holes only into the other material. We met this prerequisite by using tunneling injection through 4 (d) Vb n IX nt i (a) K MoS KWSe2 (c) ' Q' + hBN MoS2 hBN WSe2 hBN (b) graphene electrodes in vertical heterostructures. In contrary to optical injection11,14 -- 19, for which electrons and holes are almost exclusively created in the same and not in different materials, the electrical injection scheme circumvents the competing short-lived direct in- tralayer recombination, since no charge carriers of different type are present in the same material. The injection process is illustrated in Figure 1 (b) and electroluminescence (EL) becomes the method of choice to characterize the devices. Strikingly, besides the observation of the EL of the IX, we observe a large upconversion effect enabling intralayer emission of WSe2 and MoS2 at voltages well below the voltages corresponding to the respective excitonic bandgaps. RESULTS Van der Waals heterostructures for interlayer excitons A schematic drawing of a typical device is presented in Figure 1 (c). Si/SiO2 is used as the substrate for the vdW heterostructure with the following layer sequence: Gr / 3- 5 hBN / 1 WSe2 / 0-1 hBN / 1 MoS2/ 3-5 hBN / Gr. For all devices discussed in the text the outer hBN barriers were three to five layers thick. The middle thin hBN spacer was either one layer thick or completely absent. Based on the presence of this hBN spacer we can divide the devices studied into two groups: with and without a monolayer hBN spacer separating the TMDs. Our study comprised seven devices out of which five were fabricated with and two without a monolayer hBN spacer. The devices with hBN spacer are numbered A1 - A5 and samples without hBN spacer B1, B2. For all these devices the two TMDs were aligned under the optical microscope (estimated accuracy 2◦). An optical microscope image of such a sample is shown in Figure 1 (d). Details about the device fabrication and the basic optical properties can be found in ref. 20 and in the methods section. Figure 1 (b) illustrates the type-II band alignment of a WSe2/MoS2 vdW heterostructure with a monolayer hBN spacer and the tunneling pathways for charge carriers upon application of a voltage between the top and bottom graphene electrodes. The reported values for the band offsets of monolayer WSe2 and MoS2 are in the range of ∼ 0.6−0.7 eV21 -- 23 for the conduction band offset, between ∼ 0.8 − 1.1 eV21,22,24 for the valence band offset and in the range of ∼ 0.9 − 1.3 eV21 -- 23 for the interlayer bandgap. For the case depicted in the sketch, the 5 voltage is large enough to enable electron injection into the MoS2 conduction band and hole injection into the WSe2 valence band (Vb > 1 V), but the voltage is below the threshold for direct WSe2 and MoS2 electron-hole injection (Vb < 1.7 V and 1.9 V, respectively)25. This selective injection of a given carrier type into only one material together with the large band offsets results in an extremely large charge build up at the interface and hence facilitates large IX populations. Tunable electroluminescence of interlayer excitons First, we discuss the experimental results typical for the devices (B1 and B2) without hBN spacer between the TMDs. Figure 2 (a) shows the evolution of the EL as a function of bias voltage in a broad energy range for sample B1. The EL spectra, Figure 2 (b), show three contributions: an emission originating from MoS2 at around 1.9 eV, a broad band at 1.7 eV related to WSe2 and a third peak at around 1.2-1.3 eV, which we attribute to IX emission. We can pinpoint this peak to originate from IX since we observe a strong blueshift as a function of bias voltage. This blueshift is a consequence of the electric field that builds up in our vertical tunneling structure upon applying the bias voltage, which results in an increase of the distance between the conduction band edge of MoS2 and the valence band edge of WSe2 (see Figure 1 (b)). This behavior was universal for all devices studied showing linear shifts with slopes in the range of 90-200 meV/V (see Figure 2 (b)). The different slopes are the result of the different effective thicknesses of the barriers. The shift of the IX is a measure of the electric field and can be used to estimate the IX density (see Supplementary Note 4). To determine the total IX shift one has to estimate the energy of a presumptive IX without electrically injected carriers. To this end, photoluminescence and reflectance contrast measurements (see Supplementary Note 4) were used to extract the threshold voltages for carrier injection (red shaded area in Figure 2 (b)). The extrapolation of the linear fits to this voltage yields an energy of about 1.08 eV for a presumptive IX with- out electrically injected carriers, in good agreement with an interface bandgap of 1.08 eV, recently extracted from transport measurements23. The IX emission energy in vertical vdW heterostructures is easily tunable, which is an interesting feature for many prospective ap- plications. In contrast, no significant shift of the intralayer transitions as a function of bias voltage can be observed, in agreement with the fact that transitions in the same material 6 FIG. 2. Interlayer excitons. (a) False color contour plot of the EL spectra as a function of bias voltage for sample B1 without monolayer hBN spacer. (b) EL spectra for biases in the range of Vb=2.0 - 2.2 V extracted from (a). The spectra are vertically shifted for clarity. The inset in panel (b) shows the peak position of the IX as a function of bias for five different samples and linear fits to the dependencies. The gray dashed line marks the energy of 1.08 eV, which is an estimation for the energy of a presumptive IX without electrically injected carriers. We obtain this value by using threshold voltages of ∼ 0.6 V−0.7 V (red shaded area) for measurable carrier injection extracted from photoluminescence and reflectance contrast measurements (see Supplementary Note 4). are not sensitive to relative band movements caused by the electric field. For both devices without an hBN spacer (results for device B2 are shown in Supplemen- tary Note 3) the different contributions emerge one after another with increasing bias voltage. The IX emission appears first (Vb ∼ 1.5 V), as the difference between the conduction band of MoS2 and the valence band of WSe2 constitutes the lowest barrier for electron-hole in- jection (see Figure 1 (b)). At larger voltages (Vb ∼ 1.75 V) the WSe2 emission emerges, which is a result of tunneling into the intralayer excitonic states. Finally, at (Vb ∼ 2.1 V) MoS2 emission becomes observable in accordance to the larger band-gap. It is interesting to note that the voltages for which EL can be observed correspond to the exciton emission energy rather than to the single particle band gap in agreement to what has been observed for vdW heterostructures with a single WSe2 monolayer25. 7 MoS2 2.2 V 2 V WSe2 1.08 eV A1 A2 A5 B1 B2 1.4 1.2 1.0 IX (eV) 0.0 0.5 2.0 1.0 1.5 Vb(V) IX (b) 15900 EL (a.u.) 2002 252 32 4 EL (ct/s) IX WSe2 MoS2 1.2 1.4 1.6 1.8 1.2 1.4 E(eV) 1.6 E (eV) 1.8 (a) 2.2 2.0 1.8 Vb(V) 1.6 1.4 FIG. 3. Upconverted electroluminescence. (a) False color contour plot of the EL spectra as a function of bias voltage for sample A1 with a monolayer hBN spacer. (b) EL spectra for seven different bias voltages extracted from (a). The spectra are vertically shifted for clarity. For a voltage of Vb=1.32 V emission at energies up to around 1.9 eV are observed, clearly illustrating the large upconversion effect. (c) Comparison of the integrated EL intensity in the spectral range of intralayer emission (1.32 - 2.37 eV) as a function of bias voltage for sample A1 (red circles) and B1 (black squares). A background signal from the response at Vb=0 V was subtracted for each spectrum. The integrated EL intensity at voltages below the onset of observable emission corresponds to the noise level of our setup of around 2ct/s per pixel (integrated over about 4000 pixels). Upconverted emission of intralayer excitons The situation is different for the devices with a monolayer hBN spacer for which one expects a further reduced radiative recombination. The results are exemplary shown for device A1 in Figure 3 (a) (data for other devices is presented in Supplementary Note 2). Strikingly, one observes emission from both MoS2 and WSe2 at bias voltages as low as 1.3 V. The emitted photons at that voltage have an energy of around 1.7 eV (WSe2) and 1.9 eV (MoS2), which constitutes a remarkable upconversion of ∼ 0.6 eV. These measurements reveal electrically driven upconversion for light-emitting devices based on vdW heterostruc- tures. An additional interesting observation is that the onset of emission is virtually the same for both MoS2 and WSe2. Moreover, we note that the emission from WSe2 and MoS2 emerges at lower applied bias voltages than the emission from the IX in contrast to samples without 8 A1 B1 (c) 107 106 105 104 Integrated Intensity x4 x8 x16 x16 x16 x16 WSe2 MoS2 20 ct/s IX 1.66 V 1.56 V 1.46 V 1.36 V 1.32 V 1.30 V 1.26 V (b) 457 EL (ct/s) 70 11 2 0 EL (ct/s) 1.2 1.4 1.6 1.8 2.0 1.2 1.4 E (eV) 1.6 E(eV) 1.8 2.0 1.2 1.4 1.6 1.8 Vb(V) (a) IX WSe2 MoS2 1.8 1.6 Vb(V) 1.4 1.2 an hBN spacer (compare to Figure 2 (a)). The EL spectra presented in Figure 3 (b) also clearly show that for the sample with an hBN spacer the contribution of the intralayer excitonic emission of WSe2 and MoS2 is much stronger than the IX emission. These major observations allow drawing the following important conclusions. First, the process seems not to be dependent on the upconversion energy, since it shows equal onsets and intensities for both WSe2 and MoS2. Second, the upconversion process must be almost equally probable for electrons and holes, since in order to observe these intralayer excitons one has to lift holes into the valence band of MoS2 and electrons into the conduction band of WSe2 (compare Figure 1 (b)). Third, although the IX has a lower emission energy of 1.2-1.3 eV compared to around 1.9 eV (1.7 eV) for the intralayer excitons of MoS2 (WSe2), these intralayer excitons emerge at lower applied bias voltages in the EL spectra. Energetic upconversion was reported in literature for different inorganic semiconducting26 -- 30 as well as organic1 -- 3 materials. In the case of upconversion beyond the energy-scale of phonons commonly Auger-processes were identified to be responsible for the effects. The observation of substantially different onset voltages for measurable intralayer emission in samples A1 and B1 is highlighted in Figure 3 (c) (see also Supplementary Figure 8). Clearly, the intralayer emission appears in our spectra at bias voltages as low as Vb ∼ 1.3 V for sample A1, whereas this emission becomes visible at larger voltages Vb ∼ 1.75 V in sample B1. One must however note that the definition of these onset voltages is somewhat arbitrary, defined by the actual experimental conditions/sensitivity (the background signal). As discussed below and illustrated in Figure 3 (c), one should not expect a complete disappearance of Auger processes in sample B1, but their efficiency being orders of magnitude smaller in sample B1 as compared to sample A1. Auger processes and upconversion mechanism Thanks to the purely electrical carrier injection in our devices we can rule out nonlinear effects involving photon absorption. It is therefore intuitive to investigate whether Auger processes can account for the above described observations. Such an Auger process would proceed in the following manner: (i) An electron in MoS2 and a hole in WSe2 form an IX that recombines non-radiatively. (ii) The excess energy is transferred to another electron in the MoS2 conduction band. (iii) Since the transferred energy is larger than the band offset the electron can tunnel into the WSe2 conduction band. (iv) Due to the large number of 9 holes present in the valence band of WSe2, the electron can form a WSe2 intralayer exciton and recombine radiatively, which gives rise to the characteristic light emission for WSe2 monolayers. A similar process would be possible for holes in the valence band of WSe2. With such a process one can account for the energy independence of the process, which is true as long as the energy of the indirect transition is larger than the band offsets. However, the second aspect identified above, i.e. same probability for the upconversion of electrons and holes, is hard to imagine in such a case. For such an Auger process to be efficient there should be a state to which the charge carrier can be excited. Yet in the case of WSe2 there are no bands with an energy distance of the order of the IX energy in the band structure around the K point to which a hole could be excited31. Even if one includes larger momenta it is hard to imagine that such a process would be equally probable as compared to electrons for which such bands should be present. In this picture it is also difficult to explain why upconversion sets in at lower bias voltages than the IX emission. In order to overcome this conceptual discrepancies, we propose an excitonic Auger process instead of the above described single particle considerations in agreement with the exciton-exciton interactions observed in optical pump-probe experiments5,7 -- 10. Figure 4 presents the proposed mechanism responsible for the upconverted EL in the two-particle picture. The grey (red, blue) solid parabola represents the exciton dispersion of the IX (WSe2, MoS2). The dashed lines represent exited states and the shaded filled area stands for the excitonic continuum. Due to the lattice mismatch, the parabola describing the IX is shifted towards higher momenta. If there is an additional misalignment angle this will further increase the distance Q' (see Figure 1 (a)). As a result of the effective selective injection of carriers the population of IX increases. The distribution function of the population is schematically depicted by the shading of the grey circles. The orange circles stand for excitons that possess almost zero center of mass velocity and hence can recombine radiatively (light cones). In this picture, the IX have to compensate for the additional momentum Q' first in or- der to recombine radiatively. This can be achieved in two ways: (I) by increasing the IX population at larger voltages. Since with increasing bias electrons and holes that form the IX possess larger and larger momenta, a significant amount of the IX population will as well extend more and more towards larger momenta. The momentum range of efficient ra- diative recombination is determined by the photon dispersion, as indicated in Figure 4. If 10 FIG. 4. Mechanism of upconverted emission in the two-particle picture. The solid lines represent the excitonic ground state dispersion n1 of the IX (grey), MoS2 (blue) and WSe2 (red). The circles stand for excitons. Q0 is the momentum mismatch as defined in Figure 1 (a). The dashed lines indicate excited states n and the shaded area marks the excitonic continuum n∞. The grey-scale shading of circles schematically pictures the momentum distribution of excitons. A bright shading indicates less excitons for a given momentum than a dark shading. The photon dispersion is overlaid (grey lines) to mark the region of effective radiative recombination (orange circles). For the situation depicted, the bias voltage is below the threshold for direct intralayer charge injection. Mechanism (I) illustrates radiative IX emission facilitated by an increasing number of IX with large momenta. Mechanism (II) depicts excitonic Auger processes. The grey ellipse schematically highlights the interaction between two exemplary excitons. As a result of the interaction one exciton recombines non-radiatively and transfers the energy to the other exciton (arrows with dotted lines). Relaxation: (i) describes relaxation back to the IX ground state (exciton-exciton annihilation), (ii) and (iii) relaxation to MoS2 and WSe2 respectively, which leads to upconverted intralayer emission. 11 Q' Q E photon (ii) (i) (iii) (II) MoS2 WSe2 (I) IX-IX interaction nn n1 IX their momentum roughly equals Q', the IX can recombine radiatively. Otherwise, the IX remains optically silent and gives rise to an increasing population that enables mechanism (II) through IX-IX collisions. As a result, one IX recombines non-radiatively and transfers the energy and momentum to a second IX (dotted arrows in Figure 4). This IX is excited into the continuum of excitonic states and has now three different decay channels: (i) the IX relaxes back into the IX ground state. This scenario is also called exciton-exciton anni- hilation. (ii) corresponds to the case where the now almost delocalized exciton relaxes into a state that corresponds to the MoS2 intralayer exciton. Please note that the relaxation process shown here is indirect in real space, i.e. that in the single particle picture this would correspond to a tunneling of the electron into the MoS2 conduction band. Similarly possible is (iii) for which the excited exciton relaxes to a WSe2 exciton state. For both (ii) and (iii), the relaxation continues via intralayer excitonic states until the exciton reaches the bottom of the dispersion relation, where it can recombine radiatively. Within this picture we can imagine (ii) and (iii) to be equally probable, since there is a plethora of excited excitonic states for both materials, in contrast to the lack of higher energetic hole bands around the K point in the single particle picture. One can also explain the lower voltages for the onset of upconversion compared to the onset for the IX emission. In order to recombine radiatively the IX have to compensate for the momentum Q' (I), but below this threshold there is already a significant population of optically silent IX present that can give rise to the upconversion (II). This difference in onsets should strongly depend on the misorientation angle that adds to the momentum Q' due to lattice mismatch. Based on literature values for the effective masses and lattice parameters32 and by using an estimation of the momentum mismatch of 0.04 · ΓK for zero misalignment and 0.1 · ΓK for 5◦ misalignment33 we can estimate the additional kinetic energy needed to recombine radiatively to be ∼ 10 meV and ∼ 60 meV, respectively. These energies indicate that such an emission could be temperature activated, but depends sensitively on the misalignment angle. It has been shown theoretically that for such structures one should expect six light cones irrespective of the layer misalignment angle, which are situated at a non-zero center of mass velocity33. Consequently, the IX have to posses a large kinetic momentum or have to be scattered by defects or phonons into the light cones to recombine radiatively. Otherwise these IX remain dark as shown in Figure 4. This fact renders the observation of PL of the IX very ineffective18,19,23, but on the other hand offers an ideal platform to study many- 12 body interactions allowing more easily reaching the large IX population regime. In the more widely studied system of MoSe2/WSe2 the lattice mismatch is small (< 0.1%) which makes the PL of IX observable allowing to study the properties of the IX more directly14 -- 17,34, but consequently not allowing for large IX populations. For our samples measured at low temperatures no PL of the IX could be observed in agreement with other recent reports18,19,23 even with additional applied voltages. Please note that we only consider K − K interlayer excitons where the electron and hole stem from the K points of the respective TMD layers. Recently, Γ − K interlayer excitons were observed and identified in the photoluminescence of MoS2/WSe2 vdW heterostructures19. These Γ − K interlayer excitons appear at larger energies of about 1.6 eV, which agrees with other reports of interlayer excitons in this material system11,35,36. In this work we were able to reveal the emission spectra of K − K interlayer excitons which are not observable in PL. This observation is enabled thanks to our selective electrical injection mechanism, leading to large IX populations. DISCUSSION Out of the seven devices studied, three showed upconverted EL emission. An additional device showed similar behavior -- e.g. intralayer emission before IX -- but at larger voltages suggesting that not the whole applied voltage dropped across the active region of the vdW heterostructure (see more detailed information on every device in Supplementary Note 1). All of these devices feature a monolayer hBN spacer. The two devices without hBN spacer did not show upconversion, as well as one sample with hBN spacer for which, however, charge injection was low and asymmetric. In our picture, we can also account for the behavior of the devices that do not show upconversion. Here, the consecutive appearance of different contributions with increasing bias voltage (compare Figure 2 (a)) is reflected by the increasing energetic positions of bright excitons around zero momentum (see Figure 4) into which the charge carriers can tunnel directly at larger voltages. Our results suggest that the probability for Auger processes and upconversion versus radiative IX recombination can be effectively tuned by changing the thickness of the middle hBN barrier. The presence of an hBN spacer between the TMDs strongly influences the radiative recombination rate for IX emission in the vicinity of the light cone. By introducing an hBN spacer we increase the distance between the electron and the hole and consequently 13 decrease the wave-function overlap. We can therefore expect the radiative recombination rate to be quenched in the case of devices from group A. Indeed, we observe that the upconverted intralayer emission is more intense than the IX emission (see Figure 3). Devices with an hBN spacer hence favor larger IX densities, since the radiative IX recombination at almost zero momentum becomes less effective. To experimentally confirm this conjecture we estimated the IX density with two different methods: (i) by analyzing the apparent evolution of the exciton versus trion resonances in the MoS2 layer in photoluminescence (PL) and reflectance contrast (RC) measured as a function of bias voltage and (ii) by applying a simple capacitor model to account for the observed blueshift of the IX peak in the EL measured as a function of applied bias voltage, thus deducing the built-in electron-hole charge. Both methods (see Supplementary Note 4 for details on measurements and analysis) yield IX densities in the range of 1012 cm−2 for sample A1, but distinctly smaller densities, in the range of 1011 cm−2, for sample B1. With these findings we indeed confirm our expectations that a larger accumulation of interlayer excitons appears in samples with an hBN spacer, favoring the observation of Auger processes in such structures. However, the presence or absence (strong or weak efficiency) of the Auger processes should also be discussed with respect to an apparent ratio d/aB of the inter exciton distance d to the exciton Bohr radius aB. With n = 3.7 · 1012 cm−2 for sample A1 and n = 6.1 · 1011 cm−2 for sample B1 (see Supplementary Note 4) we obtain dA1 = 2/√πn ∼ 5.9 nm and dB1 ∼ 14.4 nm, respectively for samples A1 and B1. To estimate aB we note that the Bohr radius of an interlayer exciton is expected to be about a factor of two larger than that of the intralayer exciton37. The exciton Bohr radius for intralayer excitons of semiconducting monolayer TMDs is known to be sensitive to the dielectric surrounding and varies in the range of 1 - 1.7 nm38,39. Thus, given their hBN encapsulation, we roughly estimate aB ∼ 3 nm for the interlayer exciton Bohr radius in our structures and conclude that dA1/aB ∼ 2 for sample A1 and a significantly larger value dB1/aB ∼ 5 for sample B1. The conditions met by sample A1 are close to the Mott transition for interlayer excitons, which enables the observation of many body effects in this case38. In addition, one may speculate that d/aB can be further reduced in sample A1, as the introduction of an hBN spacer could lead to the reduction of the binding energy of the intralayer exciton in sample A1, as compared to sample B1. However, this reduction was calculated to be rather small37, thus charge accumulation seems 14 to be a decisive factor for the observation of the Auger-type processes in our samples. An expected super-linear (ideally quadratic) behavior of the intensity of the Auger up- converted emission as a function of the IX density is another issue to be considered. It is tempting to trace this behavior by investigating the dependence of the emission intensity versus driving current. However, this dependence may also be super-linear due to trivial effects not related to the Auger process, like leakage currents due to device imperfections or exciton trapping centers, which are also observed in III-V quantum well structures40. For most of our devices we obtain an overestimation of the super-linear behavior of emission intensity as a function of driving current, due to current contributions that are not related to the emission process. Instead, for device A1, showing excellent I-V characteristics, the extracted super-linear trend in the regime of the Auger upconverted emission appears re- liable, since in the case of this device the trend becomes practically linear in the direct intralayer injection regime, when the applied voltages exceed the excitonic bandgaps of the TMD monolayers (see Supplementary Figure 9). It is clear that more work on the role and nature of defects and a better control of the interfaces in vdW heterostructures is needed to tackle this issue for all devices. In conclusion, we studied Auger processes in especially chosen type-II vdW heterostruc- tures that enable large populations of optically silent IX. The double indirect nature of these excitons allows recovering a part of the excited Auger carriers through relaxation in opti- cally active states. These states lie higher in energy than the initial IX, which results in the emission of upconverted photons. The large IX populations established thanks to the purely electrical tunnel injection allowed us to observe EL of otherwise silent IX, which shifts due to the electric field. It is shown that the IX energy can be tuned by up to 200 meV/V. A purely excitonic Auger process is proposed that, in contrary to Auger processes based on the single-particle picture, accounts for the major characteristics of the EL as a function of bias voltage. The results suggest that the efficiency of the upconversion mechanism depends on the hBN layer between the active TMDs. The revealed variable nature of the IX-IX interactions is of key importance for future TMD based optoelectronic device engineering and is important for any attempt aiming towards exciton condensation or super-fluidity in such structures. 15 METHODS Device Fabrication All devices were fabricated on doped silicon substrates covered with silicon oxide 90 nm thick. Top graphene flakes were exfoliated on a silicon substrate spin coated with PMGI (polymethylglutarimide) and PMMA (polymethyl methacrylate). After dissolving PMGI, PMMA membranes with graphene were used to pick up other flakes. MoS2 and WSe2 were exfoliated on a silicon oxide substrate (290 nm thick) spin coated with PPC (polypropylene carbonate). Thin boron nitride and bottom graphene were exfoliated on silicon oxide sub- strates 70 and 90 nm thick, respectively. In some devices, we used substrate hBN about 30 nm thick exfoliated on silica substrates 90 nm thick. Light-emitting diodes were assem- bled by picking TMDs and hBN flakes one after another with the top graphene flake and transferring the resulting stack on top of bottom graphene. The PMMA membrane was then dissolved in acetone. In some devices, the flakes were picked up by top graphene and peeled onto the substrate hBN. After the assembly, electron-beam lithography was used to define a mask for contacts followed by evaporation of Chromium/Gold (Cr/Au, 3/50nm). Optoelectronic measurements The optoelectronic measurements were performed using two different helium flow cryostats (Janis ST-500, Oxford Instruments MicrostatHires) and a helium bath cryostat. All the measurements shown in the main text were taken at liquid helium temperature. The signal was collected using a 0.5 m long spectrometer equipped with liquid nitrogen cooled charge- couple-device (CCD) camera. Electrical measurements were performed using a Keithley 2450 source-measure unit synchronized with the spectrometer. A diode type behavior was observed for all devices. 16 Data availability The data supporting the findings of this work are available from the corresponding author upon reasonable request. 1 Pandey, A. K. & Nunzi, J.-M. Upconversion injection in rubrene/perylene-diimide- heterostructure electroluminescent diodes. Appl. Phys. Lett. 90, 263508 (2007). 2 Qian, L. et al. Electroluminescence from light-emitting polymer/ZnO nanoparticle heterojunc- tions at sub-bandgap voltages. Nano Today 5, 384 -- 389 (2010). 3 He, S.-J., Wang, D.-K., Jiang, N., Tse, J. S. & Lu, Z.-H. Tunable excitonic processes at organic heterojunctions. Adv. Mater. 28, 649 -- 654 (2016). 4 Iveland, J., Martinelli, L., Peretti, J., Speck, J. S. & Weisbuch, C. Direct measurement of Auger electrons emitted from a semiconductor light-emitting diode under electrical injection: identification of the dominant mechanism for efficiency droop. Phys. Rev. Lett. 110, 177406 (2013). 5 Sun, D. et al. Observation of rapid exciton -- exciton annihilation in monolayer molybdenum disulfide. Nano Lett. 14, 5625 -- 5629 (2014). 6 Yuan, L. & Huang, L. Exciton dynamics and annihilation in WS2 2D semiconductors. Nanoscale 7, 7402 -- 7408 (2015). 7 Pollmann, C. et al. Resonant internal quantum transitions and femtosecond radiative decay of excitons in monolayer WSe2. Nat. Mat. 14, 889 (2015). 8 Yu, Y. et al. Fundamental limits of exciton-exciton annihilation for light emission in transition metal dichalcogenide monolayers. Phys. Rev. B 93, 201111 (2016). 9 Aivazian, G. et al. Many-body effects in nonlinear optical responses of 2D layered semiconduc- tors. 2D Mater. 4, 025024 (2017). 10 Kulig, M. et al. Exciton diffusion and halo effects in monolayer semiconductors. Phys. Rev. Lett. 120, 207401 (2018). 11 Unuchek, D. et al. Room-temperature electrical control of exciton flux in a van der Waals heterostructure. Nature 560, 340 -- 344 (2018). 17 12 Wu, F.-C., Xue, F. & MacDonald, A. Theory of two-dimensional spatially indirect equilibrium exciton condensates. Phys. Rev. B 92, 165121 (2015). 13 Fogler, M., Butov, L. & Novoselov, K. High-temperature superfluidity with indirect excitons in van der Waals heterostructures. Nat. Commun. 5, 4555 (2014). 14 Rivera, P. et al. Observation of long-lived interlayer excitons in monolayer MoSe2 -- WSe2 heterostructures. Nat. Commun. 6, 6242 (2015). 15 Rivera, P. et al. Valley-polarized exciton dynamics in a 2D semiconductor heterostructure. Science 351, 688 -- 691 (2016). 16 Nagler, P. et al. Interlayer exciton dynamics in a dichalcogenide monolayer heterostructure. 2D Mater. 4, 025112 (2017). 17 Miller, B. et al. Long-lived direct and indirect interlayer excitons in van der Waals heterostruc- tures. Nano Lett. 17, 5229 -- 5237 (2017). 18 Zhu, H. et al. Interfacial charge transfer circumventing momentum mismatch at two-dimensional van der Waals heterojunctions. Nano Lett. 17, 3591 -- 3598 (2017). 19 Kunstmann, J. et al. Momentum-space indirect interlayer excitons in transition metal dichalco- genide van der Waals heterostructures. Nat. Phys. 14, 801 -- 805 (2018). 20 Withers, F. et al. Light-emitting diodes by band-structure engineering in van der Waals het- erostructures. Nat. Mat. 14, 301 -- 306 (2015). 21 Kang, J., Tongay, S., Zhou, J., Li, J. & Wu, J. Band offsets and heterostructures of two- dimensional semiconductors. Appl. Phys. Lett. 102, 012111 (2013). 22 Gong, C. et al. Band alignment of two-dimensional transition metal dichalcogenides: Applica- tion in tunnel field effect transistors. Appl. Phys. Lett. 103, 053513 (2013). 23 Ponomarev, E., Ubrig, N., Guti´errez-Lezama, I., Berger, H. & Morpurgo, A. F. Semiconducting van der Waals interfaces as artificial semiconductors. Nano Lett. 18, 5146 -- 5152 (2018). 24 Chiu, M.-H. et al. Determination of band alignment in the single-layer MoS2/WSe2 heterojunc- tion. Nat. Commun. 6, 7666 (2015). 25 Binder, J. et al. Sub-bandgap voltage electroluminescence and magneto-oscillations in a WSe2 light-emitting van der Waals heterostructure. Nano Lett. 17, 1425 -- 1430 (2017). 26 Jones, A. M. et al. Excitonic luminescence upconversion in a two-dimensional semiconductor. Nat. Phys. 12, 323 -- 327 (2015). 18 27 Manca, M. et al. Enabling valley selective exciton scattering in monolayer WSe2 through upconversion. Nat. Commun. 8, 14927 (2017). 28 Seidel, W., Titkov, A., Andr´e, J., Voisin, P. & Voos, M. High-efficiency energy up-conversion by an "Auger fountain" at an InP-AlInAs type-II heterojunction. Phys. Rev. Lett. 73, 2356 (1994). 29 Baumgartner, A., Chaggar, A., Patan`e, A., Eaves, L. & Henini, M. Upconversion electrolumi- nescence in InAs quantum dot light-emitting diodes. Appl. Phys. Lett. 92, 091121 (2008). 30 Potemski, M. et al. Auger recombination within Landau levels in a two-dimensional electron gas. Phys. Rev. Lett. 66, 2239 (1991). 31 Danovich, M., Z´olyomi, V., Falko, V. I. & Aleiner, I. L. Auger recombination of dark excitons in WS2 and WSe2 monolayers. 2D Mater. 3, 035011 (2016). 32 Korm´anyos, A. et al. K· p theory for two-dimensional transition metal dichalcogenide semicon- ductors. 2D Mater. 2, 022001 (2015). 33 Yu, H., Wang, Y., Tong, Q., Xu, X. & Yao, W. Anomalous light cones and valley optical selection rules of interlayer excitons in twisted heterobilayers. Phys. Rev. Lett. 115, 187002 (2015). 34 Ross, J. S. et al. Interlayer exciton optoelectronics in a 2D heterostructure pn junction. Nano Lett. 17, 638 -- 643 (2017). 35 Fang, H. et al. Strong interlayer coupling in van der Waals heterostructures built from single- layer chalcogenides. Proc. Natl. Acad. Sci. U.S.A. 111, 6198 -- 6202 (2014). 36 Chiu, M.-H. et al. Spectroscopic signatures for interlayer coupling in MoS2-WSe2 van der Waals stacking. ACS Nano 8, 9649 -- 9656 (2014). 37 Latini, S., Winther, K. T., Olsen, T. & Thygesen, K. S. Interlayer excitons and band alignment in MoS2/hBN/WSe2 van der Waals heterostructures. Nano Lett. 17, 938 -- 945 (2017). 38 Wang, G. et al. Colloquium: Excitons in atomically thin transition metal dichalcogenides. Rev. Mod. Phys. 90, 021001 (2018). 39 Stier, A. V. et al. Magnetooptics of exciton rydberg states in a monolayer semiconductor. Phys. Rev. Lett. 120, 057405 (2018). 40 Combescot, M., Combescot, R. & Dubin, F. Bose -- einstein condensation and indirect excitons: a review. Rep. Prog. Phys. 80, 066501 (2017). 19 ACKNOWLEDGMENTS This work was supported by the EC-FET European Graphene Flagship (no.785219), the ATOMOPTO project (TEAM programme of the Foundation for Polish Science co-financed by the EU within the ERDFund), the European Research Council (Synergy Grant Het- ero2D), the Royal Society (UK), the Royal Academy of Engineering (UK), the Engineer- ing and Physical Sciences Research Council (UK) grants EP/N010345/1, EP/S019367/1, EP/P026850/1, the US Army Research Office (W911NF-16-1-0279), the Elemental Strategy Initiative conducted by the MEXT, Japan and the CREST (JPMJCR15F3), JST. AUTHOR CONTRIBUTIONS J.B. carried out the optoelectronic measurements, analyzed and interpreted experimental data, participated in discussions, developed the model and wrote the draft of the manuscript. A.K. produced experimental devices, participated in the measurements, analyzed and in- terpreted experimental data, contributed to writing the manuscript, J.H. produced experi- mental devices and participated in the measurements, F.W. produced experimental devices, M. R. M. participated in the measurements, K.W. and T.T. grew high-quality hBN, C.F., A.W., M.D., V.I.F. and A.G. participated in discussions, K.S.N. and M.P. initiated the project, analyzed experimental data, participated in discussions, contributed to writing the manuscript. COMPETING INTERESTS The authors declare no competing interests. 20 Supplementary Information Upconverted electroluminescence via Auger scattering of interlayer excitons in van der Waals heterostructures Johannes Binder et al. arXiv:1905.10076v1 [cond-mat.mes-hall] 24 May 2019 1 Supplementary Note 1: Studied devices, their I-V characteristics and comments on their performance Supplementary Figure 1. Current-voltage characteristics and additional information for all seven devices studied. 2 IV curve hBN spacer Upconversion comments A1 A2 A3 A4 A5 B1 B2 YES YES low leakage currents; steep onset in IV curve YES YES large leakage and total currents; no pronounced PL and EL signals from the MoS2 monolayer YES YES (at larger voltages) YES YES YES NO parasitic voltage drop; larger voltages needed to obtain an upconversion-like behaviour: WSe2, MoS2 intralayer excitons emerge at lower voltages low leakage currents; steep onset in IV curve but at larger voltages; than IX low leakage currents; steep onset in IV curve only room temperature data; contacts ceased to work during thermal cycling very low currents; PL shows asymmetric charge injection; charge injection not effective at positive bias voltages (in the range where upconverted EL is observed for other devices) NO NO results are similar to sample B2; good reproducibility of EL(Vb) dependence NO NO results are similar to sample B1; good reproducibility of EL(Vb) dependence Altogether seven devices were studied (see Supplementary Figure 1). Five samples (A1, A2, .. A5) comprise an hBN spacer between the TMD monolayers, whereas the two other samples (B1 and B2) were fabricated without an hBN spacer. Each device was electrically tested and current-voltage (I-V) characteristics (see Supplementary Figure 1) were recorded. The I-V curves for all seven samples display tunneling diode operation, however the actual quality of the device performance, mostly determined by leakage currents, varies from sample to sample. Close to optimal is the device A1, with low leakage currents. Device A4 has only been tested at room temperature due to contact failure while cooling down the sample. The I-V characteristic of device A5 is quite distinct from the characteristics of the other samples. It is asymmetric and displays surprisingly low currents. Device A5 is the only sample with an hBN spacer that did not show upconverted emission. Both samples without an hBN spacer (B1 and B2) did also not show any upconverted emission. These two samples display very similar results when studied with electroluminescence (EL), but also photoluminescence (PL) and reflectance contrast (RC) experiments. 3 Supplementary Note 2: Upconverted emission in other devices In this section we present the EL as a function of bias voltage for other samples that show upconverted emission. As shown in Supplementary Figure 2, sample A2 is another device which clearly shows the sub-bandgap EL. We note, however, that in this sample the emission, both in EL and PL, is only due to the WSe2 monolayer whereas no signal associated with MoS2 monolayer could be observed. Supplementary Figure 2. Sample A2 with a monolayer hBN spacer showing the upconverted EL. For this device no intense signal from MoS2 has been observed, both in EL and PL. The monolayer- like emission, including the upconverted emission, is mainly associated to the WSe2 monolayer. In agreement with the results shown in the main text, the emission of the IX is emerging at larger voltages ∼ 1.6 V as compared to the WSe2 signal at ∼ 1.3 V. To clearly demonstrate the onset of the "sub-bandgap" EL, we show a series of EL spectra measured in the vicinity of the threshold voltage, both for sample A1 (from the main text) and sample A2. For device A1 we start to see EL corresponding to the A-exciton resonance of the WSe2 and MoS2 monolayers at voltages as low as Vb = 1.30 V. For device A2, even at voltages of Vb = 1.28 V electroluminescence of the WSe2 monolayer can be observed. 4 1.7 E(eV) 1.8 1.4 1.2 1.8 1.6 (V) b V 1.3 E (eV) 1.8 Vb(V) 1.6 Supplementary Figure 3. Comparison of the EL spectra close to the onset of the upconverted emission for (a) sample A1 and (b) sample A2. For sample A1 (A2) EL can be observed at voltages as low as Vb = 1.30 V (Vb = 1.28 V). The spectra are shifted vertically for clarity. 5 1.36 V 1.34 V 1.32 V 1.30 V 1.28 V 1.26 V (b) 0.5 ct/s EL(ct/s) 1.36 V 1.34 V 1.32 V 1.30 V 1.28 V 1.26 V 0.5 ct/s (a) EL (ct/s) 1.2 1.4 1.6 1.8 E (eV) 2.0 2.2 1.65 1.70 E (eV) 1.75 The EL of device A3, shown in Supplementary Figure 4, shows similar characteristic features like device A1, discussed in the main text. The EL of the WSe2 and MoS2 contri- butions emerge at lower bias voltages than the IX emission and the onset of the emission due to the A-exciton resonance of WSe2 and MoS2 is virtually the same. The only major difference is that larger voltages are needed to observe emission. We believe that the larger voltages can be explained by an additional voltage drop on our device, so that the overall applied voltage does not correspond to the voltage dropping across the active area. This conclusion agrees well with the currents at larger voltages (Vb > 3 V) observed in the IV characteristics, which are similar to the currents for other devices biased at lower voltages (Vb < 2 V), see Supplementary Figure 1. Given the fact that besides the parasitic voltage drop all other characteristics are similar, we refer to this behavior as "upconversion-like". Please note that for the devices that did not show upconverted emission, (i.e., the device B1 from the main text and sample B2 discussed in the next section) the EL of the monolayer TMDs emerges according to the magnitude of the bandgap, but the IX appears first at lower bias voltages. Supplementary Figure 4. Sample A3 with a monolayer hBN spacer. Similar to the other samples for which upconversion is observed the signals from WSe2 and MoS2 emerge at lower voltages than the emission corresponding to the IX. Most probably the voltage does not solely drop across the active area. This explains the larger voltages needed to observe the effects. 6 1.6 1.7 1.8 1.9 E (eV) 3.8 3.6 3.4 3.2 3.0 2.8 2.6 2.4 Vb(V) 1.3 1.4 E (eV) 3.8 3.6 3.4 Vb(V) Supplementary Figure 5. Sample A4 with a monolayer hBN spacer showing upconversion. For this sample only measurements at room temperature were possible. At room temperature a broad upconverted peak corresponding to WSe2 can be clearly seen. EL at energies around 1.64 eV is observed at bias voltages as low as 1.1 V. Supplementary Figure 5 presents results for device A4. Unfortunately, this device could only be measured at room temperature, since the electric contacts ceased to operate during thermal cycling. The initial measurements at room temperature, however, already show a broad upconverted emission corresponding to the A-exciton resonance of WSe2 at around 1.64 eV for voltages as low as 1.1 V. The I-V characteristics of this device show a steep onset and low leakage currents as shown in Supplementary Figure 1. 7 1.7 V 1.6 V 1.5 V 1.4 V 1.3 V 1.2 V 1.1 V 1000 100 10 1 0.1 0.01 0.001 EL (ct/s) 1.60 1.64 1.68 E(eV) Supplementary Note 3: Electroluminescence of devices not showing upconverted emission Device B2, with a similar design to sample B1 discussed in the main text (no hBN spacer), shows practically the same response as B1 in our experiments. The "regular" not upconverted behavior of the EL signal in sample B2 is illustrated in Supplementary Figure 6. The IX electroluminescence of sample B2 appears at bias voltages around 1.65 V whereas the monolayer emission due to WSe2 and MoS2 emerges progressively at higher voltages, 1.85 V and 2.15 V, correspondingly. Supplementary Figure 6. Sample B2 without hBN spacer. Similar to the sample in the main text, no upconverted emission was observed. The evolution of the contributions of the IX, WSe2 and MoS2 is similar to that for the device B1 discussed in the main text. The two graphs show two consecutive measurements sweeps. 8 1.2 1.4 1.6 1.8 2.0 E (eV) 2.3 2.2 2.1 2.1 2.0 1.9 1.8 1.7 1.6 Vb(V) Vb(V) Supplementary Figure 7 presents the results of EL measurements for device A5. Although this device comprises a monolayer hBN spacer, no upconverted emission was observed. De- vice A5 is, however, very distinct from all other devices studied. It shows low overall currents and a strong asymmetry of the I-V characteristics, with larger currents in the negative volt- age direction (Supplementary Figure 1). PL measurements as a function of bias voltage indicate a very asymmetric charge injection, with a trion peak emerging only in one voltage direction. These findings show that the injection through one graphene layer is much less effective and it can hence be understood that such a large difference between electron and hole concentrations does not allow to establish IX densities large enough to observe the excitonic Auger process and hence upconverted emission. Supplementary Figure 7. Sample A5 with a monolayer hBN spacer. Two EL spectra for different bias voltages indicate that no upconverted emission was observed for this sample. In contrast to the samples showing upconverted emission, the emission from the IX emerge first. This sample showed overall low currents and PL measurements indicated a very asymmetric carrier injection. Inset: Map of IX emission as a function of bias voltage. 9 1.3 1.4 E (eV) 2.0 1.8 1.6 Vb(V) 1.4 2V 1.4 1.6 1.8 2.0 E (eV) 20 1.6 V 0 1.2 80 60 40 EL (ct/s) Supplementary Figure 8. Side by side comparison of the contour maps of the EL spectra measured for devices (a) A1 and (b) B1. The Supplementary Figures present the same data as shown in Figures 3 (a) and 2 (a) in the main text, but both with the same contrast (thus restricted range of bias voltages) and in the same spectral range. Supplementary Figure 8 illustrates the different voltage onsets for intralayer emission for devices A1 and B1 on the same false color and bias voltage scale. The integrated electroluminescence intensities corresponding to these graphs were presented in Figure 3 (c) in the main text. 10 EL (ct/s) 550 76 10 1 0 no data 1.2 1.4 1.6 1.8 2.0 E (eV) (b) 1.8 1.4 1.2 1.6 Vb(V) EL (ct/s) 550 76 10 1 0 1.2 1.4 1.6 1.8 2.0 E (eV) (a) 1.8 1.4 1.2 1.6 Vb(V) Supplementary Note 4: Electroluminescence, reflectance contrast and photolumi- nescence versus injected current/bias voltage: Supplementary data and estimation of interlayer exciton densities 1. EL intensity versus driving current Supplementary Figure 9 presents the integrated intralayer A-exciton EL of both, MoS2 and WSe2, as a function of driving current for sample A1 in a log-log representation. The Figure depicts two regimes: the first regime, below a voltage of Vb=1.72 V (corresponding to the intralayer exciton emission energy of 1.72 eV), corresponds to upconverted EL. Above this voltage, direct injection into intralayer excitonic states becomes possible and we enter the second regime. Supplementary Figure 9 shows a super-linear trend for the upconverted EL. Super-linear trends are expected due to the many body nature of the process. At larger voltages, in the second regime, we observe an almost linear behavior, which agrees well with the expectations of direct charge carrier injection of intralayer excitons. Although the extracted quantities are reasonable and within the expectations they have to be taken with care. The problem is that the driving current (x-axis of the graph) is strongly influenced by leakage currents which especially gain importance for low injection. In Supplementary Figure 9 one can observe this behavior at low currents (I < 0.5 µA). In this regime one obtains larger exponents, which constitute an overestimation of the actual super-linear trend. The current-voltage characteristics for sample A1 (Supplementary Figure 1) show a very pronounced diode-like behavior with small currents at low voltages and a very steep onset at the threshold for direct carrier injection. However, not all the devices measured do show such an almost ideal behavior (see Supplementary Figure 1). The observed leakage leads to the above described overestimation of the super-linearity also for large voltages and currents. For device B1 for example, we can extract a dependency of ∼I2.8 in the regime of direct carrier injection, which is in contrast to the almost linear behavior seen in this regime for sample A1, only enabled by the excellent IV characteristics of this sample. In our two-terminal devices it is very difficult to establish the actual nature of the leakage contributions, given the many possible different leakage channels for which in general one cannot assume a linear dependence on voltage. Although we observe a super-linear trend for the upconverted intralayer EL and an almost linear behavior in the direct injection 11 Supplementary Figure 9. Integrated EL of the intralayer exciton emission as a function of the driving current (log-log plot) for device A1. Quadratic (∼I2) and linear (∼I1) trends are presented for comparison. regime for sample A1, an estimation relying on this measure may, however, be questionable. Optionally, we propose two other approaches to estimate the carrier concentration based on additional optoelectronic measurements. 2. Carrier injection probed with RC measurements Supplementary Figure 10 presents the results of reflectance contrast (RC) measurements as a function of the applied bias voltage together with the simultaneously measured IV curve for device A1. The RC spectra were obtained by focusing a white light spot on the active area of the device followed by the subtraction of the reference spectrum of the bare SiO2/Si substrate. In order to illustrate the changes more markedly we plot the 2nd derivative of the RC signal. A clear change in the absorption spectrum can be observed at about 0.6 - 0.7 V. The A-exciton resonance of monolayer MoS2 is quenched in this region and a kink can be ob- 12 V b =1.72 V ~I1 regime of direct intralayer exciton injection 10-5 upconversion regime Vb< 1.72 V ~I2 107 106 105 104 Integrated EL (ct/s) 103 10-7 10-6 I(A) Supplementary Figure 10. Reflectance contrast measurements as a function of bias voltage for device A1. The false color map shows the 2nd derivative of the measured RC signal. The A exciton resonance could be resolved for both MoS2 and WSe2. A clear change in the spectra is visible at about 0.6 - 0.7 V for both monolayers. The upper panel presents the I-V characteristics recorded simultaneously with the RC data. served for the A-exciton resonance of monolayer WSe2. Due to the band alignment we know that electrons are injected via tunneling into the MoS2 monolayer and holes into the WSe2 monolayer. The quenching of the A-exciton resonance of MoS2 shows the common behavior, for example also observed in dual-gated MoSe2 monolayers1. The quenching of the A-exciton in the WSe2 monolayer is somewhat less abrupt but the progressive change of its spectrum is clearly indicative of the effective injection of holes into the valence band of WSe2. The 13 WSe2 05 15 10 -5 -10 -15 -20 I (μA) upper panel in Supplementary Figure 10 shows the current-voltage characteristics measured simultaneously with the the RC measurement. For negative voltages, we see a good corre- lation between the current onset and changes in the RC. However, in the positive direction, for which upconverted emission was observed, the current remains very small although the optical absorption measurement shows a clear change in carrier concentration. This behav- ior can be taken as a sign of carrier accumulation into states with a long lifetime resulting in a small current caused by carrier recombination. At voltages above 1.7 V, a direct car- rier injection into intralayer excitonic states becomes possible and hence we expect a large recombination current component due to the effective radiative recombination and intense light emission in this region. The latter behavior is similar to conventional light emitting diodes. 3. PL versus bias voltage and estimation of densities of the injected carriers Besides the RC measurements, which provide us with information about light absorption, we also performed the photoluminescence (PL) measurements as a function of bias voltage. Supplementary Figure 11 presents such a measurement for device A1. Supplementary Figures 10 and 11 are in very good agreement. They show a quenching of the free exciton contribution at a voltage of about 0.6 - 0.7 V. This quenching allows us to conclude that at this voltage charge carriers are started to be effectively injected into states with a long lifetime yielding to a large charge build-up across the interlayer bandgap. Moreover, the PL of the MoS2 monolayer reveals a trion resonance at voltages exceeding 0.6 - 0.7 V, which further indicates that charge carriers are injected. Supplementary Figure 12 presents a PL sweep for device B1, without hBN spacer, hence for a sample that does not show upconversion. Again a trion peak emerges, at voltages of about 0.6 - 0.7 V. Both RC and PL measurements provide evidence for charge carrier injection at voltages well below the thresholds for intralayer exciton injection (∼ 1.7 V for WSe2) in agreement to the large expected charge build-up described in the main text. To obtain an estimation of the carrier concentration we use the dependencies presented in Ref. 2 allowing to translate the shift in energy of the trion ∆E=EA−EA' , where EA is the energy of the neutral exciton and EA' the energy of the trion, into a shift of the Fermi level Ef. 14 Supplementary Figure 11. Photoluminescence as a function of bias voltage for device A1 for an excitation wavelength of 514 nm. The false colour map depicts the behavior of the A-exciton resonance of MoS2. At a voltage of about 0.6 - 0.7 V a trion peak emerges and the free excitonic emission is quenched. ∆E(V ) = 1.2 · Ef (V ) + Eb (1) where Eb is the trion binding energy. Assuming an effective mass for electrons of m∗ = 2 we can calculate the carrier concentration from the Fermi level with the formula: 0.35 me n(V ) = ∆E(V ) − Eb 1.2 m∗ π¯h2 ≈ · ∆E(V ) − Eb 1.2 · 1.46 · 1014 1 eV · cm−2. (2) From Supplementary Figures 11 and 12 we can extract ∆EA1 ∼ 60 meV and ∆EB1 ∼ 35 meV at a voltage of 1.4 V and a binding energy of roughly 30 meV in agreement with literature data3,4, which yields a carrier concentration in MoS2 of nA1 ∼ 3.7 · 1012 cm−2 at a voltage of 1.4 V. A lower concentration of nB1 ∼ 6.1· 1011 cm−2 can be estimated for sample B1, without hBN spacer, not showing upconversion. 15 ΔE (1.4 V) ≈ 60 meV Supplementary Figure 12. Photoluminescence as a function of bias voltage for device B1 for an excitation wavelength of 514 nm. The false colour map depicts the behavior of the A-exciton resonance of MoS2. At a voltage of about 0.6 - 0.7 V a trion peak emerges and the free excitonic emission is quenched. 4. Blueshift of the interlayer exciton with applied voltage and estimation of interlayer exciton density The second measure allowing to estimate the carrier concentration is the blue-shift of the IX, which is a result of the relative band movements caused by an increasing electric field. The PL and RC measurements indicate that at voltages as low as 0.6 - 0.7 V charge carriers are started to be injected, hence we can conclude that at a voltage of about 1.5 V, for which the IX emerges in the spectrum, a large charge build-up must already be present at the interface between the TMDs. Based on this observation one can assume that the shift of the IX is mostly governed by the charge carriers at this interface and a simple parallel plate capacitor model can be used to the estimate a carrier concentration n: n = ∆EIX · 0 e2 · r d (3) where ∆ EIX is the absolute blue-shift, 0 the vacuum permittivity, r the relative permit- 16 ΔE (1.4 V) ≈ 35 meV tivity and d the distance between the capacitor plates. To estimate the carrier concentration we have to determine the absolute blue-shift ∆ EIX. We can extract the energy of the IX as a function of voltage EIX(V) from our EL maps revealing a linear behavior. The only unknown parameter is the energy of the IX without injected charge carriers EIX(V0). As mentioned in the main text, no PL of the IX could be observed at zero voltage so we cannot directly measure this quantity. However, the PL and RC measurements provide us with an estimation concerning the minimal voltage threshold for carrier injection into the TMDs. For device A1 this would be around VthresA1 = 0.6 V. We can hence use this value and extrap- olate the fitted IX positions to this voltage which gives an energy of EIXA1(0.6 V ) ≈ 1.08 eV. In the case of device A1 we obtain the simple formula ∆ EIX(V)=EIX(V)−EIXA1(0.6 V). The general formula for all devices is n(V ) = (EIX(V ) − EIX(Vthres)) · 0 e2 · r d (4) where EIX(Vthres) is the threshold voltage for charge accumulation extracted from PL. EIX(V) is presented for several devices in the inset of Figure 2 in the main text. The linear trends for devices B1 and B2 have similar slopes, however the trend for device B1 is shifted to larger energies. The extrapolation for this device deviates from the others and for an extracted voltage of VthresB1 = 0.6 V we obtain EIXB1(0.6 V)≈ 1.11 eV for the presumptive energy of IX emission without injected carriers. The estimated carrier concentrations for device B2 are very similar, since due to the parallel slope, the difference EIX(Vthres) is compensated by lower values of EIX(V). Besides the variation of ∆EIX the carrier concentration will also be strongly influenced by the ratio of r and d. For the distance d we use values from Ref. 5 of dWSe2/hBN = 0.528 nm, dMoS2/hBN = 0.508 nm and dWSe2/MoS2 = 0.651 nm giving a distance for device group A with an hBN spacer of dwith hBN = 0.528 nm+0.508 nm= 1.036 nm and for device group B without hBN spacer of dwithout hBN = 0.651 nm. For samples with hBN we assume rwith hBN = 4 and for samples without hBN we assume a vdW gap, hence rwithout hBN = 1. This simple estimation yields a carrier concentration of nA1 ∼ 2.8· 1012 cm−2 for device A1 and nB1 ∼ 8.3· 1011 cm−2 for device B1 at a voltage of 1.6 V. Supplementary Figure 13 presents the results for devices A1, A2 and B1, B2 as a function of bias voltage. The Figure only depicts points based on values extracted from the measurement and no interpolation. It can clearly be seen that both devices with an hBN spacer show carrier 17 Supplementary Figure 13. Carrier concentrations as a function of bias voltage. The graph sum- maries the results of the estimation using the blueshift of the IX emission for devices A1, A2, B1 and B2. concentrations in the 1012 cm−2 range whereas devices without hBN are in the range of 1011 cm−2. Both presented methods, i.e. the MoS2 trion shift and IX blueshift, are in good agreement and show that the charge carrier concentrations are much larger in the case of the samples with an hBN spacer. 18 B2 Vb (V) Carrier concentration (cm-2) Supplementary References 1 Wang, Z., Zhao, L., Mak, K. F. & Shan, J. Probing the spin-polarized electronic band structure in monolayer transition metal dichalcogenides by optical spectroscopy. Nano Lett. 17, 740 -- 746 (2017). 2 Mak, K. F. et al. Tightly bound trions in monolayer MoS2. Nat. Mater. 12, 207 -- 211 (2013). 3 Ganchev, B., Drummond, N., Aleiner, I. & Falko, V. Three-particle complexes in two-dimensional semiconductors. Physical review letters 114, 107401 (2015). 4 Jadczak, J. et al. Probing of free and localized excitons and trions in atomically thin WSe2, WS2, MoSe2 and MoS2 in photoluminescence and reflectivity experiments. Nanotechnology 28, 395702 (2017). 5 Latini, S., Winther, K. T., Olsen, T. & Thygesen, K. S. Interlayer excitons and band alignment in MoS2/hBN/WSe2 van der Waals heterostructures. Nano Lett. 17, 938 -- 945 (2017). 19
1302.4829	1	1302	2013-02-20T08:15:13	The effect of electron--hole scattering on the transport properties of a 2D semimetal in a HgTe quantum well	[ "cond-mat.mes-hall" ]	The influence of e-h scattering on the conductivity and magnetotransport of 2D semimetallic HgTe is studied both theoretically and experimentally. The presence of e-h scattering leads to the friction between electron and holes resulting in a large temperature-dependent contribution to the transport coefficients. The coefficient of friction between electrons and holes is determined. The comparison of experimental data with the theory shows that the interaction between electrons and holes based on the long - range Coulomb potential strongly underestimates the e-h friction. The experimental results are in agreement with the model of strong short-range e-h interaction.	cond-mat.mes-hall	cond-mat	The effect of electron -- hole scattering on the transport properties of a 2D semimetal in a HgTe quantum well M.V. Entin1, L.I. Magarill1,2, E.B. Olshanetsky1, Z.D. Kvon1,2, N.N. Mikhailov1, S.A. Dvoretsky1 1Institute of Semiconductor Physics, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, 630090, Russia 2 Novosibirsk State University, Novosibirsk, 630090, Russia Phys.Rev.B The influence of e-h scattering on the conductivity and magnetotransport of 2D semimetallic HgTe is studied both theoretically and experimentally. The presence of e-h scattering leads to the friction between electron and holes resulting in a large temperature-dependent contribution to the transport coefficients. The coefficient of friction between electrons and holes is determined. The comparison of experimental data with the theory shows that the interaction between electrons and holes based on the long - range Coulomb potential strongly underestimates the e-h friction. The experimental results are in agreement with the model of strong short-range e-h interaction. PACS numbers: Keywords: Introduction Recently a 2D semimetal has been shown to be present in undoped 18-21 nm HgTe quantum wells with an in- verted energy spectrum and various surface orientations (013), (112) and (100) [1 -- 3]. It has been shown that this semimetallic state is due to the overlap by an order of several meVs of the conduction band minimum in the center of the Brillouin zone and the valence band several maxima (the exact number and configuration depending on the well surface orientation) situated at some distance away from the Brillouin zone center. The Fermi energy residing inside the energy interval corresponding to this overlap results in a simultaneous existence of 2D electrons and holes in the QW. The technology of low-temperature growth of a composite (SiO2/Si3N4) dielectric layer on top of the QWs has allowed the fabrication of electro- static top gate. Using this gate makes it possible to ob- tain and study 2D semimetal states with any desired ratio of electron and hole densities. The study conducted in (013)-oriented HgTe wells has revealed certain features which are peculiar to the transport in a 2D semimetal and may be attributed to the electron -- hole scattering in- side the QW [4]. The present work presents a detailed theoretical and experimental study of electron -- hole scat- tering in a 2D semimetal. It is well known that in monopolar systems the in- terelectron scattering does not affect the low-field con- ductivity. The scattering between particles of the same kind conserves the total momentum of the system. The momentum generated by an external electric field does not dissipate, unless the impurities or phonons are in- volved. As a result, in a system with a simple elec- tronic spectrum the conductivity does not depend on the electron -- electron scattering. This is not the case in a multi-component system [5], [6]. In the absence of mu- tual collisions, the components drift in an external elec- tric field with different velocities. The scattering between particles of different sorts leads to the additional friction in the whole system. In semimetal, electrons and holes are accelerated by the electric field in the opposite di- rections and the collision between particles slowers the motion of both electrons and holes. At low temperature the e-h scattering is limited (both for electrons and holes) to the kT interval near the Fermi surface resulting in the temperature dependence of the probability of e-h scatter- ing and of the corresponding corrections to conductivity ∝ T 2 [4]. (However, in systems with degenerate spec- trum, e.g., in a 2D system in quantizing magnetic fields the e-h scattering is not frozen out down to zero temper- ature, see [7]). The paper is organized as follows. Section I contains the theory of electron -- hole scattering in a 2D semimetal. Section II deals with the experimental de- tails. In Section III the comparison between the theory and experiment is discussed. I. THE THEORY OF ELECTRON -- HOLE SCATTERING Kinetic equation solution We consider a 2D semimetal with the ge equivalent electron valleys and gh equivalent hole valleys centered in points pe,i and ph,i, correspondingly. In particular, as will be discussed in the next section, for the (013) HgTe QW studied in the experiment we have a single conductance band valley in the center of the Brillouin zone and two valence band valleys situated along the [0¯31] direction, Fig.1. The conduction bands with energy p = (p−pe,i)2/2me overlaps with valence bands spectra εe p = p2/2mh (Eg > 0). Hole mass mh is as- Eg−εp−ph,i, εh sumed to be much larger than the electron mass me. The distances between electron and hole extrema ph,i − pe,j are supposed to be large to suppress the electron -- hole re- combination. At the same time, the scattering between electrons and holes changing momenta near the extrema is permitted. Without the loss of generality, further we will count the momenta from the band extrema and re- place p − ph,j → p, p − pe,i → p. D EF FIG. 1: The energy band structure in a 20 nm (013) HgTe quantum well. The system of kinetic equations for the electron and hole distribution functions f e,h p reads 2 where index ν = (+,−) numerates holes (h) and elec- trons (e), respectively, e± = ±e, −e is the electron charge, Jν is the collision integral of holes (electrons ) with impurities, Iν,ν ′ are the inter-particle collision inte- grals, ων = eν H/mνc, ων are cyclotron frequencies. In the linear conductivity problem the collision integrals for particles of the same sort (Iν,ν ) give no contribution to the conductivity. Hence, the summation over ν ′ can be omitted replacing ν ′ by ¯ν = −ν. eν E∇pf ν p + [p, ων]∇pf ν p = Xν ′ Iν,ν ′ + Jν , (1) The hole-electron collision integral has the form 2π Ihe = S2 2ge Xp′,q,k,′ uq2δp′,p+qδk′,k+qδ(εh k) − f h k′(1 − f e p − εh p′(1 − f h p′)f e (cid:2)f h p (1 − f h p )f e k′ − εe k(1 − f e k) × k′)(cid:3) . p′ + εe Here uq is the Fourier transform of the electron -- hole in- teraction potential u(r), S is the system area. Value Ieh can be obtained from Eq.(4) by exchange e ⇆ h. We shall study linear (in E) transport. Introducing lin- p to equilibrium distribution functions ear corrections φν f 0,ν p and linearizing the kinetic equations, we obtain eν(E∇p)f 0ν p + ([p, ων]∇p)φν p = δIν,¯ν + δJν , where δIν,¯ν and δJν are linearized collision integrals, nφν p(cid:2)(1 − f 0ν k′(cid:2)f 0ν p′ )f 0¯ν p (1 − f 0ν k′ (1 − f 0¯ν p′ )(1 − f 0¯ν φ¯ν k ) + f 0ν k (1 − f 0¯ν p′ f 0¯ν p′ (1 − f 0ν k ) + f 0ν 2π δIν,¯ν = p − εν S2 2g¯ν Xp′,q,k,′ uq2δp′,p+qδk′,k+qδ(εν k′ )(cid:3) − φν k (cid:3) − φ¯ν p )f 0¯ν k ) + (1 − f 0ν k′ (1 − f 0¯ν p′ )f 0¯ν p + f 0ν k′ f 0ν p f 0¯ν p′(cid:2)f 0ν k(cid:2)(1 − f 0ν p )f ¯ν0 p′ (1 − f 0ν k (1 − f ¯ν0 p )(1 − f 0¯ν p′ + ε¯ν k′ − ε¯ν k) × k′ )(cid:3) + k′ )(cid:3)o. (2) (3) (4) (6) The solution of the system of kinetic equations can be searched in the form of φν p = Aν (εp)p, Aν(εp) ∝ E. This substitution results in the system of integral equations for Aν (εp). Instead of solving this system we use approximation p ≈ −pVν ∂εν φν p f (0ν) p , (5) where Vν are the average velocities of particles. To find the values Vν one should integrate the kinetic equations with the momentum p. The impurity collision term gives the rate of irretrievable momentum loss. The h-e colli- sion term determines the rate of momentum transfer be- tween holes and electrons (the force between subsystems of holes and electrons) f = 2ghXp pIhe. The considered procedure is equivalent to the algebraiza- tion of the collision terms δJν = − φν τν , δIν,¯ν = φ¯ν τ¯νν − φν τν ¯ν , where τν is the transport relaxation for elastic scattering on impurities. Relaxation times τhe and τeh of interpar- ticle scattering satisfy the relation mh Nsτhe = me Psτeh = η. Quantity η can be considered as the coefficient of liquid friction between the subsystems of holes and electrons: the force between electrons and holes is SNsPsη(Ve − Vh). As a result we come to the system of hydrodynamic equations [5, 6] for Vν The system (7) can be written in the matrix form as Ω · V = E, eν mν E + [Vν , ων] − Vν τν − η n¯ν mν (Vν − V¯ν ) = 0. (7) where V = (V e x , V h Ω = τe y , V h x , V e −( 1   ) y ), E = e(Ex/me, Ey/me,−Ex/mh,−Ey/mh), + 1  τeh −ωe  1 τeh 0 + 1 τhe −ωh 0 1 τeh ωh + 1 τhe ωe + 1 τeh 0 1 τhe −( 1 −( 1 −( 1 1 τhe 0 τh τh τe ) ) ) 3 (8) (9) With the use of the solution of Eq.(9) V = Ω−1E we obtain jx = e(V h x Ps − V e x Ns), jy = e(V h y Ps − V e y Ns) and σxx = N1/D, σyx = N2/D, N1 = e2(cid:16)memh(cid:0)mePsτh(cid:0)τ 2 h + 1(cid:1)(cid:1) + η(cid:16)2memhτeτh(cid:0) (Ns − Ps) 2 + e ω2 e + 1(cid:1) + mhNsτe(cid:0)τ 2 hω2 e + 1(cid:1)(cid:1)(cid:17) + e ω2 h (cid:0)τ 2 eτ 2 hω2 e (cid:0)τ 2 h + 1(cid:1) + m2 NsPs (τeτhωeωh + 1)(cid:1) + NsPs(cid:0)m2 hτ 2 η2(Ns − Ps)2τeτh (mhPsτe + meNsτh)(cid:17), N2 = −e2(cid:16)memh(cid:0)mhNsωe(cid:0)τ 2 hω2 h + 1(cid:1) τ 2 e + mePsτ 2 h (Ns − Ps) 2 (meNsωe + mhPsωh)(cid:17), e τ 2 (Ns − Ps) (Nsτeωe − Psτhωh) + η2τ 2 D = m2 e )(1 + ω2 η2((mhPsτe + meNsτh)2 + τ 2 h (cid:0)τ 2 e ω2 e + 1(cid:1) ωh(cid:1) + 2ηmemhτeτh × h) + 2ηmemh(meNsτh(1 + ω2 h(meNsωe + mhPsωe)2). e ) + mhPsτe(1 + ω2 h(1 + ω2 hτ 2 e τ 2 h)) + em2 hτ 2 e τ 2 e τ 2 (10) For components of resistivity tensor one can write ρxx = N1D 1 + N 2 2 N 2 , ρxy = N2D 1 + N 2 2 N 2 . (11) At zero magnetic field ρxy = 0, and the temperature- dependent correction to the resistivity are simplified δρ(T )/ρ(T = 0) = mhNsτe + mePsτh memh memh + η (mhPsτe + meNsτh) mhNsτe + mePsτh + ητeτh (Ps − Ns) 2 − 1 (12) Electron -- hole relaxation time mined by substitution of the distribution functions of Eq.(5) into Eq.(4). We arrive at The mean force acting between electron and hole sub- systems f , Eq.(6), in the Born approximation is deter- f = 2πgh (4π2)4 Z dpZ dp′Z dkZ dk′up−p′2δ(p′ − p + k − k′)δ(εh n(p − p′, p)Vh(−∂εh (k − k′, k)Ve(−∂εe p′ + f (0e) k′ + f (0h) (1 − f (0e) (1 − f (0h) p − εh (1 − f (0e) (1 − f (0h) )hf (0e) )hf (0h) )f (0h) )f (0e) p′ + εe k′ − εe k) ) − )io k′ p′ )(1 − f (0h) )(1 − f (0e) f (0h) p f (0e) k p k p k p k′ p′ k′ p′ p 4 (13) (14) The integral over p can be presented as R dp = mhR dεpR dϕp and similarly for the integral over other momenta. Calculating integrals over energies in the low temperature limit we obtain the following expression for the mean free time between collisions of holes with elec- trons 1 τhe ≡ Nsη mh = T 2m2 emh (4π2)3 ζ2 2Ps7 Z 2π 0 dφdϕdϕ′(1 − cos φ)δ(ζ(cos φ − 1) + cos ϕ − cos ϕ′) × δ(ζ sin φ + sin ϕ − sin ϕ′)upF h(1−cos φ)2, (15) where ζ = pF h/pF e, pF h, pF e are the Fermi momenta of holes and electrons respectively. Expression (15) can be transformed to 1 τhe = m2 e 12π3gh5 T 2 ǫF h ζZ x0 0 dx Here x0 = min(1, 1/ζ). xu2pF hx2 √1 − x2p1 − ζ2x2 .(16) The Fourier transform of the Coulomb e-h interaction uq depends on the structure of the system and the screen- ing. In the simple 2D model of electron gas the Coulomb interaction with linear screening reads as This expression is valid in the linear screening approx- imation that needs smallness of κ as compared to the transmitted momentum min(pF,e, pF,e). Besides, here we neglect the width of the quantum well. As a result, the potential becomes independent of the HgTe dielectric constant. This 2D consideration loses applicability in the specific system under the consideration where the quan- tum well width d & 1/κ. In fact, the screening radius 1/κ should be limited from below by d. Accounting for finite width of the quantum well leads to replacement of the 2D potential by uq = 2πe2 1 χ q + κ , (17) uq = 2πe2 F (qd) χ q + κF (qd) , q < 2(pF e, pF h). (18) where the screening constant is collected from the in- dividual screening constants of electron and hole gases, κ = κe + κh = 2(gh/aB,h + ge/aB,e), aB,e = 2χ/mhe2 and aB,h = 2χ/mee2 are the Bohr radiuses of electrons and holes, respectively; χ is the effective dielectric con- stant. where the function F (qd) follows from the solution of electrostatic interaction problem of two singly charged particles placed inside a layer of width d between two semi-infinite dielectrics. Using planar Fourier transform, we have for the interaction of two point charges located in the points z, z ′, d/2 > z > z ′ > −d/2: − e−q(z+z')(cid:0)edq(r + 1) − e2qz(r − 1)(cid:1)(cid:0)−r + eq(d+2z')(r + 1) + 1(cid:1) 2χq (e2dq(r + 1)2 − (r − 1)2) . (19) For z < z ′ it is necessary to do replacements z ↔ z ′ in Eq.(19). Here r = χ/χHgTe, χ is the dielectric constant of external layers (CdT e). To find the func- tion F (x), one should integrate the potential (19) with the squares of electron and hole transversal wave func- tions ψe(z)2 and ψh(z ′)2. For a well with hard walls ψe,h(z) = p2/d cos(πz/d). In this case, we find F (x) = r (x2 + 4π2)2 (cid:18)x(cid:0)3x2 + 20π2(cid:1) + 32π4 (−1 + ex) x + r (ex(x − 2) + x + 2) (r + ex(r + 1) − 1) x2 Assembling the previous expressions we get the τhe for the Coulomb scattering 1 τhe = mee4 6πχ2gh3 me mh T 2 ǫ2 F h ζ Z x0 0 dx xF 2(wx) √1 − x2p1 − ζ2x2(2x + ξF (wx))2 , (cid:19) 5 (20) (21) where w = 2pF hd, ξ = κ/pF h. In the strict 2D case w = 0, the potential converts to Eq.(17) and F (wx) should be replaced by 1. It should be emphasized that in the real case the parameter ξ is large, so, the mean free time ceases to depend on F . This conclusion is valid in the linear screening theory. Careful examination shows the necessity of revision of this approach. Quantity 1/τhe is proportional to T 2. This results in a similar temperature dependence of the correction to the residual resistivity at low temperature. Short-range interaction Together with the long-range Coulomb part the in- teraction between electrons and holes contains also the short-range kernel interaction. Large dielectric constant of HgTe and CdHgTe leads to the dielectric screening of the Coulomb contribution. In that case the on-site e-h interaction can prevail. To estimate the kernel contribu- tion one can replace u2pF hx by a constant: u2pF hx = π2 me + mh memh Λ; Λ = (22) Dimensionless quantity Λ describes the strength of the contact e-e-interaction. As a result we find for η : memh me + mh 1 2π Z u(r)dr. momenta of electrons and holes. The divergency origi- nates from the probability of two Fermi particle backscat- tering with conservation of their individual energies. For the isotropic Fermi surfaces such processes occur for all electrons on the Fermi surface. Anisotropic spectrum In fact, the holes in the system under consideration have anisotropy. That evidently limits kinematically the possibility of backscattering and the divergency. This makes it necessary to take into account the hole spec- trum anisotropy neglected before. It can be done in the relaxation time approximation for elliptic hole spectrum for the case of zero magnetic field. The anisotropy of the spectrum results in the anisotropy of temperature corrections. In accordance with the experimental situa- tion, we shall consider the electric field applied along the symmetry axes, say i. In this case Eq.(7) is modified as e mi Ei − V h i τh − η Ps mi (V h i − V e i ) = 0. (24) Here, subscript i marks the specific direction of the field and the same component of the hole mass. Friction coefficient ηi for the same direction of electric field reads η = (me + mh)2Λ2T 2 24π23NsPs ln 1 + ζ 1 − ζ (23) ηi = mh Nsτhe,i = T 2(me + mh)2Λ2 6π33NsPs f (αi, ζ), (25) In accordance with (15), the model of isotropic energy spectrum leads to a logarithmic divergency of the tem- perature corrections to the conductivity at equal Fermi where mh = √m1m2 is the mass of the hole density of states, αi = pmi/mh, f (α, ζ) = (αζ)2 32 Z 2π 0 dϕdϕ′dφdφ′(cos ϕ − cos ϕ′)δ(αζ(cos ϕ′ − cos ϕ) + cos φ − cos φ′) ×δ((ζ/α)(sin ϕ′ − sin ϕ) + sin φ − sin φ′). Integrals over three angles can be evaluated and we arrive at f (α, ζ) = α4Z 1 0 x2dx √1 − x2(1 + x2(α4 − 1)) ln(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α − ζp1 + x2(α4 − 1)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) α + ζp1 + x2(α4 − 1) . 6 (26) (27) Fig.2 shows the dependence of f (α, ζ) on ζ for different α. All curves contain the limited singularities corresponding to the equality of the hole Fermi ellipses axes to the di- ameter of the electron Fermi circle. The exception is the case of α = 1, when f (1, ζ) ∝ − lnζ − 1 at ζ → 1. In this case, the divergency can be limited by the finite temperature or collision widening. f Α,Ζ 4 3 2 1 0 1 2 3 Ζ 4 FIG. 2: Dependence of temperature correction in the anisotropic case on the electron -- to -- holes concentration ratio and the hole masses ratio via parameters ζ and α. Param- eter α runs values 0.2,0.6,1,1.4,1.8,2.2,2.6. The irection of α growth is shown by arrow. II. EXPERIMENT Samples The Hg0.3Cd0.7T e/HgT e/Hg0.3Cd0.7T e quantum wells with the (013) surface and the thickness of 20.5 nm were prepared by molecular beam epitaxy. The details of the structure growth process are described in [8, 9]. The QW cross-section and the energy diagram of the structures investigated is shown in Fig.3a and 1b, respectively. To perform magnetotransport measure- ments, the samples based on these quantum wells were prepared by standard photolithography in the form of 50 µm wide Hall bars with the voltage probes spaced 100 µm apart. The ohmic contacts to the two-dimensional gas were formed by the in-burning of indium. To change and control the electron and hole densities in the QW, the electrostatic top gate has been supplied. For this purpose, a dielectric layer containing 100 nm SiO2 and 200 nm Si3N4 was first grown on the structure using the plasma-chemical method. Then, the TiAu gate was deposited. The schematic drawing of the devices prepared in this way is shown in Fig.3c. The magne- totransport measurements in the described structures were performed in the temperature range of 0.2 -- 4.1 K in magnetic fields up to 5 T by the standard four-point circuit at the 12 -- 13 Hz ac signal with the current of 1 -- 10 nA through the sample, which is sufficiently low to avoid the overheating effects. Experimental results To gather information about the structures properties and to determine the main transport parameters of the system corresponding to different gate voltages, the mag- netic field dependences of the diagonal ρxx(B) and Hall ρxy(B) components of the resistance tensor were mea- sured. These functions show a strong dependence on the magnitude and sign of the gate voltage applied to the sample. Fig.4a,b,c present the curves measured at gate voltages -3, -1.84 and -0.5 V respectively. One can see that an alternating-sign Hall effect and strong positive magnetoresistance are observed at Vg = 3 and -1.84 V (see Fig.4a,b). Meanwhile, at Vg = -0.5 V (see Fig.4c), there is a weak negative magnetoresistance at low fields and positive magnetoresistance at higher fields, and the magnetic field dependence of the Hall resistance is linear with its slope opposite to that of ρxy(B) at Vg = 3 and ( )a CdTe 40 nm Hg Cd Te 30 nm 0.65 0.35 HgTe QW 20.5 nm Hg Cd Te 0.35 0.65 9 nm 9 -2 10 cm In-doping 12nm Hg Cd Te 0.35 0.65 9 nm CdTe 5.5 μ m ZnTe 10 nm In contacts TiAu gate ( )c 3 4 2 SiO +Si N Cd Hg Te HgTe QW Cd Hg Te 1-x x 1-x x i-GaAs substrate ( )b e T g H e T d C e T 5 6 . 0 d C 5 3 . 0 g H G 6 e T d C G 8 e T 5 6 . 0 d C 5 3 . 0 g H 1,0 0,5 0,0 -0,5 ) V e ( y g r e n e (013) GaAs substrate -60 -40 -20 0 20 40 60 z (nm) FIG. 3: The quantum well layer structure - (a), the quantum well energy diagram -(b), and the cross section of the samples studied. -1.84 V and B >0.1 T and B >0.4 T, respectively. FIG. 4: Magnetic field dependences ρxx(B) and ρxy(B) for the 2D electron-hole system in the HgTe quantum well at T = 0.19 K for three gate voltages: (a)- Vg = −3 V; (b)- Vg = −1.84 V and (c)-Vg = −0.5 V; (d)- the energy band diagrams with approximate positions of the Fermi energy corresponding to the curves on the left side. The described behavior suggests that, by varying the gate voltage we change the carrier type content in the quantum well. This conclusion is further supported by the ρxx(Vg) and ρxy(Vg) traces measured at a constant 7 magnetic field B=2 T corresponding to the quantum Hall effect regime, Fig.5. The quantum Hall plateaux in ρxy(Vg) and minima in ρxx(Vg) are well developed for filling factors ν = 1 − 10 on the electron side and for ν = 1 − 4 on the hole side indicating a very high quality of the samples investigated. At Vg ≈ −1.8V a dramatic change of the sign of ρxy takes place signifying a change of the predominant carrier type in the well. This trans- formation in ρxy(Vg) is accompanied by a sharp peak in ρxx(Vg). The behavior of our system in the quantum Hall effect regime has been studied earlier [10]. 50000 40000 30000 20000 10000 ) ( x x B= 2 T T= 0.19 K 30000 20000 10000 ) ( y x 0 -10000 -20000 0 -8 -6 -4 -2 0 Vg (V) 2 4 6 -30000 8 FIG. 5: Gate voltage dependences ρxx(B) and ρxy(B) for the 2D electron-hole system in the HgTe quantum well at T = 0.19 K and magnetic field B = 2 T. By fitting the dependences similar to those presented in Fig.4a,b,c using the formulas of the standard classical transport model in the presence of two groups of carri- ers of opposite signs [11], we can determine the types of charge carriers involved in the transport, as well as their mobilities and densities. Fig.6 presents these parameters as functions of the gate voltage. We first consider the gate voltage dependences of the electron and hole densi- ties shown in Fig.6a. For Vg ≥ −1 V, the experimental curves (see, e.g., Fig.4c) are adequately described by the transport model involving only electrons as charge car- riers. Although holes can be present in this case with a density much lower than the electron density, their con- tribution to the transport is immaterial due to their lower mobility. As would be expected, the gate voltage depen- dence of the electron density is linear with a slope of 8.12 × 1014m−2V −1 corresponding to the capacitance of the dielectric. An absolutely different pattern is observed for Vg ≤ −1.5 V. To describe the dependences similar to those presented in Fig.4a,b two types of carriers, elec- trons and holes, should be taken into account. Fig.6a shows the electron and hole densities as functions of the gate voltage for Vg ≤ −1.5 V obtained from the process- ing of the experimental data. Clearly, as the negative gate bias increases, the hole density increases and the electron density decreases linearly with slopes 7.9 × 1014 and 0.7 × 1014 m2 V 1, respectively. We note that the sum of the magnitudes of these slopes is about the mag- nitude of the slope of Ns(Vg) for Vg ≥ −1 V, as would be expected, because electrons are the only observable type of carriers for Vg ≥ −1 V. Moreover, the slope ratio Ps(Vg)/Ns(Vg) ≈ 11.3 for Vg ≤ −1.5 V should corre- spond to the ratio of the densities of states of holes and electrons. Then, if holes fill two valleys (as expected for a (013) 20 nm HgTe QW) and electrons fill only one valley, then the hole mass is mh ≈ 0.15m0 if we take the elec- tron mass me ≈ 0.025m0. These values are close to those determined from the cyclotron resonance measurements [12]. Processing diagonal ρxx(B) and Hall ρxy(B) depen- dences in the vicinity of the gate voltages where the electron and hole densities are close is rather difficult. However, extrapolating linear dependences Ns(Vg) and Ps(Vg), we find their crossing point, where the electron and hole densities are equal, Vg ≈ −1.3 V - the so called charge neutrality point (CNP). Using the electron and hole density versus gate voltage dependences obtained above we can plot the opposite to Fig.4a,b,c qualitative energy band diagrams with the cor- responding Fermi level positions and the conduction and valence band occupation, Fig.4d. As mentioned above for the (013) HgTe QWs we have a single conductance band minimum in the center of the Brillouin zone and two valence band maxima situated along the [0-13] direc- tion. According to the CNP electron and hole densities and their mass values determined above the conductance and the valence bands overlap in our samples is about 10 meV. s Now, we consider the behavior of the electron and hole mobilities as Vg varies (see Fig.6b). The lines are drawn through the experimental points for visualization. In the range of -1.5 V ≤ Vg ≤ +3 V, a decrease in the elec- tron density is accompanied by a marked decrease in their mobility roughly as ∼ N 3/2 . Similar dependency of mobility on density is frequently observed in other two- dimensional structures as well, where the carrier density is controlled by the electrostatic gate. It results from the transport time for impurity scattering depending on the carrier density as τtr ≈ N α s , where α = 1 − 2. In the gate voltage range of −2 ≤ Vg ≤ −1.5 V corresponding to the approximate equality of the electron and hole den- sities, a sharp jump in the electron mobility is observed (see the dotted line in Fig.6b). A further increase in the magnitude of the negative gate bias slightly reduces the electron mobility and weakly increases the hole mobility. Of the most interest is the jump in electron mobility at −2 ≤ Vg ≤ −1.5 V. This jump coincides with the gate voltage range where the hole density first equals and then begins to exceed the electron density: Ps ≥ Ns. We sug- gest that this jump may be accounted for by the holes screening and, therefore, reducing impurity scattering of electrons. As shown in Section I, in a bipolar system with two types of charge carriers of the opposite sign, momentum 8 Electrons Holes (a) CNP -6 -4 -2 0 2 4 (b) Electrons Holes -6 -4 -2 Vg (V) 0 2 4 4,0 3,5 3,0 2,5 2,0 1,5 1,0 0,5 0,0 50 40 30 20 10 0 ) 2 - m 5 1 0 1 ( s P , s N ) s V / 2 m ( p , n FIG. 6: (a)- the electron Ns and hole Ps densities versus gate voltage; (b)- the electron µn and hole µp mobilities versus gate voltage; T = 0.19 K. -1.84 V -2 V B=0 T=0.19 - 7 K -3 V -4.5 V -6 V CNP ) ( 900 800 700 600 500 400 300 200 100 0 -6 -5 -4 -3 -2 -1 Vg (V) 0 1 2 3 FIG. 7: Gate voltage dependences ρ(Vg) at B = 0 and various temperatures T = (from bottom to top) T= 0.2; 0.5; 1; 1.5; 2; 2.5; 3; 3.6; 4.1; 5; 6; 7 K. relaxation can be caused, in addition to other factors, by their mutual scattering (friction) [5]. Since only the particles of both kinds that fall into the kT interval in the vicinity of the Fermi level are involved in this momentum relaxation mechanism, the corresponding relaxation time is expected to change with temperature as ∼ T −2. Fig.7 presents the gate voltage dependences of our sam- ple resistance in the zero magnetic field for a number of temperatures in the interval T = 0.19 - 7 K. Each ρ(Vg) curve has a pronounced maximum. At the lowest tem- perature T = 0.19 K, the position of this maximum al- most coincides with the gate voltage at which the hole and electron densities are equal (CNP). Another inter- esting feature of the curves in Fig.7 is their asymmetric temperature dependence with respect to the gate volt- age. One can see that for Vg ≥ −1 V, i.e. when the electrons are the only detectable carriers in the system, there is only a weak dependence of resistance on temper- ature. Meanwhile, at Vg ≤ −1 V, i.e. when the holes begin to populate the valence band, a considerable in- crease (by a factor of 1.5 - 3) in resistance is observed as the temperature increases from 0.2 K to 7 K. It is max- imal in the range of −3V ≤ Vg ≤ −1 V and decreasing for higher negative gate biases. Also, with the temper- ature increasing the maximum of the ρ(Vg) curve shifts by about 0.5 V to negative gate voltages. In the follow- ing section we analyze the observed behavior using the theory of electron -- hole scattering developed in Section I. III. THEORY VERSUS EXPERIMENT Due to the electron -- hole scattering, there would be a stronger temperature dependence of the resistivity in the gate voltage range where both holes and electrons are present Vg ≤ −1 V compared to Vg ≥ −1 V, where the electrons are the only charge carriers (see Fig.6a). To an- alyze the system behavior at Vg ≤ −1 V, we use Eq.(12) from Section I, obtained for the temperature dependence of resistance in a system with two types of charge carri- ers when the momentum relaxation is due to their mutual scattering, which we rewrite in the following form: ρ(T ) = ρ0 1 + (η/e)(Nsµp + Psµn) 1 + (η/e)(Ns − Ps)2µnµp/(Nsµn + Psµp) (28) Here, ρ0, Ns, Ps, µn, and µp are the system resistance, the electron and hole densities and mobilities at T = 0, respectively and η is the electron-hole friction coefficient defined in Sec.I. Irrespective of the details of the scat- tering mechanism, at specified values of the electron and hole densities the probability of electron -- hole scattering decreases as the square of temperature, η = Θ·T 2, where Θ is a certain T -independent function of Ns and Ps to be determined. In Fig.8a we plot with closed circles the ρ(T ) depen- dences obtained from the experimental curves in Fig.7 at Vg = −1.84;−2;−3;−4.5;−6 V (in Fig.7 these gate voltages are marked with arrows). For all these gate volt- ages, the resistivity temperature dependence saturates at T ≤ 0.5K. This allows us to use the values of the electron and hole mobilities and densities at these temperatures as zero-T quantities in Eq.(28) when fitting it to the ex- perimental data in Fig.8a. For each of the specified gate 9 600 500 400 300 200 30 25 20 15 10 5 0 ) ( ) 2 K / s J 8 3 - 0 1 ( (a) Vg= - 1.84 V - 2.0 V - 3.0 V - 4.5 V - 6.0 V 0 1 2 T (K) 3 4 35 30 25 20 15 10 -3,0 -2,5 -2,0 -1,5 -1,0 -6 -5 -3 -4 Vg (V) (b) -2 FIG. 8: (a)-Temperature dependences ρ(T ) obtained from Fig.7 for Vg = 1.84, 2, 3, −4.5, −6 V (marked with arrows in Fig.7). The lines are the fitting by Eq.(28) (see the text); (b)- Parameter Θ obtained from fitting the experimental data in Fig.7 with Eq.(28). The solid line represents the theory given by Eqs.(25-27) with α = 1.2 and the contact interaction con- stant Λ = 1.36. Insert: similar experimental data from an- other sample published previously [4], the line corresponds to Eqs.(25-27) with α = 1.2 and the contact interaction constant Λ = 1.64. voltages these zero-T parameters were independently ob- tained from the magnetotransport data as described in the above discussion of the curves in Fig.4a,b,c. There- fore, the fitting procedure for each value of the gate volt- age in Fig.8a depends on a single parameter Θ in the expression for η. The fitting of Eq.(28) to the data is shown in Fig.8a by lines. The temperature range for fit- ting was chosen as 0.19-4.1 K. It was found that Eq.(28) does not fit well the experimental points for the tem- peratures higher than 4.1 K possibly because of other temperature dependent scattering mechanisms emerging at these temperatures. The points in Fig.8b show fitting parameter Θ as a function of the gate voltage. In the insert we show, for statistics, a similar data for another sample that was published previously [4]. Let us now apply the theoretical results obtained in Sec.I for the analysis of the gate voltage dependence of 10 ifies the Coulomb interaction. The long-range compo- nents of the Coulomb interaction are suppressed, while a short-range 2D scattering amplitude is formed on the scale of the well width. Under these circumstances we can consider the e-e scattering in the simplest way, and turn to the short- range potential model represented by Eq.(23) for the isotropic case and by its extension (Eqs.(25-27)) for the anisotropic spectrum. Then we have only a single fitting parameter Λ, the value of which can not be found in the 2D model developed here. The solid lines in Fig.8b and the Insert are the fitting of Eqs.(25-27) to the experimental dependences of Θ(Vg) with Λ = 1.36 (1.64 for the data in the Insert) and the hole mass anisotropy coefficient α = 1.2 in both cases. A more universal way to present the data is to plot quan- tity which case the experimental points for both of the sam- ples should fall on the same curve f (α, ζ). As one can see in Fig.9, this is indeed the case. B (me+mh)2Λ2 as a function of ζ = pPs/2Ns, in Θh3NsPs 2k2 The values of Λ obtained from the fitting in Fig.8b appear to be too large if we assume that it represents Coulomb interaction with the dielectric constant of HgTe equal to 12-15. At the same time, these values are in good agreement with the short-range model considerations. Let us discuss the origin of short-range interaction in more detail. In quantum wells of conventional semi- conductors the subbands are formed from the simple envelope-function states. On the contrary, in a HgTe quantum layer the size quantization and the formation of a gap occur simultaneously. The e-h interaction modifies the bands, and the value of the effective e-h interaction is inevitably related to the structure of the states. This determines the characteristic energy and spatial scales of the e-h interaction, namely, the gap as the characteristic energy scale and the width of the quantum layer as the length scale. That results in a value of Λ comaparble with the extracted from the experimental data. It should be emphasized, that a large e-e interaction constant means the inapplicability of the Born approx- imation for the electron -- hole pair scattering and of the Fermi gas concept. However, constant Λ can be treated as a low-energy limit of a dimensionless scattering am- plitude that preserves the above-mentioned estimates. In fact our results constitute an evidence that even in the case of σ ≫ e2/h a 2D e-h system in HgTe QW should be considered as a strongly correlated 2D e-h liquid rather than a 2D e-h gas. IV. CONCLUSIONS We have developed the theory of temperature- dependent corrections to the conductivity and magne- totransport coefficients in a 2D semimetal. These cor- rections are caused by friction between electrons and holes. The corrections obey the quadratic temperature dependence at the low temperature limit. The friction Θh3NsPs B (me+mh)2Λ2 plotted as a FIG. 9: Dimensionless quantity function of ζ = pPs/2Ns for the two samples data in Fig.8b. The solid line represents the theory, Eq.(27). 2k2 Θ = η/T 2. Before we begin, it is vital to note that parameter Θ gives a first hand information about the in- terparticle interaction which makes our situation rather unique. Indeed, in the general case of a 2D electron sys- tem with σ ≫ e2/h this information can only be obtained form the study of quantum corrections which, apart from being only few percent of the total conductivity depend on the interaction in an indirect and complicated form [13 -- 15]. First of all we notice that the weak e-h interaction ap- proximation of our theory, given by Eq.(21), does not seem to be applicable in our case. Indeed, a direct calcu- lation of Θ using Eq.(21) for our system parameters and the corresponding values of Ns, Ps yields Θ about two orders of magnitude less than that observed experimen- tally (Fig.8b). The reason for this is probably related to the following fact. In the carrier density range investi- gated, the ratio of the screening constant to the Fermi wave vector κ/min(pF h, pF e) ≈ 20 ≫ 1 in which case treating the e-h interaction as weak becomes unjustified. Besides, 1/κ = 1.3 nm ≪ d, where d= 20 nm is the QW width, and the 2D consideration loses applicability. We have taken into theoretical consideration a vari- ety of factors affecting the e-h interaction, except for its large strength. Besides the general difficulties associated with the consideration of strong interaction in a simple 2D case, there are complications due to the specificity of HgTe quantum wells. In fact, the individual electron energy levels and wave functions in a narrow-gap semiconductor are obtained from the size quantization of a many-component wave function, that results in a complicated space dependence of electron density. Owing to the large strength of e-h in- teraction, its essential part is accumulated on distances comparable to the well width. This factor strongly mod- coefficients are found for the linear-screened Coulomb electron -- hole interaction and the real spatial structure of the system. Besides, calculations have been made for the core electron -- hole scattering in the assumption that the Coulomb potential is completely screened. The ex- periments were performed in 20 nm (013) HgTe QW. We found that the conductivity variation with temper- ature due to electron -- hole scattering is very large (2-3 times higher than the conductivity in the zero temper- ature limit). This gives us the opportunity to obtain a direct information about interparticle interaction in a 2D electron -- hole system with a high (σ ≫ e2/h) conduc- tivity value. It has proved to be impossible to explain the observed strong temperature dependent variation of conductivity as a consequence of electron -- hole scattering due to Coulomb interaction. Instead, the short-range e- h scattering model was found to be satisfactory for the explanation of the observed large friction strength. 11 [1] Z. D. Kvon, E. B. Olshanetsky, D. A. Kozlov, et al., Pisma Zh. Eksp. Teor. Fiz. 87, 588 (2008) [JETP Lett. 87, 502 (2008)]. [2] Z. D. Kvon, E. B.Olshanetsky, E.G. Novik, D. A. Kozlov, et al., Phys.Rev.B 83, 193304 (2011). [3] E. B.Olshanetsky, Z.D.Kvon, N.N. Mikhailov, et al., Solid State Comm. 152, 265 (2012). [4] E. B.Olshanetsky, Z. D. Kvon, M.V.Entin et al., Pisma Zh. Eksp. Teor. Fiz. 89, 338 (2009); [JETP Lett. 89, 290 (2009)]. [5] V. F. Gantmakher, I. B. Levinson, Carrier Scattering in Metals and Semiconductors (Nauka, Moscow, 1984; North Holland, New York, 1987). [6] V. F. Gantmakher, I. B. Levinson, Zh. Eksp. Teor. Fiz. 74, 261 (1978) [JETP 47, 133 (1978)]. [7] M.V.Entin, L.I.Magarill, Eur. Phys. J. B 81, 225 (2011). [8] N.N. Mikhailov, R.N. Smirnov, S.A. Dvoretsky et al., Int.J. Nanotechnology 3, 120 (2006). [9] E.B. Olshanetsky, S.Sassine, Z.D. Kvon et al., Pisma Zh. Eksp. Teor. Fiz. 84, 661 (2006) [JETP Lett. 84, 565 (2007)]. [10] G. M. Gusev, E. B. Olshanetsky, Z. D. Kvon, A. D. Levin, N. N. Mikhailov, and S. A. Dvoretsky,Phys. Rev. Lett.108, 226804 (2012). [11] V. L. Bonch-Bruevich and S. G. Kalashnikov, Semicon- ductor Physics (Nauka, Moscow, 1990) [in Russian]. [12] D. A. Kozlov, Z. D. Kvon, N. N. Mikhailov, S. A. Dvoret- skii, and J. C. Portal, Pisma Zh. Eksp. Teor. Fiz. 93, 186 (2011), [JETP Lett. 93, 170 (2011)]. [13] B.L. Altshuler, A.G. Aronov, Electron -- electron interac- tions in disordered systems, Edited by A.L. Efros, M. Pollak, Elsevier Science Publishers B.V., 1985. [14] A.M. Finkelstein, JETP Lett. 37, 517 (1983). [15] G. Zala, B.N. Narozhny, and I.L. Aleiner, Phys. Rev. B64, 214204, (2001).
1802.01665	3	1802	2018-10-30T11:13:27	Particle Production in Ultra-Strong Coupling Waveguide QED	[ "cond-mat.mes-hall", "quant-ph" ]	Understanding large-scale interacting quantum matter requires dealing with the huge number of quanta that are produced by scattering even a few particles against a complex quantum object. Prominent examples are found from high energy cosmic ray showers to the optical or electrical driving of degenerate Fermi gases. We tackle this challenge in the context of many-body quantum optics, as motivated by the recent developments of circuit quantum electrodynamics at ultrastrong coupling. The issue of particle production is addressed quantitatively with a simple yet powerful concept rooted in the quantum superposition principle. This key idea is illustrated by the study of multi-photon emission from a single two-level artificial atom coupled to a high impedance waveguide. We find surprisingly that the off-resonant inelastic emission lineshape is dominated by broadband particle production, due to the large phase space associated with contributions that do not conserve the number of excitations. Such frequency conversion processes produce striking signatures in time correlation measurements, which can be tested experimentally in quantum waveguides. These ideas open new directions for the simulation of a variety of physical systems, from polaron dynamics in solids to complex superconducting quantum architectures.	cond-mat.mes-hall	cond-mat	Particle Production in Ultra-Strong Coupling Waveguide QED Nicolas Gheeraert,1 Xin H. H. Zhang,2 Th´eo S´epulcre,1 Soumya Bera,3 Nicolas Roch,1 Harold U. Baranger,2 and Serge Florens1 1Institut N´eel, CNRS and Universit´e Grenoble Alpes, F-38042 Grenoble, France 2Department of Physics, Duke University, P. O. Box 90305, Durham, North Carolina 27708, USA 3Department of Physics, Indian Institute of Technology Bombay, Mumbai 400076, India (Dated: October 31, 2018) Understanding large-scale interacting quantum matter requires dealing with the huge number of quanta that are produced by scattering even a few particles against a complex quantum object. Prominent examples are found from high energy cosmic ray showers, to the optical or electrical driving of degenerate Fermi gases. We tackle this challenge in the context of many-body quantum optics, as motivated by the recent developments of circuit quantum electrodynamics at ultrastrong coupling. The issue of particle production is addressed quantitatively with a simple yet powerful concept rooted in the quantum superposition principle of multimode coherent states. This key idea is illustrated by the study of multi-photon emission from a single two-level artificial atom coupled to a high impedance waveguide, driven by a nearly-monochromatic coherent tone. We find surprisingly that the off-resonant inelastic emission lineshape is dominated by broadband particle production, due to the large phase space associated with contributions that do not conserve the number of excitations. Such frequency conversion processes produce striking signatures in time correlation measurements, which can be tested experimentally in quantum waveguides. These ideas open new directions for the simulation of a variety of physical systems, from polaron dynamics in solids to complex superconducting quantum architectures. I. INTRODUCTION Exploring the quantum world [1] is an ongoing quest fu- eled by the search for fundamental understanding, which has enabled the creation of unexpected technologies. The advent of lasers and semiconducting microelectronics has indeed crucially relied on building blocks that are de- termined at the microscopic level by quantum effects. Whether intrinsically quantum effects such as entangle- ment can provide further practical scientific developments is at present intensely investigated. However, addressing increasingly complex quantum systems is pushing the boundaries of what simulations can cope with on present day hardware, due to the exponential complexity growth when working with states from the Hilbert space. This question is certainly very acute when dealing with the temporal driving of large-scale quantum circuits, which can lead to a rapid proliferation of propagating quanta. How to encode quantum information efficiently in such a situation, using only available classical computers, is a very general challenge in contemporary physics. Because quantum many-body scattering is relevant for a wide range of physical systems (solid state materials, cold atomic gases, high energy collisions in particle ac- celerators), fruitful concepts are best developed with the relevant physics at hand. For this reason, we focus in this article on the topic of many-body quantum optics, which combines discrete atomic states (the scatterer) with broadband photonic fields (leading to a huge Hilbert space of quanta). Historically, light-matter interaction has been thoroughly studied in the regime of standard quantum op- tics [2, 3], where the combination of small atomic dipoles and perturbative fine structure constant αQED (cid:39) 1/137 leads to small radiative corrections, such as the famous Lamb shift at order [αQED]3 (in units of the atomic fre- quencies). Quantum electrodynamics (QED) corrections to the bare atom picture also control the natural linewidth of atomic transitions [3, 4] associated to vacuum fluctua- tions of the electromagnetic field, occurring also at third order in αQED. As a consequence, the electromagnetic modes that may strongly interact with an atom are limited to those very close to its resonance frequency. A variety of strategies are being pursued in atomic quantum optics in order to enhance the strength of light-matter coupling. First, there is extensive work on confining light to a cavity in order to increase the magnitude of the electric field [1, 5, 6]; however, in this case, interesting effects involving a photon continuum are discarded. Under strong pump- ing, multi-photon non-resonant contributions can become sizable, but this suffers from the same problem of rather limited bandwidth. Finally, several strategies involving photonic crystals or Rydberg atoms are being pursued in which a collective light-matter coupling is made strong by using a large number of weakly coupled components [7 -- 10]. We wish to address, however, regimes where radiative effects become of order one in a system with a single emit- ter and a broad continuum of photonic modes, an area known as ultra-strong coupling waveguide quantum elec- trodynamics (wQED). Access to this regime is becoming possible [11, 12] through circuit quantum electrodynamics in which artificial superconducting atoms interact on-chip with microwave transmission lines (see Refs. [10, 13] for general reviews on the topic); in fact, in cavities, ultra- strong coupling has been achieved in this system [14 -- 19]. In ultra-strong wQED, many-body phenomena are ex- pected to occur that have no counterpart in standard quantum optics [2, 3] or in low-coupling superconduct- 8 1 0 2 t c O 0 3 ] l l a h - s e m . t a m - d n o c [ 3 v 5 6 6 1 0 . 2 0 8 1 : v i X r a ing transmission lines [20 -- 26]. A non-exhaustive list of theoretical predictions includes giant Lamb shifts [27 -- 31], single-photon down-conversion [32, 33], non-RWA transmission lineshapes [28, 34, 35], multi-mode entan- glement [36 -- 38], and non-classical emission [39]. The key element in all of the novel many-body phenomena in ultra-strong wQED is that the number of excitations is no longer conserved because the rotating-wave approxi- mation is not legitimate anymore. It is worthwhile then to focus directly on this non-conservation. We show here that a key signature of scattering or excitation in the ultra-strong regime is broadband photon production: a greater number of photons come out than go in, even in the very low power single-photon excitation regime. In contrast to previous studies which focused on effects that become prominent when the light-matter coupling α reaches values of order one (the so-called Kondo regime), we investigate here many-body effects that are realistically observable when entering the ultra-strong coupling regime, with typically 0.1 (cid:46) α (cid:46) 0.3. These many-body effects are nevertheless dramatic and have the additional advantage that they may be probed experimentally in the very near future. This regime is characterized by a qubit linewidth Γ that is a sizeable fraction of its resonance frequency ∆, owing to the perturbative relation Γ (cid:39) πα∆. For α (cid:28) 1, it is widely believed in the quantum optics context that dominant physical processes are well captured by the so-called rotating wave approximation (RWA), upon which non-resonant transitions are discarded from the outset. While it is true that RWA provides quantitatively accurate results for the linear response of an atom weakly coupled to a waveguide, we find that low-power non-linear scattering properties are however dominated by non-RWA contributions, even for arbitrarily small coupling in which the RWA is thought to become exact. The scenario that we consider is shown in Fig. 1. A right-going coherent state pulse is injected into a waveg- uide. The waveguide and qubit are initially in their ground state, implying that the qubit is non-perturbatively dressed by a cloud of waveguide photons [37]. The incom- ing coherent state pulse then scatters from this dressed state, leading to outgoing transmitted and reflected pulses, that have acquired on general grounds a many-body char- acter [40]. Our goal here is two-fold. First, we uncover new phys- ical effects in non-linear many-body photon scattering by analyzing the photonic content of non-resonant emis- sion spectra. One major observation is that significant non-linear emission arises from both RWA and non-RWA pathways. In light of standard knowledge in quantum optics, it comes as a surprise that non-RWA processes are found to dominate in magnitude the RWA non-linear response when off resonance. Indeed, in the regime of ultra-strong coupling, the linewidth of the qubit broadens substantially, leading to important non-resonant inelastic contributions to the scattering cross-section. Under a drive that is detuned in frequency above the resonance of the qubit, for instance, inelastic down-conversion occurs 2 FIG. 1. Illustration of the setup considered in this paper: a long transmission waveguide (grey horizontal line) is side- coupled to a two-level system, allowing the measurement of multi-photon scattering matrices in a typical two-terminal ge- ometry from the reflection and transmission (outgoing states with arrows pointing outward) of a coherent state Gaussian wavepacket (incoming state with arrow pointing inward). A many-body polarization cloud lives in the central region (tied to the qubit) [37]. Frequency-conversion processes are ex- tracted by a spectral and number state analysis of the outgoing wavepackets. by the "splitting" of an incoming photon into several lower energy ones [32]. For larger power, there are simi- lar processes involving an increasing number of incoming photons, all of which are described by counter-rotating terms. These processes are favored by a wide continuum of available outgoing multi-photon states. The surprising dominance of non-RWA processes can thus be interpreted as a consequence of the larger phase space of outgoing states for particle production. Another dramatic many-body effect is uncovered by studying the correlations in time. We find that ultra- strong coupling leads to striking qualitative signatures in the photon statistics of a single emitter, namely incom- plete anti-bunching on resonance at zero time-delay and strong bunching at finite delay, that are very prominent in the off-resonant case. These previously unrecognized fea- tures are quantitatively different from RWA results and constitute important signatures of particle production from an experimental point of view. A further indica- tion of interesting many-body effects is that perturbative expansions for the elastic and inelastic emission spectra cannot be captured quantitatively. Our second objective is to provide a general and pow- erful simulation toolbox to access the non-linear and inelastic processes involved to any order in the incoming beam power. This methodology relies on an expansion of the full many-body wavefunction (qubit+waveguide) in terms of multi-mode coherent states, using quantum superpositions of several classical-like configurations. It was introduced recently as a numerically controlled tech- nique to capture the ground state [30, 36] or quenched dynamics [39] in ultra-strong coupling wQED. Two orig- inal developments are made in the present manuscript. First, a new and more numerically efficient algorithm is proposed, which allows for the first time to tackle in a controlled way the many-body dynamics in waveguides composed of several thousands of modes. Second, we develop a many-body scattering protocol, that can be k01/¯nx0RINOUTOUTCLOUDused to simulate realistic scattering setups, as shown in Fig. 1, allowing to deal with the challenging problem of many-body particle production in quantum optics. The resulting multi-photon emission processes in the output field are characterized precisely. Despite the coupling being weak to intermediate in magnitude, non-RWA con- tributions to these multi-photon processes open the door to a tremendously large Hilbert space. Typically our calculations manage up to five photons in the outgoing beam, which, for a long waveguide accounting for about 1000 environmental modes, leads to an effective Hilbert space of order 1015 (cid:39) 250. It is quite remarkable that a quantum superposition of classical-like multi-mode co- herent states can be harnessed as an efficient computing resource to address quantum many-body problems that are currently well beyond the reach of any brute force numerical method. Regarding waveguides, several strategies are possible in order to bring these systems into a truly many-body territory, such as using the inductive coupling of a flux qubit to a low-impedance coplanar waveguide transmis- sion line [11, 41], or tayloring a charge-like qubit with a capacitive coupling to a high impedance metamate- rial [28, 32, 37]. For this latter purpose, long chains of Josephson junctions [42 -- 45] constitute a promising platform that is currently under investigation [12] in the context of multi-mode ultra-strong coupling quantum op- tics. In any case, it remains challenging at present to control experimentally a strongly non-linear element con- stituting a true two-level system (such as the Cooper pair box or a flux qubit) that is also very well coupled to a designed environment, because non-linearity brings a high sensitivity to external noise sources. Designs based on a weakly non-linear qubit, such as a transmon [46] ultra- strongly coupled to a waveguide [12, 47] could offer an interesting alternative for high precision measurements, at the expense however of weakening the sought-after non-linear effects. The paper is organized as follows. We first review in Sec. II the basic model of waveguide quantum electrody- namics, and develop a general many-body wavefunction approach for the study of inelastic photon emission by a single two-level system. Section III presents detailed inelastic emission spectra, in connection with the relevant physical processes. Section IV provides a comparison to standard results in quantum optics, based on the RWA, which can only account for processes in which two input photons are inelastically scattered, keeping the number of outgoing photons equal to two. This section closes with a discussion of the temporal correlations of the emitted light, showing several qualitative features of ultrastrong coupling. Finally, the perspectives section, Sec. V, dis- cusses prospects for experimental measurements of these effects in superconducting circuits, and the need for de- veloping further our theoretical tools in order to capture realistic aspects of Josephson waveguides beyond the spin- boson limit. Appendices contain technical derivations that should make the manuscript self-contained, and present details on the new algorithm proposed in this work. 3 II. MANY-BODY COHERENT STATE SCATTERING FORMALISM A. Modeling a two-level system coupled to a waveguide The main assumption that will be made in this study is the restriction of the atom to a perfect two-level sys- tem. This hypothesis is perfectly legitimate for strongly non-linear qubits, such as the Cooper pair box or the flux qubit [11, 20, 48, 49], although these devices typically experience more strongly charge or flux noise compared to a transmon qubit (which is however weakly non-linear). Focusing on a two-level system aims to capture the maxi- mum inelastic scattering cross-sections, due to its intrinsi- cally high non-linearity. It is thus an excellent testbed to examine physics that is already quite rich, and to develop state-of-the-art methodologies in the most challenging sit- uation from a computational point of view. Following this path, a qubit coupled to a full one-dimensional waveguide is quite generically expressed by the so-called spin-boson Hamiltonian (setting to unity): H = ∆ 2 σx − σz 2 (cid:88) k∈R † gfull k (ak + a k) + † ωka kak , (1) (cid:88) k∈R with ∆ the bare splitting of the qubit levels. We stress that we do not work in the qubit eigenbasis here, but rather in a basis that makes the qubit-waveguide cou- pling diagonal, as described by the σz term above (this corresponds for instance to the charge basis for a Cooper pair box that is capacitively coupled to a waveguide). This choice allows a natural description of the driving force behind the entanglement between the qubit and the waveguide, and sets the natural language for our numeri- cal technique based on coherent states. The momentum † dependence of the coupling constant gfull to mode a k of the full waveguide depends on the device geometry and its physical parameters, such as inter-island capacitances, ground capacitances and inter-island Josephson energy. In the case where the waveguide is constructed from a Josephson junction array, Refs. [32, 37, 50] proposed ex- plicit microscopic derivations of the coupling constants based on rather different designs. Similarly, the momen- tum dispersion of the eigenfrequencies ωk of the photonic modes is determined by the microscopic details of the waveguide. k In what follows we will consider for simplicity a lin- ear dispersion relation given by ωk = k (taking the speed of light in the metamaterial c = 1), and a simple parametrization of the coupling constant. For this pur- pose, and in order to simplify the problem, we start by folding the bosonic modes of the full waveguide onto a half-line, by defining even and odd modes: ae k = 1 √2 (ak + a−k) and ao k = 1 √2 (ak − a−k) , (2) (cid:88) so that the Hamiltonian (1) can be rewritten as H = ∆ 2 σx− σz 2 gk(ae k + ae† k ) + ωk[ae† k ae k + ao† k ao k ], k>0 k>0 (3) with the coupling constant to the even modes, gk = √2gfull k . We choose a parametrization of the effective cou- pling constant gk given by the following spectral function: J(ω) = πg2 kδ(ω − ωk) = 2πα ω e −ω/ωc . (4) (cid:88) (cid:88) k>0 This form of spectral function, although not completely generic, contains the main realistic ingredients of the qubit-waveguide interaction, such as a linear ohmic fre- quency dependence at low energy, and a rapid falloff near the plasma edge ωc, that we assume to be exponential in form. For a discretized momentum grid, we deduce that the coupling constant gk to even modes reads: gk = 2 α ωk δk e−ωk/ωc , (5) (cid:112) where δk is the wave-number spacing corresponding to the discretisation of the continuous momentum integral. In the form of Hamiltonian (3), only the even modes are interacting with the qubit, while the odd modes are freely propagating. This allows us to write the state vector ψ(cid:105) as the direct product of the even sector ψe(cid:105) and the odd sector ψo(cid:105): ψ(cid:105) = ψe(cid:105)e ⊗ ψo(cid:105)o = ψe(cid:105)e ψo(cid:105)o (6) provided the initial state can be decomposed accordingly. The dynamics in the odd sector is essentially trivial, while many-body effects have to be considered to capture the dynamics in the even sector, a topic that we address now. B. Many-body quantum dynamics with multi-mode coherent states The rationale behind the multi-mode coherent state (MCS) expansion is as follows. The only source of non- linearity in Hamiltonian (3) is the two-level system, and this non-linearity is transferred from a single degree of freedom (the qubit) to a large number of degrees of free- dom (the modes of the waveguide). A first effect of this coupling is to dress the two qubit states by displacing the oscillators, as is clear from the σz term in Eq. (3). This picture, which is only approximate when a single coherent state displacement is used, becomes quantitavely exact for the many-body ground state when superposing a small set of coherent states [36]. Regarding the quantum dynamics, an input coherent state (as is relevant in our description of 4 (cid:105) Ncs(cid:88) (cid:104) m=1 the scattering problem) remains stable only when turning the coupling to zero (classical-like propagation). At finite coupling, quantum fluctuations of the output field around the dominant classical trajectory are again accounted for by the superposition of additional Gaussian states. The strategy is thus to write the state vector in the even sec- tor as a coherent state expansion, also referred to in the following as the multi-mode coherent state (MCS) ansatz [30, 36, 37]: Ψe(t)(cid:105) = pm(t)fm(t)(cid:105) ↑(cid:105) + qm(t)hm(t)(cid:105) ↓(cid:105) , (7) k ]0(cid:11) where we have introduced the complex and time- dependent amplitudes pm(t) and qm(t) for each qubit component, with m an index that labels the states used in the superposition. These multi-mode coherent states also occur as two discrete sets of states (one for each qubit component): fm(t)(cid:11) = Nmodes(cid:89) e[fk,m(t)ae† k −f ∗ k,m(t)ae (8) k=1 and similar for hm(t)(cid:11). Due to the completeness of the co- herent state basis on a discrete von-Neumann lattice [51], which naturally extends to the case of many modes, this discrete decomposition can target in principle an arbitrary state of the full Hilbert space for Ncs → ∞. However, for a fixed choice of Gaussian states, this leads to the un- fathomable exponential cost that is typical of many-body quantum mechanics. The advantage of the MCS ansatz (7) lies in the variationally optimized time-dependent dis- placements fk,m(t), which allows one to track with high precision and low numerical cost the dynamics of the full state vector. What is truly remarkable about such a multi-component multi-mode wavefunction is the relatively small number of coherent states Ncs that are necessary to capture both the static many-body ground state [36] and the complex dynamics resulting from quantum quenches [39], even deep in the ultra-strong coupling regime. The method works efficiently from the case of single mode cavities [52 -- 54] up to the challenging situation of an infinite continuum [55]. As we will see later, addressing frequency conversion brings an additional difficulty in that non-linear emission signals are extremely faint when driving off resonance compared to the dominant elastic contributions, which requires very careful convergence of the numerics. In principle, the exact Schrodinger dynamics, controlled by the Hamiltonian (3), can be derived from the real Lagrangian density: L =(cid:10)Ψ(t) ←−∂t − HΨ(t)(cid:11), i 2 −→∂t − i 2 δ(cid:82) dtL = 0, upon arbitrary variations of the state vec- by applying the time-dependent variational principle [56], tor (7) with respect to its set of variational parameters. (9) 5 zx = (cid:19) 1 4 (cid:18) 1 2πσ2 This minimization obviously provides Euler-Lagrange equations d dt ∂L ∂ v = ∂L ∂v (10) for the set of variables v = {pm, qm, fk,m, hk,m}, which can be solved by numerical integration [34, 39, 57 -- 59]. The detailed form of the dynamical equations is provided in Appendix A 1. A new numerical algorithm, which supersedes the one proposed in Ref. [39] and allows to deal with up to thousands of modes, is presented in Ap- pendix A 2. The second step in the scattering picture of Fig. 1 is to include a wavepacket (with arrow pointing inward) impinging on the dressed ground state. We will work in what follows with a single coherent state pulse as input, which is realistic in terms of the classical sources used in actual experiments. Let us denote zk the displacement of the incoming wave-packet in mode k of the physical waveguide, and zx its Fourier transform to real space: We choose to use here a Gaussian-shaped wavepacket e+ikxzk. (12) (cid:90) ∞ −∞ dk √2π C. General coherent state scattering formalism zk = √¯n − (k−k0)2 4σ2 e −i(k−k0)x0 e −ik0x0/2, (13) e Now that we have obtained exact dynamical equa- tions for the time evolution under the spin-boson Hamil- tonian (1), we need to prepare our initial state in or- der to perform scattering simulations according to the scheme in Fig. 1. The generic difficulty is that the qubit is dressed non-perturbatively by a cloud of pho- tons [27, 29, 31, 33, 36, 37] in the ultra-strong coupling regime, so that this ground state assumes a many-body character. Thanks to the MCS ansatz introduced in eq. (7), we can efficiently express the static ground state of the joint qubit and waveguide system in terms of multi- mode coherent states: cs(cid:88) N GS (cid:104) f GS m (cid:105)e ↑(cid:105) − − f GS (cid:105) m=1 (11) pGS m 0(cid:105)o, m (cid:105)e ↓(cid:105) ΨGS(cid:105) = where we enforced the Z2 spin-symmetry of the spin-boson Hamiltonian (3) to simplify the expression. We have also used the fact that the odd modes do not interact with the qubit, so that the ground state displacements include only even modes in Eq. (11), and the odd modes are placed in the vacuum state. By implementing numerically a variational optimization [30, 36], one can determine the set of weights pGS m and displacements f GS k,m, and thus obtain a nearly exact result for the ground state up to negligible numerical error. Conveniently, only a small number of coherent states N GS cs , typically less than 10, are required in the realistic domain of parameters of the spin-boson model. corresponding to a signal initially centered around po- sition x0 in the waveguide, with mean wavenumber k0, spatial extent 1/σ, and total intensity corresponding to ¯n photons on average, as illustrated on Fig. 1. The associated real space wavepacket is then (cid:19) 1 (cid:18) 2σ2 zx = √¯n (cid:82) ∞ −∞ dxzx2 =(cid:82) ∞ π 4 −(x−x0)2σ2 e+ik0(x−x0)e+ik0x0/2. e (14) Note that these amplitudes are both normalized so that −∞ dkzk2 = ¯n. The even and odd parts of the incoming wavepacket are then defined strictly for k > 0 as ze k = 1 √2 (zk + z−k) and zo k = 1 √2 (zk − z−k) . (15) Since even and odd modes commute, we can then define a displacement operator D(z) for the initial incoming wave-packet which verifies: ze, zo(cid:105) = D(z)0(cid:105) = D(ze)D(zo)0(cid:105) k>0(zo k −ze∗ k>0(ze kae† k ae (cid:80) (cid:80) k)e = e (16) kao† k −zo∗ k ao k) 0(cid:105) . The final step in the initialization of the state vector is to combine the incoming wavepacket coherent state zk with the displacements entering the full many-body ground state (11). Straightforward calculations, shown in Appendix A 3, lead to the following explicit expression for the input state: ΨIN(cid:105) = cs(cid:88) N GS m=1 pGS m 1 2 −↓(cid:105) e (cid:80) (cid:104) ↑(cid:105) e (cid:80) k>0(−ze 1 2 (cid:80) k,m)e (cid:80) k>0(ze kf GS∗ k,m −ze∗ k f GS k>0[(f GS k,m+ze k)ae† k −(f GS ∗ k,m+ze k) ae k] k](cid:105) kf GS∗ k,m +ze∗ k f GS k,m)e k>0[(−f GS k,m+ze k)ae† k −(−f GS ∗ k,m+ze k) ae 0(cid:105)e zo(cid:105)o . (17) The many-body scattering theory thus amounts to use state (17) as the initial condition for the dynamical equa- tions of motion (10) performed in the even sector, see Eqs. (A1)-(A2) for their full explicit form. During the dynamics, as the incoming wavepacket impinges on the qubit, the necessary number of coherent states Ncs will sensibly grow from the initial value N GS cs due to non- classical emission, therefore requiring to add extra coher- ent states to the state vector when needed (the procedure is detailed in Appendix A 5). In the odd sector, which is completely decoupled from the qubit, the related dis- placements are trivially evolving in time according to i zo k, and a single coherent state is enough for the whole time-evolution. k = ωkzo After a given time T long enough to ensure interac- tion of the wavepacket with the qubit and subsequent decoupling of the two outgoing wavepackets from the many-body cloud surrounding the qubit (in the reflection and transmission channel of the full 1D waveguide), one expects on general grounds (since the spin-boson model is non-integrable with a realistic dispersion) a factorization of the final wavefunction as: Ψ(T )(cid:105) = ΨGS(cid:105) ⊗ ΨOUT(cid:105) , (18) where ΨGS(cid:105) is the many-body ground state of the spin-boson model and ΨOUT(cid:105) a many-body outgoing wavepacket that contains a non-trivial decomposition of the emitted signal in terms of a large number of multi- mode coherent states (typically N OUT cs(cid:88) N OUT Nmodes(cid:89) cs ∼ 20 − 30): k,m ae† k −f OUT∗ k,m ae k ]0(cid:11). e[f OUT ΨOUT(cid:105) = pOUT m m=1 k=1 m (19) The extraction procedure for the outgoing weights pOUT and displacements f OUT k,m is given in Appendix A 3. The factorization property (18) occurs because the spin-boson model (with a macroscopic number of modes) is a truly dissipative system, always showing a path for relaxation. In practice, this hypothesis can be checked from the nu- merical calculations by observing that the dressed qubit does not show correlations with the outgoing photons. In- deed, any observable of the qubit relaxes back to its initial equilibrium value at the end of the scattering protocol. Also, the nature of the scattered photons does not depend on how one traces out the qubit density matrix. We now proceed to the analysis of the transmission and spectral properties of this scattered many-body wave-packet. III. MULTIPHOTON INELASTIC SCATTERING A. Elastic emission and high power saturation As a first illustration for our dynamical many-body scattering method, we investigate the elastic reflection as a function of the frequency and power of the incoming signal. This problem is particularly challenging because of the combination of non-perturbative ultra-strong cou- pling with non-equilibrium effects that arise at finite input power. Ultra-strong coupling scattering at non-vanishing power has been addressed previously with approximate 6 techniques [28, 34, 35, 49] and with more advanced nu- merical methods [29, 33]. However, systematic extraction of many-body scattering matrices has not been performed to our knowledge. Our calculation scheme proceeds similarly to an ex- perimental setup: the incoming Gaussian coherent-state wavepacket, shown schematically as the incoming distri- bution of photons in real space in Fig. 1, is initialized to the left of the qubit. The qubit is placed at position x = 0 as seen from its sharply decreasing photonic cloud [37] which is present in both the input and output ports, but remains statically bound to the central impurity. After propagation towards the qubit and subsequent interac- tion, the photon flux decouples at long times, and is separated into a reflected left going (k < 0) signal and a transmitted right going (k > 0) signal, both shown with arrows pointing outward in Fig. 1. Note that in all the simulations made in this paper, we have consid- ered the linewidth σ of the wavepacket in k-space to be smaller than the qubit linewidth Γ (in order to achieve high spectroscopic resolution), but large enough to keep the simulations on a reasonable system size (typically we consider from Nmodes = 1000 to Nmodes = 3000 modes for the chain used in the even sector). All calculations are done in the units of the plasma frequency ωc as defined in the spectral density (4), and the wavepacket linewidth appearing in Eq. (13) is taken as σ = 0.005ωc, unless indicated otherwise. We define the reflection and transmission coefficients (cid:80) (cid:80) † k<0 (cid:104)a kak (cid:105)out † k>0 (cid:104)a kak (cid:105)in , in the following way: (cid:80) (cid:80) † k>0 (cid:104)a kak (cid:105)out † k>0 (cid:104)a kak (cid:105)in T = and R = (20) where we have denoted (cid:104). . .(cid:105)in the average over the state vector corresponding to the coherent incoming wave- packet before scattering, and (cid:104). . .(cid:105)out the average over the many-body outgoing wave-packet after scattering. Both are obtained from the full state vector (19) by simply fil- tering out in real-space the polarization cloud associated with the ground state, as explained in Appendix A 3. Results for different values of the incoming power are shown in Fig. 2. The probability of reflection generally increases on resonance; indeed, for elastically scattered photons, interference effects cause almost complete re- flection when exactly on resonance. For small values of the incoming power (¯n = 0.01 and ¯n = 0.1), for which the initial coherent state wavepacket has a very small probability of containing Fock states with more than 1 photon, one can note that the reflection only reaches R (cid:39) 0.9 at peak value. This incomplete reflection of the photons arises from the finite linewidth of the incoming wavepacket, and not from inelastic losses. Since our in- coming Gaussian pulse is not perfectly monochromatic, the modes at the edge of a resonant incoming beam (cen- tered at k0 = ∆R) are slightly off-resonant and do not get fully reflected by the qubit. Even in the present case of a relatively small light matter coupling α = 0.1, many-body 7 FIG. 2. Saturation effects in the reflection coefficient Eq. (20) as a function of incoming wavevector k0, for three different amplitudes ¯n = 0.01, 0.1, 2.0 of the input, with a wavepacket width σ = 0.005ωc. These curves correspond to converged numerical data with up to Ncs = 16 coherent states in the MCS wavefunction (7). The bare qubit frequency is ∆ = 0.1ωc and the dimensionless light-matter interaction is α = 0.1, leading to a sizeable renormalized qubit frequency ∆R (cid:39) 0.08ωc. effects due to the ultra-strong coupling are apparent in the reflection curve of Fig. 2. First, a non-Lorentzian asymmetric lineshape is obtained, with a high energy tail more prominent than at low energy. In addition, we clearly observe a substantial renormalization of the qubit frequency ∆R (cid:39) 0.08ωc from its bare value ∆ = 0.1ωc. For higher incoming power, one physically expects satu- ration effects to take place, and these are clearly evidenced by the curve with average number of photons ¯n = 2.0 in Fig. 2. We stress that converging such computations in the high power regime is quite challenging, and approxi- mate techniques such as a single coherent state truncation lead to uncontrollable noise levels, as found in previous work [34]. We show in detail in Appendix A 4 that the reflection curve converges smoothly at ¯n = 2.0 for about Ncs = 16 coherent states in the MCS state vector (7). This is also confirmed by a systematic control of the error, as done previously for quantum quench protocols [39]. B. Off-resonant frequency-conversion spectra We now turn to analyzing the emitted radiation in the off-resonant case, in which the system is excited at a frequency k0 above the renormalized qubit transition frequency ∆R. A typical inelastic spectrum is shown in Fig. 3, here for k0 = 0.16ωc, ∆R = 0.08ωc, and an injected ¯n of 0.5. The stronger transmission relative to reflection (upper panel) simply reflects the off-resonant situation k0 ≈ 2∆R, in agreement with the reflection curve in Fig. 2. The vertical scale is expanded in the lower panel, so that the inelastic contributions are made apparent at the foot of the large reflection and transmission elastic peaks located at ±k0. Note that the actual linewidth of this elastic peak, FIG. 3. Mean density of photons at momentum k in the outgoing wave-packet (the incoming wavepacket is displayed as a dashed line); bottom panel is a zoom of top. The total outgoing signal (gray shaded) is decomposed into Fock resolved excitations (in full lines) with N = 2 (bottom curve), N = 3 (top curve) and N = 4 (middle curve) photons, in order to highlight the processes of Fig. 4. The parameters for this simulation are the incoming wave-packet wave-number k0 = 0.16ωc and linewidth σ = 0.005ωc, the mean photon number ¯n = 0.5, the qubit bare energy ∆ = 0.1ωc, and the coupling strength α = 0.1. Simulations were performed by considering a wavefunction containing Ncs = 30 coherent states, and Nmodes = 1200 modes. set by σ = 0.005ωc, is in fact much smaller than what the lower panel seems to indicate, because the maximum peak amplitude is 2000 times higher that the scale of the graph. The gray-shaded curve displays the expectation value of † the total number of outgoing photons (cid:104)a kak (cid:105)out while the dashed line indicates the total number of incoming † photons (cid:104)a kak (cid:105)in centered around k0. The first striking result is the broad spectrum of emission extending from the qubit frequency ∆R all the way down to k = 0. The full lines display how the total outgoing photon contribution is distributed among different Fock states N(cid:105) with photon number N = 2, 3, 4, allowing us to assess the nature of particle production. Our method to obtain these photon-number resolved amplitudes by considering the probability of all the possible single- and multi-photon states for a given momentum k is explained in Appendix A 6. Note that the majority of the inelastic emission involves 3 and 4 photon contributions. Since the incoming average photon number is only 0.5, clearly substantial particle production is occurring! Both the broad inelastic spectrum and particle production are quintessentially ultra-strong coupling phenomenona. 8 FIG. 4. Diagrammatic representation of some non-linear pho- ton processes occurring during scattering onto a two-level sys- tem. Panel (a) is a RWA frequency exchange process restricted at the two photon level, which shows two sharp emission lines at ∆R and 2k0 − ∆R (due to energy conservation). Panel (b) shows the non-RWA one photon to three photon conver- sion [32], which leads to a broad emission continuum, sharply peaked at the resonance ∆R in the case of an off-resonant drive. Panel (c) is a similar particle producing process, now with two input photons, one being down converted to three photons, the other photon being elastically scattered. For an initial understanding of the various contributions to this spectrum, we consider a schematic diagrammatic perturbation theory as shown in Fig. 4. For two incoming photons, an inelastic RWA process can occur by distribut- ing the total incoming energy 2k0 into a resonant photon at ∆R and another at 2k0 − ∆R as shown in panel (a). However, for a single incoming photon with momentum k0, since the emission is still maximum at the (renormal- ized) resonant qubit frequency ∆R, an excess energy of k0 − ∆R must be distributed between two extra outgoing photons (in order to properly relax to the ground state). The accessible non-resonant states thus lead to the non- RWA 3-photon emission process shown in panel (b). In general, the two extra photons that are produced are not resonant, and the amplitude of the total process is sizable only because of the ultra-strong coupling regime. Indeed, the elastic reflection curve of Fig. 2 is spectrally very broad, and emission does not necessarily occur strictly on resonance. The non-RWA nature of the particle production process is obvious from the non-conservation of excitations: the middle outgoing arrow in Fig. 4(b) corresponds to the emission of a photon upon excitation of the two-level system (instead of the usual de-excitation). Four photon production is also displayed in panel (c) for an input state with two photons. In this case, one input photon is elastically scattered at k0, while the second input photon splits into three photons similar to the process in panel (b). Since the RWA 2 → 2 process in panel (a) and the non- RWA 2 → 4 process in panel (c) come at the same order FIG. 5. Frequency conversion spectrum by going deeper in the ultra-strong coupling regime (α = 0.1, 0.2, 0.3, bottom to top curves) for an off-resonant incoming wavevector k0 = 0.2ωc. Parameters are otherwise taken as in Fig. 3. Although the res- onance frequency ∆R and inelastic linewidth γinel R were fitted, one observes increasing deviations to the fitting formula (B1) at larger α. Enhanced scattering of low-energy modes, pre- cursor of the Kondo regime, originate from non-perturbative many-body corrections beyond the lowest order perturbation theory of Ref. [32]. in the input power, they can be used to directly compare the relative strength of RWA and non-RWA processes. All three processes of Fig. 4 are clearly observed in the spectrum shown in Fig. 3, as the emission amplitude is decomposed into photon number states N = 2, 3, 4. In view of the wide use of the RWA in the quantum optics context, the main surprise in these results (to be discussed in more detail below) is that non-RWA processes strongly dominate in amplitude the RWA processes. Some of the off-resonant processes were previously pre- dicted perturbatively by Goldstein et al. [32] in the α → 0 limit and at the Toulouse limit, and we are able to char- acterize quantitatively the non-linear emission for the first time at finite α values, as seen in Fig. 5. The main effect brought by stronger coupling is a further renor- malization of the spontaneous emission line ∆R down to lower values, as well as a global increase of the probability for inelastic conversion. Interestingly, we find that the perturbative formula (B1) cannot quantitatively describe our data anymore in this regime, even when allowing to fit the inelastic linewidth. Perturbation theory thus fails to capture the pile-up of low-energy photons found in the numerical simulations, which signals the approach to the incoherent Kondo regime, in which the qubit reso- nance is fully washed out. A detailed study of non-linear spectra as a function of incoming momentum is given in Appendix B 1. k0gigik0k0R⌦k0R⌦k0gigieiRk0k0eigigigigik0k0k0gi(a)(b)(c)eieigigieieieigi2k0RRWAbeyond RWAbeyond RWAk1k2k0k1k2k1k2k0k1k2k3k39 FIG. 6. Left panel: probability distribution of the two-photon states αk1,k22/dk2 corresponding to the 2-photon curve in Fig. 3. The clear inelastic sidebands correspond to frequency exchange between 2 photons. Right panel: probability density distribution αk1,k2,k3=∆R2/dk3 of the three-photon states in which one photon is at the resonance frequency k3 ≈ ∆R, corresponding to the 3-photon curve in Fig. 3. The emission continuum associated to particle production is revealed by the diamond-shaped line. C. Particle production processes We now investigate more precisely the photonic content of the emitted radiation in the inelastic channel. Let us start with the 2-photon particle-conserving RWA contribu- tion (bottom full line N = 2 in Fig. 3) forming two lobes symmetrically arranged around the main elastic peak (at k0 = 0.16ωc). The lowest energy lobe is centered around k (cid:39) 0.08ωc (cid:39) ∆R corresponding to the spontaneous ree- mission of the qubit, while the high energy lobe is located around k (cid:39) 0.24ωc (cid:39) 2k0 − ∆R, as expected from energy conservation [panel (a) in Fig. 4]. A closer view into this two-photon joint emission process is given by the com- plete two-photon probability distribution αk1 k22 that is plotted in the top panel of Fig. 6 (see Appendix A 6 for details). The main 2-photon elastic peaks are the white disks located at [±k0,±k0], that have been cut off in order to magnify the small inelastic contributions. From the lateral inelastic peaks, one can immediately read-off the two-photon frequency conversion process in which two photons with energy k0 redistribute their energy into one photon with momentum ∆R and another with energy 2k0 − ∆R. The inelastic spectrum originating from the conversion of a single incoming photon into three outgoing photons, with probability (cid:104)nk(cid:105)3photon of measuring one of these photons at energy k, is represented by the middle full line in Fig. 3. This inelastic lineshape presents quite unusual features: a sharp resonance at the qubit frequency ∆R, a broad continuum extending from zero energy up to the foot of the elastic peak, and a small lobe at the same energy 2k0 − ∆R as the previous two-photon conversion process. The latter is easily understood as an input of three photons with momentum k0, out of which one photon is elastically scattered, while the other two are RWA frequency converted to ∆R and 2k0 − ∆R (similar to the previous 2 → 2 RWA process). We have checked that this 3 → 3 RWA process becomes relatively weaker in amplitude as the input power ¯n is turned down, and is indeed associated to a three-photon input. The broad low-energy continuum is readily explained by the 1-photon to 3-photon non-RWA conversion process shown in panel (b) of Fig. 4. This interpretation is backed up by studying in the right panel of Fig. 6 the probability distribution αk1k2k3=∆R2 of 3-photon outgoing states for which one of the three outgoing modes is resonant, k3 = ∆R. To understand this diamond-shaped pattern, one can observe that the process leading to the diagonal line in the top-left quadrant can easily be parametrized as: k0(cid:105) → ∆R(cid:105)Ω(cid:105)k0 − ∆R − Ω(cid:105) with Ω ∈ [0 : k0 − ∆R]. which basically expresses the conservation of energy be- tween the input and the output. Note that in the on- resonant situation (or for a drive at frequency below ∆R), all emitted photons present energies below the qubit fre- quency. The next section investigates how the complete inelastic emission spectra compare with the standard RWA pre- diction in quantum optics. This comparison will provide not only a benchmark of our simulations, but also several physical signatures that cannot be captured without the inclusion of particle production processes. 1e-3Type to enter text2k0R<latexit sha1_base64="IlbOT2QklPUrr+wLCpZNth27AbI=">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</latexit><latexit sha1_base64="IlbOT2QklPUrr+wLCpZNth27AbI=">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</latexit><latexit sha1_base64="IlbOT2QklPUrr+wLCpZNth27AbI=">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</latexit><latexit sha1_base64="IlbOT2QklPUrr+wLCpZNth27AbI=">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</latexit>R<latexit sha1_base64="2KkK8gRAjr7ETFgZeCvjHMvrA/g=">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</latexit><latexit sha1_base64="2KkK8gRAjr7ETFgZeCvjHMvrA/g=">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</latexit><latexit sha1_base64="2KkK8gRAjr7ETFgZeCvjHMvrA/g=">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</latexit><latexit sha1_base64="2KkK8gRAjr7ETFgZeCvjHMvrA/g=">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</latexit>IV. SUCCESS AND FAILURE OF THE RWA FOR NON-LINEAR EMISSION A. RWA inelastic conversion To highlight particle production that arises at ultra- strong coupling, we now compare our MCS simulations to a direct treatment within the RWA, an approximation which conserves the number of excitations. Transport under the RWA is obtained in the framework of input- output theory. Within the RWA, it is convenient to work in the basis that diagonalizes the qubit. After applying the rotating wave approximation to the Hamiltonian (1) and assuming a frequency-independent coupling constant gk = √α∆R, one finds that the system is described by the Hamiltonian (cid:90) dω ω(cid:0)r ∆σz + dω H = 1 2 (cid:90) + (cid:2)σ+(rω + lω) + h.c.(cid:3) g 2 (cid:1), † ωrω − l † ωlω (21) (cid:112) where σ+ is the raising operator of the qubit and rω(lω) is the annihilation operator for the right(left)-going mode of frequency ω. We adapt standard input-output theory for a monochromatic input [60 -- 63] to our case of an incoming wavepacket with finite energy resolution. The input-output relation remains the usual one, rout(t) = π/2 g σ−(t) and similarly for the left-going rin(t) − i field lin/out. This allows one to find the properties of the outgoing field from a master equation for the qubit. In this way the power spectrum is calculated through the first- † order correlation function G(1)(t1, t2) = (cid:104)a out(t1)aout(t2)(cid:105) by a Fourier transform (cid:90) T (cid:90) T S[ω] = 1 2π dt1 0 0 dt2 G(1)(t1, t2)eiω(t2−t1). (22) We assume that the qubit is located at x = 0 while the input and output ends are located at x = −T /2 and T /2 respectively (c = 1). From the definition (13), we can write the wavepacket in frequency as z(ω) = √¯n (cid:18) 1 (cid:19) 1 in(ω) − h.c.(cid:3) z+(cid:105) = exp(cid:2)(cid:82) dk z(ω) r − (ω−k0)2 e 2πσ2 4σ2 4 † through which the input coherent state is defined as † in(ω) is the standard monochromatic input operator [60 -- 63] of input-output theory. The input operator describing our wavepacket then satisfies 0(cid:105), where r ei(ω−k0)T /2, (23) A(t)e 1 √2π −ik0t z+(cid:105) and lin(t)z+(cid:105) = 0, rin(t)z+(cid:105) = (24) with A = √2¯nσ(2π)1/4 is where A(t) = Ae−σ2(t−T /2)2 the change of driving amplitude on the qubit with time as the Gaussian wavepacket passes by. A master equation for the qubit density matrix ρs is then obtained by transforming to the Schrodinger picture, ∂ ∂t ρs = − i(cid:2) δ + πg2(cid:0)σ−ρsσ+ − 2 (cid:3) 1 2{ρs, σ+σ−} (cid:1), σz + gA(t)σ+ + h.c. , ρs 10 (25) (cid:90) T (cid:90) T−t dt −t 0 where a rotating frame given by k0σz/2 has been used. Note that decay rate is Γ = πg2 = πα∆R. For the reflected light, the power spectrum can be shown to be SL[ω] = g2 dτ (cid:104)σ+(t)σ−(t + τ )(cid:105) ei(ω−k0)τ ; (26) two additional interference terms appear in the power spectrum for the transmitted light and are not given here. The desired correlation function (cid:104)σ+(t)σ−(t + τ )(cid:105) can be calculated through the master equation (25) and the quantum regression theorem [64]. FIG. 7. Comparison of MCS simulations to RWA input-output theory with regard to frequency conversion spectra for the off-resonant case (upper panel) and on-resonant case (lower panel). The parameters are the same as in Figs. 3 and 12. This confirms the previous interpretation that RWA inelastic processes dominate only on-resonance, and miss the main contributions to the off-resonant signal. Comparison to the RWA power spectrum in the off- resonant and resonant cases is shown in the upper and lower panels of Fig. 7 respectively. In making this com- parison, we used as input to the RWA calculation the numerically found renormalized level spacing and width, ∆R and ΓR = πα∆R, as this is essential to get the elastic peak correctly. The dominant inelastic process within the RWA is the scattering of two incoming photons into two outgoing photons, see panel (a) in Fig. 4. One sees that this process explains most of the total scattered spec- trum in the resonant case. Indeed, the lower panel in Fig. 7 shows that the RWA and numerical MCS results are nearly identical on the scale shown. In particular both the overall width and shape of the inelastic power spectrum agree well. However, it is clear that the RWA prediction is only a small fraction of the total inelastic scattering in the off-resonant case (upper panel in Fig. 7), as particle production leads to qualitatively different and much larger cross-sections. Thus for these parameters, the RWA fails badly, even though the coupling constant α = 0.1 is not very large. B. Temporal correlations associated to particle production It is interesting to study photon number temporal cor- relations, a standard measure of non-linearities, but now in light of the large inelastic effects that we uncovered in the ultra-strong coupling regime. We have computed the photon-number autocorrelation function g2(τ ) of the reflected signal (x < 0, k < 0), defined by † g2(τ ) = (cid:104)a† x+τ ax+τ ax(cid:105) xa † † xax(cid:105) x+τ ax+τ(cid:105)(cid:104)a (cid:104)a , (27) where x is a point within the left-going wavepacket, such that both x and x + τ are within the wavepacket. In principle, g2(τ ) also depends on x, but this dependence is weak provided the wavepacket is almost monochromatic, and the location x is taken deep within the outgoing photon wavepacket. Details of the computation in the context of an MCS expansion are given in Appendix A 7. We find that temporal correlations are a very sensitive measure of ultra-strong coupling effects. In the resonant case (see the top panel of Fig. 8), the correlations are typical of single photon emission. The comparison to the RWA is globally quantitative, as expected from the pre- vious agreement in the inelastic spectrum on-resonance (small oscillations at long time in g2(τ ) reflect the im- proper convergence of our MCS numerics near the edges of the outgoing wavepacket). In disagreement with the RWA however, we notice that the numerical data shows partial antibunching at zero delay, g2(0) > 0, signaling the production of particles, as was revealed by the low energy spectrum in Fig. 12. Thus particle production leads to physical effects that are potentially observable experimen- tally even when on resonance. This offset, which is zero in the RWA, is found to increase with α (see the upper mid- dle of Fig. 8 for k0 = ∆R). The incomplete cancellation 11 FIG. 8. Second order correlation function g2(τ ) in reflection, at α = 0.1 (upper panel) and α = 0.2 (middle panel). The dip at τ = 0 is a standard signature of anti-bunching, but multiple photon emission seen in Fig. 3 at ultra-strong coupling leads to an incomplete cancellation, g2(0) > 0. In the off-resonant case (k0 = 1.2∆R), particle production is enhanced relative to the single photon reflection, resulting in a stronger bunching (g2(τ ) > 1) than predicted in the RWA. The MCS simulations were performed with the same parameters as in Fig. 12, except for the stronger coupling α = 0.2, σ = 0.004ωc and a hard cut-off that was used (see Appendix A 7). Bottom panel: Real space probability distribution of the two-photon states αx1,x22/dx2 at the initial (left panel) and final (right panel) times of the simulation. One remarks the absence of reflection for two photons arriving at the same time on the qubit, as seen by the dip within the probability distribution located in the bottom left quadrant of the right panel. here can be readily interpreted as a probability of emitting many-photon states due to frequency down-conversion. Even more striking is the appearance of a large bunching signal at intermediate times in the off-resonant case (see the middle panel of Fig. 8 for k0 = 1.2∆R), which was not reported to our knowledge for the radiation of a single level qubit (bunching can be observed in spontaneous emission from multilevel atoms [65], due to a simpler cas- cade effect [66], or from multiqubit systems [67 -- 69]). Here, bunching originates from the single-shot emission of three photons by the two-level system, a property that is only allowed at ultra-strong coupling. The bunching signal becomes sizable in the off-resonant case, even though the particle production is comparable to that in the resonant case, because the reflection amplitude for single photon emission is reduced. As a nice illustration of the partitioning of the incoming beam by the two-level system, we show in the bottom panels of Fig. 8 the real space probability distribution of the two-photon states αx1,x22/dx2 at the beginning and at the end of the time evolution. These results were obtained in the on resonant case with α = 0.1, by Fourier transforming to real space the k-space displacements. One can clearly see within the reflected signal (bottom left quadrant in the right panel) a deep trench on the diag- onal x1 = x2 < 0 with vanishing photon content (the incoming coherent state is shown in the bottom left panel for comparison). Two photons impinging simultaneously on the qubit have thus very low likelihood of both being reflected. This provides a direct visualization of photon anti-bunching, which arises because a single emitter can only reflect one photon at a time. V. CONCLUSION AND PERSPECTIVES In this work we have developed a powerful methodol- ogy, namely the MCS technique, based on multi-mode and multi-configuration coherent state wavefunctions, to address many-body scattering properties of a two-level system that is embedded in a waveguide in the regime of ultra-strong coupling. This problem is intrinsically non-perturbative in nature due to the large production of particles, and cannot be reliably addressed by standard methods in quantum optics. Our main finding is that excitation-preserving processes, described by the rotating wave approximation (RWA), dominate the inelastic spectrum only in the resonant situation. In contrast, when the frequency of the incom- ing photons is larger than the renormalized transition frequency of the two-level system, particle production be- comes very favorable and dominates the inelastic signal. We have been able to characterize precisely the output field, by decomposing the reflected and transmitted pho- ton wavepackets into Fock states, and also by computing temporal correlations. The main results are as follows. (i) The process by which one photon is absorbed and three photons are emitted dominates in the off-resonant low power limit and leads to a broad spectrum of emis- sion extending from zero frequency to the renormalized qubit frequency. (ii) Even in the resonant case, while the dominant inelastic emission near the resonant frequency is captured by the rotating wave approximation, there is still a broad spectrum of weak inelastic transmission produced by the counter-rotating terms. (iii) The corre- lation function g2(τ ) in reflection is a sensitive measure of ultra-strong coupling physics. In particular, particle production implies that it needs not vanish at zero delay, g2(0) > 0, and that it shows a strong bunching effect at a delay of order the inverse lifetime. (iv) Finally, we 12 have found that perturbative predictions for the inelastic response [32] cannot be used simply by renormalizing the bare qubit resonance frequency and linewidth when the coupling becomes ultrastrong. A more consistent theory including self-energy effects should be developed for the future. All our quantitive predictions have relevance for the ongoing experimental effort in pushing waveguide quan- tum electrodynamics to the ultra-strong coupling regime. The connection to future experiments opens in addition various research directions. One important issue is that superconducting qubits are rarely operated as truly per- fect two-level systems. Reducing the non-linearity of the qubit is typically important to minimize the effect of ran- dom noise from the circuit, but this strongly diminishes of course the amplitude of the interesting non-linear signals. Thus, extending our methodology to fully realistic super- conducting quantum circuits will be crucial to address whether particle-production can be sizeable in practice. The ability of the multi-coherent state method to deal naturally with coherent state pulses and open environ- ments is also relevant for the large interest in quantum manipulation within complex architectures. It would thus be very useful to adapt techniques from signal treat- ment in order to numerically optimize the quantum evo- lution of the displacements that are used to simulate the Schrodinger dynamics of the complete system. Such devel- opments will certainly be useful, because the description of strongly driven open quantum systems is a very impor- tant topic currently. Based on the physical artifacts that we can observe in our simulations of the scattering prob- lem when the wavefunction is far from being converged, we suggest that the description of non-linear effects in quantum circuits for arbitrary pulse sequences is a very delicate subject that has to be examined with advanced and reliable many-body techniques. ACKNOWLEDGMENTS We thank Manuel Houzet and Izak Snyman for stimu- lating discussions. N. G. acknowledges support from the Fondation Nanosciences de Grenoble. S. B. acknowledges support from DST, India, through Ramanujan Fellowship Grant No. SB/S2/RJN-128/2016. H. U. B. acknowl- edges travel support from the Fondation Nanosciences de Grenoble under RTRA contract CORTRANO. The work at Duke was supported by U.S. DOE, Division of Materials Sciences and Engineering, under Grant No. de- sc0005237. S. F. and N. R. are supported by the ANR contract CLOUD (project number ANR-16-CE24-0005). pmκmj Mjm, (A1) j Appendix A: Technical aspects of the simulations 1. Dynamics of the MCS state vector The multi-mode coherent state decomposition (7) leads to compact Euler-Lagrange equations (10) that determine the full quantum dynamics the spin-boson model (1): (cid:19) 1 2 pm − (cid:18) (cid:18) (cid:16) fk,mf (cid:63) pm fk,m + (cid:18) k,m + f (cid:63) (cid:88) (cid:88) (cid:88) m m k Pj = F k j = κij = (cid:19) (cid:19) (cid:17) . 1 2 fk,m pmκmj pm − k,mfk,m − 2f (cid:63) (cid:80) k[fk,i2+fk,j2−2f − 1 fk,m k,j 2 (cid:17) (cid:16) ∂E ∗ k,ifk,j ] corre- Here, Mij = (cid:104)fifj(cid:105) = e sponds to the overlap between two multi-mode coherent states, and arises in the equations because of the over- completeness of the coherent state basis. Identical equa- tions (up to a minus sign in all terms containing gk) are obtained for the variables qn and hk,n. We have denoted respectively in Eq. (A1) and Eq. (A2): Pj = −i ∂E and j = −i ∂E j − i F k (cid:16) (cid:88) the average energy, which reads explicitely: (cid:16) (cid:88) (cid:88) ∗ nqm(cid:104)fnhm(cid:105) + pmq ∗ (cid:16) npm(cid:104)fnfm(cid:105)W f ∗ npm(cid:104)fnfm(cid:105)Lf p ∗ nqm(cid:104)hnhm(cid:105)W h ∗ nqm(cid:104)hnhm(cid:105)Lh f k j , with E = (cid:104)ΨHΨ(cid:105) ∗ n(cid:104)hnfm(cid:105) (cid:17) (cid:17) j + ∂E p(cid:63) ∂pj nm − q n,m 1 2 nm + q (cid:17) ∆ 2 E = ∂f (cid:63)k − ∂p(cid:63) j ∂p(cid:63) j n,m pj nm nm + p p 2 , nm =(cid:80) nm =(cid:80) (cid:80) (cid:80) k>0 ωkh∗ k>0 gk(h∗ k>0 ωkf∗ where we have defined W f nm = k,nhk,m, Lf nm = k,n + hk,m). We now proceed with the imple- mentation of a new and efficient numerical solution of the dynamical equations. k,nfk,m, W h k,n + fk,m), Lh k>0 gk(f∗ (A4) n,m 2. New integration algorithm One can note that the dynamical equation (A2) for the displacement field fk,m(t) is not yet in the proper form where a unique time derivative fk,m(t) is extracted on one side of the set of equations. Achieving such a decomposition is required for efficient time integration, but considering that the system under study will require Nmodes (cid:39) 1000 (for accurate spectral resolution) and Ncs (cid:39) 40 (for convergence of the quantum many-body state), a brute force inversion of equations (A1-A2) would scale prohibitively as (Ncs × Nmodes)3 operations for each time step. A more efficient algorithm, allowing to cope with a few hundred modes was proposed in Ref. [39], used 13 an inversion technique with only (Ncs)6 operations, which is favorable provided Ncs (cid:28) Nmodes. We present here an improved version of this algorithm, which enables us to reach the realistic situation of several thousands of modes. The first step is to multiply Eq. (A1) by M−1, with M the overlap matrix Mij = (cid:104)fifj(cid:105): M −1 nj Pj = pn − pmκmjM −1 mj Mjn (A5) (cid:88) (cid:88) mj pn 1 2 1 2 Mjm, (A2) + (A3) = pn − (cid:88) mjq ( fq,mf (cid:63) q,m + f (cid:63) q,mfq,m) q −1 nj Mjmf (cid:63) q,j pmM fq,m, (A6) (cid:88) (cid:88) and similarly for Eq. (A2): −1 nj F k M j = pn fk,n + pnfk,n (cid:88) j − + 1 2 (cid:88) jmq pnfk,n ( fq,nf (cid:63) q,n + f (cid:63) q,nfq,n) q −1 nj Mjmfk,mf (cid:63) q,j pmM fq,m. (A7) We now substitute Eq. (A5) in Eq. (A7): pn fk,n = (cid:88) (cid:88) j (M − mjq −1 nj F k j − fk,nM −1 nj Pj) (A8) −1 nj Mjmpmf (cid:63) q,j M fq,m (fk,m − fk,n) , which allowed to eliminate the complex conjugate time derivative f (cid:63) q,m. Eq. (A8) is not yet in explicit form since time derivatives of all possible displacement fields appear in the right hand side. We define the mode-independent f (cid:63) k,ifk,n and quantities ain = pn fk,n and bin =(cid:80) (cid:80) f (cid:63) k,i k k solve for ain by inserting Eq. (A8) in its expression: ajmM (cid:0)M −1 nj Mjm (bim − bin) = Ain, −1 nj F k −1 nj Pj (cid:1). j − fk,nM jk f (cid:63) k,i (A9) After solving the linear system (A9) with the (Ncs)2 unknown parameters ain, the evolution equation for each displacement field is then cast into explicit form: (cid:88) with Ain =(cid:80) ain + mj pn fk,n = −1 nj F k j − fk,nM −1 nj Pj) M −1 nj Mjmajm(fk,m − fk,n), (A10) (cid:88) (cid:88) jk (M − mj −1 (cid:80) (cid:88) which can be integrated numerically using an RK4 method. The numerical inversion of the system (A9) can be sped up below the naive (Ncs)6 cost by defining din = −1 in Mln(blm − bln), so that we can solve a linear system for din: −1 nj MjmAjm (A11) nj Mjmajm and αinm =(cid:80) (cid:88) (δijδnm + αinmδnj) dmj = j M l M M mj j which assumes a sparse form suitable for Krylov based methods (provided a good preconditioner can be found). 3. Incoming and outgoing many-body states Combining the incoming coherent state, described by the displacement zk in Eq. (13), with the static polariza- tion cloud wavefunction Eq. (11) can be done by trans- forming the incoming signal in the even/odd basis (see Sec. II C). For the spin-up projection of the wavefunction, we readily find: cs(cid:88) Ψ↑(cid:105) = D(ze)D(zo)ΨGS↑ (cid:105) (A12) (cid:80) (cid:80) N GS k>0 f GS k,mae† k>0 ze kae† k −c.c. = e k −c.c. 0(cid:105)e zo(cid:105)o pGS m e m which can be recombined using the standard relation eAeB = eA+Be 2 [A,B], valid as the commutator here is only a number. The initial state associated to the ↑ qubit state thus reads: 1 N GS cs(cid:88) (cid:80) m Ψ↑(cid:105) = (cid:80) pGS m e 1 2 k>0(ze kf GS∗ k,m −ze∗ k f GS k,m) (A13) k)ae† k −(f GS × e k,m+ze k,m+ze k) k>0[(f GS k] 0(cid:105)e zo(cid:105)o . For the spin-down projection, one simply replaces f GS k,m by −f GS k, so that our total initial wavefunction is given by Eq. (17). k,m without changing the sign of ze ae ∗ The outgoing wavepacket is constructed in a similar spirit: ΨOUT(cid:105) = cs(cid:88) N OUT m=1 pOUT m e 1 2 (cid:80) x−f OUT∗ † x>0(f OUT x,m a x,m ax) 0(cid:105) , (A14) where we have written the displacements of the outgoing state in real space because they have no spatial overlap with the real space modes that populate the many-body ground state (working in momentum space would compli- cate the analysis). This decoupling occurs in fact when the wavepacket reaches distances away from the qubit that are larger than the inverse Kondo energy [37], or said otherwise, that are larger than the entanglement cloud around the qubit. Clearly the quantum many-body character of the scattering process is encoded in the sum over more than a unique coherent state, in contrast to the incoming wavepacket (16) that is characterized by a single coherent state (namely a classical-like signal). Con- trarily to the driven dynamics for an isolated few level quantum system, this long time equilibration between the many-body ground state and the wavepacket is physically expected because the waveguide acts as a bath for the dressed two-level system, and thus provides a natural pathway for relaxation, even in a many-body system. Extracting the wavepacket contribution (A14) from the long-time wavefunction (18) can be performed as follows. 14 k,n(T ), he The complete set of displacements {f e k,n(T )} in the even sector at a fixed long time T for the full wave- function (7) are first Fourier transformed to real space using (12). The local photon density n(x) associated to these displacements is sketched in Fig. 1: photons are either bound statically near the qubit (associated to the dressed vacuum) or travel in the outgoing wavepackets. The displacements are then simply set to zero in the region surrounding the qubit, and Fourier transformed back to the momentum basis. Due to factorization (18), the outgoing wavefunction is recovered, up to a normal- ization factor, which is supplemented accordingly. The even modes thus obtained and the trivial odd mode wave- functions are finally combined together in the case of the incoming wavepacket, allowing reconstruction of the full outgoing wavefunction for the physical waveguide. 4. Convergence properties Assessing the good convergence of the numerical results is important to gain confidence in the time-dependent variational MCS technique Indeed, we find that using too few variational parameters imposes strong constraints on the dynamics, which may result in unphysical behavior and numerical artifacts. One delicate test is the strong power saturation spectrum shown in Fig. 2 of the main text. Indeed, the calculations that use only a single coherent state, as done in a previous publication [34], are found to be problematic in the strong power regime. This behavior is illustrated in the top panel of Fig. 9, showing the power reflection spectrum as a function of incoming frequency at a strong input power (¯n = 2) for three different values of the number of coherent states Ncs = 1, 4, 16. The computation with Ncs = 1 is indeed quite noisy and imprecise, and a smooth and converged curve is only obtained at Ncs = 16. We find that the inelastic spectra shown in Fig. 3 are also delicate to compute, because they consist of a tiny fraction of the total signal, and encode complex quantum states. A relatively large number of coherent state is also necessary here for success, even at small input power. An unbiased criterion for the convergence of our algo- rithm for this non-equilibrium many-body dynamics is also shown in the lower panel of Fig. 9. Here we demon- strate that the error with respect to the exact Schrodinger dynamics vanishes with the number of coherent states. The error is defined [39] by the squared norm of the auxiliary state Φ(t)(cid:11) Err(t) ≡ (cid:10)Φ(t)Φ(t)(cid:11) ≡ (i∂t − H)Ψ(t)(cid:11). Indeed, (A15) this error decreases steadily and scales as [Ncs]−2. For the off-resonant case of Fig. 3 (see bottom curve in the lower panel of Fig. 9) we managed to reach an error of the order of 10−7. 15 FIG. 10. Number of used coherent states in the MCS wave- function defined in Eq. (7) as a function of time, for the off-resonant k0 = 0.16ωc simulation. The initial 6 coherent states present are those required for describing the dressed ground state of the spin-boson model as well as the incoming wave-packet. Subsequent additions of coherent states start only as the wave-packet starts to impinge on the qubit. The addition process stops as the wavepackets exit the interaction region. the addition of a new set of variational displacement does not immediately affect the dynamics, but provides the necessary freedom to our variational algorithm for main- taining a minimal error at later times. This procedure is illustrated in Fig. 10. m = hk The right time for a coherent state to be added is found by monitoring the error, which was defined by Eq. (A15), and by defining an error increment Errmax at which the coherent state should be added. Whenever Err(t)− Errref > Errmax during the time evolution, where Errref is the error right after the previous coherent state was added, one simply adds a coherent state with dis- m = 0 and weights pm = qm (cid:39) 10−6 placement f k in Eq. (7). The near-zero amplitude ensures that this new coherent state only changes the wave-function neg- ligibly at the time it is added. Empirically, we find the value of Errmax = 10−7 to be adequate. The system, through the variational principle, will subsequently have the possibility to increase the displacements and weights according to the requirements of the quantum trajectory. As an example, a plot of the number of coherent states as a function of time for the off-resonant k0 = 0.16ωc simulation in Fig. 3 is given in Fig. 10. 6. Calculation of Number Resolved Spectra To assess the nature of particle production in the scat- tering process, we analyze the inelastic spectrum in terms of Fock states N(cid:105). First, consider the general expansion of the multi-mode outgoing wavefunction (19) in terms of FIG. 9. Top panel: Power reflection spectrum shown for different number of coherent states Ncs = 1, 4, 16 included in the MCS wavefunction (7), with the same parameters as in Fig. 2 in the case of ¯n = 2 photons in the incoming beam. Bottom panel: Convergence of the error defined in the text at the final time Tfinal of the simulations, as a function of coherent state number in the wavefunction for various incoming momenta and power. 5. Protocol for adding coherent states during the time evolution Because the coherent state basis is over-complete, all the coherent states required for good convergence (typ- ically Ncs > 16) cannot be initialised simultaneously at the initial time. Indeed, two coherent states with identical displacements will result in a singularity in the matrices to be inverted for solving the dynamics, due to a vanishing determinant. During the initial stage of the dynamics, this is not an issue, as only a small number of coherent states (typically 6 to 10) is needed to describe the static many-body cloud and the incoming coherent state. Af- ter some time however, the wavepacket starts to interact with the dressed qubit, which would increase the error should the number of coherent states remain the same. Therefore, to account for the emerging complexity of the many-body scattered state, we progressively increase the number of coherent states Ncs in the MCS state vector (Eq. (7)), initializing the newly added coherent states in a bosonic vacuum configuration with zero weight. Thus, (cid:88) number states: (cid:88) ΨOUT(cid:105) = γ 0(cid:105) + + k1,k2,k3 (cid:88) † k 0(cid:105) + αka † αk1,k2,k3 a k1a k † † k2 0(cid:105) k1a αk1,k2a k1,k2 † † k3 0(cid:105) + ... (A16) k2a It can then easily be verified that the 1-photon amplitude is given by: αk = (cid:104)0 ak ΨOUT(cid:105) = pnfk,n (cid:104)0fn(cid:105) , (A17) (cid:88) n and that the scattering amplitude for a generic N -photon state is: αk1,...,kN = 1 N ! (cid:104)0 ak1 ... akN ΨOUT(cid:105) , (A18) which can be obtained straightforwardly from the alge- braic identities of coherent states. For the sake of clarity, we have dropped the OUT labels on pn and fn,k. From the multi-photon amplitudes, we can then compute the probability distribution for finding a photon in a given k mode, according to the various Fock contents of the total wavefunction: (cid:88) (cid:104)nk(cid:105)1photon = αk2, (cid:88) αk,k12, (cid:104)nk(cid:105)2photon = 4 k1 (cid:104)nk(cid:105)3photon = 18 k2,k3 αk,k2,k32. (A19) These Fock resolved inelastic contributions (cid:104)nk(cid:105)Nphoton, with N = 2, 3, 4 are displayed as full lines in Fig. 3 (note that the outgoing N = 1 process is purely elastic and is not shown). 7. Calculation of g2(τ ) In this appendix, we give some details on the calculation of the correlation function g2(τ ) when using the MCS approach. First, since we take the speed of light c = 1, τ is just the distance traveled by radiation in time τ . Inserting the MCS expansion Eq. (7) into definition (27), we obtain a compact expression for the autocorrelation function in terms of the real space displacements f x n : npm(f x n )∗(f x+τ )∗f x+τ (cid:104)n(x)(cid:105)(cid:104)n(x + τ )(cid:105) n m f x m (cid:104)fnfm(cid:105) , (A20) with the local photon number † xax(cid:105) = (cid:104)n(x)(cid:105) = (cid:104)a ∗ ∗ npm(f x n ) p f x m (cid:104)fnfm(cid:105) . (A21) (cid:88) m,n (cid:80) m,n p∗ g2(τ ) = In the simulations performed to compute this quantity we used a sharp cutoff Θ(ωc − ω) for the dispersion relation 16 instead of the exponential cutoff e−ω/ωc which we defined in Eq. (5). Note that using the hard cut-off results in a slightly lower value of the renormalised qubit energy ∆R, than with the exponential cutoff. This allowed us to decrease the numerical cost and therefore attain a higher number of coherent states, Ncs = 40, which was necessary because second-order correlations are more challenging to converge than average photon numbers. The simulations were stopped at a timescale T = 1250/ωc long enough that the wavepacket is located far away from the dressed qubit, and we chose the spacial point x = −681 in Eq. (27), so as to keep the range of the function near the center of the wavepacket. We finally note that spurious effects associated with the finite spatial extension of the wavepacket (due to σ (cid:54)= 0) lead to the small oscillations seen in Fig. 8 at longer times. Appendix B: Further analysis of non-linear emission 1. Detailed off-resonant conversion spectra FIG. 11. Low-energy inelastic spectrum for several off-resonant values of the incoming momentum k0 = 0.16, 0.20, 0.25ωc ob- tained using the MCS technique (with the same parameters as in Fig. 3), together with a comparison to the analytical for- mula (B1), using a fitted and momentum-dependent linewidth γR(k0). We proceed here with a systematic study of particle production spectra in the off-resonant case, as a func- tion of incoming momentum k0 (see Fig. 11). A weak coupling calculation of the one photon to three photon conversion process (see Fig. 4) was given in the α → 0 limit in Ref. [32]. We have found that this theory can quantitatively account for our data at small α upon two important modifications. First, as already seen by the frequency shift in the reflection spectrum in Fig. 2, one must replace the bare qubit frequency ∆ by the renor- malized quantity ∆R within the analytical results given by the perturbative approach. Second, the golden rule value for the qubit linewidth appearing in the transmis- R sion lineshape, given by Γ = πα∆ at small α, cannot be used. For the elastic response, one can use reliably ΓR = πα∆R up to moderate values of α. However, we find that the renormalized broadening parameter γinel entering the inelastic response function for fixed value of the incoming momentum k0 is not given by ΓR, but rather displays a strong momentum dependence, γinel R = γR(k0). This is not completely unexpected, since a consistent cal- culation should include the full momentum variation of the self-energy, and we found that the theory of Ref. [32] is very sensitive to the way the inelastic regularization is implemented. For the present purpose, we will only use a phenomelogical model that uses (as fitting parameters) only two renormalized quantities ∆R and γR(k0) within the perturbative formula: (cid:90) k0−k dk1 kk0k1k2 (cid:104)nk(cid:105)3photon = (cid:12)(cid:12)(cid:12) k0kk1k2 − k2 ∆2 R α4 8 0 ∆(k2 + k2 1 + k2 0 − k2 ∆)(k2 (k2 − k2 2 + kk1 + kk2 + k1k2) + 3k4 ∆ ∆)(k2 ∆)(k2 2 − k2 ∆) 1 − k2 (B1) (cid:12)(cid:12)(cid:12)2 , 17 in the N = 2 curve of Fig. 12. As in the off-resonant case, the resonant scattered spectrum also presents a 3- photon low energy continuum, as can be seen from the inset. The shape however does not present any sharply peaked feature, since this time the continuum does not contain the resonant frequency k = ∆R at which the qubit spontaneously reemits. Instead, the spectrum is more flat, implying the single photon splits more uniformly into all the possible (k1, k2, k3) allowed by the 1 → 3 process of Fig. 4. Interestingly, the magnitude of this 3-photon continuum is of the same order of magnitude as in the off-resonant case of Fig. 3, since non-linear processes are here intensified by having an on-resonant input, which compensates for the absence of an enhancing resonant frequency in the output below k0. Again, this particle production process dominates the RWA contribution, here only away from the probe frequency. with k2 = k0−k1−k and k∆ = ∆R +iγR(k0), with ∆R the renormalized qubit frequency and γR(k0) the linewidth describing the inelastic spectrum, which is fitted from our numerical data. The resulting comparison is shown in Fig. 11, with excellent quantitative agreement. 2. Detailed on-resonant conversion spectra We consider here the detailed photonic content of the emission spectra in the resonant case where the incoming photon energy k0 = ∆R matches the renormalized atomic transition energy. Fig. 12 shows the total transmitted signal as well as its decomposition in terms of number states with N = 1, 2, 3 photons. Not surprisingly, the two- photon amplitude in this regime is strongly enhanced with respect to the off-resonant situation of Fig. 3. One-photon contributions are also observed as two side-bands away from the resonance ∆R, which are due to the finite width of the incoming wave-packet. The resonant 1-photon states (at exactly k = k0) are completely reflected, as expected. In the resonant case, 2 → 2 RWA frequency conversion gives rise to the broader wings (extending clearly beyond the linewidth σ of the pump), as seen FIG. 12. On-resonance transmission spectrum for an incoming photon energy that matches the renormalized qubit excitation, k0 = 0.08ωc = ∆R. The simulation required 2500 modes with the use of 34 coherent states (all other parameters are identical to the ones in Fig. 3). The black dashed curve corresponds to the nearly monochromatic incoming wave-packet. Because of the wavepacket finite linewidth σ, a small fraction of one- photon states is still transmitted (dot-dashed line), despite being on resonance. The two-photon contribution (top full line) presents wider inelastic wings, that extend beyond the width σ, and that are parametrically larger in amplitude than the off-resonant signal of Fig. 3. The 3-photon continuum is magnified in the inset. [1] S. Haroche and J.-M. Raimond, Exploring the Quantum (Oxford Univ. Press, Oxford, UK, 2006). [2] P. Meystre and M. Sargent, Elements of quantum optics (Springer-Verlag Berlin, Heidelberg, 2010). [3] R. Loudon, The Quantum Theory of Light, 3rd ed. (Oxford University Press, New York, 2003). [4] K. S. E. Eikema, J. Walz, and T. W. Hansch, "Continuous coherent lyman- α excitation of atomic hydrogen," Phys. Rev. Lett. 86, 5679 -- 5682 (2001). [5] H. Mabuchi and A. C. Doherty, "Cavity quantum electro- dynamics: Coherence in context," Science 298, 1372 -- 1377 (2002). [6] Serge Haroche, "Nobel Lecture: Controlling photons in a box and exploring the quantum to classical boundary," Rev. Mod. Phys. 85, 1083 -- 1102 (2013). [7] Michael J. Hartmann, "Quantum simulation with inter- acting photons," J. Opt. 18, 104005 (2016). [8] C.M. Haapamaki, J. Flannery, G. Bappi, Maruf R. Al, S.V. Bhaskara, O. Alshehri, T. Yoon, and M. Bajcsy, "Mesoscale cavities in hollow-core waveguides for quantum optics with atomic ensembles," Nanophotonics 5, 392 -- 408 (2016). [9] Pablo Solano, Jeffrey A. Grover, Jonathan E. Hoffman, Sylvain Ravets, Fredrik K. Fatemi, Luis A. Orozco, and Steven L. Rolston, "Optical nanofibers: A new platform for quantum optics," in Advances In Atomic, Molecular, and Optical Physics, Vol. 66, edited by Ennio Arimondo, Chun C. Lin, and Susanne F. Yelin (Academic Press, 2017) pp. 439 -- 505. [10] Dibyendu Roy, C. M. Wilson, and Ofer Firstenberg, "Col- loquium: Strongly interacting photons in one-dimensional continuum," Rev. Mod. Phys. 89, 021001 (2017). [11] P. Forn-D´ıaz, J. J. Garc´ıa-Ripoll, B. Peropadre, J.-L. Or- giazzi, M. A. Yurtalan, R. Belyansky, C. M. Wilson, and A. Lupascu, "Ultrastrong coupling of a single artificial atom to an electromagnetic continuum in the nonpertur- bative regime," Nat. Phys. 13, 39 (2017). [12] J. Puertas Mart´ınez, S. L´eger, N. Gheereart, R. Dasson- neville, N. Planat, F. Foroughi, Y. Krupko, O. Buis- son, C. Naud, W. Guichard, S. Florens, S. Snyman, and N. Roch, "Probing a transmon qubit via the ultra- strong coupling to a josephson waveguide," arXiv preprint arXiv:1802.00633 (2018). [13] X. Gu, A. Frisk Kockum, A. Miranowicz, Y.-X. Liu, and F. Nori, "Microwave photonics with superconducting quantum circuits," Phys. Rep. 718, 1 (2017). [14] M. Devoret, S. Girvin, and R. Schoelkopf, "Circuit-QED: How strong can the coupling between a josephson junction atom and a transmission line resonator be?" Ann. Phys. (Leipzig) 16, 767 (2007). [15] R. Schoelkopf and S. M. Girvin, "Wiring up quantum systems," Nature 451, 664 (2008). [16] J. Bourassa, J. M. Gambetta, A. A. Abdumalikov, O. Astafiev, Y. Nakamura, and A. Blais, "Ultrastrong coupling regime of cavity QED with phase-biased flux qubits," Phys. Rev. A 80, 032109 (2009). [17] T. Niemczyk, F. Deppe, H. Huebl, E. P. Menzel, F. Hocke, M . J. Schwarz, J. J. Garc´ıa-Ripoll, D. Zueco, T. Hummer, E. Solano, A. Marx, and R. Gross, "Circuit quantum electrodynamics in the ultrastrong-coupling regime," Nat. Phys. 6, 772 (2010). [18] P. Forn-D´ıaz, J. Lisenfeld, D. Marcos, J. J. Garc´ıa-Ripoll, E. Solano, C. J. P. M. Harmans, and J. E. Mooij, "Ob- servation of the bloch-siegert shift in a qubit-oscillator system in the ultrastrong coupling regime," Phys. Rev. Lett. 105, 237001 (2010). [19] F. Yoshihara, T. Fuse, S. Ashhab, K. Kakuyanagi, S. Saito, and K. Semba, "Superconducting qubit-oscillator circuit beyond the ultrastrong-coupling regime," Nat. Phys. 13, 44 (2017). [20] O. Astafiev, A. M. Zagoskin, A. A. Jr Abdumalikov, Y. A. Pashkin, T. Yamamoto, K. Inomata, Y. Nakamura, and J. S. Tsai, "Resonance fluorescence of a single artificial atom," Science 327, 840 (2010). [21] A. A. Abdumalikov, O. V. Astafiev, Yu. A. Pashkin, Y. Nakamura, and J. S. Tsai, "Dynamics of coherent and incoherent emission from an artificial atom in a 1d space," Phys. Rev. Lett. 107, 043604 (2011). [22] Io-Chun Hoi, C. M. Wilson, Goran Johansson, Tauno Palomaki, Borja Peropadre, and Per Delsing, "Demon- stration of a single-photon router in the microwave regime," Phys. Rev. Lett. 107, 073601 (2011). 18 [23] Io-Chun Hoi, Tauno Palomaki, Joel Lindkvist, Goran Johansson, Per Delsing, and C. M. Wilson, "Generation of nonclassical microwave states using an artificial atom in 1d open space," Phys. Rev. Lett. 108, 263601 (2012). [24] Io-Chun Hoi, C M Wilson, Goran Johansson, Joel Lind- kvist, Borja Peropadre, Tauno Palomaki, and Per Delsing, "Microwave quantum optics with an artificial atom in one- dimensional open space," New Journal of Physics 15, 025011 (2013). [25] A. F. van Loo, A. Fedorov, K. Lalumi`ere, B. C. Sanders, A. Blais, and A. Wallraff, "Photon-mediated interac- tions between distant artificial atoms," Science 342, 1494 (2014). [26] Neereja M. Sundaresan, Yanbing Liu, Darius Sadri, L´aszl´o J. Szocs, Devin L. Underwood, Moein Malekakhlagh, Hakan E. Tureci, and Andrew A. Houck, "Beyond strong coupling in a multimode cavity," Phys. Rev. X 5, 021035 (2015). [27] A. J. Leggett, S. Chakravarty, A. T. Dorsey, Matthew P. A. Fisher, Anupam Garg, and W. Zwerger, "Dynamics of the dissipative two-state system," Reviews of Modern Physics 59, 1 -- 85 (1987). [28] Karyn Le Hur, "Kondo resonance of a microwave photon," Phys. Rev. B 85, 140506 (2012). [29] B. Peropadre, D. Zueco, D. Porras, and J. J. Garcia- Ripoll, "Nonequilibrium and nonperturbative dynamics of ultrastrong coupling in open lines," Phys. Rev. Lett. 111, 243602 (2013). [30] Soumya Bera, Serge Florens, Harold U. Baranger, Nicolas Roch, Ahsan Nazir, and Alex W. Chin, "Stabilizing Spin Coherence Through Environmental Entanglement in Strongly Dissipative Quantum Systems," Physical Review B 89, 121108(R) (2014). [31] Guillermo D´ıaz-Camacho, Alejandro Bermudez, and Juan Jos´e Garc´ıa-Ripoll, "Dynamical polaron ansatz: A theoretical tool for the ultrastrong-coupling regime of circuit QED," Phys. Rev. A 93, 043843 (2016). [32] Moshe Goldstein, Michel H. Devoret, Manuel Houzet, and Leonid I. Glazman, "Inelastic Microwave Photon Scat- tering off a Quantum Impurity in a Josephson-Junction Array," Physical Review Letters 110, 017002 (2013). [33] E. Sanchez-Burillo, D. Zueco, J. J. Garcia-Ripoll, and L. Martin-Moreno, "Scattering in the ultrastrong regime: Nonlinear optics with one photon," Phys. Rev. Lett. 113, 263604 (2014). [34] Soumya Bera, Harold U. Baranger, and Serge Florens, "Dynamics of a qubit in a high-impedance transmission line from a bath perspective," Physical Review A 93, 033847 (2016). [35] T. Shi, Y. Chang, and J. J. Garcia-Ripoll, "Ultrastrong coupling few-photon scattering theory," Phys. Rev. Lett. 120, 153602 (2018). [36] Soumya Bera, Ahsan Nazir, Alex W. Chin, Harold U. Baranger, and Serge Florens, "Generalized multipolaron expansion for the spin-boson model: Environmental en- tanglement and the biased two-state system," Physical Review B 90, 075110 (2014). [37] Izak Snyman and Serge Florens, "Robust Josephson- Kondo screening cloud in circuit quantum electrodynam- ics," Physical Review B 92, 085131 (2015). [38] T. Shi, E. Demler, and J. I. Cirac, "Variational study of fermionic and bosonic systems with non-gaussian states: Theory and applications," Annal. Phys. 245, 390 (2018). [39] Nicolas Gheeraert, Soumya Bera, and Serge Florens, "Spontaneous emission of Schrodinger cats in a waveguide at ultrastrong coupling," New J. Phys. 19, 023036 (2017). [40] Tao Shi, Darrick E. Chang, and J. I. Cirac, "Multiphoton- scattering theory and generalized master equations," Phys. Rev. A 92, 053834 (2015). [41] Max Haeberlein, Frank Deppe, Andreas Kurcz, Jan Goetz, Alexander Baust, Peter Eder, Kirill Fedorov, Michael Fis- cher, Edwin P. Menzel, Manuel J. Schwarz, Friedrich Wulschner, Edwar Xie, Ling Zhong, Enrique Solano, Achim Marx, Juan-Jos Garca-Ripoll, and Rudolf Gross, "Spin-boson model with an engineered reservoir in circuit quantum electrodynamics," arXiv:1506.09114 [cond-mat, physics:quant-ph] (2015), arXiv: 1506.09114. [42] Nicholas A. Masluk, Ioan M. Pop, Archana Kamal, Zlatko K. Minev, and Michel H. Devoret, "Microwave Characterization of Josephson Junction Arrays: Imple- menting a Low Loss Superinductance," Physical Review Letters 109, 137002 (2012). [43] M. T. Bell, I. A. Sadovskyy, L. B. Ioffe, A. Yu. Kitaev, and M. E. Gershenson, "Quantum Superinductor with Tun- able Nonlinearity," Physical Review Letters 109, 137003 (2012). [44] Carles Altimiras, Olivier Parlavecchio, Philippe Joyez, Denis Vion, Patrice Roche, Daniel Esteve, and Fabien Portier, "Tunable microwave impedance matching to a high impedance source using a Josephson metamaterial," Applied Physics Letters 103, 212601 (2013). [45] T. Weissl, G. Rastelli, I. Matei, I. M. Pop, O. Buisson, F. W. J. Hekking, and W. Guichard, "Bloch band dynam- ics of a Josephson junction in an inductive environment," Physical Review B 91, 014507 (2015). [46] Jens Koch, Terri M. Yu, Jay Gambetta, A. A. Houck, D. I. Schuster, J. Majer, Alexandre Blais, M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, "Charge-insensitive qubit design derived from the cooper pair box," Phys. Rev. A 76, 042319 (2007). [47] C. K. Andersen and A. Blais, "Ultrastrong coupling dy- namics with a transmon qubit," New J. Phys. 19, 023022 (2017). [48] A. A. Abdumalikov, O. Astafiev, A. M. Zagoskin, Yu. A. Pashkin, Y. Nakamura, and J. S. Tsai, "Electromagnet- ically induced transparency on a single artificial atom," Phys. Rev. Lett. 104, 193601 (2010). [49] L. Magazz`u, P. Forn-D´ıaz, R. Belyansky, J.-L. Orgiazzi, M. A. Yurtalan, M. R. Otto, A. Lupascu, C. M. Wilson, and Grifoni M., "Probing the strongly driven spin-boson model in a superconducting quantum circuit," Nat. Com- mun. 9, 1403 (2018). [50] A. Parra-Rodriguez, E. Rico, E. Solano, and I. L. Egusquiza, "Quantum networks in divergence-free circuit QED," Quantum Sci. Technol. 3, 024012 (2018). [51] M. Boon and J. Zak, "Discrete coherent states on the von neumann lattice," Phys. Rev. B 18, 6744 -- 6751 (1978). [52] S. Bera, S. Florens, H. Baranger, N. Roch, A. Nazir, and A. Chin, "Unveiling environmental entangle- ment in strongly dissipative qubits," arXiv preprint arXiv:1301.7430 (2013). [53] Lei Cong, Xi-Mei Sun, Maoxin Liu, Zu-Jian Ying, and Hong-Gang Luo, "Frequency-renormalized multipolaron expansion for the quantum rabi model," Phys. Rev. A 95, 19 063803 (2017). [54] C. Leroux, L. C. G. Govia, and A. A. Clerk, "Enhanc- ing cavity quantum electrodynamics via antisqueezing: Synthetic ultrastrong coupling," Phys. Rev. Lett. 120, 093602 (2018). [55] Serge Florens and Izak Snyman, "Universal spatial correla- tions in the anisotropic kondo screening cloud: Analytical insights and numerically exact results from a coherent state expansion," Phys. Rev. B 92, 195106 (2015). [56] P. Kramer and M. Saraceno, Geometry of the Time- Dependent Variational Principle in Quantum Mechanics (Springer-Verlag Berlin, Heidelberg, 1981). [57] Irene Burghardt, Mathias Nest, and Graham A. Worth, "Multiconfigurational system-bath dynamics using Gaus- sian wave packets: Energy relaxation and decoherence induced by a finite-dimensional bath," The Journal of Chemical Physics 119, 5364 -- 5378 (2003). [58] Yao Yao, Liwei Duan, Zhiguo L, Chang-Qin Wu, and Yang Zhao, "Dynamics of the sub-Ohmic spin-boson model: A comparison of three numerical approaches," Physical Review E 88, 023303 (2013). [59] Nengji Zhou, Lipeng Chen, Yang Zhao, Dima Mozyrsky, Vladimir Chernyak, and Yang Zhao, "Ground-state prop- erties of sub-Ohmic spin-boson model with simultaneous diagonal and off-diagonal coupling," Physical Review B 90, 155135 (2014). [60] D. F. Walls and G. J. Milburn, Quantum Optics, 2nd ed. (Springer, 2007). [61] Shanhui Fan, S¸ukru Ekin Kocaba¸s, and Jung-Tsung Shen, "Input-output formalism for few-photon transport in one-dimensional nanophotonic waveguides coupled to a qubit," Phys. Rev. A 82, 063821 (2010). [62] S¸ukru Ekin Kocaba¸s, Eden Rephaeli, and Shanhui Fan, "Resonance fluorescence in a waveguide geometry," Phys. Rev. A 85, 023817 (2012). [63] Kevin Lalumi`ere, Barry C. Sanders, A. F. van Loo, A. Fe- dorov, A. Wallraff, and A. Blais, "Input-output theory for waveguide QED with an ensemble of inhomogeneous atoms," Phys. Rev. A 88, 043806 (2013). [64] H. J. Carmichael, An Open Systems Approach to Quantum Optics (Springer-Verlag, Berlin, 1993) pp. 41 -- 46. [65] E. S´anchez-Burillo, L. Mart´ın-Moreno, J. J. Garc´ıa-Ripoll, and D. Zueco, "Full two-photon down-conversion of a single photon," Phys. Rev. A 94, 053814 (2016). [66] S. Gasparinetti, M. Pechal, J.-C. Besse, M. Mondal, C. Eichler, and A. Wallraff, "Correlations and entangle- ment of microwave photons emitted in a cascade decay," Phys. Rev. Lett. 119, 140504 (2017). [67] Huaixiu Zheng and Harold U. Baranger, "Persistent quantum beats and long-distance entanglement from waveguide-mediated interactions," Phys. Rev. Lett. 110, 113601 (2013). [68] Matti Laakso and Mikhail Pletyukhov, "Scattering of two photons from two distant qubits: Exact solution," Phys. Rev. Lett. 113, 183601 (2014). [69] Yao-Lung L. Fang and Harold U. Baranger, "Waveguide QED: Power spectra and correlations of two photons scattered off multiple distant qubits and a mirror," Phys. Rev. A 91, 053845 (2015).
0910.4413	2	0910	2010-07-26T20:23:59	Rippled Graphene in an In-Plane Magnetic Field: Effects of a Random Vector Potential	[ "cond-mat.mes-hall" ]	We report measurements of the effects of a random vector potential generated by applying an in-plane magnetic field to a graphene flake. Magnetic flux through the ripples cause orbital effects: phase-coherent weak localization is suppressed, while quasi-random Lorentz forces lead to anisotropic magnetoresistance. Distinct signatures of these two effects enable an independent estimation of the ripple amplitude and correlation length.	cond-mat.mes-hall	cond-mat	Rippled Graphene in an In-Plane Magnetic Field: Effects of a Random Vector Potential Department of Physics and Astronomy, University of British Columbia, Vancouver, BC V6T 1Z4, Canada Mark B. Lundeberg∗ and Joshua A. Folk (Dated: October 30, 2018) We report measurements of the effects of a random vector potential generated by applying an in-plane mag- netic field to a graphene flake. Magnetic flux through the ripples cause orbital effects: phase-coherent weak localization is suppressed, while quasi-random Lorentz forces lead to anisotropic magnetoresistance. Distinct signatures of these two effects enable an independent estimation of the ripple amplitude and correlation length. 0 1 0 2 l u J 6 2 ] l l a h - s e m . t a m - d n o c [ 2 v 3 1 4 4 . 0 1 9 0 : v i X r a Graphene is a one-atom-thick carbon sheet with unusual electronic properties due to its two-dimensional honeycomb lattice1,2. As an ultrathin membrane, graphene easily wrin- kles into the third dimension, with nanometer-scale ripples that have been observed in microscopic studies3 -- 6. Strains as- sociated with rippling are expected to modify electronic trans- port by generating random scalar and vector potentials, which would affect transport by suppressing anti-localization7 and increasing the scattering rate8. Until this time, however, no transport measurements have directly probed graphene's rip- ples. Here, we report a magnetotransport measurement of graphene ripples using an in-plane magnetic field. In general, in-plane fields do not affect electron transport directly because electronic motion couples only to the component of magnetic field perpendicular to the graphene sheet. As illustrated in Fig. 1, however, a magnetic field aligned in the average plane of a graphene sheet (B(cid:107)) includes an inhomogeneous perpen- dicular component (δB⊥), which depends on the local slope of the graphene flake. Ripples allow the in-plane field to af- fect transport directly by converting it to an inhomogeneous out-of-plane magnetic field, a form of random vector potental (RVP). We observe two distinct effects of the B(cid:107)-controlled RVP that each depend on the ripples' rms height, Z, and correla- tion length, R. Random Aharonov-Bohm phases break time- reversal symmetry and suppress weak localization9 -- 11 (WL) with an effective dephasing rate12 proportional to B2(cid:107)Z 2R. Random Lorentz forces13 lead to anisotropic momentum scat- tering by the in-plane field14 and anisotropic magnetoresis- tance ∝ B2(cid:107)Z 2/R. The effects are distinguished by measur- ing conductance in an additional uniform perpendicular field (B⊥) -- an independent experimental knob that can be used to suppress time-reversal symmetry. Figure 1b shows a schematic of the graphene flakes and measurement set-up. Two flakes (called A and B) were pre- pared on SiO2, electrically contacted, then cooled to an elec- tron temperature of 40 mK. A two-magnet system provided independent control over the uniform magnetic field compo- nents B(cid:107) and B⊥. The carrier density n of the graphene was controlled capacitively using the Si back-gate (not shown in Fig. 1b), and the electrical conductance G(n) showed a typical density-dependence (Fig. 1c) with mesoscopic fluctuations. The conductance fluctuations of flakes A and B showed spin- split character when an in-plane field was applied, as reported in our earlier work15. Here, we instead report the magnetic field dependence of the mean conductance (cid:104)G(cid:105)n, averaging FIG. 1: (a) Simulation of a rippled graphene sheet with a correlation length, R, ten times its rms height, Z. The uniform in-plane field B(cid:107), applied to the rippled topography of graphene, leads to a ran- dom surface-normal field δB⊥. (b) Schematics of graphene devices A (left) and B (right), showing orientation of applied fields B(cid:107), B⊥, and electrical measurement setups (two-probe for A, multi-probe for B). Unused/broken electrodes are indicated by dashed edges. Scale bars are 5 µm. (c) Conductance G(n) for B=0, flake B. (d) Weak- localization B⊥-magnetoconductance of flake B at low density, mea- sured at different in-plane fields: zero (+), −4 T ((cid:3)), +4 T ((cid:4)). Two fits to (1) are shown as solid lines. Applying the in-plane field dulls the central dip and decreases overall conductance. These are attributed to dephasing and scattering by the random δB⊥. out (over intervals in n) the conductance fluctuations, follow- ing the procedure of Ref. 6. Figure 1d shows typical (cid:104)G(cid:105)n(B⊥) curves for B(cid:107) = 0 as well as B(cid:107) = 4. We first discuss the conventional case, B(cid:107) = 0. The conductance can be separated into Drude (cid:112)πn/h, depends on the momentum scattering time and WL contributions, (cid:104)G(cid:105)n(B⊥) = W L [gD + gWL(B⊥)], L . The Drude conductivity, gD = with device aspect ratio W 2e2vτm τm and Fermi speed v ≈ 106 m/s and does not change sig- nificantly with B⊥ for the fields used in this experiment. The W [(cid:104)G(cid:105)n(B⊥) − (cid:104)G(cid:105)n(0)], is thus B⊥-magnetoconductivity, L determined entirely by the WL component ∆gWL(B⊥). Weak localization is a phase-coherent back-scattering ef- fect originating from the interference of a closed path with its reversed counterpart. Paths that contribute to WL are lim- ited in size by accumulated phases that are not symmetric un- der time (and path) reversal, coming from inelastic scatter- ing, magnetic fields and some types of disorder. Graphene's chiral charge carriers are intrinsically anti-localized, so gWL is positive as long as chiral memory is maintained along a path. When chiral memory is lost due to intervalley scatter- ing, however, conventional weak localization decreases gWL. Taking these effects together, a non-monotonic magnetocon- ductance curve is typically observed: large B⊥ probes only short paths where chiral memory is retained, giving gWL > 0, whereas small B⊥ also includes the contribution of longer paths for which chiral memory is lost, giving a conductance dip at B⊥ = 0. For 2D diffusive graphene, the WL contribution has been calculated to be:16 ∆gWL(B⊥) = e2 πh F (cid:20) (cid:16) (cid:16) τ−1 B φ τ−1 − 2F (cid:17) − F (cid:16) (cid:17) (cid:17)(cid:21) τ−1 B τ−1 φ + 2τ−1 τ−1 τ−1 φ + τ−1 i + τ−1∗ B i , (1) i i z + 1 B , τ−1 φ , τ−1 2 ) for digamma function ψ(x). where F (z) = ln(z) + ψ( 1 , τ−1∗ } Equation (1) depends on four rates {τ−1 characterizing different mechanisms that suppress WL. The diffusive accumulation of Aharonov-Bohm phases from uni- form B⊥ gives τ−1 B = 2v2τmeB⊥/. The dephasing (τ−1 φ ), inter-valley scattering (τ−1 ), and intra-valley scattering (τ−1∗ ) rates each originate from scattering processes that each break different time-reversal symmetries.16 The WL scattering rates were extracted from measured (cid:104)G(cid:105)n(B⊥) curves by fitting to Eq. 1, and are listed in Table I for the B(cid:107) = 0 case. Values of τm (used to scale τ−1 B ), as com- puted from gD ≈ W L (cid:104)G(cid:105)n, were the primary systematic error since W L was difficult to determine (see Methods section). The longest time-scale, τφ, may have been saturated by the device The other characteristic lengths Li ∼ 600 nm, L∗ ∼ 100 nm, and vτm < 100 nm, were not influenced by the sample geom- etry. Adding an in-plane field, B(cid:107) = 4T, changed (cid:104)G(cid:105)n(B⊥) in two distinct ways (Fig. 1d). First, the dephasing rate was in- creased, visible as a suppression of the WL dip at small B⊥. Second, the overall (Drude) conductance was reduced, caus- ing an overall downward shift in the conductance at large B⊥. These effects can both be attributed to the ripple-induced RVP, and are discussed in order. dimensions at high doping, with Lφ = v(cid:112)τmτφ/2 ∼ 2 µm. 2 Parameter Units n W/L τ−1 m −1 τ φ −1 τ i τ−1∗ 1011/cm2 - 1012/s 109/s 109/s 1012/s Flake B Flake A Hole Hole Low density Electron -13...-5 -5...-3 5...13 2.4 ± 0.8 2.0 ± 0.7 1.6 ± 0.6 0.7 ± 0.3 15 ± 5 15 ± 5 11 ± 5 11 ± 1 11 ± 2 11 ± 1 70 ± 50 120 ± 80 20 ± 10 5.3 ± 0.4 2.7 ± 0.5 2.1 ± 0.4 4.0 ± 0.3 -2...2 20 ± 10 35 ± 8 170 ± 70 TABLE I: Rates extracted from (cid:104)G(cid:105)n(B⊥, B(cid:107) = 0) at 40 mK (e.g. Fig. 1d); (cid:104)G(cid:105)n was averaged over the specified density ranges. FIG. 2: (a,b) B⊥-magnetoconductance for small B⊥, at various val- ues of B(cid:107). Fits to Eq. 1 were computed assuming τi, τ∗, and τm are −1 φ was a free parameter. independent of B(cid:107) (as in Table I), while τ (a) and (b) correspond to the low density (Fig. 1d) and hole-doped −1 regions, respectively. (c) For both flakes, extracted values of τ φ increase in proportion with B2(cid:107), as predicted from Eq. 2. Error bars L . plotted here do not include the larger systematic error from W The additional dephasing effect of an in-plane field due to Gaussian-correlated ripples was calculated in Ref. 12. Whereas a uniform B⊥ affects WL through the diffusive rate τ−1 the RVP affects WL as a micro-scattering rate B , (τ−1 φ ) since the ripples are uncorrelated beyond short dis- tances (R (cid:28) vτm ∼ 100 nm): √ (2) Eq. (1) was fit to multiple (cid:104)G(cid:105)n(B⊥) curves at finite B(cid:107), al- π(e2/2)vZ 2RB2(cid:107). φ → τ−1 τ−1 φ + φ L . Although the WL rate τ−1 lowing only τ−1 to change from the B(cid:107) = 0 fits (Fig. 2a,b), in order to extract the B(cid:107) effect. Figure 2c confirms the φ (B(cid:107)) ∝ B2(cid:107) dependence in Eq. 2, with a density- ∆τ−1 independent Z 2R = 1.7 ± 0.5 nm3 extracted for flakes A and B (Fig. 2); the uncertainty is dominated by uncertainty is commonly associated in W with inelastic scattering and loss of phase information, in this case it is enhanced by elastic scattering from the RVP which only scrambles phase information deterministically. The dis- tinction can be seen in conductance fluctuations, which are softened by decoherence but only scrambled by elastic scat- tering. As reported in our previous work on flakes A and B, fluctuations were scrambled by B(cid:107), and only decreased by 1 2 in variance due to broken spin symmetry.15 φ Besides suppressing localization, B(cid:107) caused an overall downward shift in the B⊥-magnetoconductance trace away from zero field (Fig. 1d). This shift indicates a change in the Drude conductivity; it was isolated from the dephasing effect by examining the B(cid:107)-magnetoconductance for values of B⊥ > 50 mT (Fig. 3), where the total WL correction16 gWL is essentially unaffected by the changes in τ−1 φ . The ef- fect of RVP on Drude conductivity can understood as a de- crease in τm due to scattering by the Lorentz forces from the RVP (Fig. 3a). A similar random-field resistivity has been ob- served in 2D systems subject to RVPs originating from nearby magnetic particles or superconducting vortices13,14,17. expected magnetoresistivity ∆ρ(B(cid:107)) The L [1/G(B(cid:107)) − 1/G(0)] for Gaussian ripples due to Lorentz W forces from the RVP can be calculated by a Boltzmann approach, assuming kF R (cid:29) 1 (high doping): = ∆ρ(n, θ, B(cid:107)) = sin2 θ + 3 cos2 θ 4 1 n3/2 Z 2 R B2(cid:107), (3) where θ is the angle of the current flow relative to (cid:126)B(cid:107). Equa- tion 3 is derived in the Appendix, and is most conveniently measured as a change in resistivity (rather than conductance). The threefold anisotropy predicted by Eq. (3) was observed in a GaAs 2DEG when in-plane field lines were rippled by nearby ferromagnets14. In that case, as with rippled graphene, each individual magnetic ripple includes equal parts positive and negative magnetic flux, with a y-oriented zero-field chan- nel (for (cid:126)B(cid:107) along x; see Fig. 3a). In the strong-field limit, such channels might form magnetic waveguides18,19, but we observed a weaker form: the x-component of the velocity of a diffusive ensemble is randomized faster than the y-component because y-moving particles are deflected more (Fig. 3a).14 The density dependence, ∆ρ(n) ∝ n−3/2, predicted from Eq. 3 can be seen in the experimental data (Fig. 3b), on top of phase-coherent conductance (resistance) fluctuations due to the in-plane field15. The resistivity saturated for n (cid:46) 1012 cm−2, perhaps due to the breakdown of classical scat- tering when kFR (cid:46) 1. The fluctuations in Fig. 3b were minimized by measuring at 4K, and averaging the quantity n3/2∆ρ(n, (cid:126)B(cid:107)) over n = (−3.5 . . .−1) × 1012 cm−2, al- lowing fits to Eq. 3 (Fig. 3c,d). It was confirmed that the mag- nitude of the effect did not change from 4K down to 40mK, 3 FIG. 3: Anisotropic in-plane magnetoresistivity at B⊥ > 50 mT in flake A at 4K. (a) Upper: a simulated symmetric bump in the graphene sheet, with uniform field B(cid:107) applied in the x direction. Lower: The resulting surface-normal field, δB⊥, is antisymmetric (positive on the right). Simulated trajectories show how an elec- tron's x-velocity is randomized more quickly than its y-velocity. (b) Density-dependence of ρ(8 T) − ρ(0 T) at B⊥ = 50 mT, θ = 20◦. Predicted large-n behaviour in (3) for Z 2/R = 0.15 nm shown as dashed curve. (c) Averages (cid:104)n3/2∆ρ(n, (cid:126)B(cid:107))(cid:105)n show the depen- dence on the magnitude and direction of (cid:126)B(cid:107). The current path (de- picted in inset) was measured at in-plane field orientations θ ≈ 20◦ (◦) and θ ≈ 70◦ (+). (d) Measurements on a different pair of elec- L ≈ 1.6) confirm that the anisotropy depends on the anglu- trodes ( W lar difference, θ, between field and current. though the amplitude of the fluctuations increased at low tem- perature as expected. Flake A was measured with two current paths along θ ≈ 20◦ and 70◦. The device was then re-cooled in a 90◦-rotated orientation, to change θ → θ + 90◦. Fits of magnetoresistance curves to Eq. 3 gave a range Z 2/R ∼ 0.05 -- 0.2 nm for flake A (Fig. 3cd). The measured anisotropy ∆ρ(70◦)/∆ρ(20◦) was approximately 0.13 ± 0.01 for one current path (Fig. 3c) and 0.26 ± 0.03 for the other (Fig. 3d), whereas Eq. (3) predicts 0.44. In the single measurement of flake B, Z 2/R ≈ 0.02 -- 0.04 nm. Using the value Z 2R ≈ 1.7 nm3 from the analysis in Fig. 2 and the range of values of Z 2/R reported above, Z = 0.6 ± 0.1 nm and R = 4 ± 2 nm are extracted for Gaussian-correlated rippling of Flake A (the spread in Z 2/R is incorporated into uncertainties for Z and R). The values for Z and R from this work can be compared to values ob- tained from scanning probe measurements on these flakes, and to values reported in the literature for graphene flakes on SiO2. After the measurements described above, room temperature atomic force microscope (AFM) measurements were performed on flake A using an Asylum MFP3D-SA, after annealing the flake at 400◦C in a low pressure N2/H2 gas mixture to remove resist residues4. These measurements gave Z = 0.13 ± 0.02 nm and R = 10 ± 5 nm; limi- tations of vibration and drift prevented more accurate mea- surements. AFM measurements on similar flakes in Ref. 4 gave Z = 0.19 nm and R = 32 nm, whereas scanning tunnelling microscope (STM) measurements in Ref. 5 gave Z = 0.35 nm, R ≈ 5 nm20. A consistent discrepancy is noted between scanning probe measurements and topographic parameters extracted from transport: the observed magneto- resistance in flakes A and B is nearly a factor of 100 larger than would be expected from our own AFM measurements on flake A. Although the right-hand side of Eq. (3) is enhanced by a factor between one and two for realistic non-Gaussian correlations in ripples (see Appendix), this is far too small to explain the measured magneto-resistance. The in-plane field couples to spins as well as to the mo- tion of charges, leading to the possibility of spin-related ef- fects on transport.15 Three different types of spin-related ef- fects are considered. First, Zeeman splitting of the Fermi level implies altered populations of spin-up and spin-down electrons15 that screen charged impurities less efficiently, leading to a magnetoresistance21 ρ(B(cid:107))/ρ(0) ≈ (2 × 105 cm−2 T−2)B2(cid:107)/n for densities larger than the impurity broadening. Quantitatively, however, this effect would be too small to be observed in our measurements. Second, spin- flip scattering off magnetic impurities can lead to decoher- ence, adding to τ−1 φ . This effect would be field dependent, as B (cid:38) Bi will freeze impurities into their ground state22, disabling spin-flip dephasing at Bi = kBT /g∗µB ∼ 100 mT. Such an effect would show up as a peak in τ−1 at zero field in data such as Fig. 2c. Based on the apparent absence of this peak, impurity-induced spin flips must be uncommon for electrons traversing our devices. Finally, magnetic impurities may also generate an RVP by their localized magnetic fields, but the strength of this RVP (and hence its contribution to τφ) would not change when the impurities align to B(cid:107). Finally, we turn to an important analogy that can be drawn between the effects of an in-plane field and those of strain due to ripples. Since both the in-plane field and ripple strain generate random vector potentials that are directly correlated with ripple topography7,12, similar effects can be expected. In particular, strain is commonly believed to suppress weak anti- localization6,7,16, but there is widespread disagreement about how to estimate the magnitude of the effect. We argue that the suppression occurs through a short-range dephasing process much like that in Eq. 2, and that strain-induced dephasing can fully explain the observed suppression of anti-localization. φ Strain RVP affects each valley oppositely with a magnitude that depends on R and Z,7 corresponding to a fictitious valley- dependent in-plane field of size Z/(ea0R2), where the lat- tice constant is a0 = 0.14 nm. The resulting valley-dependent dephasing affects τ−1∗ ,16 and by analogy with Eq. 2 we expect τ−1∗ ≈ vZ 4/(a2 0R3). Using this expression, the ripple dimen- sions extracted from our in-plane field measurements or from STM measurements5 may fully explain the large τ−1∗ ∼ 1 -- 10 ps−1 observed in most graphene WL magnetoresistance 4 experiments (see Table I and Refs. 6,23,24). This result contrasts with previous estimates of the intra- valley effect of ripple strain, which have assumed that the strain-induced effective magnetic field is truly random, with no requirement for flux compensation over multiple correla- tion lengths.6,7,25 That assumption would imply long range correlations in the RVP. The dephasing measurements in the first half of this paper show that ripples and resulting RVP cor- relations are instead short-range, as expected for adhesion to a polished wafer, and the analysis for τφ in Eq. 2 should apply also to τ∗. In conclusion, transport measurements of graphene flakes in an in-plane magnetic field showed effects due to the mag- netic flux threaded through the ripples. The use of an auxil- iary out-of-plane field allowed two distinct effects to be sepa- rated: weak localization suppression (by dephasing) and over- all anisotropic magnetoresistance (by Lorentz-force scatter- ing). Besides allowing a determination of the ripples' typical height and length scale, these measurements provide insight as to how other short-range random vector potentials (such as that due to ripple strain) might affect transport in graphene. Experimental methods Silicon wafers with a ∼300 nm wet thermal oxide were thinned to ∼260 nm oxide thickness by a CF4/O2 plasma, then cleaned using a standard SC-1/SC-2 process. Within an hour of cleaning, flakes of graphene were deposited us- ing the mechanical exfoliation technique2 then located in an optical microscope. The Cr/Au electrodes were deposited in an e-beam lithography process using PMMA resist. Im- mediately before the cooldowns for electrical measurements, devices were baked on a 125◦C hot plate. Graphene flakes were confirmed to be single layers by quantum Hall effect measurements15; the backgate capacitance ne/(VG − V0) = 8.0±0.1×1010 cm−2e/V was calculated from resistivity min- ima in this data. In each cooldown, the gate offset V0 was determined by requiring the conductance minimum to occur at n = 0. For Figs. 1 and 2, V0 = 1 V in flake A15 and V0 = 11.5 V in flake B. The quality of flake A had decayed before the 4 K measurements of Fig. 3, so that V0 varied from 13 to 23 V depending on cooldown and the current path. Two magnets provided fields up to 120 mT oriented out- of-plane, and 12 T oriented roughly in-plane. A ∼1◦ mis- alignment of the in-plane axis was corrected by biasing the out-of-plane magnet: the reported B⊥ values have been man- ually offset at each B(cid:107) so that the WL dip always occurs at B⊥ = 0. Conductance was measured by lock-in techniques with a 10 nA current bias (Fig. 1b). In flake B, conductance was measured in a four-probe geometry (Fig. 1b), and we fur- ther reduced the effects of conductance fluctuations by aver- aging over two opposing sets of voltage probes with the same aspect ratio (Fig. 1b). In flake A, two-terminal conductance was calculated after subtracting a 3.2 kΩ contact resistance. Aspect ratios of flake A were computed by a 2D Laplace equa- tion solver, and aspect ratios of flake B were estimated from the device geometry. In either case, aspect ratios were com- plicated by the invasive nature of the contacts26. The differ- ent aspect ratios for hole and electron doping (Table I) were determined by requiring the conductivity to be n-symmetric: ( L W G)(n) = ( L W G)(−n). Appendix: B(cid:107) anisotropic magnetoresistance bution f ((cid:126)k), with evolution d for D[S, f ]((cid:126)k) = (cid:82) d2k(cid:48) We model diffusive particle motion by a probability distri- dt [f ((cid:126)k)] = D[S0 + S(cid:107), f ]((cid:126)k), (2π)2 S((cid:126)k, (cid:126)k(cid:48))(f ((cid:126)k(cid:48)) − f ((cid:126)k)). The zero- field scattering matrix is S0, assumed to have the elastic and isotropic form S0((cid:126)k, (cid:126)k(cid:48)) = s0(k, q)δ(k(cid:48) − k), where (cid:126)q = (cid:126)k(cid:48) − (cid:126)k. Transport occurs by modes fx(θ) = 1√ π cos(θ) and fy(θ) = 1√ π sin(θ) which are eigenmodes of D[S0, fi] = −τ−1 ii fi for all k. The zero-field transport time is then τxx = τyy = τm, with τxy = 0. The B(cid:107)-induced scattering matrix is calculated by Fermi golden rule as S(cid:107)((cid:126)k, (cid:126)k(cid:48)) = 2πV (cid:104)(cid:126)k(cid:48)U ((cid:126)r)(cid:126)k(cid:105)2δ(Ek(cid:48) − Ek). The potential is U ((cid:126)r) = −ev (cid:126)A((cid:126)r) · (cid:126)σ for Dirac fermions with Pauli matrices (cid:126)σ operating on chirality. Given B(cid:107) ori- ented along x, the vector potential is12 (cid:126)A((cid:126)r) = −B(cid:107)h((cid:126)r)y for rippled graphene with out-of-plane displacements denoted by h((cid:126)r). We consider just single-valley chiral plane waves (cid:126)k(cid:105) = 1√ ei(cid:126)k·(cid:126)r[e−iθk/2, eiθk/2]T under the assumption that U does not mix valleys, yielding 2V 2πe2vB2(cid:107) S(cid:107)((cid:126)k, (cid:126)k(cid:48)) = c((cid:126)q) sin2 θk(cid:48) + θk δ(k − k(cid:48)), 2 2 (A.1) for height correlator c((cid:126)r) = (cid:104)h((cid:126)r0)h((cid:126)r0 + (cid:126)r)(cid:105). This scat- tering depends on the sum of θk(cid:48) and θk, and therefore may anisotropically break the degeneracy of the transport eigen- modes. As a perturbation of the evolution, this modifies the scattering rates as δτ−1 −π dθk fi(θk)D[S(cid:107), fj](θk), to first order. ij = (cid:82) π Here we examine only the case of isotropic ripples c((cid:126)q) = c(q), for which ρxy, ρyx remain zero. After some calculation, the anisotropy is found to be exactly threefold: ∆ρxx = 3 2 ρ(cid:107), 5 (cid:90) ∞ 0 πB2(cid:107) and ∆ρyy = 1 integral of the height correlator: 2 ρ(cid:107), where ρ(cid:107)(k) may be written as a real-space dr rW (kr)c(r), (A.2) ρ(cid:107)(k) = for W (z) =(cid:82) 2π r √ 0 dr 1 2 ) sin2 φ 0 dφ J0(2z sin φ 2 . We have confirmed the threefold anisotropy of relaxation by simulating a classical charge moving in the x-y plane with out-of-plane magnetic field B = −B(cid:107) dh dx z, for random isotropic h(r). Simulation and experiment in Ref. 14 also show ∆ρxx = 3∆ρyy. W (z) is approximately constant for z (cid:28) 1, so ρ(cid:107)(k) plateaus at a resistivity ∼ (ZRB(cid:107))2/ at low densities where k (cid:28) 1/R. At higher density (k (cid:29) 1/R), the oscillatory W (z) may be integrated out via a Hankel transform. This yields the classical result ρ(cid:107) ∝ k−3 of Eq. 3 with an effective π, which is exactly Z 2/R [Z 2/R]eff = −(cid:82) ∞ d dr c(r)/ if the ripples are Gaussian. For realistic non-Gaussian correlation functions, and for k ≈ 1/R, Eq. 3 may be used for Z and R statistically de- fined as in Ref. 4, though with some error. For more accuracy, Eq. A.2 should be computed directly. Using the correlation function of Ref. 4, Eq. A.2 gives a magnetoresistance approxi- mately 80% larger than that expected from Eq.3, due to fractal scaling of ripples which causes [Z 2/R]eff > Z 2/R. For the R ≈ 5 nm correlation function20 of Ref. 5, k ≈ 1/R for the range of n examined in Fig. 3, and Eq. A.2 gives the magne- toresistance a k-dependence slightly weaker than the classical k−3. Averaging over n in the method of Fig. 3 yields a result in agreement with Eq. 3, though strong disagreement arises for R < 4 nm as ρ(cid:107) saturates to its maximum. Acknowledgments We acknowledge V. F'alko, M. Fuhrer, D. Goldhaber- Gordon, C. Marcus, and especially C. Lewenkopf for help- ful discussions. Graphite provided by S. Fain and D. Cobden. MBL acknowledges a PGS-D from NSERC; work funded by CIFAR, CFI, and NSERC. ∗ to whom correspondence should be addressed: [email protected] 1 Y. Zhang, Y.-W. Tan, H. L. Stormer, and P. Kim, Nature (London) 438, 201 (2005). 2 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004). 3 J. C. Meyer, A. K. Geim, M. I. Katsnelson, K. S. Novoselov, T. J. Booth, and S. Roth, Nature (London) 446, 60 (2007). Savchenko, Phys. Rev. Lett. 100, 056802 (2008). 7 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). 8 M. I. Katsnelson and A. K. Geim, Phil. Trans. R. Soc. A 366, 195 9 P. M. Mensz, R. G. Wheeler, C. T. Foxon, and J. J. Harris, Appl. (2008), 0706.2490. Phys. Lett. 50, 603 (1987). 4 M. Ishigami, J. H. Chen, W. G. Cullen, M. S. Fuhrer, and E. D. 10 D. F. Sullivan, B. E. Kane, and P. E. Thompson, Appl. Phys. Lett. Williams, Nano Lett. 7, 1643 (2007). 85, 6362 (2004). 5 V. Geringer, M. Liebmann, T. Echtermeyer, S. Runte, M. Schmidt, R. Ruckamp, M. C. Lemme, and M. Morgenstern, Phys. Rev. Lett. 102, 076102 (2009). 6 F. V. Tikhonenko, D. W. Horsell, R. V. Gorbachev, and A. K. 11 G. M. Minkov, O. E. Rut, A. V. Germanenko, A. A. Sherstobitov, B. N. Zvonkov, V. I. Shashkin, O. I. Khrykin, and D. O. Filatov, Phys. Rev. B 70, 035304 (2004). 12 H. Mathur and H. U. Baranger, Phys. Rev. B 64, 235325 (2001). 6 13 A. K. Geim, S. J. Bending, I. V. Grigorieva, and M. G. Blamire, Phys. Rev. B 49, 5749 (1994). 14 A. W. Rushforth, B. L. Gallagher, P. C. Main, A. C. Neumann, M. Henini, C. H. Marrows, and B. J. Hickey, Phys. Rev. B 70, 193313 (2004). 15 M. B. Lundeberg and J. A. Folk, Nature Phys. 5, 894 (2009). 16 E. McCann, K. Kechedzhi, V. I. Fal'ko, H. Suzuura, T. Ando, and B. L. Altshuler, Phys. Rev. Lett. 97, 146805 (2006). 17 S. Wada, N. Okuda, and J. Wakabayashi, Physica E: Low- dimensional Systems and Nanostructures 42, 1138 (2010), ISSN 1386-9477, 18th International Conference on Electron Properties of Two-Dimensional Systems. 18 L. Oroszl´any, P. Rakyta, A. Korm´anyos, C. J. Lambert, and J. Cserti, Phys. Rev. B 77, 081403 (2008). 19 T. K. Ghosh, A. De Martino, W. Hausler, L. Dell'Anna, and R. Eg- ger, Phys. Rev. B 77, 081404 (2008). 20 We thank the authors of Ref. 5 for providing raw STM data for statistical analysis, used to extract Z and R by the criteria given in Ref. 4. 21 E. H. Hwang and S. D. Sarma, Phys. Rev. B 80, 075417 (2009). 22 J. S. Meyer, V. I. Falko, and B. L. Altshuler, NATO Science Series II, Vol. 72 (Kluwer Academic, Dordrecht, 2002). 23 J. Eroms and D. Weiss, New Journal of Physics 11, 095021 (2009), 0901.0840. 24 Y. Chen, M. Bae, C. Chialvo, T. Dirks, A. Bezryadin, and N. Ma- son, Journal of Physics Condensed Matter 22, 205301 (2010), 0910.3737. 25 X. Wu, X. Li, Z. Song, C. Berger, and W. A. de Heer, Phys. Rev. Lett. 98, 136801 (2007). 26 B. Huard, N. Stander, J. A. Sulpizio, and D. Goldhaber-Gordon, Phys. Rev. B 78, 121402 (2008).
1203.1249	4	1203	2013-08-05T06:22:27	Magnetic-non-magnetic superlattice chain with external electric field: Spin transport and the selective switching effect	[ "cond-mat.mes-hall" ]	Based on Green's function formalism, the existence of multiple mobility edges in a one-dimensional magnetic-non-magnetic superlattice geometry in presence of external electric field is predicted, and, it leads to the possibility of getting a metal-insulator transition at multiple values of Fermi energy. The role of electric field on electron localization is discussed for different arrangements of magnetic and non-magnetic atomic sites in the chain. We also analyze that the model quantum system can be used as a perfect spin filter for a wide range of energy.	cond-mat.mes-hall	cond-mat	Magnetic-non-magnetic superlattice chain with external electric field: Spin transport and the selective switching effect Moumita Dey,1 Santanu K. Maiti,2, ∗ and S. N. Karmakar1 1Theoretical Condensed Matter Physics Division, Saha Institute of Nuclear Physics, Sector-I, Block-AF, Bidhannagar, Kolkata-700 064, India 2Physics and Applied Mathematics Unit, Indian Statistical Institute, 203 Barrackpore Trunk Road, Kolkata-700 108, India Based on Green's function formalism, the existence of multiple mobility edges in a one-dimensional magnetic-non-magnetic superlattice geometry in presence of external electric field is predicted, and, it leads to the possibility of getting a metal-insulator transition at multiple values of Fermi energy. The role of electric field on electron localization is discussed for different arrangements of magnetic and non-magnetic atomic sites in the chain. We also analyze that the model quantum system can be used as a perfect spin filter for a wide range of energy. PACS numbers: 73.63.Nm, 72.20.Ee, 73.21.-b, 73.63.Rt I. INTRODUCTION Quantum transport in low-dimensional systems has been a topic of interest within the past few decades due to its potential applicability in the field of nanoscience and nanotechnology. Exploitation of the spin degree of free- dom adds a possibility of integrating memory and logic into a single device, leading to remarkable development in the fields on magnetic data storage application, de- vice processing technique, quantum computation1, etc. Naturally a lot of attention has been paid to study spin transport in low-dimensional systems both from experi- mental2 -- 4 and theoretical5 -- 18 points of view. The understanding of electronic localization in low- dimensional model quantum systems is always an inter- esting issue. Whereas, it is a well established fact that in an infinite one-dimensional (1D) system with random site potentials all energy eigenstates are exponentially lo- calized irrespective of the strength of randomness due to Anderson localization19, there exists another kind of lo- calization, known as Wannier-Stark localization, which results from a static bias applied to a regular 1D lat- tice, even in absence of any disorder20. Till date a large number of works have been done to explore the under- standing of Anderson localization and scaling hypoth- esis in one- and two-dimensional systems21. Similarly, Wannier-Stark localization has also drawn the attention of many theorists22 -- 26 as well as experimentalists27. For both these two cases, viz, infinite 1D materials with ran- dom site energies and 1D systems subjected to an exter- nal electric field, one never encounters any mobility edge i.e., energy eigenvalues separating localized states from the extended ones, since all the eigenstates are localized. But there exist some special types of 1D materials, like quasi-periodic Aubry-Andre model and correlated disor- dered systems where mobility edge phenomenon at some particular energies is obtained28 -- 37. Although the studies involving mobility edge phenomenon in low-dimensional systems have already generated a wealth of literature35 -- 41 there is still need to look deeper into the problem to ad- dress several interesting issues those have not yet been explored. For example, whether the mobility edges can be observed in some other simple 1D materials or the number of mobility edges separating the extended and localized regions in the full energy band of an 1D mate- rial can be regulated, are still to be investigated. To address these issues in the present article we in- vestigate two-terminal spin dependent transport in a 1D mesoscopic chain composed of magnetic and non- magnetic atomic sites in presence of external electric field. To the best of our knowledge, no rigorous effort has been made so far to explore the effect of an ex- ternal electric field on electron transport in such a 1D magnetic-non-magnetic superlattice geometry. Here we show that, depending on the unit cell configuration, a 1D superlattice structure subjected to an external elec- tric field exhibits multiple mobility edges at different val- ues of the carrier energy. We use a simple tight-binding (TB) framework to illustrate the model quantum sys- tem and numerically evaluate two-terminal spin depen- dent transmission probabilities through the superlattice geometry based on the Green's function formalism. From our exact numerical analysis we establish that a sharp crossover from a completely opaque to a fully or partly transmitting zone takes place which leads to a possibility of tuning the electron transport by gating the transmis- sion zone. In addition to this behavior we also show that the magnetic-non-magnetic superlattice structure can be used as a pure spin filter for a wide range of energy. These phenomena enhance the prospect of such simple superlat- tice structures as switching devices at multiple energies as well as spin filter devices, the design of which has sig- nificant impact in the present age of nanotechnology. With an introduction in Section I, we organize the pa- per as follows. In Section II, first we present the model, then describe the theoretical formulation which include the Hamiltonian and the formulation for transmission probabilities through the model quantum system. The numerical results are illustrated in Section III and finally, in Section IV, we draw our conclusions. II. THEORETICAL FRAMEWORK Let us start with Fig. 1 where a 1D mesoscopic chain composed of magnetic and non-magnetic atomic sites is attached to two semi-infinite 1D non-magnetic elec- trodes, namely, source and drain. The chain consists of p (p being an integer) number of unit cells in which Source Conductor Drain Magnetic Atom Non−magnetic Atom FIG. 1: (Color online). A 1D mesoscopic chain composed of magnetic (filled magenta circle) and non-magnetic (filled green circle) atomic sites is attached to two semi-infinite 1D non-magnetic metallic electrodes, namely, source and drain. each unit cell contains n and m numbers of magnetic and non-magnetic atoms, respectively. Both the chain and side-attached electrodes are described by simple TB framework within nearest-neighbor hopping approxima- tion. The Hamiltonian for the entire system can be written as a sum of three terms as, H = Hc + Hl + Htun. (1) The first term represents the Hamiltonian for the chain and it reads i i (ǫi + ~hi.~σ)ci +Xi hc† i↓ (cid:17) ; ci =(cid:18) ci↑ tci+1 + h.c.i 0 ǫi (cid:19); ci↓ (cid:19) ; ǫi =(cid:18) ǫi 0 i↑ c† (2) c† Hc =Xi where, c† i =(cid:16) c† 0 1 (cid:19) ; and t = t(cid:18) 1 0 ~hi.~σ = hi(cid:18) cos θi sin θie−jφi sin θiejφi − cos θi (cid:19). Here, ǫi refers to the on-site energy of an electron at the site i with spin σ (↑, ↓), t is the nearest-neighbor hopping strength, ciσ† (ciσ) is the creation (annihilation) operator of an electron at the ith site with spin σ and hi is the strength of local magnetic moment where hi = 0 for non-magnetic sites. The term ~hi.~σ corresponds to the interaction of the spin of the injected electron with the local magnetic moment placed at the site i. The direction of magnetization in each magnetic site is chosen to be arbitrary and specified by angles θi and φi in spherical polar co-ordinate system for the ith atomic site. Here, θi represents the angle between the direction of magnetization and the chosen Z axis, and φi represents the azimuthal angle made by the projection of the local moment on X-Y plane with the X axis. In presence of bias voltage V between the source and 2 drain an electric field is developed, and therefore, the site energies of the chain becomes voltage dependent. Mathematically we can express it as ǫi = ǫ0 i + ǫi(V ), where ǫ0 is the voltage independent term. The voltage i dependence of ǫi(V ) reflects the bare electric field in the bias junction as well as screening due to longer range electron-electron interaction. In the absence of such screening the electric field varies uniformly along the chain and it reads ǫi(V ) = V /2 − iV /(N + 1), where N corresponds to the total number of atomic sites in the chain. In our present work, we consider both the linear and screened electric field profiles. As illustrative 0.5 0 y g r e n e e t i S -0.5 0 20 40 60 80 100 Site index FIG. 2: (Color online). Voltage dependent site energies in a 1D chain considering 100 atomic sites for three different electrostatic potential profiles when the bias voltage V is set equal to 1. example, in Fig. 2 we show the variation of voltage dependent site energies for three different electrostatic potential profiles for a chain considering 100 atomic sites and describe the nature of electronic localization for these profiles in the forthcoming section. The second and third terms of Eq. 1 describe the TB Hamiltonians for the 1D semi-infinite non-magnetic elec- trodes and the chain-to-electrode coupling. These Hamil- tonians are written as follows. Hl = Xα=S,D"Xn c† n ǫlcn +Xn (cid:2)c† n tlcn+1 + h.c.(cid:3)# (3) and, Htun = Htun,S + Htun,D = τs[c† 1 c0 + h.c.] + τd[c† N cN +1 + h.c.]. (4) The summation over S and D in Eq. 3 implies the incor- poration of both the two electrodes, viz, source and drain. ǫl and tl stand for the site energy and nearest-neighbor coupling, respectively. The electrodes are directly cou- pled to the chain through the lattice sites 1 and N , and the coupling strengths between these electrodes with the chain are described by τs and τd, respectively. To obtain spin resolved transmission probabilities of an electron through the source-chain-drain bridge system, we use Green's function formalism. The single particle Green's function operator representing the entire system for an electron with energy E is defined as, G = (E − H + iη)−1 (5) where, η → 0+. Following the matrix form of H and G the problem of finding G in the full Hilbert space H can be mapped exactly to a Green's function Gef f corresponding to an effective Hamiltonian in the reduced Hilbert space of the chain itself and we have, c G = Gef f c =Xσ where, (E − H c − Σσ S − Σσ D)−1 , (6) 3 flipping can be explained from the following arguments. The operators σ+ (= σx + iσy) and σ− (= σx − iσy) asso- ciated with the term ~hi.~σ in the TB Hamiltonian Eq. 2 are responsible for the spin flipping, where ~σ being the Pauli spin vector with components σx, σy and σz for the injecting electron. In our present model since we consider that all the magnetic moments are aligned along +Z di- rection, the term ~hi.~σ (= hixσx + hiyσy + hizσz) gets the form hizσz, and accordingly, the Hamiltonian does not contain σx and σy and so σ+ and σ− do not appear, Σσ S(D) = H † tun,S(D) GS(D)H tun,S(D). (7) These ΣS and ΣD are the self-energies introduced to in- corporate the effect of coupling of the chain to the source and drain, respectively. Using Dyson equation the ana- lytic form of the self energies can be evaluated as follows, 4 a T 2 V=0 n=1 m=4 p=80 TE ADOS Σσ S(D) = τ 2 s(d) E − ǫl − ξl (8) 0 -2 -1 0 Energy 1 2 S O D A S O D A S O D A 2 b T 1 V=0.2 n=1 m=4 p=80 TE ADOS 0 -2 -1 0 Energy 1 2 2 c T 1 V=0.5 n=1 m=4 p=80 TE ADOS where, ξl = (E − ǫl)/2 − ipt2 Following Fisher-Lee relation, the transmission proba- bility of an electron from the source to drain is given by the expression, l − (E − ǫl)2/4. Tσσ′ = Tr[Γσ S Gr Γσ′ D Ga]. (9) where, Γσ S(D)'s are the coupling matrices representing the coupling between the chain and the electrodes and they are defined as, Γσ S(D) = ihΣσ (10) S(D) − Σσ† S(D)i . k and Σσ† Here, Σσ k are the retarded and advanced self- energies associated with the k-th (k = S, D) electrode, respectively. Finally, we determine the average density of states (ADOS), ρ(E), from the following relation, ρ(E) = − 1 N π Im [Tr[G]] . (11) In what follows we limit ourselves to absolute zero tem- perature and use the units where c = e = h = 1. For the numerical calculations we set t = 1, ǫ0 i = 0 ∀ i, hi = 1 for the magnetic sites, θi = φi = 0, ǫl = 0, tl = 1 and τs = τd = 0.8. The energy scale is measured in unit of t. III. NUMERICAL RESULTS AND DISCUSSION Throughout our numerical calculations we assume that the magnetic moments are aligned along +Z direction (θi = φi = 0), which yields vanishing spin flip transmis- sion probability, viz, T↑↓ = T↓↑ = 0, across the bridge sys- tem. The net transmission probability is therefore a sum T (E) = T↑↑(E) + T↓↓(E), and the origin of this zero spin 0 -2 -1 0 Energy 1 2 FIG. 3: (Color online). Transmission probability T and ADOS as a function of energy for a 1D magnetic-non- magnetic superlattice geometry considering a linear bias drop along the chain, as shown by the pink curve in Fig. 2, where (a)-(c) correspond to the results for three different values of bias voltage V . which leads to the vanishing spin flip transmission proba- bility across the 1D chain. Below, we address the central results of our study i.e, the possibility of getting multiple mobility edges in 1D magnetic-non-magnetic superlattice geometries and how such a simple model quantum sys- tem can be used as a perfect spin filter for a wide range of energy. In Fig. 3 we show the variation of total transmission probability T along with the average density of states for a 1D magnetic-non-magnetic superlattice geometry considering a linear bias drop. Here we consider a 400- site chain in which each unit cell contains one magnetic and four non-magnetic sites and the results are shown for three different bias voltages. For the particular case when the chain is free from external electric field i.e., V = 0 electronic conduction through the bridge takes place for the entire energy band as shown in Fig. 3(a) which pre- dicts that all the energy eigenstates are extended in na- ture. The situation becomes really very interesting when the superlattice geometry is subjected to an external elec- tric field. It is illustrated in Figs. 3(b) and 3(c). From 4 a T 2 V=0 n=1 m=4 p=80 TE ADOS 0 -2 -1 0 Energy 1 2 2 b V=0.2 n=1 m=4 p=80 T 1 TE ADOS 0 -2 -1 0 Energy 1 2 2 c T 1 V=0.5 n=1 m=4 p=80 TE ADOS S O D A S O D A S O D A 4 energy zone where T drops to zero an insulating phase will appear, while for the other case, where T is finite, a metallic phase is observed and it leads to the possibility of controlling the electronic transmission by gating the transmission zone. The width of the localized regions between the band of extended regions increases with the strength of the electric field as clearly shown by com- 4 a T 2 V=0 n=1 m=4 p=80 TE ADOS 0 -2 -1 0 Energy 1 V=0.2 3 b n=1 m=4 p=80 TE ADOS T 1.5 S O D A 2 3 S O D A 0 -2 -1 0 Energy 1 2 3 c T 1.5 V=0.5 n=1 m=4 p=80 TE ADOS S O D A 0 -2 -1 0 Energy 1 2 0 -2 -1 0 Energy 1 2 FIG. 4: (Color online). Transmission probability T and ADOS as a function of energy for a 1D magnetic-non- magnetic superlattice geometry when the electrostatic poten- tial profile varies following the green curve shown in Fig. 2, where (a)-(c) represent the identical meaning as in Fig. 3. these spectra we notice that there are some energy re- gions for which the transmission probability completely drops to zero which reveals that the eigenstates associ- ated with these energies are localized, and they are sepa- rated from the extended energy eigenstates. Thus, sharp mobility edges are obtained in the spectrum, and, the total number of such mobility edges separating the ex- tended and localized regions in a superlattice geometry in presence electric field strongly depends on the unit cell configuration and it can be regulated by adjusting the number of magnetic and non-magnetic sites. This phenomenon describes the existence of multiple mobility edges in a superlattice geometry under finite bias con- dition. Now if the Fermi energy is fixed at a suitable (Color online). Transmission probability T and FIG. 5: ADOS as a function of energy for a 1D magnetic-non- magnetic superlattice geometry when the electrostatic poten- tial profile varies following the blue curve shown in Fig. 2, where (a)-(c) represent the identical meaning as in Fig. 3. paring the spectra given in Figs. 3(b) and 3(c), and, for strong enough field strength almost all energy eigenstates are localized. In that particular limit metal-to-insulator transition will no longer be observed. The above results are analyzed for a particular (linear) variation of electric field along the chain. To explore the sensitivity of getting metal-to-insulator transition on the distribution of electric field, in Figs. 4 and 5 we present the results for two different screened electric field pro- files taking the identical chain length. From the spectra we clearly observe that the width of the localized region gradually disappears with the flatness of the electric field profile in the interior of the bridge system. If the poten- tial drop takes place only at the chain-to-electrode inter- faces, i.e., when the potential profile becomes almost flat along the chain the width of the localized region almost vanishes and the metal-to-insulator transition is not ob- served, as is the case for the zero bias limit. Finally, we illustrate how such a simple magnetic-non- magnetic superlattice geometry can be utilized as a per- fect spin filter for a wide range of energy in absence of any 1.0 a (cid:173) (cid:173) T 0.5 0.0 -2 1.0 b n=1 m=4 p=80 ¯ ¯ T 0.5 0.0 -2 4.0 c 2.0 S O D A 0.0 -2 V=0 n=1 m=4 p=80 -1 0 Energy 1 2 V=0 -1 0 Energy 1 2 V=0 n=1 m=4 p=80 -1 0 Energy 1 2 FIG. 6: (Color online). T↑↑, T↓↓ and ADOS as a function of energy for a 1D magnetic-non-magnetic superlattice geometry in absence of external electric field. external electric field. As illustrative example, in Fig. 6 we present the transmission probabilities for up and down spin electrons together with the average density of states as a function of energy for a 1D magnetic-non-magnetic superlattice geometry. From the spectra we observe that the up and down spin electrons follow two different chan- nels while traversing through the superlattice geometry, since the spin flipping is completely blocked for this con- figuration. This splitting of up and down spin conduc- tion channels is responsible for spin filtering action and the total number of these channels strongly depends on the unit cell configuration. From Figs. 6(a) and (b) we clearly see that for a wide range of energy for which 5 the transmission probability of up spin electrons drops to zero value, shows non-zero transmission probability of down spin electrons. Therefore, setting the Fermi energy to a suitable energy region we can control the transmis- sion characteristics of up and down spin electrons, and, a spin selective transmission is thus obtained through the bridge system. Before we end, we would like to point out that since the overlap between the up and down spin conduction channels depends on the magnitudes of the local magnetic moments, we can regulate the spin degree of polarization (DOP) simply by tuning the strength of these magnetic moments and for a wide range of energies it (DOP) almost reaches to 100%. Thus, our proposed magnetic-non-magnetic superlattice geometry is a very good example for designing a spin filter. IV. CONCLUSION To conclude, in the present work we investigate in de- tail the spin dependent transport under finite bias con- dition through a 1D magnetic-non-magnetic superlattice geometry using Green's function formalism. We use a simple TB framework to describe the model quantum system where all the calculations are done numerically. From our exact numerical analysis we predict that in such a simple 1D magnetic-non-magnetic superlattice ge- ometry multiple mobility edges separating the localized and extended regions are obtained in presence of exter- nal electric field and the total number of mobility edges in the full energy spectrum can be controlled by arrang- ing the unit cell configuration. This phenomenon reveals that the superlattice geometry can be used as a switching device for multiple values of Fermi energy. The sensitiv- ity of metal-to-insulator transition and vice versa on the electrostatic potential profile is thoroughly discussed. Fi- nally, we analyze how such a superlattice geometry can be utilized in designing a tailor made spin filter device for wide range of energies. Setting the Fermi energy at a suitable energy zone, a spin selective transmission is obtained through the bridge system. All these predicted results may be utilized in fabricating spin based nano electronic devices. The results presented in this communication are worked out for absolute zero temperature. However, they should remain valid even in a certain range of finite temperatures (∼ 300 K). This is because the broaden- ing of the energy levels of the chain due to the chain-to- electrode coupling is, in general, much larger than that of the thermal broadening42. ∗ Electronic address: [email protected] 1 S. A. Wolf, D. D. Awschalom, R. A. Buhrman, J. M. Daughton, S. von Moln´ar, M. L. Roukes, A. Y. Chtchelka- nova, and D. M. Treger, Science 294, 1488 (2001). 2 L. P. Rokhinson, V. Larkina, Y. B. Lyanda-Geller, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 93, 146601 (2004). 3 S. Sahoo, T. Kontos, J. Furer, C. Hoffmann, M. Graber, A. 6 Cottet, and C. Schonenberger, Nature Phys. 1, 99 (2005). 4 N. Tombros, C. Jozsa, M. Popinciuc, H. T. Jonkman, and B. J. van Wees, Nature 448, 571 (2007). 5 I. A. Shelykh, N. T. Bagraev, N. G. Galkin, and L. E. Klyanchkin, Phys. Rev. B 71, 113311 (2005). 6 H. W. Wu, J. Zhou, and Q. W. Shi, Appl. Phys. Lett. 85, 1012 (2004). Brezini, Z. Phys. B 99, 381 (1996). 26 J. R. Borysowicz, Phys. Lett. A. 231, 240 (1997). 27 C. Hamaguchi, M. Yamaguchi, H. Nagasawa, M. Morifuji, A. Di Carlo, P. Vogl, G. Bohm, G. Trankle, G. Weimann, Y. Nishikawa and S. Muto, Jpn. J. Appl. Phys. 34, 4519 (1995). 28 D. H. Dunlap, H.-L. Wu, and P. Phillips, Phys. Rev. Lett. 7 D. Frustaglia, M. Hentschel, and K. Richter, Phys. Rev. 65, 88 (1990). Lett. 87, 256602 (2001). 29 A. S´anchez, E. Maci´a, and F. Dom´ınguez-Adame, Phys. 8 R. Ionicioiu and I. D'Amico, Phys. Rev. B 67, 041307(R) Rev. B 49, 147 (1994). (2003). 30 F. A. B. F. de Moura and M. L. Lyra, Phys. Rev. Lett. 9 A. A. Shokri, M. Mardaani, and K. Esfarjani, Physica E 81, 3735 (1998). 27, 325 (2005). 31 F. M. Izrailev and A. A. Krokhin, Phys. Rev. Lett. 82, 10 M. Mardaani and A. A. Shokri, Chem. Phys. 324, 541 4062 (1999). (2006). 11 A. A. Shokri and A. Daemi, Eur. Phys. J. B 69, 245 (2009). 12 A. A. Shokri and A. Saffarzadeh, J. Phys.: Condens. Mat- ter 16, 4455 (2004). 13 M. Dey, S. K. Maiti, and S. N. Karmakar, Eur. Phys. J. B 80, 105 (2011). 14 M. Dey, S. K. Maiti, and S. N. Karmakar, J. Appl. Phys. 109, 024304 (2011). 32 F. Dom´ınguez-Adame, V. A. Malyshev, F. A. B. F. de Moura, and M. L. Lyra, Phys. Rev. Lett. 91, 197402 (2003). 33 S. Aubry and G. Andr´e, in Group Theoretical Methods in Physics, Annals of the Israel Physical Society Vol. 3, edited by L. Horwitz and Y. Neeman (American Institute of Physics, New York, 1980), p. 133. 34 S. Sil, S. K. Maiti, and A. Chakrabarti, Phys. Rev. B 79, 15 M. Dey, S. K. Maiti, and S. N. Karmakar, Phys. Lett. A 193309 (2009). 374, 1522 (2010). 35 C. M. Soukoulis and E. N. Economou, Phys. Rev. Lett. 48, 16 M. Dey, S. K. Maiti, and S. N. Karmakar, J. Comput. 1043 (1982). Theor. Nanosci. 8, 253 (2011). 36 S. Das Sarma, S. He, and X. C. Xie, Phys. Rev. Lett. 61, 17 S. Bellucci and P. Onorato, Phys. Rev. B 78, 235312 2144 (1988). (2008). 37 M. Johansson and R. Riklund, Phys. Rev. B 42, 8244 18 S. Bellucci and P. Onorato, J. Phys.: Condens. Matter 19, (1990). 395020 (2007). 19 P. W. Anderson, Phys. Rev. 109, 1492 (1958). 20 G. H. Wannier, Phys. Rev. 117, 432 (1960). 21 E. Abrahams, P. W. Anderson, D. C. Licciardello, and T. V. Ramakrishnan, Phys. Rev. Lett. 42, 673 (1979). 22 H. M. James, Phys. Rev. 76, 1611 (1949). 23 D. Emin and C. F. Hart, Phys. Rev. B 36, 7353 (1987). 24 R. Ouasti, N. Jekri, A. Brezini, and C. Depollier, J. Phys.: Condens. Matter 7, 811 (1995). 25 N. Zekri, M. Schreiber, R. Ouasti, R. Bouamrane, and A. 38 A. Eilmes, R. A. Romer, and M. Schreiber, Eur. Phys. J. B 23, 229 (2001). 39 S. Sil, S. K. Maiti, and A. Chakrabarti, Phys. Rev. Lett. 101, 076803 (2008). 40 S. Sil, S. K. Maiti, and A. Chakrabarti, Phys. Rev. B 78, 113103 (2008). 41 S. K. Maiti and A. Nitzan, Phys. Lett. A 377, 1205 (2013). 42 S. Datta, Electronic transport in mesoscopic systems, Cam- bridge University Press, Cambridge (1995).
1501.06710	3	1501	2015-07-17T08:13:02	Majorana states and magnetic orbital motion in planar hybrid nanowires	[ "cond-mat.mes-hall", "cond-mat.quant-gas" ]	The Majorana phase boundaries in planar 2D hybrid (semiconductor-superconductor) nanowires are modified by orbital effects due to off plane magnetic components. We show that Majorana zero modes survive sizable vertical field tiltings, uncovering a remarkable phase diagram. Analytical expressions of the phase boundaries are given for the strong orbital limit. These phase boundaries can be fulfilled with attainable setups, such as an InAs nanowire of $150\, {\rm nm}$ in transverse width.	cond-mat.mes-hall	cond-mat	Majorana states and magnetic orbital motion in planar hybrid nanowires 1Institut de F´ısica Interdisciplin`aria i de Sistemes Complexos IFISC (CSIC-UIB), E-07122 Palma de Mallorca, Spain 2Departament de F´ısica, Universitat de les Illes Balears, E-07122 Palma de Mallorca, Spain (Dated: May 27, 2015) Javier Osca1, ∗ and Lloren¸c Serra1, 2 The Majorana phase boundaries in planar 2D hybrid (semiconductor-superconductor) nanowires are modified by orbital effects due to off plane magnetic components. We show that Majorana zero modes survive sizable vertical field tiltings, uncovering a remarkable phase diagram. Analytical expressions of the phase boundaries are given for the strong orbital limit. These phase boundaries can be fulfilled with attainable setups, such as an InAs nanowire of 150 nm in transverse width. PACS numbers: 73.63.Nm,74.45.+c I. INTRODUCTION The physics of 2D electron gases in magnetic fields has proved invaluable for the Condensed Matter field with, e.g., the celebrated quantum Hall effect1 as well as with many devices based on quantum wells, wires and dots.2 On the other hand, Majorana zero modes in quasi 1D wires have recently attracted strong interest, both as exotic quasi-particles and as candidates for topological quantum computing.3–6 In this article we relate 2D-gas properties and Majorana physics, showing the remark- able role of the orbital motions characteristic of 2D sys- tems in magnetic fields. Majorana modes in quasi 1D wires are effectively chargeless, zero-energy quasiparticles. They arise from the splitting, through a phase transition, of bulk elec- tronic states into pairs of quasiparticles on the wire ends, each one being its own antiparticle.7–18 An important feature of the Majorana mode is that it appears only when a critical value of the external magnetic field, a phase-transition threshold, has been surpassed. Several experiments with hybrid superconductor-semiconductor nanowires using tunneling spectroscopy from a normal conductor to the nanowire have observed a zero bias peak consistent with a Majorana state.19–22 The observed peak height is, however, an order of magnitude lower than the quantized value 2e2/h. This discrepancy is not yet well understood, as it might be due to effects ranging from fi- nite temperature, experimental and tunneling resolutions to other low-energy subgap states and possible inelastic and renormalization processes.23,24 In practice, distinguishing zero-bias peaks due to Ma- jorana modes from other potential sources relies on the detailed knowledge of the phase diagram in each partic- ular physical realization. Therefore, it is highly relevant knowing how Majorana physics is affected by the extra dimension in 2D, with respect to 1D. This question has been addressed with quasi-1D multiband wires,25–28 mod- elling transverse modes as mutually coupled 1D wires. This leads to essentially the 1D physics with only one Majorana mode at each nanowire end when an odd num- ber of transverse modes are above their critical magnetic field. However, the role of the magnetic orbital mo- FIG. 1. Schematic of a 2D planar nanowire showing the axis definitions. A magnetic field in a tilted direction is included. The density distribution of Majorana modes on the wire ends is qualitatively shown. tion has been usually disregarded. Addressing Majorana physics in 2D systems with orbital motion is therefore rel- evant as a way to discard alternative scenarios that have been suggested, such as attributing the observations to Kondo-like interaction effects.28 We want to clarify too that the above mentioned experiments19–22 used cylin- drical nanowires, for which the orbital effects may be different from those discussed here. In this article we show that in a planar nanowire the or- bital motion strongly affects the phase transition bound- aries. We show how the Majorana phase survives sizable vertical field tiltings (Fig. 1), even reaching the purely perpendicular orientation in some cases. This is not an intuitive result since the electronic orbital motion might lead to a gap closing, allowing edge propagating solu- tions, or it might totally change the character of the topological states. In this sense we note that the s-wave nanowires inside a magnetic field have an associated di- rectionality with their Majoranas located at the left and right edges (Fig. 1), while a 2D p-wave nanowire has their edge states located all along the 1D perimeter. In the strong orbital limit, the phase transitions occur for critical values of the polar angle, following a simple analytical law that does not depend on sample details. With parallel (x) field orientation the transition law is also analytical, while for intermediate regimes the phase transitions are obtained numerically. We assess the con- sistency between the phase diagram and direct calcula- tions of the Majorana modes in semi-infinite and finite 2D wires, emphasizing the importance of covariant grid discretizations for the latter.30,31 The work is organized as follows. Section II introduces the physical system in detail while Sec. III contains the algorithm used to determine the phase-transition bound- aries. In Sec. IV we discuss the phase diagrams for out-of-plane tiltings of the magnetic field with different strengths of the spin-orbit coupling. Sections V and VI deal with the actual Majorana solutions in semi-infinite and finite 2D Majorana nanowires, respectively. Section VII contains the conclusions of the work. II. PHYSICAL MODEL The combination of s-wave superconductivity, Rashba interaction and an external magnetic field is a well known source of Majorana Fermions.7 We consider electronic motion restricted to the x (longitudinal) and y (trans- verse) directions in presence of these three effects. The homogeneous magnetic field points in an arbitrary direc- tion and the edges are modeled as infinite square well potentials in the longitudinal and transverse directions (Fig. 1). The nanowire physics is described by a Hamiltonian of the Bogoliubov-deGennes kind, split in the following way HBdG = H0 + HZ + HR + Horb . (1) The successive contributions to Eq. (1) are the zero-field and superconducting energies. H0 = p2 x + p2 y 2m + V (x, y) − µ! τz + ∆s τx ; (2) the Zeeman term HZ = ∆B (sin θ cos φ σx + sin θ sin φ σy + cos θ σz) ; the Rashba coupling term HR = α ( pxσy − pyσx ) τz , (3) (4) and, finally, the magnetic orbital terms Horb = 2 2ml4 z y2τz − ml2 z ypx − α l2 z yσy . (5) In Eqs. (2)-(5) we used the following Nambu-spinor con- vention, relating discrete components with spin (↑↓) and isospin (⇑⇓) as Ψ ≡ (Ψ↑⇑, Ψ↓⇑, Ψ↓⇓, −Ψ↑⇓)T . The contributions in Eq. (2) are, in left to right order, the kinetic, electrical potential V , chemical potential µ and superconducting ∆s energies. The Pauli operators for isospin (particle-hole) are represented by τx,y,z while those for spin are σx,y,z. The superconductor term repre- sents an effective mean field approximation to more com- plicated interactions with a nearby s-wave superconduc- tor. The Zeeman term, Eq. (3), depends on parameter 2 ∆B and models the coupling of the spin with a magnetic field of arbitrary polar and azimuthal angles (θ, φ) ≡ n. The Rashba coupling Eq. (4) is the result of the self- interaction between the quasiparticle spin with its own motion. This interaction is due to the presence of a trans- verse electric field representing an internal asymmetry in the confinement along z that may be either intrinsic or externally induced. The first Rashba contribution, de- pending on pxσy is called the 1D Rashba term while the second one, pyσx, is the Rashba mixing term. si1/2 B,n = h(µ − ǫn)2 + ∆2 The joint effects of superconductivity, Zeeman and 1D Rashba terms give rise to independent Majorana states, one from each transverse band like in the 1D model. Each one of the modes has a different critical magnetic field ∆(c) , with n = 1, 2, . . . , and ǫn the transverse mode energies. Adding the Rashba mix- ing term to this scenario changes the critical fields due to the coupling between different transverse bands. It effec- tively allows only one Majorana zero mode in parameter regions where the 1D Rashba term would yield an odd number of them (even-odd effect).27,32 This is further dis- cussed in the Appendix as a particular analytical limit of the most general case presented below. Besides, in 1D wires the effect of tilting the magnetic field implies the additional requirement of the so-called projection rule ∆B sin θ sin φ < ∆s.33,36 This is due to the indirect gap closing of the infinite wire energy bands at ±kc due to the tilted field, where kc is a non vanishing wave num- ber. As discussed in Sec. IV, in 2D nanowires we find strong modifications of the critical magnetic fields, but the projection rule still applies. In a planar nanowire the perpendicular component of the magnetic field induces orbital motions of the nanowire quasiparticles. The magnetic orbital terms, Eq. (5), describe this motion and their effect on the Majorana states is the central point of this article. These terms de- pend on the magnetic length lz, defined as l2 z = c/eBz, and they stem from the kinetic and Rashba energies with the magnetic substitution px → px − y/l2 z and adding the required Pauli matrix τz for proper particle-hole sym- metry. In Eq. (5) we assumed the Landau gauge centered on yc = 0, although our results are independent on this choice as discussed below. All parameters of the complete Hamiltonian are con- stant inside the nanowire, modeled as a perfectly confin- ing box with Lx ≫ Ly. The numerical results of this work are presented in characteristic units of the problem obtained by taking , m and the width of the nanowire Ly as reference values. That is, our length and energy units are, respectively, LU ≡ Ly and EU ≡ 2/mL2 y. A spin- orbit length Lso is usually defined as Lso = 2/mα but, as explained below, the Hamiltonian orbital terms will in- troduce an effective transversal confinement. Therefore, the comparison between its characteristic length and the one of the nanowire real confinement is relevant. We no- tice that in our convention, the numerical value of α is precisely the ratio of transverse and spin-orbit lengths, α/EU LU = Ly/Lso. III. THE MATCHING METHOD In topological systems it is in general possible to re- late the states of the bulk with those at the boundaries, a consequence of the bulk-to-edge correspondence prin- ciple. In our particular case this means that the semi- infinite Majorana solution Ψ will be expressed as a linear superposition of the infinite-nanowire eigensolutions Ψ(k) (i.e., for the same Hamiltonian but disregarding the left and right edges), Ψ(x, y, ησ, ητ ) =Xk Ck Ψ(k)(x, y, ησ, ητ ) , (6) where spin and isospin variables are indicated with ησ and ητ , respectively. Notice that the infinite-nanowire solutions Ψ(k) are characterized by a wave number k that, accounting for evanescent waves, may be a complex quan- tity. In a given spin-isospin basis, χsσ (ησ) and χsτ (ητ ), with sσ = ± and sτ = ±, the infinite-wire solutions read Ψ(k)(x, y, ησ, ητ ) = Xsσ sτ Ψ(k) sσ sτ (y) eikx χsσ (ησ) χsτ (ητ ) , (7) sσ sτ (y) is a 1D 4-component spinor characteristic where Ψ(k) of the infinite-wire solution with wave number k. It has been demonstrated that a Majorana phase tran- sition occurs in a semi-infinite nanowire when the propa- gating bands for the corresponding infinite nanowire per- form a gap closing and reopening when increasing the magnetic field, at vanishing energy and wave number.8,33 Therefore, to determine the phase boundaries we only need to investigate the band structure at k = 0 and E = 0. However, a full determination of the band spec- trum for every set of Hamiltonian parameters by diago- nalization is time consuming and computationally inef- fective. In accordance with this, it has been pointed out that in spite of the non locality of the topological states a full knowledge of the band spectrum is not necessary in general to determine the phase of a topological system, but the only relevant regions are those near the Dirac cones that appear at the phase transitions.34 In our case, this implies searching the solutions of hsσsτ hs′ σs′ τ i Ψ(0) s′ σ s′ τ (y) = 0 , (8) Xs′ σs′ τ where h is obtained neglecting all px-dependent terms in Eq. (1), h = p2 y 2m + V (y) − µ! τz + ∆s τx − α h pyσxτz +∆B σ · n + 2 2ml4 z y2τz − α l2 z yσy . (9) 3 Notice that with Eq. (8) we achieved a reduction to an effective 1D problem. We can use the algorithm devised in Refs. 32 and 35 to study the solutions of Eq. (8). The particular parameter sets allowing such a solution will signal the gap closing of the original 2D Hamiltonian of Eq. (1) and thus the phase transition we are looking for. The algorithm con- sists in solving Eq. (8) in a 1D grid as a linear system, assuming vanishing boundary conditions at y = 0 and y = Ly. However, due to the homogeneous character of this linear system the trivial solution Ψ(0) sσsτ (y) = 0 is al- ways possible. The algorithm discards the trivial solution by means of a matching point ym, where for an arbitrary pair of components (sσsτ ) = (st) a non vanishing wave function is imposed. In addition, continuity of the first derivative at the matching point is also imposed for the components other than (st), Ψ(0) st (ym) = 1 , (10) (cid:18) d(U) dy − d(L) dy (cid:19) Ψ(0) sσsτ (ym) = 0 , (sσ, sτ ) 6= (s, t) , (11) where d(U,L)/dy denote grid derivatives using only upper (U) or lower (L) y-grid neighbors. Equations (10) and (11) are used at ym in place of the Bogoliubov-deGennes ones. In particular, Eq. (10) makes the system no longer homogenous and such that it always admits a solution. The algorithm does not ensure the first-derivative continuity for the component (st) at the matching point. Therefore, this condition is used to distinguish the physical from the non-physical solutions with the continuity measure (cid:18) d(U) dy − F =(cid:12)(cid:12)(cid:12)(cid:12) 2 . (12) d(L) st (ym)(cid:12)(cid:12)(cid:12)(cid:12) dy (cid:19) Ψ(0) As mentioned, the F zeroes will signal the desired gap closing boundaries of Eq. (1). Further details about the algorithm can be found in Ref. 35. IV. PHASE DIAGRAMS Figure 2b shows the phase transition boundaries ob- tained with the matching method for an Ly = 150 nm nanowire with material parameters typical of InAs and a magnetic field strength between 0 and 6 T (∆B = 0 − 25 EU ). Phase boundaries are signaled by zero values of F (red lines). Only F = 0 values represented by red lines have physical sense and the color scale is only indi- cating the measure deviation from zero. As mentioned, the energy unit scales as L−2 y , such that in a 300 nm wire 25EU would correspond to 1.5 T. The other panels of Fig. 2 correspond to lower (2a) and higher (2c,d) values of the Rashba coupling strength. We will explicitly calcu- late the zero modes for particular sets of parameters be- low. Here, let us anticipate that the orbital terms do not change the even-odd effect of multiband nanowires.25–28 4 boundaries in particular limits (see Appendix). For θ = 90◦ and φ = 0 the critical magnetic fields read ∆(c) B,n =s(cid:18)µ − ǫn + mα2 22 (cid:19)2 + ∆2 s , (13) where n = 1, 2 . . . and ǫn are the (transverse) square well eigenenergies. This analytical result extends recent find- ings from other authors25,28 who assumed that the contri- bution in parenthesis in Eq. (13) is an effective chemical potential from subband n, without specifying its detailed α dependence. Analogously, in the strong orbital limit ∆B >> (2/mL2 y, mα2/2, ∆s, µ) the critical angles are θ(c) n = arccos(cid:18) gm∗ 4(n − 1 2 )(cid:19) , (14) where g is the gyromagnetic factor and m∗ = m/me the ratio between the electron effective and bare masses (m and me respectively). In this limit quasiparticles are con- fined by the effective ∆B-dependent harmonic potential caused by the first and second terms of Eq. (5), inde- pendently of the real nanowire transversal boundaries. Quite remarkably, Eq. (14) is independent of the mag- netic field, Rashba and superconductivity strengths, as well as on the specific wire width Ly. In this sense, the critical angles are rather universal. The values of Eqs. (13) and (14) for the particular parameters used in Fig. 2 are overprinted as vertical and horizontal dashed lines, respectively. As shown, the numerical values match the analytical ones in their corresponding limits. In Figs. 2a,b the transitions boundaries do not deviate substan- tially from the analytical laws in all the plot, while Figs. 2c,d show large differences for intermediate values of the parameters. The structure of the phase transitions in Figure 2a is typical for cases when the kinetic orbital effects already dominate around the first transition boundary with in- creasing ∆B. The phase boundaries just bend from a vertical to a horizontal line due to the effective change of the transversal confinement in the nanowire, from square well to a ∆B-dependent harmonic confinement [Eq. (5)]. Assuming µ ≈ 0, the conditions for this simpler phase diagrams (lz shortest scale) can be written as a triple inequality ∆(c) B,1 cos θ(c) 1 ≫ gm∗ 4 (cid:18)EU , α2 EU L2 U , ∆s(cid:19) . (15) Figure 2b (InAs with Ly = 150 nm) still presents a phase diagram qualitatively similar to Fig. 2a although the second inequality of Eq. (15) is not well satisfied (see Tab. I in Appendix). On the other hand, Figures 2c and d show the modifications of the phase diagram as α increases in effective units. As anticipated, the deviations can even allow a Majorana state in perpendicular field (θ = 0) in Fig. 2d. However, so large spin orbit strengths FIG. 2. F measure in a color (gray) scale as a function of ∆B and polar angle θ. The azimuthal angle remains φ = 0. Zero values of F (red) signal the phase transition boundaries. Phases with a Majorana mode are labeled with an M. From top to bottom the panels correspond to α = 0.1 EU LU (a), 2 EU LU (b), π EU LU (c), 4 EU LU (d). Dashed lines indi- cate the analytical limits. We have assumed ∆s = 3 EU and typical InAs parameters g = 15, m∗ = 0.033. Panel (b) corrresponds to an InAs nanowire with α = 30 meVnm, ∆s = 0.3 meV and Ly = 150 nm in a magnetic field range from zero to 6 T. The phases in Fig. 2 contain either no Majoranas or at most one Majorana mode in regions labeled with an M. The main result of this article is that orbital effects do not destroy Majoranas into other phases. However, they do lead to characteristic phase maps where the Ma- jorana states survive sizable vertical tiltings (even up to θ = 0 in Fig. 2d). The change in transition boundaries is caused by two reasons. First, by the change in the ef- fective transversal confinement due the first two terms of Eq. (5). Note that the first contribution leads to a har- monic confining. Second, the third term of Eq. (5) can be understood as an effective inhomogeneous magnetic field pointing in y direction due to the combination of Rashba and orbital effect. It is possible to give analytical expressions of the phase give rise to complicated phase maps that strongly deviate from the analytical limits. z of Eq. (5). The strong-α effects seen in the lower panels of Fig. 2 are caused by the term −αyσy/l2 In- deed, this term effectively adds a component along y to the magnetic field. Therefore, the effective angle θe is such that θe > θ, thus explaining the downwards shift of the lower phase transition boundary in Figs. 2c and d. Figures 2a,b do not change with other azimuthal an- gles while in the strong-α diagrams (Fig. 2c,d) φ mod- ifies the precise boundary positions for θ = 90◦ (ver- tical lines), but not the horizontal asymptotes and the overall qualitative behavior. Note that, as mentioned in Sec. II, with φ 6= 0 there is an additional require- ment for the existence of Majorana modes, the projection rule ∆B sin θ sin φ < ∆s.33,36 However, the effective tilt- ing of the magnetic field towards y caused by the term −αyσy/l2 z does not modify the projection rule because this effective tilting is in opposite directions for positive and negative y's while the projection rule applies only to homogeneous magnetic fields. V. SEMI-INFINITE NANOWIRES Explicit zero-energy eigenstates can be obtained in semi-infinite and finite nanowires in order to confirm the above phase diagrams. We have checked that either one or none Majoranas are obtained in the corresponding re- gions of Fig. 2. In the semi-infinite system exact zero- energy solutions can be studied with the complex band structure method of Refs. 32 and 33. The calculation is carried out solving the boundary condition CkΨ(k) sσ ,sτ (y) = 0 , (16) Xk where the allowed complex wave numbers {k} and the Ψ(k) sσ sτ (y) functions are obtained with the matching method discussed previously in Sec. III, with the only difference that k now is not vanishing. Equation (16) can be reworked into Mk′kCk = 0 , Xk with the matrix Mk′k = Xsσ,sτZ dy Ψ(k′)∗ sσ,sτ (y)Ψ(k) sσ ,sτ (y) . (17) (18) Equation (17) shows that, when enough wave numbers are included, each Majorana state is represented by a null-space eigenvector of matrix M. In Fig. 3a we can see the convergence of the M eigenvalues with the cut off in wave number for a particular point of Fig. 2c. Clearly, the lower eigenvalue vanishes asymptotically indicating that for this point of the phase diagram a Majorana mode is present as expected. In Fig. 3b we can see the corre- sponding density function, confirming the edge character of the mode, as also expected for a Majorana. a) b) c) s e u l a v n e g i e M 103 100 10–3 10–6 6 0.5 0.0 –0.5 0.5 0.0 ) U L ( y ) U L ( y –0.5 0 5 18 k ( L U ) −1 30 1 2 3 x ( L U ) FIG. 3. a) Evolution of the lower eigenvalues of matrix M when increasing the number of evanescent modes, as given by a cutoff k. b) Majorana density function in a semi-infinite nanowire for the null eigenvector of M. c) Majorana den- sity function in a finite nanowire with Lx = 20LU calculated by direct diagonalization of the Hamiltonian using covari- ant derivative discretization. The three panels correspond to ∆B = 10 EU , θ = 75◦ and the rest of parameters as in Fig. 2c. A. Decay lengths Within our complex-band-structure approach to the semi-infinite nanowire we can estimate the length of the Majorana decay tail from the imaginary part of the al- lowed wave numbers. The lower the imaginary part, the longer the Majorana decay tail (and thus the required length of the nanowire to contain it without distortion). Figure 4 shows a typical evolution of the wave numbers (red islands) in the complex plane as the polar angle is ap- proaching the critical value. In the sequence from upper to lower panels, one of the wave numbers moves along the imaginary axis towards the origin; the phase transition being signaled by one mode touching the origin (lower panel). We calculate the required nanowire length with the smallest imaginary wave number of the set of all allowed wave numbers {k(m)}. However, as shown in Fig. 4 the smallest imaginary part ℑ(k(m)) ≡ k(m) changes from an approximately fixed mode to the one touching the origin when approaching the phase transition. We define the i 6 FIG. 5. Mode lengths Lm (defined in Sec. V A) for the two wave numbers shown in Fig. 4. Note that the required nanowire length at each θ is the higher of both curves. aries become blurred due to the finite size effect. Figure 3c shows the density of the finite nanowire Majorana cor- responding to the semi infinite one of Fig. 3b. Differences are small, just a slight distortion and a somewhat longer decay tail of the finite-nanowire density. In the finite nanowire diagonalization with orbital terms we have found it crucial using a covariant grid discretization.30,31 Otherwise, numerical artificial biases wrongly suggest that Majoranas are always destroyed by orbital terms,29 in clear contradiction with the phase di- agram (Fig. 2) and the semi-infinite wire analysis. In essence, the covariant discretization amounts to express- ing the canonical momentum components as symmetry- like transformations. For instance, Πx ≡ −i ∂ ∂x − y l2 z = eiyx/l2 z(cid:18)−i ∂ ∂x(cid:19) e−iyx/l2 z . (19) Although these two representations of the canonical op- erator are equivalent in the continuous limit, they are not on a discrete grid. As demonstrated in Ref. 31 the covariant derivative preserves by construction the gauge invariance of the so- lutions while a non-covariant treatment only does that for extremely fine discretizations, unfeasible in our case. Changing the gauge origin usually constitutes a severe difficulty for numerical discretizations not using covariant derivative formulations. In our case, we can introduce an arbitrary gauge center yc for the canonical momentum, generalizing Eq. (19) to Πx = ei(y−sτ yc)x/l2 z(cid:18)−i ∂ ∂x(cid:19) e−i(y−sτ yc)x/l2 z , (20) FIG. 4. F measure in a color (gray) scale showing the position of the allowed wave numbers as zeroes (red islands). The parameters used are g = 15, m∗ = 0.033, α = π EU LU , ∆s = 3 EU , ∆B = 10 EU . Panels from top to bottom are for different polar angles: θ = 68◦ (a), θ = 67◦ (b), θ = 66◦ (c). mode length Lm as two times the length needed for the wave function to drop to one percent of its maximum, that is e−k(m) i Lm/2 = 0.01. An estimate of the nanowire length for undistorted Majoranas is simply the maximum of all mode lengths. In Fig. 5 we show the mode lengths of the two allowed wave numbers of Fig. 4. As we decrease the polar angle from 90 degrees, the needed nanowire length (the higher of the two curves) remains more or less stable until θ approaches the critical value. A few degrees before the transition the Majorana contracts before diverging to in- finity at the phase transition angle. VI. FINITE NANOWIRES The phase diagram can also be checked with full di- agonalizations of nanowires with large, but finite, Lx. Though more realistic, this approach is conceptually more qualitative, since finite-nanowire Majoranas are not exact zero modes but small energy modes (the smaller the energy the larger Lx). Equivalently, the phase bound- where the isospin sign sτ = ± is introduced in order to preserve the particle-hole symmetry of the Bogoliubov- deGennes equation. We have checked that our numerical results for the finite nanowire diagonalization as, e.g., in the lower panel of Fig. 3, do not depend on the choice of yc, thus proving the gauge invariance of the finite sys- tem results. We have also obtained good agreement of the finite nanowire diagonalizations and the results of the semi-infinite system regarding the existence or absence of a zero mode in the different regions of the phase diagrams (Fig. 2), again proving the reliability of the method. No- tice that the semi-infinite solution, being purely 1D, can be obtained in very dense y grids, while the finite system 2D diagonalization requires much coarser xy grids. VII. CONCLUSIONS In this work we have shown that the orbital motions caused by perpendicular components of the magnetic field in planar 2D nanowires give rise to a rich phase diagram, with regions containing Majoranas for sizable vertical tilting of the magnetic field. In fact, with proper parameters, it is possible to find Majoranas in a fully per- pendicular field. We have developed a general numerical method to obtain the Majorana phases in nanowires in a computer efficient way and we have checked this method against alternative calculations for semi-infinite and fi- nite nanowires. Analytical expressions of the transition boundaries in asymptotic regions have been found. For realistic parameter values (weak α) these analytical ex- pressions are a good approximation in general and not only asymptotically. In the strong orbital limit the criti- cal angles are independent of sample details. Finally, the relevance of the covariant grid discretization for the finite nanowire diagonalization has been pointed out. ACKNOWLEDGMENTS We acknowledge useful discussions with R. L´opez and D. S´anchez. This work was funded by MINECO- Spain (grant FIS2011-23526), CAIB-Spain (Conselleria d'Educaci´o, Cultura i Universitats) and FEDER. We hereby acknowledge the PhD grant provided by the Uni- versity of the Balearic Islands. Appendix: Analytical limits 1. Longitudinal magnetic field When the magnetic field is along x (see axis orienta- tions in Fig. 1) the phase transition law is fully analytical. As discussed in Sec. III, finding the phase transition im- plies searching for the zero energy eigenstates of the sim- plified Hamiltonian h given in Eq. (9). For an x-oriented field l−2 z = 0 and all orbital terms vanish. Assuming also a vanishing bottom potential for V (y) we have h = p2 y 2m − µ! τz + ∆s τx +(cid:16)∆B − α pyτz(cid:17) σx . (A.1) 7 FIG. 6. a) Phase transition boundaries for a 2D planar nanowire with a longitudinal (x) magnetic field. We have as- sumed ∆s = 3EU and µ = 0. The shaded regions correspond to topological phases with a Majorana zero mode. interaction can be absorbed in the kinetic term h = p2 y 2m − mα2 22 − µ! τz + ∆s τx + ∆Bσx , (A.2) where py = py − mασx/. Using a basis of square-well eigenstates of energies ǫn = 2π2n2/(2mL2 y), the matrix to diagonalize is h ≡ ǫn − µ − mα2 ∆s 22 ∆s −(ǫn − µ − mα2 22 )! . (A.3) The diagonalization of this simplified Hamiltonian yields the eigenenergies Ens1s2 = s1∆B + s2s(cid:18)µ − ǫn + mα2 22 (cid:19)2 + ∆2 s , (A.4) with n = 1, 2, . . . , s1 = ±1 and s2 = ±1. Of the four eigenenergies only the two with opposite s1 and s2 can lead to a zero energy solution at the critical values ∆(c) B,n =s(cid:18)µ − ǫn + mα2 22 (cid:19)2 + ∆2 s . (A.5) Equation (A.5) with n = 1, 2, . . . gives the critical Zee- man parameter of phase transitions for a two dimensional nanowire in parallel magnetic field. Notice that in Eq. (A.5) n has to be interpreted simply as an ordering in- dex of the successive transitions, and not as a label of independent transverse modes. These latter interpreta- tion would be wrong, since different transverse modes are coupled through the Rashba mixing term and one can not associate a particular transverse mode with a particular transition point. Shaded regions in Fig. 6 contain one Majorana mode, while white regions have none. There are no regions with multiple Majoranas due to the en- ergy splittings induced by the Rashba mixing in planar nanowires.27,32 2. Strong orbital limit The eigenstates of Eq. (A.1) can be obtained analyt- ically noticing that the linear py term from the Rashba When the kinetic orbital effect overcomes both the con- finement by the transverse square well and the Rashba Panel a) b) c) d) ∆(c) B,1 cos θ(c) 1 1.43 1.03 0.74 1.06 gm∗ 4EU 0.12 0.12 0.12 0.12 U gm∗ α2 4E2 U L2 0.001 0.48 1.22 1.98 gm∗∆s 4 0.36 0.36 0.36 0.36 TABLE I. Numerical values in effective units of the inequal- ities, Eq. (11) of the paper, corresponding to the four panels in Fig. 2 of the paper. term, the magnetic length lz ≡pc/eBz is smaller than ly (here we define ly = Ly of Fig. 1) and also smaller than the Rashba length lα ≡ 2/mα. In the limit lz ≪ (ly, lα) it is possible to derive an analytical expression of the transition boundaries. Neglecting the square well V (y) and the Rashba terms in Eq. (9) we find h = p2 y 2m + 2 2ml4 z y2 − µ! τz + ∆B ~σ · n + ∆s τx . (A.6) The eigenvalues of Eq. (A.6) are straightforward in a basis nsσsτ i, where n = 1, 2, . . . represent now harmonic oscillator eigenstates, sσ = ± indicates spin eigenstates in direction n, while sτ = ± indicates isospin in direction z. Since the h matrix is diagonal in spin, we can diag- onalize each subspace independently. For instance, the matrix for sσ = + reads n − µ + ∆B ǫ(ho) ∆s ∆s n − µ(cid:17) + ∆B ! , −(cid:16)ǫ(ho) (A.7) n = (n − 1/2)2/ml2 with ǫ(ho) z. The eigenvalues of Eq. (A.7) are easily found, as well as those of the analogous matrix for spin sσ = −. ∗ [email protected] 1 T. Chakraborty and P. Pietilainen, The quantum Hall ef- fects (Springer, 1995). 2 T. Ihn, Semiconductor nanostructures (Oxford, 2010). 3 J. Alicea, Rep. Prog. Phys., 75, 076501 (2012). 4 C. W. J. Beenakker, Annu. Rev. Condens. Matter Phys., 4, 113 (2013). 5 T. D. Stanescu and S. Tewari, J. Phys. Condens. Matter, 25, 233201 (2013). 6 M. Franz, Nature Nanotechnology, 8, 149 (2013). 7 R. M. Lutchyn, J. D. Sau, and S. Das Sarma, Phys. Rev. Lett., 105, 077001 (2010). 8 Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett., 105, 177002 (2010). 9 T. D. Stanescu, R. M. Lutchyn, and S. Das Sarma, Phys. Rev. B, 84, 144522 (2011). 10 K. Flensberg, Phys. Rev. B, 82, 180516 (2010). 11 S. Gangadharaiah, B. Braunecker, P. Simon, and D. Loss, Phys. Rev. Lett., 107, 036801 (2011). 8 The null-eigenvalue condition for h is now ∆B =s(cid:20)(n − 1/2) 2 ml2 z − µ(cid:21)2 + ∆2 s , (A.8) that looks similar to Eq. (A.5). An essential difference, however, is that the r.h.s. in Eq. (A.8) depends itself on the Zeeman parameter ∆B through lz. It is 2 ml2 z = 4 gm∗ ∆B cos θ , (A.9) where m∗ is the ratio of effective to bare mass, m = m∗me, while g is the gyromagnetic factor defined from the Zeeman parameter by ∆B ≡ gµBB/2. From Eq. (A.8) we finally arrive at the following relation cos θ = . (A.10) B − ∆2 s + µ gm∗ 4 p∆2 (cid:0)n − 1 2(cid:1) ∆B For large enough ∆B, as compared to ∆s and µ, this leads to the prediction of field-independent critical angles , (A.11) cos θ(c) n = as given in Eq. (14). gm∗ 4(cid:0)n − 1 2(cid:1) The triple inequality lz ≪ (ly, lα, ls), where we define ls ≡ p2/m∆s, leads, when written in effective units, to Eq. (15). In this situation the phase diagram does not deviate much from the straight lines of the analytical limits, Eqs. (A.5) and (A.11). Table I contains the nu- merical values of the inequality sides for the four panels in Fig. 2. While panel a) fulfills all conditions, for the rest of panels the second inequality degrades as α increases from panel b) to d). This explains the deviations in those panels from the analytical limits. 12 C.-H. Lin, J. D. Sau, and S. Das Sarma, Phys. Rev. B, 86, 224511 (2012). 13 J. Klinovaja and D. Loss, Phys. Rev. B, 86, 085408 (2012). 14 D. Sticlet, C. Bena, and P. Simon, Phys. Rev. Lett., 108, 096802 (2012). 15 X.-J. Liu and A. M. Lobos, Phys. Rev. B, 87, 060504 (2013). 16 L. Jiang, D. Pekker, J. Alicea, G. Refael, Y. Oreg, A. Brataas, and F. von Oppen, Phys. Rev. B, 87, 075438 (2013). 17 F. Pientka, G. Refael, New Journal of Physics, 15, 115001 (2013). L. Y. Oreg, Jiang, D. Pekker, and F. J. Alicea, von Oppen, 18 P. San-Jose, E. Prada, and R. Aguado, Phys. Rev. Lett., 108, 257001 (2012). 19 V. Mourik, K. Zuo, S. Frolov, S. Plissard, E. Bakkers, and L. Kouwenhoven, Science, 336, 1003 (2012). 20 M. T. Deng, C. L. Yu, G. Y. Huang, M. Larsson, P. Caroff, and H. Q. Xu, Nano Letters, 12, 6414 (2012), pMID: 23181691, http://dx.doi.org/10.1021/nl303758w. 21 A. Das, Y. Ronen, Y. Most, Y. Oreg, M. Heiblum, and H. Shtrikman, Nature Physics, 8, 887 (2012). 22 A. D. K. Finck, D. J. Van Harlingen, P. K. Mohseni, K. Jung, and X. Li, Phys. Rev. Lett., 110, 126406 (2013). 23 Y. Peng, F. Pientka, L. I. Glazman, and F. von Oppen, Phys. Rev. Lett., 114, 106801 (2015). 24 S. D. Sarma, H.-Y. Hui, P. M. R. Brydon, and J. D. Sau, arXiv:1503.00594v1 (2015). 25 R. M. Lutchyn, T. D. Stanescu, and S. Das Sarma, Phys. Rev. Lett., 106, 127001 (2011). 26 A. C. Potter and P. A. Lee, Phys. Rev. Lett., 105, 227003 (2010). 27 A. C. Potter and P. A. Lee, Phys. Rev. B, 83, 094525 (2011). 28 P. San-Jose, E. Prada, and R. Aguado, Phys. Rev. Lett., 9 112, 137001 (2014). 29 J. S. Lim, L. Serra, R. Lopez, and R. Aguado, Phys. Rev. B, 86, 121103 (2012). 30 M. Governale and C. Ungarelli, Phys. Rev. B, 58, 7816 (1998). 31 S. Janecek and E. Krotscheck, Phys. Rev. B, 77, 245115 (2008). 32 L. Serra, Phys. Rev. B, 87, 075440 (2013). 33 J. Osca, D. Ruiz, and L. Serra, Phys. Rev. B, 89, 245405 (2014). 34 A. R. Akhmerov, J. P. Dahlhaus, F. Hassler, M. Wimmer, and C. W. J. Beenakker, Phys. Rev. Lett., 106, 057001 (2011). 35 J. Osca and L. Serra, Phys. Rev. B, 88, 144512 (2013). 36 S. Rex and A. Sudbø, Phys. Rev. B, 90, 115429 (2014).
1507.02161	3	1507	2019-06-05T05:25:47	Generalization of Laughlin's Theory for the Fractional Quantum Hall Effect	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	Motivated by the quasiparticle wavefunction in the composite fermion (CF) theory for fractional quantum Hall filling factor $\nu = 1/m$, I consider a suitable quasiparticle operator in differential form, as a modified form of Laughlin's quasiparticle operator, that reproduces quasiparticle wave function as predicted in the CF theory. I further consider the conjugate of this operator as quasihole operator for obtaining a novel quasihole wave function for $1/m$ state. Each of these wave functions is interpreted as expelled electron into a different Hilbert subspace from the original Hilbert space of Laughlin condensate while still maintaining its correlation (although changed) with the electrons in the condensate such that the expelled electron behaves as a CF with respect to the electrons in the condensate. With this interpretation, I show that the ground state wavefunctions for general states at filling fractions $\nu_{n,m}^{\pm} = n/[n(m\mp 1)\pm 1]$, respectively, can be constructed as coherent superposition of $n$ coupled Laughlin condensates and their conjugates, formed at different Hilbert subspaces. The corresponding wave functions, specially surprising for $\nu_{n,m}^{-}$ sequence of states, are identical with those proposed in the theory of noninteracting CFs. The states which were considered as fractional quantum Hall effect of interacting CFs, can also be treated in the same footing as for the prominent sequences of states describing as the coupled condensates among which one is a non-Laughlin condensate in a different Hilbert subspace. Further, I predict that the half filling of the lowest Landau level is a quantum critical point for phase transitions between two topologically distinct phases each corresponding to a family of states.	cond-mat.mes-hall	cond-mat	Generalization of Laughlin's Theory for the Fractional Quantum Hall Effect Sudhansu S. Mandal 9 1 0 2 n u J 5 ] l l a h - s e m . t a m - d n o c [ 3 v 1 6 1 2 0 . 7 0 5 1 : v i X r a Department of Physics and Centre for Theoretical Studies, Indian Institute of Technology Kharagpur, West Bengal 721302, India Abstract Motivated by the structure of the quasiparticle wavefunction in the composite fermion (CF) theory for fractional quantum Hall filling factor ν = 1/m (m odd), I consider a suitable quasi- particle operator in differential form, as a modified form of Laughlin's quasiparticle operator, that reproduces quasiparticle wave function as predicted in the CF theory, without a priori assumption of the presence of CFs. I further consider the conjugate of this operator as quasihole operator for obtaining a novel quasihole wave function for 1/m state. Each of these wave functions is in- terpreted as expelled electron into a different Hilbert subspace from the original Hilbert space of Laughlin condensate while still maintaining its correlation (although changed) with the electrons in the condensate such that the expelled electron behaves as a CF with respect to the electrons in the condensate. With this interpretation, I show that the ground state wavefunctions for general n,m = n/[n(m ∓ 1) ± 1], respectively, can be constructed as coherent su- states at filling fractions ν± perposition of n coupled Laughlin condensates and their "conjugates", formed at different Hilbert subspaces. The corresponding wave functions, specially surprising for ν− n,m sequence of states, are identical with those proposed in the theory of noninteracting CFs. The states which were consid- ered as fractional quantum Hall effect of interacting CFs, can also be treated in the same footing as for the prominent sequences of states describing as the coupled condensates among which one is a non-Laughlin condensate in a different Hilbert subspace. Further, I predict that the half filling of the lowest Landau level is a quantum critical point for phase transitions between two topologically distinct phases each corresponding to a family of states: One consists of large number of coupled Laughlin condensates of filling factor 1/3 and the other corresponds to large number of coupled conjugate Laughlin condensates of filling factor 1, which may be distinguished, respectively, by the absence and presence of upstream edge modes. 1 I. INTRODUCTION Correlated quantum liquids formed by two-dimensional electrons exposed to a large mag- netic field manifest through the phenomenon of fractional quantum Hall effect (FQHE)1, characterized by different topological quantum numbers labeled2 with the corresponding filling fractions. The single particle wave function for two-dimensional electrons in the low- est Landau level (LL) subjected to a magnetic field B perpendicular to the plane of the system is ψl(z) = zl exp[−z2/4] with complex coordinate z = (x− iy)/(cid:96)B, angular momen- tum of the degenerate orbitals l = 0, 1, 2,··· , and magnetic length (cid:96)B = (c/eB)1/2. The manybody wave function describing any correlated liquid, in principle, will be an appro- priate linear combination of wave functions for occupied orbitals; however this will not be useful for large number of electrons as we do not have any general procedure for extract- ing any meaningful convenient form of the wave function from it. Nonetheless Laughlin3, based on certain general principles, proposed very accurate ground state wave function for 1/m =(cid:81) 1/m =(cid:81) k<l zm for each electron) as ΨL N electrons at the filling fraction ν = 1/m (hitherto suppressing ubiquitous Gaussian factor kl with zkl = zk − zl, where zk is the coordinate of k-th electron. The wave functions of a quasiparticle (QP) and a quasihole (QH) excitations of the Laughlin states are ΨL,qp 1/m with their respective charges (−e/m) and e/m. The Haldane-Halperin (HH) hierarchy theory4,5 for other fractions in the lowest LL predicts the production of daughter states by the condensation of these QPs and 1/m =(cid:81) l ∂zlΨL 1/m and ΨL,qh l zlΨL QHs at certain densities. However, no simple6 -- 8 explicit wave functions for these daughter states can be formulated using QP or QH wave functions. The composite fermion (CF) theory9 -- 11, on the other hand, begins with the postulate that every electron associated with an even number (2s) of quantized vortices forms a CF (denoted by 2sCF) which becomes the effective weakly interacting QP for the FQHE. The integer quantum Hall effect12 of these QPs with filling factor ν∗ = n will produce incompressible states at filling factors n/(2sn ± 1), where the sign +(−) refers to the parallel (antiparallel) direction of effective magnetic field for the CFs with respect to the applied magnetic field. The corresponding hierarchy is robust as the states that have been observed belong to these sequences and their energy gaps decrease with the increase of both n and s. This theory naturally predicts the explicit ground state wave functions which have been shown to be fairly accurate13 for the Coulomb interaction in finite systems, and it reproduces9 Laughlin 2 wave function for ν = 1/(2s± 1) when n = 1. Further, the CF theory suggests that the CFs form gapless Fermi sea14,15 at ν = 1/(2s) that has been rigorously tested16 -- 18 along with the presence of the structure19 -- 22 of effective Landau-like levels. In spite of exemplary success of the CF theory, it is not yet clear how the CFs emerge as relevant quasiparticles for the FQHE. Further, why the general principles that fetch Laughlin theory do not reproduce other states that are described by the composite fermion theory is an outstanding issue for about last three decades. Ideally, we would like to have a connection between the CF picture and the Laughlin description which is the main focus of this study. I propose a QP operator, as a modified form of Laughlin quasiparticle operator, and its conjugate as QH operator which operating on the Laughlin wave function describing a Laughlin condensate3,23 (LC) at ν = 1/m in Hilbert-subspace of the lowest LL, generate a QP and a QH, respectively, by expelling electrons from the LC to the projected Hilbert- subspace of the second LL and its conjugate LL (conjugation of the basis states) onto the lowest LL. I argue that the expelled electrons are CFs with respect to the LC as their associated order of correlation holes either reduces or increases by one in comparison to the electrons in LC. The condensate formed by these expelled electrons into the fraction 1/m, together with its coupling to the original condensate, respectively constitute new fractions 2/[2(m − 1) + 1] and 2/[2(m + 1) − 1]. In general, hierarchically constructed fractions n,m = n/[n(m + 1) − 1], respectively, can be described as n n,m = n/[n(m − 1) + 1] and ν− ν+ coupled LCs and their conjugates of filling fraction 1/m each formed at the different Hilbert sub-spaces of analytic functions24; the corresponding wave functions are found to be identical with those proposed in the noninteracting CF theory. The wave functions for the states25 -- 27 in the range 1/3 < ν < 2/5, such as 4/11 and 5/13, can be described by the two coupled condensates among which one is a non-Laughlin condensate formed in a different Hilbert subspace. The present generalization predicts the filling fractions 1/(2s) to be the quantum critical points for phase transitions between two distinct families of topological phases. Rest of the paper is organised as follows. In Section-II, I describe a quasiparticle and a novel quasihole operators which acting on Laughlin states produce quasiparticle and quasi- hole wave functions. Both these wave functions are energetically favourable in comparison to Laughlin's quasiparticle and quasihole wave functions. By exploiting the steps of forming more and more quasiparticles, the ground state wave functions for other conventional non- Laughlin FQHE states have been constructed in Section-III. These are coherently coupled 3 either Laughlin condensates or their conjugates. The wave functions for few other states have also been constructed. Possible topological phase transitions at the even denominator filling factors have been discussed in Section-IV. A comparative study between the present and other hierarchical constructions in the literature has been discussed in section-V. Section-VI is devoted for conclusion. II. QUASIPARTICLES AND QUASIHOLES OF LAUGHLIN CONDENSATES Motivated by the form of the quasiparticle wavefunction11,28 within the composite fermion theory at ν = 1/m, I begin with proposing a modified Laughlin-like QP operator in a nonlocal form: Oqp = (cid:32) N(cid:88) j=1 (cid:89) l(cid:54)=j ∂ ∂zj ∂ ∂zlj (cid:33) (1) This QP operator reduces angular momentum of a selected electron by one unit and relative angular momentum of other electrons with respect to the selected electron also by one unit. While the Laughlin QP operator is the product of the derivatives with respect to all the particle coordinates, this operator is the sum of the products of derivatives with respect to a selected particle coordinate and relative coordinates of other particles. The action of this QP operator on the Laughlin state at ν = 1/m is to produce a state with one reduced flux quantum from the LC (Fig. 1a). A quasiparticle wave function is thus found to be N(cid:88) j=1 (cid:48)(cid:89) i<k 1/m = OqpΨL Ψqp 1/m = (−1)j ∂ ∂zj T m−1 j zm ik ≡ N(cid:88) j=1 (−1)jPjT m−1 j Ψ(j) 1/m , where (cid:48) denotes the exclusion of jth electron in the product, and Ψ(j) LC without j-th electron with Pj = (cid:80) k(cid:54)=j z−1 jk , and T m−1 j (2) 1/m =(cid:81)(cid:48) = (cid:81) k(cid:54)=j zm−1 jk k<l zm kl represent . The physical description of this QP in Eq. 2 is as follows: One of the electrons gets expelled (Fig. 1b) from the LC where remaining (N − 1) electrons rearrange themselves in their Hilbert subspace. While each of the electrons in the condensate experiences correlation holes of order m at the position of other electrons, they feel correlation holes of order (m − 1) associated with the expelled electron; thus a m−1CF emerges as QP excitation of the LC and the net increase of charge is (−e/m) due to the reduction of correlation hole by order one. Although the 4 expelled electron in Eq. (2) is shown to be in the zero angular momentum state, it can acquire states with any angular momentum without costing any extra energy. Exiting from j}, the LC in the Hilbert subspace, H1, characterized by the set of analytic functions {z(cid:96) ((cid:96) = 0, 1, 2,··· representing angular momentum of a single-particle state), the electron occupies one of the states in the Hilbert subspace, H2, characterized by a set of analytic functions24 {z(cid:96) jPj}. While the former subspace corresponds to the lowest LL, the latter may be obtained by projecting29 the second LL onto the lowest one for a fixed number of electrons. Although these analytic functions may appear to be singular when two electrons approach each other, the QP wave function is not singular due to the coupling of expelled electron , and H2 exists only when some of with the condensate in the form of Jastrow factor T m−1 the electrons occupy H1. The above functional form of Pj is fixed by the requirement of maximum possible exponent of any coordinate equal to the total number of flux quanta. The j QP wave function in Eq. (2) differs from the Laughlin QP wave function but is identical with the QP wave function in the CF theory, and thus, as shown in Ref. 28, the energy of Ψqp 1/m is lower than that of ΨL,qp 1/m . The form (2) of quasiparticle wave function can be exploited to construct (see section III-A below) the ground state wave function for FQHE state at the filling factor ν+ 2,m = 2/[2(m − 1) + 1]. FIG. 1. (colour online) Schematic diagrams for the steps of transformations on decreasing flux from a Laughlin condensate. The disk-shaped bases represent different Hilbert subspaces, ellipsoids on these bases represent LCs, lines connected to electrons (spheres) represent strong interaction, and ring-shaped structures represent coupling between the layers. The arrows represent the evolution of the condensates on decreasing flux. Once we choose a QP operator, we loose freedom of choosing a QH operator and vice 5 versa. The QH operator is then conjugate to the QP operator (1), (cid:32) N(cid:88) j=1 (cid:89) l(cid:54)=j Oqh = zj zlj (cid:33) which acting on the Laughlin wave function produces a QH wave function 1/m = OqhΨL Ψqh 1/m ≡ N(cid:88) (cid:89) l(cid:54)=j zj zlj (cid:89) i<k 2 1 P (1) P (2) ... 2 zm ik (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) N ··· 1 ··· P (1) ··· P (2) ... . . . ··· P (N−1) N N j=1 1 1 P (1) P (2) ... 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)(cid:12)(cid:12) P (N−1) 1 P (N−1) 2 which may be rewritten as N(cid:88) j=1 zj (cid:89) l(cid:54)=j zlj (cid:89) i<k zm+1 ik Ψqh 1/m = as the determinant in Eq. (5) is equivalent to (up to a constant factor) the inverse of the Vandermonde matrix, i.e.,(cid:81) (3) (4) (5) (6) (7) ln (see supplementary material). Here (cid:88) l<n z−1 P (n) j = = n(cid:88) l=1 k1(cid:54)=k2···(cid:54)=kn(cid:54)=j 1 zjk1 ··· zjkn (−1)l+1Pj,lP (n−l) j (n − 1)! (n − l)! with Pj,l =(cid:80) k(cid:54)=j( 1 zjk be re-expressed as )l and Pj,1 = P (1) because (cid:81) l(cid:54)=j z−1 lj = P (N−1) j Ψqh 1/m = and T m+1 j j ≡ Pj. The quasihole wave function (5) is then may N(cid:88) = (cid:81) . The wave function of a "conjugate" (−1)jzjT m+1 (8) Ψ(j) 1/m , j=1 j l(cid:54)=j zm+1 lj Laughlin condensate30 (CLC) without jth electron is given by (cid:48)(cid:89) i<k zm+1 ik Ψ(j) 1/m = (cid:12)(cid:12)(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 ··· P (1) j−1 ··· P (2) j−1 ··· ... ··· P (N−2) j−1 1 P (1) j+1 P (2) j+1 ... P (N−2) j+1 N ··· 1 ··· P (1) ··· P (2) ··· ... ··· P (N−2) N N 1 1 P (1) P (2) ... 1 2 1 P (1) P (2) ... 2 P (N−2) 1 P (N−2) 2 6 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (9) where (cid:48) represents the exclusion of jth particle in the product. The form (8) of quasihole wave function is novel and rich. The wave functions (2 and 8) represent excitations due to decrease and increase of one unit flux quantum from the LC at ν = 1/m, respectively. The quasihole wave function (8) can thus also be represented as an expelled electron from the Laughlin condensate, akin to the quasiparticle representation above, and the resultant quasihole wave function becomes suitable starting point for constructing (Section III B) ground state wave function which will be exactly same as the CF wave function at ν− 2,m = 2/[2(m + 1) − 1]. The wave function of a LC at ν = 1/m is identical (see supplementary material) with the wave function of a CLC at ν = 1/m when all the electrons are present in the condensate. 1 characterized by a set of additional analytic functions {P ((cid:96)) We thus can interpret the condensate at ν = 1/m also as a CLC produced in the Hilbert j }. The physical subspace H∗ description of a QH is similar to the description of a QP: Upon increasing one unit of flux at ν = 1/m, one of the electrons gets expelled from the condensate to the Hilbert subspace H∗ j }. Since the expelled electron characterized by a set of additional analytic functions {zjP ((cid:96)) is associated with correlation holes of order (m + 1) with respect to all other electrons in the 2 CLC where every electron feels correlation holes of order m in the positions of other electrons, the expelled electron can be considered as an emergent m+1CF from the condensate and the excess of one order in correlation hole for the expelled electron justifies the QH with charge e/m. The QH wave function in Eq. (8) represents a QH with zero angular momentum, but in principle, QH may have any angular momentum without costing any extra energy. While the QH wave wave functions for the Laughlin and the CF theories are identical, the QH wave function here is different and has lower energy (Fig.2) than the energy of former two QHs. Note that a QP (QH) is represented by the generation of a m−1CF (m+1CF) out of a LC at ν = 1/m. This may apparently look like an inconsistency, but it is justified with continuity as follows. While the states below 1/m are represented by the sequence ν+ n,m for m−1CFs, the states above 1/m are represented by the sequence ν− n,m for m+1CFs; the state at ν = 1/m is a dichotomy as it belongs to both the sequences that are constituted with different types of CFs. This has been elucidated in Fig. 3 for m = 3. In the next section, we construct ground state wave functions for ν+ n,m states as n coupled LCs of 1/m filling formed at different Hilbert subspaces and for ν− n,m states as n coupled CLCs of 1/m filling. The weight factor of the LCs or the CLCs as the case may be will depend on the prefactors 7 FIG. 2. (Colour online) Difference of Coulomb energies (in the unit of e2/lB) between Laughlin or CF quasiholes and the proposed quasihole (8) at ν = 1/3 calculated for different N . Here is the dielectric constant of the background of the two-dimensional electron gas and lB is the magnetic length. of the analytic functions for that particular Hilbert subspace and interconnectivity of the Hilbert spaces will be given by T m−1 for H2 and H∗ 2, respectively, in Eqs. (2) and (8). n,m states, as elucidated for ν− j for ν+ n,m states and T m+1 j FIG. 3. (Colour online) Both the sequences of filling factors n 2n+1 and n 4n−1 starting at ν = 1/3 form two legs. First few states of the sequences are shown and the ellipsis represent the continuation of the sequences. The direction of the increment of flux in the diagram is shown by an arrow. The arrows connecting two consecutive states in the sequences represent whether the states are generated by the QP (filled arrows) or QH (unfilled arrows) excitations of their predecessors. 8 0.060.080.10.120.140.161/N00.0010.0020.0030.0040.005EL/CF - EqhqhIII. CONSTRUCTION OF THE NON-LAUGHLIN STATES FROM A LAUGHLIN STATE A. States on Reducing Magnetic Flux Decrease of q flux quanta from the Laughlin ground state leads to a LC with (N − q) electrons in H1 and the remaining q electrons occupying H2 strongly interact (Fig.1c). The electrons in H2 are the composite fermions with respect to the electrons in H1 and vice versa, and these two species of electrons are correlated through (m − 1) correlation holes. In each step of the decrement of quantum of flux from its commensurate value for forming a LC for all electrons, the maximum exponent of each electron coordinate in the reduced LC is equal to the number of flux. The decrease of one more unit of flux quantum after N/2 electrons get expelled into H2, electrons will not be transfered into H2 from H1. Instead, this decrease of flux helps to form a LC for the electrons in H2, as the flux becomes commensurate for the required number of zeros to form the new LC. The commensurate flux here refers to which the maximum exponent of all the electrons in the polynomial of the corresponding wavefunction, no matter which Hilbert subspace they are situated in, is equal to the total number of flux. The coupling of these two LCs through their associated correlation holes form a topologically distinct condensate (Fig.1d) at the filling factor ν+ 2,m. Based on this description which follows from the quasiparticle wave function (2), the explicit form of the constructed ground state wave function for ν+ 2,m state thus can be written as Ψν+ 2,m =  (−1)j (cid:89) (cid:88) j1<···<jN/2 l /∈{ji} 1/m({zj}, j ∈ {ji}) Ψ[1] 1/m({zl}, l /∈ {ji}) (cid:17)  (cid:89) (cid:16)(cid:81) zm−1 j∈{ji} jl × Ψ[2] (10) which is precisely30 the Jain wave function9,11,13 when 2s = m − 1. The condensates of H1 and H2 are related via Ψ[2] 1/m({zj}) as H1 and H2, respectively, Ψ[1] j∈{ji} Pj are spanned by the functions {z(cid:96) j}. The wave function in Eq. (10) is analogous to the Halperin wave function31 (m, m, m − 1) for a bilayer system; the layers here correspond to different Hilbert subspaces. j} and {Pjz(cid:96) 1/m({zj}) = A further decrease of one unit of flux quantum right at the filling factors ν+ 2,m will force to exit one electron each from the two LCs in H1 and H2 to H3, characterized by a set analytic functions {z(cid:96) (cid:1), but remains coupled with (cid:0)z−1 j =(cid:80) j } where P (l) k1j ··· z−1 jP (2) k1,··· ,kl(cid:54)=j klj 9 both the condensates through the correlation holes. Decrease of one unit of flux will thus create excitation of charge (−ν+ 2,me). If we decrease q flux quanta from the flux which creates 2,m, the LCs of H1 and H2 will contain (N/2 − q) electrons each and 2q the condensate at ν+ electrons in H3 will strongly interact (Fig.1e). Decrease of (N/6 + 1) quanta of flux will create three LCs of ν = 1/m in H1, H2 and H3 with N/3 number of electrons each and their mutual coupling will manifest (Fig.1f) a new condensate30 at the filling factor ν+ 3,m. The wave function for this condensate is then can be constructed as (cid:88) /∈{ki}(cid:88) (cid:80) j nj (−1) Ψν+ 3,m = j1<···<jN/3 k1<···<kN/3 ×Ψ[3] 1 m ({zk}, k ∈ {ki}) Ψ[2] 1 m (−1)k (cid:89)  (cid:89) 1/m({zj}) = ((cid:81) j∈{ji} ({zj}, j ∈ {ji}) Ψ[1] k∈{ki} (−1)j (cid:89) l /∈{ki},{ji}  zm−1 jk zm−1 jl zm−1 kl 1 m ({zl}, l /∈ {{ji},{ki}}) (11) 3,m − (m + 2)], where the LC in the Hilbert j∈{ji})Ψ[1]({zj}) and nj is the having angular momentum M + 3 = (N/2)[N/ν+ subspace Hn relates with the LC in H1 as Ψ[n] j P (n−1) number of elements in the set {ki} greater than j in the set {ji}. To be general, any state in the sequence ν+ n,m = n/[n(m − 1) + 1] will consist of mutually coupled n Laughlin 1/m condensates formed in H1,··· , and Hn with N/n electrons in each. Decrease of one unit of flux quantum will expel n electrons (one from each LC) into Hn+1 creating QPs with total charge (−ν+ n,me). A decrease of N/[n(n + 1)] + 1 units of flux quanta will create coupled 1/m LCs formed in H1,··· , and Hn+1 and the system moves to a new n+1,m = (n + 1)/[(n + 1)(m − 1) + 1] . The topologically distinct state at the filling factor ν+ n,m are all identical with those ground state wave functions Ψν+ proposed in the noninteracting CF theory. The corresponding sequence for m = 3 given by 2n+1 (see Fig. 3) is the same sequence predicted by the theory of no-interacting 2CFs. All the states in this sequence are constructed using coupled 1/3 LCs formed in different at the filling factors ν+ ν = n n,m Hilbert subspaces. 1. States Between Two Consecutive ν+ n,m States The present description not only precisely reproduces the wave functions for the states described by noninteracting CFs, it can easily construct the wave functions for the states that can otherwise be understood through the model of CFs when their interactions are taken into account. The basic principle for forming any incompressible state in the range 10 n+1,m is the condensation of interacting electrons in Hn+1 into an incompressible n,m < ν < ν+ ν+ liquid and the coupled state of this condensate with n LCs formed in H1,··· ,Hn become an eigenstate of total angular momentum. As for example, when N/4 + 2 flux quanta are reduced from a LC at 1/3 filling, the coupling of LC at 1/3 filling with (3N/4− 2) electrons in H1 and an unconventional condensate32, ¯Ψ[2] 1/5, characterized by the zero-energy ground state of Haldane pseudopotential4, V3, with relative angular momentum 3, at 1/5 filling with (N/4 + 2) electrons in H2 form a condensate at ν = 4/11. The corresponding wave function can thus be expressed as  (cid:89) (−1)j (cid:89) (cid:88) j1<···<jN/4+2 1/5({zj}, j ∈ {ji}) Ψ[1] j∈{ji} × ¯Ψ[2]  z2 jl l /∈{ji} 1/3({zl}, l /∈ {ji}) Ψ4/11 = (12) (13) The angular momentum of this state M = (N/2)[(11/4)N − 5] is consistent with the flux- particle relation for incompressible state33 in spherical geometry. This wave function is, however, different from the trial wave function33 of interacting CF theory because the state Jastrow factor(cid:81) at 1/5 filling, unlike Laughlin states, cannot be obtained by simple multiplication of the ij to the state at 1/3 filling of 2CFs for the pseudopotential V3. Further, energy of the wavefunction (12) will be less in comparison to the wavefunction proposed in i<j z2 interacting CF theory33, as 1/5 state is the zero-energy ground state32 of V3. Similarly, the wavefunctions for the incompressible states33,34 at ν = 5/13 and 3/8 can be constructed30 by the coupling of a LC at 1/3 filling formed at H1 and another condensate formed at H2 with filling fractions 2/7 and 1/4, respectively as follows. A decrease of (2N − 2)/5 units of flux quanta from ν = 1/3 state will create an uncon- ventional condensate of (2N − 2)/5 electrons in H2 at the filling factor 2/7. The ground state wave function at ν = 5/13: (cid:88)  (cid:89) (−1)j (cid:89)  z2 jl j1<···<j(2N−2)/5 j∈{ji} l /∈{ji} × ¯Ψ[2] ({zj}, j ∈ {ji}) Ψ[1] ({zl}, l /∈ {ji}) 2 7 1 3 Ψ5/13 = can then be described as coherently coupled LC at ν = 1/3 and this unconventional con- densate (characterized by V3) at ν = 2/7. The angular momentum of this state M5/13 = (N/2)[ 13 5 (N − 1)] is also consistent with the predicted flux-particle relationship33. 11 A fractional quantum Hall state at ν = 3/8 can then also be described as coherently coupled LC of (2N/3) electrons at ν = 1/3 in H1 and a condensate at ν = 1/4 with Anti- Pfaffian pair correlation35,36 for N/3 electrons in H2. Therefore the corresponding ground state wave function (cid:88)  (cid:89) (−1)j (cid:89)  z2 jl Ψ3/8 = j1<···<jN/3 j∈{ji} × ¯Ψ[2] ({zj}, j ∈ {ji}) Ψ[1] l /∈{ji} ({zl}, l /∈ {ji}) (14) has angular momentum M3/8 = (N/2)[8N/3−3] obeying consistent flux-particle relationship34. Although the ground state wave functions (Eqs.12 -- 14) which are different from the pre- 1 4 1 3 viously proposed33,34 ones, for ν = 4/11, 5/13, and 3/8 are proposed here, their validity for the Coulomb interaction has not been checked. The validity of these wave functions will be published elsewhere with the implementation of the present theory in a spherical geometry and by determining unconventional wave functions at 1/5, 2/7, and 1/4 filling factors, i.e., ¯Ψ 1 , and ¯Ψ 1 , ¯Ψ 2 . 5 7 4 B. States on Increasing Magnetic Flux By the increase of q flux into the Laughlin state at ν = 1/m, the CLC will have q fewer electrons and the q expelled electrons situating in H∗ 2 interact among themselves. When q become N/2, the CLC will have N/2 electrons and the remaining electrons will strongly interact in H∗ 2. These electrons will be benefited to form a CLC, because of the commensuration of flux required for a CLC, by the increment of one more unit of flux and the coherent superposition between these two coupled CLCs will form a condensate at the 2,m = 2/[2(m + 1) − 1]. The explicit wave functions for these states can be filling factors ν− constructed, following Eq.(8), as Ψν− 2,m = (−1)j (cid:89) (cid:88) j1<···<jN/2 l /∈{ji} 1/m({zj}, j ∈ {ji}) Ψ[1] 1/m({zl}, l /∈ {ji}) . zm+1 jl j∈{ji} × Ψ[2]  These are identical30 with the Jain wave functions11,13. Here the CLCs in H∗ related via Ψ[2] 2 are 1/m({zj}). Similarly, an increase of (N/6 + 1) flux 1/m({zj}) = 1 and H∗ j∈{ji} zj (cid:16)(cid:81) (15)  (cid:89) (cid:17) Ψ[1] 12 quanta from the ground state at ν− Hilbert subspace H∗ at ν− with angular momentum M− 1, H∗ 2,m will create three coupled CLCs of ν = 1/m in the 3 with N/3 electrons in each and form a new condensate 3(m+1)−1. The corresponding ground state wave function that can be constructed 2, and H∗ 3,m − (m − 2)] is given by 3 = (N/2)[N/ν− 3,m = 3 (−1)k (cid:89) (−1)j (cid:89) l /∈{ki},{ji}  zm+1 jk zm+1 jl zm+1 kl (cid:88) /∈{ki}(cid:88)  (cid:89) (cid:80) (−1) j nj Ψν− 3,m = j1<···<jN/3 k1<···<kN/3 × Ψ[3] 1 m ({zk}, k ∈ {ki}) Ψ[2] 1 m k∈{ki} j∈{ji} ({zj}, j ∈ {ji}) Ψ[1] ((cid:81) where the CLC in H∗ n is related with the CLC in H∗ j∈{ji}) Ψ[1]({zj}). The wave function Ψν− j zn−1 3 1 m ({zl}, l /∈ {{ji},{ki}}) (16) 1/m({zj}) = is identical with the corresponding CF wave 1 by the relation Ψ[n] functions9,11,13. Similarly, all the ground state wave functions Ψν− all of those will precisely be the CF wave functions. n can be constructed, and n will create a condensate of (n + 1) coupled CLCs at filling factor ν− In general, increase of flux by N/[n(n + 1)] + 1 from the ground state at filling factor n,m = n/[n(m + 1) − 1] consisting of coherent superposition of coupled n CLCs formed in ν− H∗ 1,··· , and H∗ n+1,m = (n + 1)/[(n + 1)(m + 1) − 1]. All the associated ground state wave functions are identical30 with those proposed in noninteracting CF theory. The sequence ν = n 4n−1 shown in Fig.3 arises due to the increase of flux in comparison to the same for Laughlin 1/3 state. Each of the state in this sequence consists of 1/3 filled CLCs which are coupled. IV. TOPOLOGICAL PHASE TRANSITIONS AT EVEN-DENOMINETOR FILL- INIGS The sequences of filling factors ν+ n,m which are n coupled LCs of individual filling factors n,m−2 which are n coupled CLCs of individual filling factors 1/(m − 2) converge 1/m, and ν− at the filling factor ν = 1/(2s) when m = 2s + 1 and n → ∞ . Every value of n on either side of ν = 1/(2s) is characterized by a Chern number. As the filling factors for large n in either side of ν = 1/(2s) is densely populated, it appears that for a finite range of filling factors on either side of ν = 1/(2s) realisation of distinct topological phases with definite Chern number should not be possible. However, the present analysis also suggests that there should be another class of topological number to distinguish the entire family of n/(2sn + 1) states from the entire family of n/(2sn − 1) states. These two distinct topological phases 13 are possibly characterized by the direction of the edge modes: while all the edge modes for n/(2sn + 1) are downstream, upstream modes37,38 exist for n/(2sn − 1) states. In addition to n-component Chern-Simons gauge theory for distinguishing two consecutive topological phases30,37,38 in any of the sequences, viz., which distinguishes topological phase of entire family in the sequence of ν+ of other family of states in the sequence of ν− n,m−2. n n(m−1)±1, there must be another gauge symmetry n,m from the phase States with ν (cid:46) 1/(2s) will be topologically distinct from a state with ν (cid:38) 1/(2s) because the former corresponds to the coherent superposition of large number of coupled LCs, formed at different Hn, of 1/(2s + 1) filling factors and the latter is formed when the coherent superposition of large number of CLCs, formed at different H∗ n, of filling factor 1/(2s − 1). The bulk energy gap closes at the filling factors 1/(2s) which separates above topologically distinct states and thus νc = 1/(2s) serve as the quantum critical points for topological phase transitions when magnetic flux is tuned around these. The closing of the bulk gap at νc is also corroborated with the formation of the Fermi sea14,15 of 2sCFs. While the Fermi surface for ν < νc corresponds to the 2sCF particles, the 2sCF holes create the Fermi surface for ν > νc because attachment of 2s vortices for these states over-screens the magnetic flux and thus their Hilbert subspace becomes conjugate to the 2sCF particles. These CF holes of second kind, however, should be differentiated with the literature of the Fermi surface made with the CF holes39 of first kind which are composite fermionized holes (absence of electrons). The possible connection of this particle-hole symmetry around ν = 1/(2s) with the proposed Dirac semi-metal40,41 of the CFs near ν = 1/(2s) cannot be ruled out. While the surface acoustic wave techniques42 can probe the length scale, the experiments on photoluminescence spectroscopy43 -- 46 developed for FQHE systems should probe the energy scale for determining the universality class of the predicted quantum critical point. V. DISCUSSION Immediately after proposal of the CF theory, Read47 argued that the HH hierarchy and the CF states for same ν are two different descriptions of the same universality class as they both describe same QP charge and braiding statistics48. On the contrary, Jain11,49 pointed out that the CF theory is fundamentally different from the hierarchy theory and they cannot be adiabatically connected as the flux attachment for CFs is nonperturbative. 14 The calculations50,51 based on the conformal field theory show that the CF wave function can be obtained as the correlators of certain operators. This construction, however, is not microscopic and either presumes the existence of the CF theory since the different sets of operators have been chosen representing the different effective Landau levels of the CFs or chooses a specific set of parameters out of the zoo of parameter space. Bonderson52 has argued that the CF description of the FQHE is hierarchical in nature and that the wave functions of a CF state can be constructed using QP wave function of the hierarchically higher CF state, generated by the CF theory. This study also explicitly assumes the existence of the CF theory. In this paper, I have proposed QP and QH operators which generate CFs, upon operating on the Laughlin states, without assuming the presence of any CF in the Laughlin states. The wave functions here are hierarchically constructed but they are fundamentally different from the HH hierarchy theory: (i) While the hierarchically lower states are formed by the condensation of QPs or QHs of its immediate predecessor in HH hierarchy theory, all the states here are formed due to the condensation of electrons into LCs or CLCs in different Hilbert subspaces; (ii) in contrary to the HH hierarchy theory, all the states in the sequence n/(4n − 1) are created by the QH excitations of their immediate predecessors; (iii) while the state 5/13 may be obtained only by the condensation of QHs created from the parent 2/5 state in HH hierarchy, this state here is constructed by the condensation of expelled electrons from 1/3 state into the filling fraction 2/7 in H2. VI. CONCLUSION I have shown that the CFs emerge as quasiparticle and quasihole excitations of LC; the wave functions for the filling factors ν = n/(2sn ± 1) in the noninteracting CF theory are identical with the proposed coherent superposition of n coupled Laughlin (conjugate- Laughlin) condensates of filling factors 1/(2s ± 1) formed at different Hilbert-subspaces, where an electron in any of the condensates is felt by all the electrons in other condensates as CFs. Therefore this work provides an unification of Laughlin and the composite fermion theories, and I believe that it would put an end to all the differences of opinions regarding hierarchy pictures. This generalized scheme predicts the ground state wave functions for all (though shown only for 4/11, 5/13, and 3/8) the fractional quantum Hall states which are in between any two prominent states described above, without considering any higher order 15 effect such as FQHE of interacting CFs. This theory further opens the possibility of future studies exploring the universality class of the quantum criticality and possible topological symmetry distinguishing two phases about the critical points at the filling factors νc = 1/(2s) in the lowest Landau level. ACKNOWLEDGEMENTS I thank S. Mukherjee for useful discussions and K. Manna for his help in drawing cartoon diagrams. 1 D. C. Tsui, H. L. Stormer, A. C. Gossard, Phys. Rev. Lett. 48, (1982) 1559. 2 R. Willett, J. P. Eisenstein, H. L. Stormer, D. Tsui, A. C. Gossard, J. H. English, Phys. Rev. Lett. 59, (1987) 1776. 3 R. B. Laughlin, Phys. Rev. Lett. 50, (1983) 1395. 4 F. D. M. Haldane, Phys. Rev. Lett. 51, (1983) 605. 5 B. I. Halperin, Phys. Rev. Lett. 52, (1984) 1583. 6 R. Morf, N. d'Ambrumenil, B. I. Halperin, Phys. Rev. B 34, (1986) 3037. 7 R. Morf, B. I. Halperin, Z. Phys. B 68, (1987) 391. 8 M. Greiter, Phys. Lett. B 336, (1994) 48. 9 J. K. Jain, Phys. Rev. Lett. 63, (1989) 199. 10 J. K. Jain, Phys. Rev. B 41, (1990) 7654. 11 J. K. Jain, Composite Fermions (Cambridge University Press, Cambridge, U.K., 2007). 12 K. Klitzing, G. Dorda, M. Pepper, Phys. Rev. Lett. 45, (1980) 494. 13 J. K. Jain, R. K. Kamilla, Int. J. Mod. Phys. B 11, (1997) 2621. 14 V. Kalmeyer, S.-C. Zhang, Phys. Rev. B 46, (1992) 9889. 15 B. I. Halperin, P. A. Lee, N. Read, Phys. Rev. B 47, (1993) 7312. 16 R. L. Willett, R. R. Ruel, K. W. West, L. N. Pfeiffer, Phys. Rev. Lett. 71, (1993) 3846. 17 W. Kang, H. L. Stormer, L. N. Pfeiffer, K. W. Baldwin, K. W. West, Phys. Rev. Lett. 71, (1993) 3850. 18 V. J. Goldman, B. Su, J. K. Jain, Phys. Rev. Lett. 72, (1994) 2065. 16 19 I. V. Kukushkin, K. v. Klitzing, K. Eberl, Phys. Rev. Lett. 82, (1999) 3665. 20 I. V. Kukushkin, J. H. Smet, V. W. Scarola, V. Umansky, K. von Klitzing, Science 324, (2009) 1044. 21 U. Wurstbauer, D. Majumder, S. S. Mandal, I. Dujovne, T. D. Rhone, B. S. Dennis, A. F. Rigosi, J. K. Jain, A. Pinczuk, K. W. West, L. N. Pfeiffer, Phys. Rev. Lett. 107, (2011) 066804. 22 T. D. Rhone, D. Majumder, B. S. Dennis, C. Hirjibehedin, I. Dujovne, J. G. Groshaus, Y. Gallais, J. K. Jain, S. S. Mandal, A. Pinczuk, L. Pfeiffer, K. West, Phys. Rev. Lett. 106, (2011) 096803. 23 S. M. Girvin and A. H. MacDonald, Phys. Rev. Lett. 58, (1987) 1252. 24 S. M. Girvin, T. Jach, Phys. Rev. B 29, (1984) 5617. 25 W. Pan, H. L. Stormer, D. C. Tsui, L. N. Pfeiffer, K. W. Baldwin, and K. W. West, Phys. Rev. Lett. 90, (2003) 016801. 26 W. Pan, K. W. Baldwin, K. W. West, L. N. Pfeiffer, and D. C. Tsui, Phys. Rev. B 91, (2015) 041301(R). 27 N. Samkharadze, I. Arnold, L. N. Pfeiffer, K. W. West, and G. A. Csathy, Phys. Rev. B 91, (2015) 081109(R). 28 G. S. Jeon, J. K. Jain, Phys. Rev. B 68, (2003) 165346. 29 The single particle eigen states in the second and higher Landau levels have one of the terms as j )λ (ignoring Gaussian factors), where λ = 1, 2,··· representing higher Landau levels. When zj z(cid:96) j j(z∗ z(cid:96) we project second and higher Landau levels onto the lowest one we transform (z∗ j → ∂λ j )λzl into an operator acting on the wave function for the filled lowest Landau level. This provides the projected second Landau level Hilbert subspace characterized by the additional functions {z(cid:96) j } apart from the functions available for projected lower Landau levels, where P (1) j = Pj. jP (λ) See Ref. 11 for further details of the procedure of projecting effective higher order Landau levels of composite fermions onto the lowest Landau level. 30 See Supplementary Material accompanying this paper. 31 B. I. Halperin, Helv. Phys. Acta 56, (1983) 75. 32 A. W´ojs, K.-S. Yi, and J. J. Quinn, Phys. Rev. B 69, (2004) 205322. 33 S. Mukherjee, S. S. Mandal, Y.-H. Wu, A. W´ojs, and J. K. Jain, Phys. Rev. Lett. 112, (2014) 016801. 34 S. Mukherjee, S. S. Mandal, A. W´ojs, and J. K. Jain, Phys. Rev. Lett. 109, (2012) 256801. 17 35 G. Moore and N. Read, Nucl. Phyys. B 360, (1991) 362. 36 M. Levin, B. I. Halperin, and B. Rosenow, Phys. Rev. Lett. 99, 236806 (2007) 236807. 37 X.-G. Wen, Int. J. Mod. Phys. B 06, (1992) 1711. 38 X.-G. Wen, Adv. in Phys. 44, (1995) 405. 39 M. Barkeshli, M. Mulligan, M. P. A. Fisher, Phys. Rev. B 92, (2015) 165215. 40 D. T. Son, Phys. Rev. X 5, (2015) 031027. 41 S. D. Geraedts, M. P. Zaletel, R. S. K. Mong, M. A. Metlitski, A. Vishwanath, and O. I. Motrunich, Science 352, (2016) 197 42 See B. I. Halperin in Perspectives of Quantum Hall Effects eds. S. D. Sarma and A. Pinczuk (John Willey and Sons, 1997). 43 B. B. Goldberg, D. Heiman, M. Dahl, A. Pinczuk, L. Pfeiffer, K. West, Phys. Rev. B 44, (1991) 4006. 44 G. Yusa, H. Shtrikman, I. Bar-Joseph, Phys. Rev. Lett. 87, (2001) 216402. 45 M. Byszewski, B. Chwalisz, D. K. Maude, M. L. Sadowski, M. Potemski, T. Saku, Y. Horayama, S. Studenikin, D. G. Austing, A. S. Sachrajda, P. Hawrylak, Nat. Phys. 2, (2006) 239. 46 S. Nomura, M. Yamaguchi, H. Tamura, T. Akazaki, Y. Hirayama, M. Korkusinski, P. Hawrylak, Phys. Rev. B 89, (2014) 115317. 47 N. Read, Phys. Rev. Lett. 65, (1990) 1502. 48 D. Arovas, J. R. Schrieffer, F. Wilczek, Phys. Rev. Lett. 53, (1984) 722. 49 J. K. Jain, Indian J. Phys. 88, (2014) 915. 50 T. H. Hansson, C.-C. Chang, J. K. Jain, S. Viefers, Phys. Rev. B 76, (2007) 075347. 51 J. Suorsa, S. Viefers, and T. H. Hansson, New J. Phys. 13, 075006 (2011); Phys. Rev. B 83, 235130 (2011) 235130. 52 P. Bonderson, Phys. Rev. Lett. 108, (2012) 066806. 18
1302.4501	1	1302	2013-02-19T01:27:14	Giant conductance and phase time anomalous events of hole quantum transport	[ "cond-mat.mes-hall" ]	Events of giant conductance and anomalies of the phase transmission time for holes, are theoretically investigated within the multicomponent scattering approach. Based on this model, new analytical expressions for unitarity relations in the uncoupled hole transport are obtained and directly applied to study the behavior of the conductance and the phase transmission time in a double barrier resonant tunneling (DBRT) and a superlattice $GaAs$-cladding layer$/(AlAs/GaAs)^{n}/GaAs$-cladding layer. Clear-signature evidences of giant conductance phenomena for hole transmission without valence-band mixing through a DBRT and a superlattice were found. The giant conductance effect losses robustness by manipulating the number of superlattice layers and by including the valence-band particles coupling as well. Phase time through the heterostructure exhibits extremal dependencies in the gaps and in the barriers, as those reported before for electrons. We have detected an earlier arrival phase time for the propagation of both flavors of holes within the barrier, in the order of few tenths of picoseconds. An appealing filter-like effect is presented, whenever a selective confinement strength arises independently for both flavors of holes in the uncoupled regime. Our results also prescribe noticeable evidences for both uncoupled and coupled hole fluxes, similar to those foretold by Hartman, upon transmission of electrons through opaque barriers.	cond-mat.mes-hall	cond-mat	Giant conductance and phase time anomalous events of hole quantum transport. S. Arias-Lasoa, L. Diago-Cisnerosb bDepartamento de Física Aplicada, Facultad de Física, Universidad de La Habana, C.P.10400, Cuba. aDepartamento de Física, ISPJAE, C.P. 19390, La Habana, Cuba. 3 1 0 2 b e F 9 1 ] l l a h - s e m . t a m - d n o c [ 1 v 1 0 5 4 . 2 0 3 1 : v i X r a Abstract Events of giant conductance and anomalies of the phase transmission time for holes, are theoretically investigated within the multicomponent scattering approach. Based on this model, new analytical expressions for unitarity re- lations in the uncoupled hole transport are obtained and directly applied to study the behavior of the conductance and the phase transmission time in a double barrier resonant tunneling (DBRT) and a superlattice GaAs-cladding layer/(AlAs/GaAs)n/GaAs-cladding layer. Clear-signature evidences of giant conductance phenomena for hole trans- mission without valence-band mixing through a DBRT and a superlattice were found. The giant conductance effect losses robustness by manipulating the number of superlattice layers and by including the valence-band particles coupling as well. Phase time through the heterostructure exhibits extremal dependencies in the gaps and in the barriers, as those reported before for electrons. We have detected an earlier arrival phase time for the propagation of both flavors of holes within the barrier, in the order of few tenths of picoseconds. An appealing filter-like effect is presented, whenever a selective confinement strength arises independently for both flavors of holes in the uncoupled regime. Our results also prescribe noticeable evidences for both uncoupled and coupled hole fluxes, similar to those foretold by Hartman, upon transmission of electrons through opaque barriers. Keywords: PACS: 73.43.Jn, 34.80.Bm tunneling, phase time, elastic quantum scattering 1. Introduction The impressive development of low-dimensional elec- tronic and optoelectronic devices, brought a new urgency to the essential measurement and modeling of charge car- riers transmission time and particularly of the tunneling time through specific potential regions. When electrons and holes are involved, the low-dimensional device response depends on the slower charge-carrier's traveling time [1]. Undoubtedly, for such systems it is crucial to study the tunneling time of holes. An equally important reason for address investigations on heavy holes (hh) and light holes (lh), is their intrinsic band mixing effects. Since the hole's transport is essentially an interconnected-multichannel pro- cess, the theoretical calculation of transmission quantities is in some sense more cumbersome than that for electrons. Currently, the research of under-synchronic anomalous scattering events, is one of the major areas in the quantum transport theory[2 -- 7]. Regarding the long-standing con- troversy around superluminal velocities and total trans- mission inconsistency, together with their wide-known ano- malous character [8, 9], they are welcome due to all pos- sible benefits they open up for design and development of nano-optoelectronic devices[2, 12, 13]. There is a large ac- cumulated knowledge about these topics -mostly dedicated Email addresses: [email protected] (S. Arias-Laso), [email protected] (L. Diago-Cisneros) to electrons, optical pulses and electromagnetic waves [5, 10 -- 15]-, accurately enough to unambiguously provide solid grounds for superluminal propagation, the paradoxical Hart- man premonition and legitimacy of the phase time ap- proach. Worthwhile noticing however that, hole transport analogous phenomena has been less studied yet. Resonant mechanisms between propagating modes os- cillations and the self oscillations of the system are the fundamental cause, which gives rise to an isolated maxi- mums of transmission, often referred to as giant conduc- tance (GG) phenomena. Events of GG have been observed for electrons and optical pulses in many finite periodic sys- tems, which possess time reversal symmetry, namely: bal- listic mesoscopic conductors, luminal pulses and superlat- tices [17, 19 -- 22]. Recent theoretical studies in molecular devices suggested that could rise up particularly large val- ues for dynamic conductance [23] and novel experimental achievements pointed out to a giant enhancement of elec- tronic tunneling in a two-dimensional lattice of graphene[24, 25]. Lately, was found a strong dependence of the conduc- tance fluctuations with the density of states in graphene nanoribbons, which represent an alternative insight to study the band structure of these physical systems[26]. Owing to its relevance for quantum transport, the tun- neling time has played a key role for a correct evaluation of electronic devices [8]. A large number of proposals in reply the challenge to settle how long does it takes to a Preprint submitted to Physica E December 16, 2017 particle to tunnel a single potential barrier [30], have been put forward [8, 9, 34]. The lively and long standing de- bate concerning this matter [8, 9, 28, 30 -- 35] -presumed not to finish completely-, it seems to have finally found its cornerstone in precise experimental measurements for photons and optical pulses [12, 13, 36 -- 38]. These gen- uine measurements put to the test, before all else, the very question loosely formulated by McColl and directly bear on truly temporal features of tunneling phenomenon. Besides, they are consistent with the phase time concep- tion derived from group delay within the stationary-phase method [28, 29, 31]. An additional motivation was the accurate description of above mentioned experiments by careful phase-time calculations [10, 11]. In this paper we will present numerical calculations of the conductance and the phase time for an arbitrary num- ber of alternate layers of III − V materials, in the frame- work of the multicomponent scattering approach (MSA)[3]. The main idea is to search evidences of anomalous events in hh and lh transmission. In Sec.2 some theoretical re- marks within the MSA are exposed. Brief complemen- tary insights to the uncoupled regime formula of scatter- ing matrix unitarity relations and phase time will also be presented. Analytical expressions for the phase time and the conductance are applied in Sec.3 to study the anoma- lous events associated with these physical quantities in the double barrier resonant tunneling (DBRT) and the super- lattice. Phase time behavior for holes shows a very good agreement with the previous results for electron-tunneling through superlattices[11]. Evidences of the giant conduc- tance phenomena for uncoupled hole transport are shown. Finally, in Sec.4 we summarize and conclude. 2. Conduction and phase time in the multicompo- nent scattering approach The MSA is based on standard (N × N ) k · p effective Hamiltonians and the multichannel transfer matrix (TM) method. In this approach the scattering amplitudes can be obtained straightforwardly [3]. The main purpose here, is to evaluate quantities of synchronic multi-mode transport for massive charge carriers moving through multi-layered structures at coupled and decoupled channel regimes. Owing to consistency, we found useful to recall few basic outlines of the Kohn-Lüttinger (KL) two-band model [39], due to its widely accepted accuracy for describing dynamics of elementary excitations as well as electronic properties in the valence band, within the framework of the multiband effective mass approach. This model, is highly common regarding to the field of hole quantum transport through periodic heterostructures. It considers in the k· p approximation the highest two valence bands degenerated in the Γ-point of the Brillouin Zone. The usual (4 × 4) KL Hamiltonian [39] in the Broido-Sham representation 1: Schematic Figure and lh quantum transport processes cladding layer(L)/(AlAs/GaAs)n/GaAs superlattice. illustration of multichannel hh through a GaAs − cladding layer(R) − [3], has the form H kl =   with 0 H11 H12 H13 H ∗ 12 H22 H ∗ 0 13 −H ∗ 0 13 H ∗ 0 −H13 H22 H12 12 H11 ,   (1) H12 = t + V (z) − B2 y − k2 H11 = A1κ2 2√3 2 m0 (cid:0)γ2(k2 2√3 2 m0 γ3(kx − iky) H13 = i ∂2 ∂z2 , x) + 2iγ3kxky(cid:1) , , ∂ ∂z ∂2 ∂z2 , H22 = A2κ2 t + V (z) − B1 being A1,2 = (γ1 ± γ2) B1,2 = (γ1 ± 2γ2), in atomic units. The matrix elements Hij show a direct dependency of the semi-empirical parameters of Lüttinger γi (i = 1, 2, 3), that typify each slab of the heterostruc- ture. While V (z) and κt, stand for the periodic poten- tial, and the two-components in-plane quasi-momentum, respectively [3]. lh−1/2, channel 3 : To study the quantum transport properties of holes through semiconductor heterostructure, we consider the system shown in Fig.(1), described by the Kohn-Lüttinger Hamiltonian in the expression (1). The four accessible channels of the system, denoted as: channel 1 : hh+3/2, lh+1/2, and channel 4 : channel 2 : hh−3/2, are considered simultaneously, following one of the bare bones of the MSA. We emphasize this upmost advan- tage, as it allows clearly identify inter- and intra-subband hole transitions together with transitions from Kramer- up toward Kramer-down states to time reversal, and the other way around. The latter is not the usual case. In this approach, the segment between zL and z3 in Fig. (1), is defined as a cell and represents the period of the het- erostructure superlattice [3]. Combining the (4 × 4) Kohn-Lüttinger Hamiltonian with the TM formalism, a direct connection with the scat- tering theory is established [3], then we have M sv(zR, zL) = 2 N −1M f d(zR, zL)N . Here, N defines a relation between the traveling wave vectors of the tunneling problem and their derivatives, with the N−dimensional state vectors that describe a propagating or evanescent mode. This crucial transformation of the MSA, connects two differ- ent kinds of TM. The first class of TM, M f d(zR, zL), re- lates the linearly independent state functions and their derivatives (f d) between two points of the heterostructure in the form M f d(zR, zL) = N (zR) · N (zL)−1, where the matrix N (zi) can be constructed once the linearly inde- pendent solutions are obtained [40]. The second kind of TM, M sv(zR, zL), whose elements describe a propagat- ing or evanescent mode depending on the energy, connects the state vectors (sv) in two different points of the sys- tem. The mathematical structure of these two matrices is conditioned by requirements of time-reversal invariance and flux conservation [41]. A very strong criterium to ver- ify the calculations is based in the determinant of these transfer matrices which must be constant and equal to the unity, mostly. Then applying the well known relations between the δ (cid:19), and the scattering matrix Msv = (cid:18) α β S = (cid:18) r γ t′ t r′ (cid:19) blocks [41], one can directly evaluate the relevant physical quantities of the scattered transport the- ory. Initially we have to obtain the complex transmission amplitudes [3] t = α − βδ−1γ, tij = tRij + itIij , (2) for each transition from the j-th incoming channel (in- cident propagating mode) to the i-th outgoing channel (transmitted or reflected propagating modes). It is worth noticing that within the MSA, none of the incoming ampli- tudes is required to cancel, as it is commonly done. Hence, an authentic multichannel and synchronic description of the hole transmission processes is naturally achieved. This is to say that, we are able to distinguish accurately the contribution from each input channel to the scattering pro- cess, modulated by the simultaneous presence of the rest of the channels, which are, in principle, equally accessible to the incident hole stream. None of the usual simplifica- tions of the original k · p Hamiltonians are invoked, thus the time reversal symmetry is assured. Given the trans- mission amplitudes tij, it is possible to evaluate the total transmission probability to channel i Gi = (e2/π) Xj tij2 , (3) the total two-probe Landauer conductance defined as [3, 42] G = (e2/π) Xi,j tij2 , (4) and the transmission amplitude phase for each of the 16 available paths ij [3], θij = arctan{tIij /tRij}. The trans- 3 mission phase time is obtained from [3] τij = ∂ ∂E θnij, (5) whose advantages and applicability criteria has been widely discussed in the literature [3, 10, 11, 13, 43]. The sums in equations (3) and (4) run over all the open channels and the expressions (3)-(5) and (8) are defined for the n cells periodic structure showed in Fig.(1). Hereafter E stands for the incoming particle's energy. Tunable parameter κt, is essential to modify the in- plane dynamics of holes, i.e. the degree of freedom trans- verse to the main transport direction, also refereed as the valence-band mixing for holes. This later is crucially rele- vant in order to understand the main problem envisioned here, and how we face it within the MSA. Whenever the quasi-momentum parallel to the interfaces, κt ≈ 0, the crossed transitions are not allowed and the system is said to yield an uncoupled hole regime. Under this particu- lar condition, only direct incoming-outgoing channels are connected, therefore (4 × 4) blocks α, β, γ and δ of the TM M sv are diagonal matrices. The later, straightfor- wardly leads us to new unitarity relationships, that satisfy the blocks of the multiband scattering matrix S under an uncoupled hole regime, acquiring a more simple form: α∗α − β∗β = I N ′ β∗δ − γα∗ = ON ′ βδ∗ − γ∗α = ON ′ δδ∗ − γγ ∗ = I N ′ , (6) where N ′ = N/2. Consequently, transmission and reflec- tion amplitudes have reduced expressions respect to (2): t = 1/α∗ r = −δ−1γ with δ(γ)ii = δ(γ)Rii + ıδ(γ)Iii with αii = αRii + ıαIii . (7) Rewriting the phase time (5) for this specific case, we ob- tain: τii = (αRii Tii ∂αIii ∂E − αIii ∂αRii ∂E ). (8) Expressions like (7) were previously reported for the reso- nant tunneling and band mixing in multichannel superlat- tices of propagating electron systems [11]. 3. Numerical results and discussion: tunneling time and giant conductance For the numerical calculations that will be presented here we shall consider our layered media with barrier thick- ness Lb = (10 − 180) Å, and well width Lw = (50, 150) Å. For the barrier height we will have either Vb = 0.498 eV or Vb = 0.550 eV . The cases where the in-plane wavenumber (parallel to the interfaces) κt ≈ 0, with vanishing band mixing effects, and the cases where κt 6= 0, with coupling phenomena effects on the quantum transport[3, 44, 45] will be separately study. Focusing to study the behavior of the Figure 2: Phase time τii for a tunneling of holes through a single barrier of Lb = 20 Å, as a function of the hole incident energy at κt = 0 with Vb = 0.498 eV . The solid (dashed) line stands for i = hh+3/2 (lh−1/2) direct path. In the broken interval a negative dip extents up to 0.4 ps. phase time τii and the free motion time τf = nlcm∗ h/(kz)h h the effective mass of holes (being lc the cell longitude, m∗ and (kz)h the eigenvalues of the scattering problem solved with the MSA[3]) as a function of the incoming energy Ei, the barriers height and thickness were fixed at Vb = 0.23 eV and Lb = 30 Å, respectively. Both, a double bar- rier (n = 2 cells) and a superlattice of n = 8 cells were considered. For consistency with the reported results for electrons[11], a 30% concentration of Al was fixed inside the barrier: (GaAs/Al0.3Ga0.7As/GaAs)n. 3.1. Uncoupled hole regime phenomena In Figure 2, we plot the phase time for lh and hh It is clear from this through a single potential barrier. figure that the lh modes (dashed line) are faster than hh modes (solid line), although the difference is not large. In an effort to validate our numerical simulation with experi- mental data, we compare the results of Figure 2 with those of Dragoman et. al. [2] The tunneling times measured for holes, for null or small magnetic field in a ballistic-regimen nano-device, are of the order of 10−13 s. To reproduce that situation, we canceled the interaction between holes by assuming κt ≈ 0 (uncoupled regimen), in the absence of magnetic field. In our calculation the phase time for hh and lh is also of the order of 10−13 s, which is in an acceptable qualitative agreement with experiment and it is just one order less than the traveling time for elec- trons through a single barrier [46], as was detected exper- imentally for a DBRT [43]. We have observed broaden 4 Figure 3: Direct phase time τii for hole tunneling through a single barrier as a function of the hole incident energy for several values of the barrier thickness at κt ≈ 0, with Vb = 0.498 eV and i = 1. peaks of the phase time for hh+3/2 direct transition at E ≈ 0.7 eV and for lh−1/2 direct transition at E ≈ 0.9 eV . We have observed, although not plotted, that the trans- mission coefficient maxima are located at comparable en- ergies. It was shown in Ref.[11] that, the tunneling time follows the band structure. This feature of hole tunneling agrees also with calculations of Ref.[27], where a particu- lar case of the envelope function approximation was used. Evaluating the tunneling times for energies in the range 0.40 eV 6 E 6 0.60 eV we have found, although not shown, a reduction of τii in 3 orders of magnitude with re- spect to that shown in Figure 2 when the in-plane momen- tum changes as ∆κt = 0.001 Å. This behavior of τii, was earlier detected in experiments for a hole passage through a slightly different system [48] -a double asymmetric quan- tum well (QW)-, and is considered to arise from valence subbands mixing [48]. Figure 3 displays a raising of the phase time while Lb increases in the case of uncoupled regime (κt ≈ 0) of holes incidence. The same is achieved for light holes, although not shown in this figure. This behavior agrees qualitatively with the results of an experiment on time-resolved lumi- nescence spectroscopy [43]. For E > 0.4 eV , τii is sensitive to the raising of Lb and it is shown an oscillating behavior of the transmission phase time as function of the incident energy. The resonances shift to smaller energies, increasing their intensities with Lb. It is assumed understood that, when the de Broglie wavelength of the hole propagating mode is a multiple of Lb, there is a resonance in the trans- -- -- Figure 4: Phase time versus barrier thickness for the direct tran- sitions hh±3/2 → hh±3/2 given as solid diamonds/circles (E < Vb/E > Vb), in the uncoupled case (κt ≈ 0). A fragmented exponen- tial fitting is shown in dashed line (with full circles) at E = 0.551 eV . The barrier height Vb = 0.498 eV . mission and correspondingly the phase time delay raises at these energies. As Lb increases, new resonances appear, thus the oscillating behavior is achieved. These oscillations correspond to the quasi-stationary levels of the virtual QW with increasing width. In that sense, from τii properties we are able to forecast a longer lifetime for the lower un- bound hole levels. On general grounds, we found this in correspondence with the idea of the quasi-stationary states lowering, whenever the virtual QW becomes wider. Figure 4 shows phase transmission time versus barrier thickness for the direct transitions hh±3/2 → hh±3/2 given as solid diamonds/circles (E < Vb/E > Vb), in the uncou- pled case. We prescribe appealing tunneling events of holes for E < Vb under uncoupled propagating regime. This fig- ure displays τii for resonant direct hh±3/2 → hh±3/2 tran- sitions (diamonds), which rapidly becomes autonomous from barrier broadening as Hartman predicted for elec- trons [29]. These results are in accordance with the tran- sit time for optical pulses through opaque barriers (which transmit about 10−4 of the incident radiation) [13] and also with theoretical predictions for the delay time in one and two dimensions [58]. The circles in Figure 4 corre- spond to numerical results within the MSA, for the tun- neling time of direct paths (indicated in the legend) for E > Vb. Here we must stress, that the resonant oscillat- ing behavior (on top of an average slope of smooth growth [59]), is an expected phenomena due to the barrier's inter- Figure 5: Direct phase time τii for hole tunneling through a DBRT structure as a function of the energy of an impinging hole stream at κt ≈ 0, with Vb = 0.550, Lb = 20 Åand Lw = 50 Å, for i = hh+3/2 (solid line), and i = lh−1/2 (dashed line). The inset displays transmission coefficients for the same direct transitions, through an identical DBRT heterostructure, as a function of the incident energy. The labelling for the hole levels follows that of the Reference [45]. ference effects with the continuum states, wide accepted as Ramsauer-Townsend oscillations [60]. For E > Vb we briefly investigated segments in Figure 4, pursuing a lo- cally exponential-like behavior of τii, as was suggested in a reported experiment with holes [43], within a rank com- parable with their measurements. The interval of barrier thickness length variance in their measurements, ranges from 30 Å up to 80 Å. Notice in Figure 4 a sectional fit- ting, using an exponential growth function [61] (dashed line with full circles). We do believe that oscillations are already intrinsically present in Heberle's, et al. accurate investigation [43] and could be shown by including more experimental points or enlarging in rank the search. Phase time through DBRT and superlattice Figure 5 is devoted to analyze briefly, the spectrum and the quasi-stationary bound states lifetime in a DBRT. The curves of Figure 5 (inset), show several transmission phase times τii resonances for the uncoupled hole regime. In par- ticular we have analyzed a semiconductor heterostructure of the form GaAs/(AlAs/GaAs)2/GaAs. It was shown in Ref. [11] that the tunneling time follows the band struc- 5 Table 1: Comparison to typical quasi-steady DBRT hole-level struc- ture with Vb = 0, 550 eV , Lb = 10 Å, and Lw = 50 Å. Levels [45] Resonances [eV ] Experiment[53]. Theory (MSA). hh1 lh1 hh2 hh3 lh2 hh4 0,028 0,073 0,111 0,245 0,299 0,428 0,027 0,071 0,110 0,247 0,307 0,434 ture profile. Figure 5 (inset) displays clearly, at E < Vb, the familiar discrete quasi-stationary hole-level spectrum of the embedded QW, widely accepted from hole quan- tum transport calculations [3, 27, 44, 49 -- 52]. We have found at low energies (E < Vb), that the τii resonances are allocated at energies which correctly reproduce the hole spectrum previously reported for the DBRT experiment [53], as expected (see Table 1). Notice the good quanti- tative agreement of our approach in Table 1, with similar values reported elsewhere [53]. A previous comparison [3] with a robust and widely accepted theoretical model [27], exhibits the advantages of the present approach. Tunneling time can not be quantitatively determined from the resonant tunneling line width, [57] nevertheless some qualitative prognosis is allowed. Notice the general trend of larger values for hh states phase time in compari- son to that of lh states. This is in correspondence with the properties we have found for the transmission-coefficient resonances depicted in the inset of Figure 5, which shows, as tendency, that hh resonances are thinner than the lh ones. This observation foretells a stronger confinement for heavy holes, and we guess it could be used in resonant- tunneling diode design and nano-modelling [45]. Among the different formulations of tunneling time with recurrent experimental agreement [34], the phase time have clearly advantages. In Reference [11], was demon- strated that the phase time associated with the passage (tunneling or not) of particles is a relevant quantity of the quantum transport, which was calculated and success- fully compared with precise experimental measurements [12, 13]. Moreover in Reference [11], some appealing effects for electron phase time were predicted, namely: superlu- minal phase propagation and resonant-band structure. In an attempt to distinguish similar behaviors for holes, we perform some numerical simulations, exhibited in the next two figures. In Figure 6 the phase transmission time τ1(2), through a single cell (n = 1) and a DBRT (n = 2) respectively, together with T2 are shown in connection with the free motion time τf (dotted line). The last, is the propagation time of the quasi-particle in absence of scattering poten- tials. The direct paths under analysis, are indicated in Figure 6: Top(Bottom) panel shows the evolution of the phase time, transmission probability, free motion time through n = 2 cells (DBRT) and the phase time through one single cell as function of the incoming energy for direct transitions of hh± 3/2 (cid:0)lh∓ 1/2(cid:1). With Vb = 0.23 eV , Lb = 30 Åand Lw = 150 Å. A 30% concentration of Al was fixed inside the barrier. the legends of each panel. A resonant behavior for τ2 is very clear for both hh states (top panel) and lh states (bottom panel). In the forbidden energy region, τ2 ap- proaches τ1 (dashed red line), which represents its lower bound as had been foretold for electrons [11]. The rela- tionship between τ2 and T2 (blue line) resonances - as one can see from (8)-, justify τ2 to accurately reproduce, the quasi-stationary states of the embedded GaAs QW. An- other appealing feature arises for forbidden energy regions (T2 vanishes), where τ2 < τf . The opposite is predicted for allowed energies. For hh states (top panel) of incom- ing energies around 100 meV , (τf − τ2) ≈ 0.4 ps , while for lh states (bottom panel) (τf − τ2) ≈ 0.1 ps. This ear- lier arrival time for hh and lh, was predicted for electron tunneling through a same heterostructure [11], and sug- gest a more speedily passage of holes through a DBRT like (GaAs/Al0.3Ga0.7As/GaAs)2. It remains to notice that (τf )hh > (τf )lh for the DBRT, having a bigger difference at low energies, with respect to that of high energies. Another remarkable distinction is the width of T2 resonances, which are noticeable widest for lh. This difference yields to a larger confinement for hh inside the embedded QW, with respect to that for lh. Thus greater lifetime for hh in the scattering region is expected. In Figure 7 the phase time τ8 and transmission coef- ficient T8, through a superlattice (n = 8) undoubtedly 6 E = 0.551 eV 20 40 60 80 100 120 140 160 180 E = 0.475 eV ) s p ( i i τ 6.0 4.0 2.0 0.0 ) s p 2 − 0 1 ( i i τ 10.0 9.0 8.0 7.0 10 20 30 40 50 60 ) s p ( i i τ ) s p 2 − 0 1 ( i i τ 2.0 1.0 0.0 8.4 8.0 7.6 7.2 E = 0.551 eV 20 40 60 80 100 120 140 160 180 E = 0.475 eV 20 40 60 80 Lb (A) 100 120 140 Figure 8: Phase transmission time as a function of the barrier thick- ness for heavy hole transition hh±3/2 → hh±3/2 within the uncou- pled regime, κT ≈ 0. Both incident energies E: less (full triangles) and greater (empty triangles) than the barrier height Vb = 0.498 eV were fixed at 0.475 eV and 0.551 eV, respectively. The two bottom and top panels represents the numerical values of this time for the DBRT and the superlattice of n = 4 cells respectively. We can also note a decrease of the phase time in the region (10 − 15) Å for the DBRT and around (10 − 17) Å for the superlattice, showing the attractive effect of the barrier, as a reminiscence of the similar behavior widely quoted in the literature for electrons through a simple barrier [8, 9]. For energies higher than the potential barrier (empty tri- angles ) τii, in the DBRT and in the superlattice, exhibits a resonant oscillating behavior similar to those of the sin- gle barrier (see Figure 4). This is a consequence of an interference effect between the reflected wave and the con- tinuum states, known as Ramsauer-Townsend oscillations as commented above. By comparing Figure 4 and Fig- ure 8, it is clear that oscillations [4, 60] remain robust for the DBRT. The robustness of this feature vanishes for the superlattice, provides that well-resolved oscillating curves are no longer observed for n = 4 cells. Figure 7: Top(Bottom) panel shows the evolution of the phase time, transmission probability, free motion time through n = 8 cells (DBRT) and the phase time through one single cell as function of the incoming energy for direct transitions of hh± 3/2 (cid:0)lh∓ 1/2(cid:1). With Vb = 0.23 eV , Lb = 30 Åand Lw = 150 Å. A 30% concentration of Al was fixed inside the barrier. follow a resonant-mini-band structure. Direct hole transi- tions were considered and appear indicated in the legends of each panel. For finite periodic systems -which is the case envisioned here-, a finite number of intra-band en- ergy levels [62] are typically observed, at variance with the continuous band structures predicted for infinite peri- odic systems. In the mini-gaps, T8 vanishes and τ8 tends to τ1, which is similar to the DBRT case discussed before. On the other hand, in the mini-bands region τ8 exhibits a resonant behavior approaching τf , which behaves as a lower bound [11]. In the scattering regions, τ8 increases as the incoming energy approximates to any allowed energy level. While, in the forbidden regions the holes propaga- tion is away faster, reflecting the evanescent behavior and the lack of hospitality of the barrier. These evidences for hh and lh phase time, resemble those for electrons as was foretold for the tunneling time in a similar superlattice [11]. Figure 8 exhibits two dissimilar behaviors of τii for di- rect transitions hh±3/2 → hh±3/2 as function of the barrier thickness. For energies under the potential barrier (full tri- angles) τii becomes independent of the barrier thickness, being consistent with Hartman's classical premonition for electrons [29]. In the DBRT, τii becomes autonomous starting from Lb ≈ 60 Å, meanwhile, in a superlattice of n = 4 cells, this quantity saturates at about Lb ≈ 50 Å. 7 Giant conductance phenomena 0.0 0.5 1.0 1.5 2.0 sine of the phase difference. One can see, that the curve for hh output channel (solid line) changes two times in π radians and in the bottom panel appear two Gj maximum correspondingly (at E ≈ 0.72 eV and at E ≈ 1.25 eV ). Meanwhile for the lh output channel (dashed line), there is only one change (at E ≈ 0.98 eV). Direct paths contribu- tions are the largest ones, as it follows from vanishing band mixing effects as κt ≈ 0. The maximum values shown in the bottom panel of Figure 9, for an even function of ∆θ (cos ∆θ), at even multiples of π [17, 19, 20, 54, 55], are not enough to confirm an abrupt increment of trans- mission. Indeed, there is nothing particularly large about hole transmission through a single cell, as that reported for electrons in finite periodic systems with time reversal symmetry [19]. Besides, in the single-cell case, does not exist a resonant mechanism that allows hole conductance to become giant through the barrier. However, it is tempt- ing to have a fairly reference frame to compare to what we will see next for a DBRT. Worthwhile noticing -although not shown here-, the largest contribution of the imaginary scattering amplitude component, that give rise Gj to max- imize in the same energy intervals, as we have shown in the Figure 9 bottom panel. Figure 9: The bottom panel plots the 4 outgoing channel conduc- tances Gj through a single barrier heterostructure with Lb = 20 Å, where i = 1, 4 (solid line) represents the hh±3/2 channels, respec- tively, and i = 2, 3 (dashed line) represents the lh±1/2 channels, respectively. The top panel displays the cosine of phase shift for the same transitions. It is quite well established in the literature the strong relation between the transmission magnitudes and the wave function's phase [17, 19, 62]. Specifically, the giant con- ductance events are nearly related with phase shifts and phase interference phenomena [19]. It has been shown for various mesoscopic systems, the correlation between the variations of the conductance and an even function of the phase shift, due to an alternatively appearance of maximum-minimum valued sharp peaks or dips, whenever an even functional of the phase shift, changes by almost π radians [17, 19, 54, 55, 63]. Here, the cosine of the phase shift was chosen to verify, whether strike maximum- transmission peaks occur or not, in a negligible conduc- tance background. Commonly, the cos(θij ) is plotted, as it is a suitable even function to show evidences of giant conductance events [17, 19]. Figure 9 (bottom panel), presents the conductance Gj to direct channels hh±3/2 → hh±3/2 (solid line), lh±1/2 → lh±1/2 (dashed line), and the cosine of phase shift for the same transitions (top panel) for the uncoupled hole regime as function of the incident energy. It is clear that the max- imum values in the conductance for both output channels are always accompanied by an extreme change in the co- Figure 10: Dependency of the 4 outgoing channels conductance and the cosine of phase shift with hole incident energy, through a DBRT and a superlattice with n = 4 cells of width Lb = 20 Å, and height Vb = 0.498 eV . Well-established time reversal symmetry of the (4 × 4) Kohn-Lüttinger model, is not fulfilled by the (2 × 2) sub- spaces as was recently demonstrated [56]. Events of GG have been observed in many finite periodic systems, which 8 possess time reversal invariance but as yet not observed in time asymmetric ones. Probably, this explains why other accurate calculations made in the (2 × 2) Kohn-Luttinger subspaces, did not report GG events for holes [43, 45, 47, 50]. Figure 10 displays the conductance and phase shift behavior as a function of the incident energy, for direct transitions to hh±3/2 (solid line) and lh±1/2 (dashed line) output channels. The Gi curves exhibit evidences pointing out the existence of well-defined resonances for E < Vb. It is relevant in this case, the presence of striking conduc- tance events, as can be clearly noticed in the panels for n = 2 of Figure 10. Indeed, peaks lh2 and hh4, have found to have a clear signature of GG phenomenon since: (i) they are sharp resonances, (ii) they keep a simple rela- tion with an even function cos(θij )n=2 of the transmission phase (θij )n=2, thereby peaks' locations at energy, are co- incident with those of (θij )n=2 whenever the last changes by almost π radians in a relative short energy interval [17 -- 19, 23, 62], and finally (iii) there is a strong phase in- terference mechanism between propagating or evanescent modes, giving rise to resonant transmission [19]. As follows from the scattering theory, the GG mechanism in a DBRT, derives from the interference between states, whenever the incoming modes energy perfectly matches with one of the quasi-stationary states within the embedded QW of the scatterer system. This agreement produces the essential conditions for energy to maximize, i.e. a resonant trans- mission occurs. Sharp peaks become visible at energies E ∼= 0.29 eV (dashed line) for direct transitions of lh, and for E ∼= 0.42 eV (solid line) for direct transitions of hh. We have disregarded the lh1 sharp resonances and oth- ers in the low energies region, because they do not reach maximum and also the corresponding phase shifts differ significantly from π radians. In addition, it is necessary to mention the loss of the GG phenomenon in the super- lattice. As one can see in the top panels of Figure 10 (for n = 4 and E < Vb), the (θij )n=4 changes are not about π radians. In this case the leak of maximum transmission for the superlattice becomes visible. We conjecture the exis- tence of a competitive mechanism when the number of cells grow, and consequently a phase coherence effect rises. For higher energies, the direct Gi might give rise to a plateau, and we foretell the probable existence of another related striking feature: the giant resistance. Despite a more am- biguous interference process in this case, a sharp dips are expected as a consequence of a transmission rate decrease, due to a resonant transfer of holes flux toward evanescent modes. It is likely to underline that the energies of the reso- nances -whether they are giant or not-, accurately repro- duce the quasi-bounded hole spectrum as was published for the DBRT [27], were another approaching frame was applied. This is to say that, none GG phenomenologi- cal analysis can possibly be complete if a resonant means is missing. In this sense, to endorse the GG evidences showed in Figure 10, the most immediate tool within the MSA, is the TM technique to estimate the hole states en- ergies at the QW of the DBRT. The basic idea here, is to search several states of the embedded finite quantum well (fQW), by cautious tuning those of the external in- finite quantum well (iQW). The last we had achieved, by imposing that zl,r slowly tends to z1,4, respectively (see Figure 11). This method approaches the iQw boundary conditions to those of the fQW, and is a reasonably good sketch for an estimation of some fQW levels. Figure 11: Schematic representation of the quasi-bound-state levels (horizontal lines) and valence band lineup, within an infinite walls quantum well. The ground state wave function (dotted line) is repre- sented qualitatively. The energy is taken as positive for convenience. Figure 11 shows schematically, holes quasi-bound-state levels (horizontal lines) and valence band lineup, within the iQW. For a numerical estimation of the quasi-stationary levels within the fQW, we consider a cell enclosed between the zL and z03 points. The barriers and well have the same dimensions of the Figure 10. Thus, it is undemanding and straightforward to reproduce a DBRT or a superlattice, by a periodic concatenations of cells from z03 [3]. The latter considers facts of equivalence for points z1(z2) and z3(z4), respectively, as they have the same materials to the left and to the right. As boundary condition for the iQW, is standard to impose the wave function goes to zero at the points (zl − ∆z) and (zr + ∆z) ϕ(zl − ∆z) ≡ ϕ(zr + ∆z) ≡ 0, (9) being ∆z the splitting distance between these points and the infinite potential. In the framework of the TM ap- proach, the latter requirement can be written as: (cid:18) 0 ϕ′(z) (cid:19)(zr+∆z) M f d(zr + ∆z, zl − ∆z)(cid:18) 0 = ϕ′(z) (cid:19)(zl−∆z) . (10) Therefore, the non trivial solutions of the expression (10) can be numerically determined, by solving the equation Det{M AD(zr + ∆z, zl − ∆z)} = 0, (11) being M ad the amplitude-derivative matrix block of the TM M f d. Up to this end, it is turn clear from Table 2, a reasonably good agreement between our estimation for the energies of several quasi-steady levels of the fQW and 9 1 0 1 0 1 ) h / 2 e 2 ( i G ) h / 2 e 2 ( i G ) h / 2 e 2 ( i G n = 9 n = 8 n = 7 hh±3/2 → hh±3/2 lh∓1/2 → lh∓1/2 0 0.2 0.25 0.3 0.35 0.4 0.45 Ei (eV ) Figure 12: Two-probe conductance through the four direct accessible channels as a function of the incoming energy on the κT ≈ 0 limit, for a superlattices of n = 7, 8, 9 cells (from bottom to top) in the range of energies between [0.2 − 0.45] eV . the resonant-band structure displayed in Figure 10. Now it is undoubted, the very existence of these fQW levels, and therefore their role as a fair resonant mechanism of the GG phenomena in the DBRT we claim for. The lh2 and hh4 energy values showed here, are a truthful prove of the GG phenomena, predicted for the uncoupled hole transport in the DBRT for sharp resonances allocated at similar energies (see Figure 10). Figure 12 presents the dependency of the conductance G of hh and lh through a superlattices of n = 7, 8, 9 cells with the energy of the incident flux. The holes travel along all four accessible direct channels hh±3/2 → hh±3/2 Table 2: Quasi-steady fQW hole-level for a DBRT structure, quoted by Eq.(11), with Vb = 0.498 eV , Lb = 20 Å, and Lw = 50 Å. Levels (Labeling from Ref.[45]) Resonances [eV ] hh2 lh2 hh4 0.1362 0,2978 0,4112 ) s p ( i e m T n o i s s i m s n a r T e s a h P hh+3/2 → lh+1/2 lh−1/2 → lh+1/2 lh−1/2 → hh−3/2 hh+3/2 → lh−1/2 2.5 2.0 1.5 1.0 5.0 0.0 0 20 40 60 80 100 120 140 160 180 Lb (A) Figure 13: Phase transmission time through one single cell versus barrier thickness for the direct and crossed passages of holes in a coupled case (κT = 5 × 10−4 Å−1) for E = 0.475 eV < Vb = 0.498 eV . (solid line) and lh±1/2 → lh±1/2 (dotted-dashed line). It is wide open visible the thinner peaks of the hh resonances, respect to those of the lh ones, since we are restricted on the uncoupled regime and thereby hh-spectrum states are the closest due to their larger effective mass. Fur- ther outstanding behavior, is the filter-like effect that can be observed on both flavors of holes in the selected re- gion. Indeed, the barriers become opaque for the hh/lh states at low/high energies, leading to a stronger confine- ment for the hh/lh at low/high energies. This fact yields a larger/shorter lifetime for hh/lh inside the allowed region of the scatterer system for low energy, at variance with the opposite for higher energies. Dealing with a superlattice of n = 9 cells (topmost panel), we found an hh − lh overlap region at the vicinity of 0.3 eV. The phenomenology de- scribed in Figure 12, allows us hypothesize it as concerns applications in hole-based electronics, pursuing maximum- minimum conductivity response as function of the charge carrier effective mass. 3.2. Coupled hole regime phenomena Let us considers first the Hartman effect under this regime. Figure 13 depicts the phase transmission time as a function of the barrier thickness for several crossed transi- tions paths, namely: hh+3/2 → lh+1/2, lh−1/2 → lh+1/2, lh−1/2 → hh−3/2 and hh+3/2 → lh−1/2, of the sixteen possible ones, through a single-cell scatterer system. This plot shows the phase transmission time curves as a func- tion of the barrier thickness Lb, given fixed E < Vb. Notice the changes of τij through each path while Lb increases, until τij reaches saturation values. This autonomous be- havior of the phase time with Lb, unambiguously gives rise to the paradoxical Hartman effect [29], which remains robust, even for a relative strong coupling of holes. Ow- ing to simplicity we dropped the remain transitions, pro- vided they follow a similar tendency. Worthwhile un- 10 lh−1/2 → hh+3/2 lh+1/2 → hh+3/2 G1 1.0 0.5 0.0 -0.5 -1.0 0.4 0.2 0.0 ) d a r π ( ) j 2 θ ( s o c ) h / 2 e 2 ( e c n a t c u d n o C lh+1/2 → lh−1/2 hh−3/2 → lh−1/2 G2 1.0 0.5 0.0 -0.5 -1.0 0.2 0.1 0.0 ) d a r π ( ) j 2 θ ( s o c ) h / 2 e 2 ( e c n a t c u d n o C 0.15 0.2 0.25 0.3 0.35 0.4 0.15 0.2 0.25 0.3 0.35 0.4 E (eV ) E (eV ) Figure 14: The bottom panel shows the crossed-path contributions to the conductance through the first outgoing channel, G1, for a DBRT with barriers of width Lb = 20 Å, and height Vb = 0.498 eV . The top panel displays the cosine of two phase shifts that match with the same outgoing channel, hh+3/2 → lh−1/2 and hh+3/2 → lh+1/2, both in the same range of energy. Figure 15: The bottom panel shows the the crossed-path contri- butions to the conductance through the second outgoing channel, G2, for a DBRT with barriers of width Lb = 20 Å, and height Vb = 0.498 eV . The top panel displays the cosine of two phase shifts that match with the same outgoing channel, lh−1/2 → lh+1/2 and lh−1/2 → hh−3/2, both in the same range of energy. derlying some characteristics of τij transient interval for several paths. Crossed transitions hh+3/2 → lh+1/2 (+) and lh−1/2 → lh+1/2 (up empty triangles), show negative slope before reach saturation. Meanwhile the crossed path hh+3/2 → lh−1/2 (down full triangles), exhibits a double change of slope. These behaviors clearly departs (in the transient interval) from the classical Hartman premoni- tion for electrons and deserve more investigation to shed light on this topic. In the presence of interference effects (κt 6= 0) at E ≈ 0.55 eV , we have found abrupt tunneling phase changes when ∆θ ≈ π radians, which seems to be an anomalous behavior of the phase time. In Figure 13, we have found an overlap of τij for several crossed tran- sitions. For example: with j : 1(2) → 3 : i at Lb ≈ 20 Å, also with j : 1(2) → 2(3) : i at Lb ≈ [110, 140] Å, and with j : 1 → 2(3) : i at Lb ≈ [70, 80] Å. In some cases the effective mass changes, and the incoming channels be- long to a so called Kramer-up degenerated hole states [3], while the outgoing channels are Kramer-low ones. It can be readily observed, similar τij matches, for other crossed transitions at different values of Lb. These coincidences relate to the strong hh− lh propagating modes mixing, to the simultaneous treatment of the 4-coupled channels of the present approach, and to hh − lh interaction with an effective potential Vef f due the barrier broadening. Ac- cordingly, whenever a match in τij takes place, the time spent by the hh and lh in the classically forbidden region, shall be the same. We turn now to discuss how robust are the hole gi- ant conductance features, in the presence of valence-band mixing. Figure 14 and Figure 15 present the conductance and the cos(θij )n=2 through the outgoing channels hh+3/2 and lh−1/2, respectively, as a function of the incoming en- ergy under interference between channels. We had taken κt = 10−3 Å−1. Despite the presence of sharp peaks at E < Vb (qualitatively similar to those showed in Figure 10), as one can observe at E ∼= 0.16 eV , E ∼= 0.22 eV and E ∼= 0.24 eV in Fig.14, it is not seen a neat concor- dance with the widely accepted GG formulation posted above [17, 19, 54, 55, 63], and illustrated in Fig.10. Firstly there is not a nice correspondence between the highest peak of G1 at the vicinity of E = 0.24 eV (see Fig.14 bottom panel) and the fluctuation of (θij )n=2, which for this peak is sensitive less than π radians (see Fig.14 top panel). On the other hand, the values of G1,2 exhibited in Fig.14 and Fig.15, respectively, are in average almost one order smaller than those shown by Fig.10, being the comparative behavior observed in Fig.15 even more restric- tive in this sense, with respect to Fig.10. Following these evidences, we conjecture so far, that the GG events are not straightforward signatures for hole tunneling travers- ing a DBRT, under a selected range of parameters in the coupled-particle regime. Additional remarkable differences become clear from a comparison of this trend in the coupled hole regime and Figure 10. For instance, there is a small shift between 11 ) h / 2 e 2 ( e c n a t c u d n o C 0.4 0.2 0.0 G2 ) j i θ ( e n s o c i i i i ) s p ( e m T n o s m s n a r T e s a h P 0.15 0.2 0.25 0.3 0.35 0.4 E (eV ) Figure 16: Full contributions to the conductance through the second outgoing channel, G2, for a DBRT with barriers of width Lb = 20 Å, and height Vb = 0.498 eV . the position in energies of the peaks and the changes in the cos(θij )n=2. We guess that the last reflects the loss of maximum transmission when the valence-band mixing, i.e. the interference between paths is incremented. Be- sides, the Gn oscillations spectra in the low-energy inter- val shown in Fig.15, seem to be of the Fano-type leading to assume the existence of interference of embedded-QW discrete levels with a continuum. The later gives rise to characteristically asymmetric peaks in transmission spec- tra [64]. The porpoise of Figure 17, is to briefly analyze fur- ther anomalous features in the transmission phase time spectra. Fig.17 displays in panel (a) the cos(θij)n=1 for the crossed lh − hh transition lh+1/2 → hh−3/2 (dashed line, i : 3 → 4 : j). In panel (a) with dotted line, we plot the difference of the the same magnitude for the path i : 3 → 3 : j with respect to the path i : 3 → 4 : j. The bottom panel (b) plots tunneling phase time for the crossed-transition i : 3 → 4 : j through a single barrier In the presence of interference effects heterostructure. (κt 6= 0) at E ≈ 0.55 eV, the phase changes abruptly when ∆θ ≈ π radians, as can be seen from the dashed line in the top panel of Figure 17. Both curve in panel (a) show a clear correspondence of abrupt changes in cos θij with observed peaks(dips) of τ ij, which seem to be an anomalous behavior of the phase time, as they resemble Fano-like resonances due theirs asymmetry hallmark [64]. In the neighborhood of 0.8 eV [see panel (b)] we have found Fano-like positive (negative) peaks (dips) of τ ij with i 6= j, displayed in the bottom panel of Figure 17. They are as- sociated with changes of the phase in almost π/2(π/3) radians, and are plotted in the top panel with dashed line. We also have observed, although will not be shown here, local increments of transmission coefficients in the energy vicinity, where negative dips appeared, though regretfully they need more investigation to derive proper conclusions. Most interestingly, though, we found indications that 12 (a) cosine ( θ -θ ) 33 34 cosine ( θ ) 34 0.5 0.6 0.7 0.8 0.9 (b) crossed transition lh to hh +1/2 -3/2 1 0 -1 4 3 2 1 0 -1 -2 0.5 0.6 0.7 E (eV) 0.8 0.9 Figure 17: The bottom panel plots tunneling phase time for the crossed-transition j : 3 → 4 : i through a single barrier heterostruc- ture at κt = 0.001 Å−1, with Lb = 20 Å and Vb = 0.498 eV. Top panel shows the cosine of phase shift for the same transition j : 3 → 4 : i (dashed line) where the first number stands for the input channel j and the other stands for the output one i. In dotted line the same magnitude is shown for the direct path j : 3 → 3 : i with respect to the crossed path j : 3 → 4 : i. might be connected to possible giant conductance events in the single-cell case, which we omit graphically here because of brevity. There were found two prompt shifts in trans- mission phase amplitude, equivalent to an even multiple of π (∆θij = 6π) at κt ≈ 0.01 Å−1, similar to those plot- ted in Figure 10. One of them, appear in lh+1/2 → lh−1/2 intra-band transition (i : 3 → 2 : j) when Lb varies within [10 − 20] Å range. The second one, was observed through hh−3/2 → lh+1/2 inter-band transition (i : 1 → 4 : j) when Lb varies within [120− 130] Å range. Then, maxima of Gj could be expected for those crossed paths, not only vary- ing the incident energy, but the barrier thickness as well. However, a clear resonant mechanism in this case remains a puzzle and further investigations need to be addressed to clarify wether the effect of giant conductance for holes, is expected to arise under valence-band coupling phenomena during transport trough a single-cell system. 4. Concluding Remarks [8] E. H. Hauge and J. A. Støvneng, Rev. Mod. Phys 61, 917 Relevant and striking quantum transport features of holes, directly connected to the phase, the phase shift and the phase resonant interference phenomena are: the giant conductance-resistance events, the Hartman effect, and the earlier phase propagation in the barrier. Sev- eral results nicely resemble those previously reported for electrons [11], while some predictions of the MSA model are in acceptable qualitative compliance with the experi- ments [2, 48, 53] and with theoretical predictions [46, 62]. Theoretical evidences of the giant conductance phenom- ena for decoupled hole transmission through DBRT were exhibited. In this case a trustworthy resonant mechanism supporting this anomaly, is the alignment of the incident propagation energy with one of the quasi-bond hole states of the embedded quantum well in the DBRT. This mech- anism was confirmed with an independent numerical cal- culations of several quasi-bound levels. The giant con- ductance effect vanishes in the presence of valence-band particles coupling and by tuning the number of superlat- tice layers as well. The hole phase transmission time for a (GaAs/Al0.3Ga0.7As/GaAs)n superlattice, shows a res- onant mini-bands structure in the scattering regions, hav- ing the free motion as an inferior limit time. Moreover, in the forbidden regions the holes propagation became away faster and was shown τii to be inferiorly bounded by the single-cell phase time τ1. In the DBRT mini-gaps regions, τii moves forward from the free motion time, suggesting a more speedily phase propagation within the barrier. The Hartman's classic prediction is observed robust for null and finite hole band mixing. Was presented an alternate- selective confinement strength independently for both fla- vors of holes, implying filter-like effects on the effective mass via the manipulation of the incident energy in the uncoupled hole regime. Acknowledgments We thanks Dr. H. Rodríguez-Coppola for clue sug- gestions in some numerical treatments. One of the au- thors (L.D-C) gratefully acknowledges the Visiting Aca- demic Program of the UIA/México and the facilities of the IFSC-USP/São Carlos, Brasil. References [1] H. Schneider, H. T. Grahn, K. Klitzing, and K. Ploog, Phys. Rev. B 40, 10040 (1989). [2] D. Dragoman and M. Dragoman, J. Appl. Phys. 93, 6133 (2003). [3] L. Diago-Cisneros, H. Rodríguez-Coppola, R. Pérez-Álvarez, and P. Pereyra, Phys. Rev. B 74, 045308 (2006). [4] L. Diago-Cisneros, H. Rodríguez-Coppola, R. Pérez-Álvarez and P. Pereyra, Rev. Mex. Fís 53, 7 (2007). [5] P. Pereyra and H. Simanjutak, Phys. Rev. E 75, 1 (2007). [6] H. G. Winful, Phys. Rev. Lett 91, 26 (2003). [7] H. G. Winful, Phys. Rev. Lett 90, 2 (2003). (1989). [9] R. Landauer and Th. Martin, Rev. Mod. Phys 66, 217 (1994). [10] P. Pereyra, V. G. Ibarra-Sierra and J. L. Cardoso, Microelec- tronics Journal 40, 779 (2009). [11] P. Pereyra, Phys. Rev. Lett. 84, 1772 (2000). [12] A. M. Steinberg, P. G. Kwiat and R. Y. Chiao, Phys. Rev. Lett 71, 708 (1993). [13] Ch. Spielmann, R. Szipöcs, A. Stingl, and F. Krausz, Phys. Rev. Lett. 73, 2308 (1994). [14] S. Chu and S. Wong, Phys. Rev. Lett 48, 738 (1982). [15] S. Bosanac, Phys. Rev. A 28, 557 (1983). [16] A. B. Shvartsburg, M. Marklund, G. Brodin and L. Stenflo, Phys. Rev. E 78, 016601 (2008). [17] A. Kadigrobov, A. Lagoskin, R. I. Shekter and M. Jonson, Phys. Rev. B 52, R8662 (1995). [18] A. Kadigrobov and M. Jonson, Applied Physics Report 94, 6 (1994). [19] P. Pereyra, Physica E 17, 209 (2003). [20] N. K. Allsopp, J. Sánchez-Cañizares, R. Raimondi and C. G. Lambert, J. Phys. Condens. Matter 8, L377 (1996). [21] Roberto Romo, Phys. Rev. B 66, 245311 (2002). [22] Jorge Villavicencio and Roberto Romo, Phys. Rev. B 68, 153311 (2003). [23] Junling Wu, Baigeng Wang, Jian Wang and Hong Guo, Phys. Rev. B 72, 195324 (2005). [24] Yuanbo Zhang, Victor W. Brar, Feng Wang, Caglar Girit, Yossi Yayon, Melissa Panlasigui, Alex Zettl and Michael F. Crommie, Nature Physics 4, 627 (2008). [25] A. K. Geim and K. S. Novoselov, Nature Materials 6, 183 (2007). [26] Guangyu Xu, Carlos M. Torres Jr., Emil B. Song, Jianshi Tang, Jingwei Bai, Xiangfeng Duan, Yuegang Zhang and Kang L. Wang, arXiv.org (to appear in Nano Letters) (2010). [27] R. Wessel and M. Altarelli, Phys. Rev. B 39, 12802 (1989). [28] E. P. Wigner, Phys. Rev. 98, 145 (1955). [29] T. E. Hartman, J. Appl. Phys 33, 12 (1962). [30] L. A. MacColl, Phys. Rev 40, 621 (1932). [31] D. Bohm, Quantum Theory, (Prentice-Hall, New York, 1951). [32] F. T. Smith, Phys. Rev. 118, 349 (1960). [33] A.I. Baz', Sov. J. Nucl.Phys. 4, 182 (1967). [34] M. Büttiker and R. Landauer, Phys. Rev. Lett. 49, 1739 (1982). [35] M. Büttiker Phys. Rev. B 27, 6178 (1983). [36] G. Nimtz, A. Haibel, and R.-M. Vetter, Phys. Rev. E 66, 037602 (2002). [37] A. Ranfagni, D. Mugnai, P. Fabeni and G. P. Pazzi, Appl. Phys. Lett. 58, 774 (1991). [38] A. Enders and G. Nimtz, J. Phys. France I2, 1693 (1992). [39] J. M. Lüttinger and W. Kohn, Phys. Rev. 97, 869 (1955). [40] L. Diago-Cisneros, Tunelaje multicanal y simetrías de los hue- cos en heteroestructuras semiconductoras, Universidad de La Habana, Facultad de Física, La Habana (2005). [41] P. A. Mello, P. Pereyra and N. Kumar, Ann. Phys. (N.Y) 181, 290 (1988). [42] J. Imry and R. Landauer, Rev. Mod. Phys. 71, 306 (1999). [43] A. P. Heberle, X. Q. Zhou, A. Tackeuchi, W. W. Rühle, and K. Köhler, Semicond. Sci. Technol. 9, 519 (1994). [44] T. Kumar, M.Cahay, and K. Roenker, Phys. Rev. B 56, 4836(1997). [45] G. Klimeck, R. Chris Bowen, and T. B. Boykin, Supperlattices and Microstructures 29, 188 (2001). [46] R. Landauer, Nature (London) 341, 567 (1989). [47] M. F. Krol, S. Ten, B. P. McGinnis, M. J. Hauduk, G. Khitrova, and N. Peyghambarian, Phys. Rev. B 52, R14344 (1995). [48] S. Ten, M. F. Krol, P. T. Guerreiro, and N. Peyghambarian, Appl. Phys. Lett. 69, 3387 (1996). [49] M. Morifuji and C. Hamaguchi, Phys. Rev. B 52, 14131 (1995). [50] M. U. Erdogan, K. W. Kim, and M. A. Stroscio, App. Phys. Lett. 62, 1423 (1993). [51] A. D. Sánchez and C. R. Proetto, J. Phys. Condens. Matter 7, 2059 (1995). 13 [52] K. V. Rosseau, K. L. Wang, and J. N. Schulman, Superlattices and Microstructures 6, 67 (1989). [53] E. E. Mendez, W. I. Wang, B. Rico, and L. Esaki, Appl. Phys. Lett 47, 415 (1985). [54] P. G. N. de Vegvar, T. A. Fulton, W. H. Mallison, and R. E. Miller, Phys. Rev. Lett. 73, 1416 (1994). [55] H. Pothier, S. Guéron, D. Esteve, and M. H. Devoret, Phys. Rev. Lett. 73, 2488 (1994). [56] L. Diago-Cisneros, H. Rodríguez-Coppola, R. Pérez-Alvarez, and P. Pereyra, Phys. Scripta 71, 1 (2005). [57] G. Nimtz, Proceedings of the Adriatico Research Conference on Tunneling and its implications, pp.223 (ICTP, Trieste, Italy, 1996). [58] A. M. Steinberg and R. Y. Chiao, Phys. Rev. A 49, 3283 (1994). [59] We have obtained the amplitude of the exponential fitting func- tion, for the phase transmission time with different sections of barrier thickness, to be constant starting from 55 Å. [60] Luis de la Peña, Introducción a la Mecánica Cuántica, Uni- versidad Nacional Autónoma de México, Fondo de Cultura Económica, 2006, Cap.6. [61] In additions of an exponential growing functions, adjustments via n-order polynomial (with n = 3, 4) and cycloid functions are suitable as well. [62] P. Pereyra and E. Castillo, Phys. Rev. B 65, 205120 (2002). [63] V. T. Petrashov, V. N. Antonov, P. Delsing and R. Claeson, Phys. Rev. Lett 70, 3 (1993). [64] U. Fano, Phys. Rev. 124, 1866 (1961). 14
1512.07801	1	1512	2015-12-24T12:08:13	Temperature regimes of formation of nanometer periodic structure of adsorbed atoms in GaAs semiconductors under the action of laser irradiation	[ "cond-mat.mes-hall" ]	The theory of nucleation of nanoscale structures of the adsorbed atoms (adatoms), which occurs as a result of the self-consistent interaction of adatoms with the surface acoustic wave and electronic subsystem is developed. Temperature regimes of formation of nanoclusters on ${n}$-GaAs surface under the action of laser irradiation are investigated. The offered model permits to choose optimal technological parameters (temperature, doping degree, intensity of laser irradiation) for the formation of the surface periodic defect-deformation structures under the action of laser irradiation.	cond-mat.mes-hall	cond-mat	Condensed Matter Physics, 2015, Vol. 18, No 4, 43801: 1 -- 8 DOI: 10.5488/CMP.18.43801 http://www.icmp.lviv.ua/journal Temperature regimes of formation of nanometer periodic structure of adsorbed atoms in GaAs semiconductors under the action of laser irradiation R.M. Peleshchak∗, O.V. Kuzyk, O.O. Dan'kiv Drohobych Ivan Franko State Pedagogical University, 24 Franko St., 82100 Drohobych, Ukraine Received October 4, 2015, in final form November 15, 2015 The theory of nucleation of nanoscale structures of the adsorbed atoms (adatoms), which occurs as a result of the self-consistent interaction of adatoms with the surface acoustic wave and electronic subsystem is developed. Temperature regimes of formation of nanoclusters on n-GaAs surface under the action of laser irradiation are investigated. The offered model permits to choose optimal technological parameters (temperature, doping degree, intensity of laser irradiation) for the formation of the surface periodic defect-deformation structures under the action of laser irradiation. Key words: nanocluster,temperature,diffusion,deformation PACS: 81.07.Bc,66.30.Lw 1. Introduction Recently, we have observed intensive researches into obtaining semiconductor structures with self- assembled nanoclusters by methods of molecular beam epitaxy [1, 2], ion implantation [3, 4] and under the action of laser irradiation [5 -- 7], as well as the possibility of controlling their physical properties. The non-linear diffusion-deformation theory of self-organization of nanoclusters of implanted impu- rities in semiconductors that considers an elastic interaction of the implanted impurities among them- selves and with atoms of a matrix was developed in [8]. The impurity which gets to a matrix, leads to changes in its volume and energy, and the initial fluctuation of deformation under certain conditions causes the emergence of deformation-induced flows of the implanted impurities. As a result, there are forces in the non-uniform deformation-concentration field proportional to gradients of concentration and deformation which further deform the matrix. These forces are caused by an increase of initial fluc- tuation and lead to self-organization of clusters of impurities. The information on nucleation (incipient state of the formation) of periodic nanostructures of ad- atoms and the implanted impurity is important for optimization of technological process and the pre- dicted controlling of physical parameters of semiconductor structures with nanoclusters [9 -- 13]. In par- ticular, to calculate the symmetry, the period and the formation time of surface structures it is sufficient to analyze only the initial (linear) stage of the development of the defect-deformation instability [14]. The theory of spontaneous nucleation of the surface nanometer lattice which is caused by instability in the system of the adatoms interacting with the self-consistent surface acoustic wave (SAW) was devel- oped in [14, 15]. Within this theory, the conditions of the formation of nanoclusters on the surface of solid states are established and the periods of a nanometer lattice as functions of concentration of adatoms and temperature are defined. However, the offered theory adequately describes the processes of the for- mation of nanoclusters only at low temperatures. It is based on the fact that this model does not consider the temperature dependence of concentration of the adatoms, as well as the interaction of electronic and defect subsystems that significantly depends on the temperature. ∗E-mail: [email protected] © R.M. Peleshchak, O.V. Kuzyk, O.O. Dan'kiv, 2015 43801-1 R.M. Peleshchak, O.V. Kuzyk, O.O. Dan'kiv Due to the high mobility of charge carriers, gallium arsenide is widely used in the production of quantum-dimensional structures and high-frequency lasers on their basis. A considerable attention of researchers has recently been paid to laser modification of the morphology of near-surface GaAs layers at which the conditions of the formation of nanoclusters are controlled [5, 6]. The purpose of this work is to study, on the initial states, the conditions of the formation of surface structures of adatoms and their periods in a semiconductor under the action of laser irradiation depending on temperature and doping degree. In this work, the temperature regimes and the influence of the degree of a doping of GaAs semiconductor on the formation of periodic nanometer structures of adatoms and their periods under the action of laser irradiation are investigated. 2. The model The equations for the displacement vectors ~u of an elastic medium is of the form [16]: ∂2~u ∂t 2 = c2 t ∆~u + (c2 l − c2 t )grad (div ~u) , (2.1) where cl and ct are longitudinal and transversal sound velocities, respectively. Let the surface of semiconductor coincide with the plane z = 0 (z-axis is directed into the crystal depth). Let us assume that the surface perturbation of an elastic medium arises along the x-axis. We rep- resent this perturbation in the form of a static SAW, which quickly fades into the depth of a semiconductor and has an amplitude growing with time [15]: ux = −iqReiqx+λt −kl z − iktQeiqx+λt −ktz , uz = klReiqx+λt −kl z + qQeiqx+λt −ktz , (2.2) (2.3) where k2 l,t plitudes. = q 2 + λ2/c2 l,t; λ is the increment of defect-deformation instability [15]; R and Q are SAW am- Then, deformation ε on the semiconductor surface (z = 0) is of the form ε = ∂ux ∂x + ∂uz ∂z = − λ2 c2 l Reiqx+λt . (2.4) Let us consider the alloyed n-GaAs semiconductor containing impurities (ionization donors, free elec- trons and adsorbed atoms of Ga which are also ionization donors). In this case, the electroneutrality condition can be presented as follows: n0(T ) = N + d (T ) + N + 0 (T ), (2.5) where Nd, n0 and N0 are the surface concentration of ionized donors, the spatially homogeneous values of the surface concentration of electrons and adatoms, respectively. Concentration of adatoms can be presented as the sum of concentration N + 0r (T ) of equilibrium defects of the defects generated under the action of laser irradiation that depends on its and concentration N + 0l power: N + 0 (T ) = N + 0r (T ) + N + 0l , (2.6) where N + constant; T is temperature; N ∞ 0r (T ) = N ∞ 0r exp (−∆U /kBT ); ∆U is the formation energy of an adatom of Ga; kB is Boltzmann 0r is the surface concentration of atoms of Ga. The direction of the surface wave is determined by anisotropic properties of crystals. In particular, in [17] it is shown that the surface diffusion coefficient of Ga on GaAs (001) in the [0¯11] direction is four times larger than the surface diffusion coefficient in the [011] direction. Periodic surface deformation leads to spatial nonuniform redistribution of adatoms N (x), the mod- ulation of the bottom of the conduction band and, respectively, to redistribution of the concentration of 43801-2 Temperature regimes of formation of nanometer periodic structure of adsorbed atoms in GaAs conduction electrons n(x) and the electrostatic potential ϕ(x): N (x) = N0 + N1(x) = N0 + N1(q)eiqx+λt , n(x) = n0 + n1(x) = n0 + n1(q)eiqx+λt , ϕ(x) = ϕ(q)eiqx+λt , (2.7) (2.8) (2.9) where N1(q), n1(q), and ϕ(q) are the amplitudes of the corresponding periodic perturbations. Then, taking into account (2.7), (2.8) and (2.9), the Poisson's equation can be presented as follows: − q 2ϕ(q) = e ε0 εa£n1(q) − N1(q)¤ , (2.10) where ε0 and ε are dielectric constant and dielectric permittivity of the medium, respectively; a is the lattice constant. The equations for concentration of the charged adsorbed atoms are of the form ∂N ∂t = D ∂2N ∂x2 + ∂ ∂xµµN ∂ϕ ∂x¶ − D θ kT ∂ ∂x ·N ∂ ∂xµε + l 2 d ∂2ε ∂x2¶¸ , (2.11) where D and µ are the surface diffusion coefficient and the mobility of adatoms which are related among themselves by Einstein's relation µ = De/(kBT ); θ is the deformation potential; ld is the characteristic length of interaction of adatoms with lattice atoms. The second term considers the interaction of adatoms with an electric field arising due to the spatial nonuniform redistribution of electric charge. The third term expresses the interaction of adatoms with the deformation field, taking into account the nonlocal interaction [15]. Taking into account (2.4), (2.7) -- (2.10) and in an approximation N1 << N0, the equation (2.11) can be written as follows: λN1(q) = −D q 2N1(q) + D N0Φ kT £n1(q) − N1(q)¤ − where Φ = e2/(ε0 εa). The density of the electron current j = nµn ∂χ ∂x , D N0θ kBT " λ2 c2 l Rq 2(1 − q 2l 2 d)# , (2.12) (2.13) where µn (T, n) is the mobility of electrons which depends on the temperature and the doping degree of semiconductor [18]; the electrochemical potential χ is defined by the relation χ(x) = kBT ln n(x) Ni − eϕ(x) + acε(x), (2.14) where Ni is the effective density of states; Ni = 2¡2πmkBT /h2¢3/2 ; ac is the constant of hydrostatic de- formation potential of the conduction band. Then, taking into account (2.13) and (2.14), the continuity equation can be presented as follows: e ∂n ∂t = kBT µn ∂ ∂x µn ∂ ∂x ln n Ni¶ − eµn ∂ ∂x µn ∂ϕ ∂x¶ + acµn ∂ ∂xµn ∂ε ∂x¶ . Taking into account (2.4) and (2.7) -- (2.10), the equation (2.15) can be written as follows: n1(q)¡eλ + kBT µn q 2 + n0µn Φ¢ = N1(q)n0µn Φ + acn0µn q 2 λ2 c2 l R . (2.15) (2.16) Solving the system of equations (2.12) and (2.16), we obtain expressions for the amplitudes of the surface concentration of adatoms N1(q) and conduction electrons n1(q). 43801-3 The spatial nonuniform distribution of adatoms modulates the surface energy F (x), which leads to the appearance of lateral mechanical tension σxz = ∂F¡N (x)¢/∂x, which is compensated by shift tension in the medium [15]. The boundary condition expressing the balance of lateral tension is as follows: R.M. Peleshchak, O.V. Kuzyk, O.O. Dan'kiv E 1 + νµ ∂ux ∂z + ∂uz ∂x ¶z=0 = ∂F¡N (x)¢ ∂x = ∂F ∂N1(x) ∂N ∂x , (2.17) where E and ν are Young's modulus and Poisson's ratio, respectively. Besides, the interaction of adatoms with atoms of a semiconductor results in the emergence of the normal mechanical tension on the surface, and the corresponding boundary condition is of the form: E 1 + νµ ∂uz ∂z + ν 1 − 2ν ∂ux ∂x ¶z=0 = θ a N1(x). (2.18) Thus, the system of homogeneous linear equations (2.17) and (2.18) for amplitudes R and Q is ob- tained and the dispersion dependencies λ(q) can be obtained from the condition of non-triviality of so- lutions (from the condition of equality to zero of the determinant of this system). 3. Calculation results and their discussion The calculations of λ(q) were carried out for the GaAs semiconductor doped by silicon at the fol- lowing values of parameters: ld = 2.9 nm; a = 0.565 nm; c l = 3500 m/s; ν = 0.318; E = 0.85 Mbar; ac = = 3 × 1014 cm−2; ∆U = 0.1 eV; −7.17 eV; θ = 10 eV; ε = 12; D = 5 × 107 exp [−5.6/(kBT )] cm2/s [19]; N ∞ 0r N + = N0Si exp[−0.058/(kBT )], where N0Si is concentration of Si impurities. Mobility of electrons as func- d tion of temperature and concentration of donors was determined by the technique given in work [18]. = 5 × 1012 cm−2 (figure 1) and N + 0l The results of calculation of the dependence of the increment of defect-deformation instability on the module of the wave vector are given in figures 1 and 2 at various values of the concentration of silicon impurities N0Si and various values of temperature at average concentration of the Ga adatoms = 2 × 1012 cm−2 (figure 2), which are generated under the action of N + 0l laser radiation and are defined by its power. Such a dependence has a maximum which is shifted towards great values of the module of the wave vector with an increasing concentration of donors. The value of qmax at which the increment of defect-deformation instability λ has a maximum, defines the period of the dominating structure d = 2π/qmax (figure 3). The negative values of the increment of defect-deformation instability λ mean the attenuation of SAW and, respectively, the impossibility of the formation of peri- odic defect-deformation structures. Except the structure dominating in the linear approximation with an increment λmax(qmax), the whole continuum of structures with λ > 0 intensifies (figure 1 and figure 2). At an exit to the stationary regime, the increase in amplitude of the dominant structure (2.4), (2.7) -- (2.9) will reach saturation due to elastic nonlinearity, but amplitudes with a smaller increment will continue to increase [20]. Thus, the spectrum of the generated modes can extend. However, in [20] it is shown that in solids with defects which are the centers of tension (θ > 0), only single-mode generation with qmax is most often realized. As seen from figure 1, at concentration of the adatoms N + 0l = 5 × 1012 cm−2, the formation of periodic structures is possible in a wide temperature range. In particular, at low temperatures T < 180 K [fig- ure 1 (a), (b) and figure 3 (a)] and high temperatures T > 460 K [figure 1 (e), (f) and figure 3 (a)] at this intensity of laser irradiation, the formation of the surface periodic structures occurs at any degree of a doping of GaAs semiconductor matrix. At a significant doping of GaAs semiconductor by donor impurities of silicon (N0Si > 1013 cm−2), the formation of periodic defect structures is possible at any temperature [figure 1 and figure 3 (a)]. At a decrease of intensity of laser irradiation, the concentration of Ga adatoms decreases. In this case, the temperature ranges in which the formation of the surface superlattice is possible are narrowed. In = 2 × 1012 cm−2, the formation of periodic structures in the particular, at concentration of adatoms N + 0l temperature range 150 K < T < 420 K is impossible at any degree of a doping of GaAs semiconductor [figure 2 and figure 3 (b)], while in undoped GaAs (curves 1), the formation of self-assembled defect structures can only occur at temperature T < 80 K [figure 2 (a) and figure 3 (b)] and T > 520 K [figure 2 (f) and figure 3 (b)]. 43801-4 Temperature regimes of formation of nanometer periodic structure of adsorbed atoms in GaAs Figure 1. The dependence of the increment of defect-deformation instability on the wave vector at con- = 5×1012 cm−2 at various values of the concentration of donors: centration of the adsorbed atoms of N + 0l 1 -- N0Si = 108 cm−2; 2 -- N0Si = 5 × 1012 cm−2; 3 -- N0Si = 2 × 1013 cm−2; 4 -- N0Si = 1014 cm−2; (a) T = 70 K; (b) T = 150 K; (c) T = 250 K; (d) T = 300 K; (e) T = 500 K; (f) T = 800 K. At an increase of concentration of donors, the acoustoelectronic effects lead to an increase of the value of the increment of defect-deformation instability. Thus, in a semiconductor with a larger degree of doping by donor impurities, the processes of the formation of nanometer periodic structures should occur quicker. The dependence of the period of the surface superlattice of GaAs adatoms on temperature is presented in figure 3. Such a dependence is of a nonmonotonic character. At high intensity of laser irradiation (N + 0l = 5 × 1012 cm−2) and at high concentration of donors [fig- ure 3 (a), curves 3 and 4], the greatest value of the period is observed at temperature T = (320 − 380) K. At T → 0 or T → ∞, the period of defect-deformation structure decreases and approaches the value 2πld. At a decrease of the degree of a doping (figure 3, curves 1 and 2), there are two critical temperatures Tc,min and Tc,max, for which only at T < Tc,min and T > Tc,max, the formation of the surface superlattice is possible. At T → Tc,min and T → Tc,max, the superlattice period increases monotonously and strives to infinity. At a decrease of intensity of laser irradiation (at a decrease of concentration of the adsorbed atoms), the temperature Tc,min decreases, and the temperature Tc,max increases [figure 3 (b)]. Our results are qualitatively consistent with the experimental results presented in works [21, 22]. In particular, in [21] it is shown that the formation of nanoclusters of gallium is observed on the laser illumi- nated part of the end surface of the quartz light guide at molecular beam epitaxy of Ga at a temperature of 100 K. At T = 300 K, the effect of self-organization of nanoclusters is not observed. In work [22] it is shown that the formation of nanoclusters at room temperature on the surface of strongly alloyed semi- 43801-5 R.M. Peleshchak, O.V. Kuzyk, O.O. Dan'kiv Figure 2. The dependence of the increment of defect-deformation instability on the wave vector at con- = 2×1012 cm−2 at various values of the concentration of donors: centration of the adsorbed atoms of N + 0l 1 -- N0Si = 108 cm−2; 2 -- N0Si = 5 × 1012 cm−2; 3 -- N0Si = 2 × 1013 cm−2; 4 -- N0Si = 1014 cm−2; (a) T = 70 K; (b) T = 150 K; (c) T = 250 K; (d) T = 300 K; (e) T = 500 K; (f) T = 800 K. conductor substrate of GaAs (∼ 1018 cm−3) is possible if the intensity of the laser radiation exceeds the defined critical value. It should be noted that at a considerable concentration of adatoms, the non-linear effects that can Figure 3. The dependence of the period of the surface defect-deformation structure on temperature at various values of the concentration of donors: 1 -- N0Si = 108 cm−2; 2 -- N0Si = 5 × 1012 cm−2; 3 -- N0Si = 2 × 1013 cm−2; 4 -- N0Si = 1014 cm−2; (a) N + 0l = 5 × 1012 cm−2; (b) N + 0l = 2 × 1012 cm−2. 43801-6 Temperature regimes of formation of nanometer periodic structure of adsorbed atoms in GaAs slightly expand the temperature range at which the formation of nanoclusters is possible [5, 23], are essential. 4. Conclusions 1. The theory of nucleation of nanoclusters on n-GaAs surface under the action of laser irradiation which considers the doping degree of the semiconductor having donor impurities is developed. 2. Within this theory, the temperature ranges of the formation of periodic defect structures on n- GaAs surface depending on the concentration of donors and on the intensity of laser radiation (concentration of the adsorbed atoms) are established. 3. It is shown that an increase of the degree of doping of a semiconductor expands temperature ranges within which the formation of periodic defect structures is possible. 4. The regularities of the effect of temperature at various values of the concentration of donors on the period of the surface defect-deformation structures of the adatoms in GaAs semiconductor are determined. References 1. Zinovyev V.A., Dvurechenskii A.V., Kuchinskaya P.A., Armbrister V.A., Teys S.A., Shklyaev A.A., Mudryi A.V., Semi- conductors, 1998, 49, 149; doi:10.1134/S1063782615020256. 2. Ipatova I.P., Malyshkin V.G., Maradudin A.A., Phys. Rev. B, 1998, 57, 12968; doi:10.1103/PhysRevB.57.12968. 3. Yamada I., Matsuo J., Tojoda N., Nucl. Inst. Meth. Phys. Res. B, 2003, 206, 820; doi:10.1016/S0168-583X(03)00857-7. 4. Boryakov A.V., Nikolitchev D.E., Tetelbaum D.I., Belov A.I., Ershov A.V., Mikhaylov A.N., Phys. Solid State, 2012, 54, 394; doi:10.1134/S1063783412020102. 5. Peleshchak R.M., Kuzyk O.V., Ukr. J. Phys., 2009, 54, 702. 6. Vlasenko A.I., Baidullaeva A., Veleschuk V.P., Mozol P.E., Boiko N.I., Litvin O.S., Semiconductors, 2015, 49, 229; doi:10.1134/S1063782615020220. 7. Veleschuk V.P., Baidullaeva A., Vlasenko A.I., Phys. Solid State, 2010, 52, 469; doi:10.1134/S1063783410030054. 8. Peleshchak R.M., Kuzyk O.V., Dan'kiv O.O., J. Phys. Stud., 2013, 17, 2601. 9. Slezov V., In: Kinetics of First-order Phase Transitions, Wiley-VCH, Weinheim, 2009; Chapter 2, 7 -- 38; doi:10.1002/9783527627769.ch2. 10. Sugakov V., J. Phys.: Conf. Ser., 2009, 148, 012048; doi:10.1088/1742-6596/148/1/012048. 11. Aaronson H., Enomoto M., Lee J.K., Mechanisms of Diffusional Phase Transformations in Metals and Alloys, CRC Press, Boca Raton, 2010. 12. L'vov P.E., Svetukhin V.V., Phys. Solid State, 2015, 57, 1213; doi:10.1134/S1063783415060232. 13. Dubrovskii V.G., Zhang X., Tech. Phys. Let., 2011, 37, 797; doi:10.1134/S1063785011090082. 14. Emel'yanov V.I., Baidullaeva A., Vlasenko A.I., Mozol' P.E., Quantum Electron., 2008, 38, 245; doi:10.1070/QE2008v038n03ABEH013644. 15. Emel'yanov V.I., Laser Phys., 2008, 18, 1435; doi:10.1134/S1054660X08120104. 16. Landau L.D., Lifshitz E.M., Theory of Elasticity, Pergamon Press, London, 1970. 17. Ohta K., Kojima T., Nakagawa T., J. Cryst. Growth, 1989, 95, 71; doi:10.1016/0022-0248(89)90354-0. 18. Mnatsakanov T.T., Levinshtein M.E., Pomortseva L.I., Yurkov S.N., Semiconductors, 2004, 38, 56; doi:10.1134/1.1641133. 19. Boltaks B.I., Diffusion and Point Defects in Semiconductors, Nauka, Leningrad, 1972 (in Russian). 20. Emel'yanov V.I., Quantum Electron., 1999, 29, 561; doi:10.1070/QE1999v029n07ABEH001533. 21. MacDonald K.F., Fedotov V.A., Pochon S., Ross K.J., Stevens G.C., Zheludev N.I., Brocklesby W.S., Emel'yanov V.I., Appl. Phys. Lett., 2002, 80, 1643; doi:10.1063/1.1456260. 22. Vintsents S.V., Zaitseva A.V., Zaitsev V.B., Plotnikov G.S., Semiconductors, 2004, 38, 245; doi:10.1134/1.1682319. 23. Peleshchak R.M., Kuzyk O.V., Dan'kiv O.O., Condens. Matter Phys., 2014, 17, 23601; doi:10.5488/CMP.17.23601. 43801-7 R.M. Peleshchak, O.V. Kuzyk, O.O. Dan'kiv Температурнi режими формування нанометрової перiодичної структури адсорбованих атомiв у напiвпровiднику GaAs пiд дiєю лазерного опромiнення Р.М. Пелещак, О.В. Кузик, О.О. Данькiв Дрогобицький державний педагогiчний унiверситет iменi Iвана Франка, вул. Iвана Франка, 24, 82100 Дрогобич, Україна Розвинуто теорiю нуклеацiї нанорозмiрних структур адсорбованих атомiв (адатомiв), яка вiдбувається в результатi самоузгодженої взаємодiї адатомiв з поверхневою акустичною хвилею та електронною пiд- системою. Дослiджено температурнi режими формування нанокластерiв на поверхнi n-GaAs пiд дiєю лазерного опромiнення. Запропонована модель дозволяє вибрати оптимальнi технологiчнi параметри (температуру, ступiнь легування, iнтенсивнiсть лазерного опромiнення) для формування поверхневих перiодичних дефектно-деформацiйних структур пiд дiєю лазерного опромiнення. Ключовi слова: нанокластер,температура,дифузiя,деформацiя 43801-8
1602.00026	2	1602	2017-02-13T19:17:49	Giant edge spin accumulation in a symmetric quantum well with two subbands	[ "cond-mat.mes-hall" ]	We have studied the edge spin accumulation in a high mobility two-dimensional electron gas formed in a symmetric well with two subbands. This study is strongly motivated by the recent experiment of Hernandez et al. [Phys. Rev. B {\bf 88}, 161305(R) (2013)] who demonstrated the spin accumulation near the edges of a bilayer symmetric GaAs structure in contrast to no effect in a single-layer configuration. The intrinsic mechanism of the spin-orbit interaction we consider arises from the coupling between two subband states of opposite parities. We obtain a parametrically large magnitude of the edge spin density for the two-subband sample as compared to the usual single-subband structure. We show that the presence of a gap in the system, i.e., the energy separation $\Delta$ between the two subband bottoms, changes drastically the picture of the edge spin accumulation. Thus one can easily proceed from the regime of weak spin accumulation to the regime of strong one by varying the Fermi energy (electron density) and/or $\Delta$. We estimate that by changing the gap $\Delta$ from zero up to $1\div 2$ K, the magnitude of the effect changes by three orders of magnitude. This opens up the possibility for the design of new spintronic devices.	cond-mat.mes-hall	cond-mat	a Giant edge spin accumulation in a symmetric quantum well with two subbands Alexander Khaetskii1 and J. Carlos Egues2 1Department of Physics, University at Buffalo, SUNY, Buffalo, NY 14260-1500 2Instituto de F´ısica de Sao Carlos, Universidade de Sao Paulo, 13560-970, Sao Carlos, Sao Paulo, Brazil (Dated: June 14, 2018) We have studied the edge spin accumulation due to an electric current in a high mobility two- dimensional electron gas formed in a symmetric well with two subbands. This study is strongly motivated by the recent experiment of Hernandez et al. [Phys. Rev. B 88, 161305(R) (2013)] who demonstrated the spin accumulation near the edges of a symmetric bilayer GaAs structure in contrast to no effect in a single-layer configuration. The intrinsic mechanism of the spin-orbit (SO) interaction we consider arises from the coupling between two subband states of opposite parities. We obtain a parametrically large magnitude of the edge spin density for a two-subband well as compared to the usual single-subband structure. We show that the presence of a gap in the system, i.e., the energy separation ∆ between the two subband bottoms, changes drastically the picture of the edge spin accumulation. The gap value governs the effective strength of the inter-subband SO interaction which provides a controllable crossover from the regime of weak spin accumulation to the regime of strong one by varying the Fermi energy (electron density) and/or ∆. We estimate that by changing the gap ∆ from zero up to 1 ÷ 2 K, the magnitude of the effect changes by three orders of magnitude. This opens up the possibility for the design of new spintronic devices. PACS numbers: 72.25.-b, 73.23.-b, 73.50.Bk Spin current and spin accumulation [1, 2] which appear due to the spin-orbit (SO) coupling in the presence of electric currents are topics of great current interest which are important for the future of spin electronics [3]. There are two distinct SO mechanisms, the extrinsic one due to the Mott asymmetry in the electron scattering off impu- rities [4 -- 7], and the intrinsic one [8, 9] due to SO induced splitting of the electron spectrum. The edge spin-density accumulation, related to either the Mott asymmetry by impurities [10] (2D electrons) or the intrinsic mechanism (2D holes) [11, 12], has been experimentally observed. It is known [2, 13 -- 15] that in the diffusive regime (and when the spin diffusion length is much larger than the mean free path) the edge spin density is entirely due to the spin flux coming from the bulk. In contrast, the physics of the edge spin-density accumulation for the in- trinsic mechanism in the opposite case of strong SO split- ting [16] only recently has been understood [17 -- 21]. This includes the experimentally important case of a diffusive sample with a large SO splitting of the spectrum so that the spin-precession length is smaller than the mean free path. This we term the quasi-ballistic regime [20,21]. In particular, for 2D holes in this regime the edge spin- density, which is due to the spin current from the bulk, is parametrically smaller than the density generated upon the boundary scattering [21]. Recently, using Kerr rotation spectroscopy, Hernan- dez et al. [22] demonstrated electric-current induced spin accumulation near the edges of a high-mobility two- dimensional electron gas in a symmetric bilayer GaAs structure in contrast to no accumulation in a single-layer configuration [23]. This result is interesting and intrigu- ing in many aspects. The observed effect is quite large despite the fact that the electric field in the high-mobility a) ! 2p 2ϕ b) ( pε ) ! 1p 1ϕ !" ε o ε e ε F ∆ p2 p1 Figure 1. Schematics of the boundary specular scattering in the presence of SO coupling. Plus and minus modes are shown for the same energy and the same wave vectors along the boundary. channel is 300÷ 400 times smaller than that in the exper- iment by Kato et al. [10], where for a GaAs sample the result was explained by the extrinsic interaction with im- purities. Note that the structure studied in [22] has inver- sion symmetry and therefore the usual Rashba term [24] is absent. On the other hand, the linear-in-momentum term [25] originating from a cubic Dresselhaus term is known not to lead to a spin current in the bulk. A signif- icant difference between the observed edge spin density in the two-subband vs. the one-subband cases has moti- vated us to look for the explanation of this phenomenon using the inter-subband Rashba-like Hamiltonian arising in two-subband wells [26],[27]. Here we follow the method proposed in [21] to calcu- late the edge spin density which appears due to bound- ary scattering [Fig. 1(a)] in the quasi-ballistic regime for a Rashba-like Hamiltonian [26],[27] describing the two- subband well [Fig. 1(b)]. In this quasi-ballistic regime the characteristic length of the spin accumulation near the boundary is smaller than the mean free path. Since the latter is around 30 µm [22], it indeed exceeds all the characteristic lengths of our theory. We have explained the experimental results, in particular, the large magni- tude of the edge spin density for the two-subband sample compared to the usual single-band structure with either the Rashba or Dresselhaus interactions. Two bands vs one band case. Interestingly, we have found that despite the problem in question resembling very much the usual Rashba problem (there are two copies of them because each state is doubly degenerate), the presence of the gap ∆ between two sub-band edges [Fig. 1(b)] changes the physics of the edge spin accumu- lation completely. This happens because the gap magni- tude governs the effective strength of the inter-subband SO interaction leading to different solutions compared to the one-band Rashba case for the occupation numbers of the incoming states participating in the boundary scat- tering, [Fig. 1(a)]. The physics is now determined by the value of the parameter ξ = 2ηkF /∆ ≡ L∆/Ls [Fig. 2a], where L∆ = vF /∆, Ls = 2/2mη are the coherence and the spin-precession lengths. Here pF = kF = mvF is the Fermi momentum, η is the inter-subband SO coupling constant [26, 27]. The parameter ξ can have an arbitrary value even for a small η since the gap ∆ can be made much smaller than the Fermi energy. First of all, the presence of the gap changes the "he- licity" direction which corresponds to the eigenvectors of the Hamiltonian (2), see Eqs. (S2), (S3) in the Sup- plemental Material (SM). For the two sub-band problem the helicity axis for a given momentum p is defined by the angle θ(p) with the z-axis (the normal to the 2D plane), cos θ = 1/p1 + (2ηp/∆)2. When ∆ is much bigger than the SO energy 2ηp/, then the helicity axis coincides with the z-axis. In the opposite case when ∆ → 0 and the SO energy dominates, the helicity axis lies within the x − y plane, as it should be for usual Rashba model (which corresponds to θ → π/2). Therefore, sin θ = (2ηp/)/p(2ηp/)2 + ∆2 determines the effec- tive strength of SO coupling, and the usual Rashba single- band model (in our case we have two copies of them) corresponds to a limit of strong SO coupling (sin θ = 1). As it has been shown in [20, 21] [see also Eq. (7) be- low], because of the unitarity of the boundary scattering the magnitude of the edge spin accumulation is propor- tional to the difference f1(εF , ky) − f2(εF , ky) between the distribution functions of the incoming electron states belonging to the sub-bands 1 and 2 for a given Fermi en- ergy and given wave vector ky along the boundary, see Fig. 1. ( In the case of one-band Rashba model the incom- 2 ing states belong to the branches of opposite helicities). These distribution functions are found from the solution of the kinetic equation for the spin-density matrix in the bulk (2D) of the sample in the presence of impurity scat- tering and an electric field [28], [29] and used as the input parameters for the boundary scattering problem. The important point is that in the one-band Rashba case the difference of the distribution functions in ques- tion is of the third order with respect to p1 − p2 = /Ls for any reasonably short-ranged impurity potential in the bulk (when the correlation radius of the impurity poten- tial d is much smaller than Ls). This has been rigorously proven in Ref. [21] (see Eqs. (14-16) in [21]). Thus, in the leading order (i.e. ∝ 1/Ls) the above mentioned dis- tribution functions take the form which they have in the absence of any spin-orbit coupling. The first order effect appears only for very smooth impurity potential when d > Ls, and the magnitude of the edge spin density for the one-band Rashba case is given by the expressions [21] hSzi ≃ kE Ls d2 L2 s at d ≪ Ls; hSzi ≃ kE Ls at d > Ls. (1) Here kE = eEτtr/, e is the modulus of the electron charge, E is the magnitude of the in-plane (driving) elec- tric field directed along the y-axis and τtr is the transport scattering time due to the impurities in the bulk of a sam- ple. Note that we consider in our paper the case d ≪ Ls, which is only the realistic one. Note that the cancellation of the effect for the Rashba model in the leading order (i.e. ∝ 1/Ls) happens when the transition rates between branches of opposite helici- ties have the same strength as the transition rates within the branch of the same helicity (one of the necessary conditions). That is why for smooth impurity potential (d > Ls) which cannot support the transitions between branches of opposite helicities, one observes the recover- ing of the first order effect, see Eq. (1). From the above considerations we can immediately un- derstand the role of the gap ∆ in the two-subband model considered in this work. As explained above, the effec- tive SO interaction decreases with increasing ∆. This causes suppression of the inter-subband transition rates, since these transitions are accompanied by spin flip, as compared to the intra-subband ones, which do not need spin flip. The suppression factor is sin2 θ, which is the probability of spin-flip (see also SM, Sec. III). Since the inter-band and intra-band rates are different now, this prevents the complete cancellation that occurs for the quantity f1(εF , ky) − f2(εF , ky) in the one-band Rashba model, and leads to the recovery of the first order effect with respect to small splitting p1 − p2 ≪ pF even for a short-ranged (d ≪ Ls) impurity potential. As we discuss later on, for a GaAs structure similar to that used in Ref. [22], it is enough to change the gap ∆ from zero up to about 1÷2 K in order to increase the mag- nitude of the effect by the three orders ("giant effect"). The most pronounced change happens at L∆ ≃ 0.5Ls, where the edge spin density Sz(x) is maximized and its order of magnitude is given by kE/Ls, which is paramet- rically larger than in the single-band case [Fig. 2a]. We consider specular scattering [i.e., a straight bound- ary for simplicity, Fig. 1(a)] and a Fermi energy much larger than the gap ∆ between the subbands, i.e., εF ≫ ∆ [30]. Moreover, the SO interaction is weak (ηkF ≪ εF ) and therefore "coherent" and spin-precession length scales are large compared to the Fermi wave length, L∆, Ls ≫ λF = 2π/kF . The ratio L∆/Ls can be arbitrary. Our calculation shows that the charac- teristic spatial scale of the edge spin density is Λ = ∆ + L2 s. L∆Ls/pL2 Model Hamiltonian. The Hamiltonian of a symmet- ric quantum well with two subbands and inter-subband- induced SO interaction resembles that of the ordinary Rashba model. In contrast to the latter, the intersub- band SO interaction is nonzero even in symmetric struc- tures with the 4 × 4 Hamiltonian is [26], [27] 3 potential, the external gate and doping potentials, and the structural quantum-well potential profile. Note that the gap is ∆ = εo − εe = 2ε−. Theoretical approach. To calculate the edge spin den- sity in the quasi-ballistic regime we follow the method de- veloped in Refs. [20, 21] for the case of the single-subband Rashba Hamiltonian. Assuming that the spatial scale of the edge spin accumulation Λ is much smaller than the mean free path l, we solve the edge spin problem by the method of scattering states, i.e., we find the exact quantum mechanical solution of the electron scattering by an impenetrable straight boundary [Fig. 1(a)] at a given Fermi energy. These solutions are then used in the calculation of the (mean) spin density profile. The popu- lations of the incoming states are found from the solution of the kinetic equation for the spin-density matrix in the bulk (2D) of the sample in the presence of electric field, see SM. The Hamiltonian (2) has 4 eigensolutions Ψi,s, Ψ1,↑, Ψ1,↓, Ψ2,↑, Ψ2,↓ (see SM for their explicit form) with the corresponding energy spectrum H = ( p2 2m + ε+)1 ⊗ 1 − ε−τz ⊗ 1 + ( η )τx ⊗ (pxσy − pyσx).(2) Here ⊗ means a direct tensor product, m is the effective mass, ε± = (εo ± εe)/2, εe and εo are quantized energies of the lowest (even) and first excited (odd) subbands, respectively, measured from the bottom of the quantum well, τx,y,z denote the Pauli matrices describing the sub- band (or pseudospin) degree of freedom, and σx,y,z are Pauli matrices referring to the electron spin. The inter- subband SO coupling η (which has the dimensionality of square of charge) is expressed [26] in terms of the gra- dients of the Hartree-type contribution to the electron p2 2m ε1,2(p) = + ε+ ∓qε2 − + η2p2/2, (3) where the subscript i = 1, 2 corresponds to the lower (higher) in energy sub-band. Each sub-band is dou- bly degenerate with respect to the "spin direction" Upon scattering by the s =↑,↓ (Kramers pairs). straight boundary where energy and momentum py along the boundary are conserved, the states in the pair Ψ1,↑(ϕ1, θ1), Ψ2,↓(ϕ2, θ2) mix up and form two scattering states, Eqs. (4),(5) [similarly for the pair Ψ1,↓(ϕ1, θ1), Ψ2,↑(ϕ2, θ2)]. For this pair of scattering states, we have the following boundary condition for the scattering by a hard wall located at x = 0 [Fig. 1(a)] Ψ1,↑(x, y)x=0 = eiky y[Ψ1,↑(π − ϕ1, θ1)e−ik1x + F 1,↑ Ψ2,↓(x, y)x=0 = eiky y[Ψ2,↓(π − ϕ2, θ2)e−ik2x + F 1,↑ 1,↑ Ψ1,↑(ϕ1, θ1)eik1x + F 2,↓ 2,↓ Ψ1,↑(ϕ1, θ1)eik1x + F 2,↓ 1,↑ Ψ2,↓(ϕ2, θ2)eik2x]x=0 = 0, 2,↓ Ψ2,↓(ϕ2, θ2)eik2x]x=0 = 0, (4) (5) y+k2 1), p2 y +k2 2 = 2(k2 1 = 2(k2 with p2 2), ε1(p1) = ε2(p2) = ε. The momenta p1, p2 describe states belonging to subbands 1 and 2 for a given energy ε, see Fig. 1(b). The angles ϕ1, ϕ2 (between the corresponding momenta and the positive direction of the x-axis) are expressed as sin(ϕ1) = ky/p1 and sin(ϕ2) = ky/p2. The angles θ1, θ2 are defined via cos θ1,2 = 1/p1 + (2ηp1,2/∆)2. The expressions for the scattering amplitudes ( F 1,↑ 1,↑ , etc.) and the corresponding components of the unitary scat- tering matrix S are presented in SM. Similar equations can be written for the pair Ψ1,↓(ϕ1), Ψ2,↑(ϕ2), and the corresponding scattering matrix elements are also deter- mined. The expectation value of the z component of the spin as a function of coordinates is given by the following ex- pression: hSz(x)i =Xi,s Z dky (2π)2 dε vx,i fi(ε, ky) ×h Ψi,s(x) Sz Ψi,s(x)i (6) Here fi(ε, ky) is the distribution function of the electron state in the sub-band i for a given energy and given wave vector ky along the boundary and the group velocity is vx,i = ∂εi/∂px. We can then calculate the most important part of the edge spin density which is smooth on the scale of the Fermi wave length [31] and involves the interference of the outgoing waves [two last terms in Eqs. (4) and (5)]. The corresponding formula for hSz(x)i valid for general values of the parameters (in the case when both subbands are occupied) is presented in the SM. In the most important case p1 − p2 ≪ pF , when the energy separation between two sub-bands p∆2 + 4η2k2 F is much smaller than the Fermi energy, the edge spin density for arbitrary values of the parameter ξ = L∆/Ls reads x hSz(x)i = − sin2 θZ dkyky (2π)2 dε εF Λp1 − (ky/kF )2! sin ×[f1(ε, ky) − f2(ε, ky)].(7) Here εF = p2 F /2m is the Fermi energy. While deriving Eq. (7), we used that θ1 − θ2 ≪ θ1,2, and ϕ2 − ϕ1 ≪ ϕ1,2. The difference of the distribution functions enter- ing Eq. (7) is calculated in the SM assuming the set of inequalities k−1 F ≪ d ≪ Ls, where d is the correlation radius of the impurity potential in the bulk of the struc- ture. The first condition means that the scattering in the bulk is of the small-angle type. Both conditions are fulfilled for a high mobility GaAs structure. The final result derived from Eq. (7) reads hSz(x)i = 3kE Ls Φ(ξ)J(x/Λ); Φ(ξ) = ξ . (2ξ2 + 1)pξ2 + 1 (8) with the spatial dependence given by the integral J( x Λ ) =Z 1 0 dzz2 π2 sin( x Λ√1 − z2 ), Λ = L∆Ls ∆ + L2 s pL2 We recall that Λ is the characteristic spatial scale of the edge spin density. For x ≪ Λ we have J(x) ∝ x/Λ. In the opposite limit x ≫ Λ, we obtain J(x) ∝ (Λ/x)3/2 cos[(x/Λ) + π/4]. Weak SO coupling: Ls ≫ L∆. To contrast our results with the usual one-band Rashba case it is instructive to consider here the limit of a weak SO coupling 2ηkF ≪ ∆. In this limit we can calculate the difference of the distri- bution functions entering Eq. (7) using their standard expressions at η = 0 (see also SM), i.e., f1,2 = (eEky/m)τtr(p1,2)∂f0/∂ε, (10) where f0 is the Fermi function, the electric field E is directed along the y-axis, and τtr(p) is the momentum- dependent transport scattering time calculated within the Born approximation due to impurity scattering in the bulk. The values of p1, p2 are related through ε1(p1) = ε2(p2) = ε = εF , Fig. 1(b). Using the condi- tion kF d ≫ 1 (small-angle scattering in bulk), we obtain (τtr(p1)−τtr(p2))/τtr ≈ 3(p1−p2)/pF ≈ (3/(kF L∆). [32] We note that compared to the usual Rashba one-band case the difference of the distribution functions consid- ered here is finite at η = 0, and is of the first order in 3Φ(ξ) (a) S z (x / L s ) (b) 4 L ∆ = 0.1L s x / L s L L s = 1.1 µm s = 2.8 µm ξ = L ∆ / L s S z (x / L s ) (c) S z (x / L s ) (d) L ∆ = 0.56L s L ∆ = 2L s x / L s x / L s Figure 2. Φ(ξ) vs. ξ a) and the edge spin density hSz(x)i in units of 106cm−2 for distinct ratios L∆/Ls and two different values of Ls b)-d), as a function of x/Ls. Note that Φ(ξ) has a maximum at ξ ∼ 0.56 = L∆/Ls. The amplitude of the oscillations is reduced as Ls increase (cf. blue and red curves in b)-d).) . (9) p1 − p2 = /L∆. Since the SO coupling is weak, the probability of the spin flip is small which shows up as the small factor sin2 θ ≈ L2 s ≪ 1 in Eq. (7), and finally we obtain ∆/L2 hSz(x)i = 3kE L∆ L2 s J(x/Λ), (11) which coincides with the result which follows from Eq.(8) in the limit ξ → 0. The calculated edge spin density Eq. (8) is maximal at L∆ ≈ Ls when it is of the order of kE/Ls. With decreas- ing the gap (Ls < L∆) the spectrum approaches the usual Rashba model type (doubly degenerate), and because of the specific cancellation inherent in that model hSz(x)i decreases in magnitude as kELs/L2 ∆ [see Fig. 2(a)], fi- nally approaching the limit calculated in Ref. 21 given by ≃ (kE/Ls)(d2/L2 s) [see also Eq. (1)]. Thus for a given strength of the SO interaction, the magnitude of the edge spin density has non-monotonic dependence as a function of the L∆ (or ∆), Fig. 2(a). We note that if one takes for the ratio d/Ls = 0.1, then the edge spin density ob- tained in Ref. [21] for the usual Rashba system with one sub-band equals in magnitude the density which follows from Eq. (8) at ξ ≈ 35, where the latter is three orders of magnitude smaller than its maximal value at ξ = 0.56. Comparison with the experiment. The experimental estimate of L∆ is ≈ 1.4 × 10−5 cm. For Ls we take two characteristic lengths 1.1µm and 2.8µm. Note that the corresponding values of η are consistent with the ones obtained from the theoretical calculations [33] for struc- tures similar to that used in the experiment of Ref. [22]. Thus the value ξ = 0.1 will reasonably correspond the above chosen lengths. Calculating τtr from the mobil- ity 1.9 × 106cm2/Vs , and using E = 0.05 mV/µm for the electric field in the quasi-ballistic region of the sam- ple (both the mobility and E are taken from Ref. [22]), we plot hSz(x)i, see Fig. 2(b). The exact experimental value of the edge spin density is not known; the authors of Ref. [22] have estimated the threshold minimal value compatible with their observation as 3×106cm−2. Hence this number is consistent with our calculation. In addi- tion, we stress that the procedure just described, i.e., comparison of our theoretical predictions for the edge spin density with the experimental value of this quan- tity, allows one to extract the value of η. In conclusion, using a Rashba-like SO interaction aris- ing from the coupling between two sub-band states of opposite parities in a symmetric two-subband quantum well, we have explained the great difference between the edge spin density in a bilayer structure as compared to the one in a single-layer configuration observed in the ex- periment of Ref. [22]. The presence of the gap between the two sub-bands governs the effective strength of the inter-subband SO interaction and changes drastically the picture of the edge spin accumulation. Thus by varying the gap value one can easily proceed from the regime of strong spin accumulation to the regime of weak spin ac- cumulation. This opens up the possibility for the design of new spintronic devices. We acknowledge financial support from FAPESP (Funda¸cao de Apoio `a Pesquisa do Estado de Sao Paulo). Helpful discussions with G. Gusev and F. G. G. Hernan- dez are greatly appreciated. A. Khaetskii is also grateful to Instituto de F´ısica de Sao Carlos of the University of Sao Paulo for the hospitality. [1] H.A. Engel, E.I. Rashba, and B.I. Halperin, in Hand- book of Magnetism and Advanced Magnetic Materials, ed. by H. Kronmuller and S. Parkin, Vol.5 (John Wiley and Sons, New York, 2007). [2] M.I. Dyakonov, and A.V. Khaetskii, "Spin Hall effect", in Spin Physics in Semiconductors, ed. by M.I. Dyakonov (Springer, Berlin, 2008). [3] I. Zuti´c, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004). [4] M.I. Dyakonov and V. I. Perel, JETP Lett. 13, 467 (1971). [5] J.E. Hirsh, Phys. Rev. Lett. 83, 1834 (1999). [6] M.I. Dyakonov, and A.V. Khaetskii, Sov. Phys. JETP 59, 1072 (1984). [7] A.V. Khaetskii, Sov. Phys. Semicond. 18, 1091 (1984). [8] S. Murakami, N. Nagaosa, S.-C. Zhang, Science 301, 5 1348 (2003). [9] J. Sinova, D. Culcer, Q. Niu, N.A. Sinitsyn, T. Jung- wirth, A.H. MacDonald, Phys. Rev. Lett. 92, 126603 (2004). [10] Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Awschalom, Science 306, 1910 (2004). [11] J. Wunderlich, B. Kaestner, J. Sinova, and T. Jungwirth, Phys. Rev. Lett. 94, 047204 (2005). [12] K. Nomura, J. Wunderlich, J. Sinova, B. Kaestner, A. H. MacDonald, and T. Jungwirth, Phys. Rev. B 72, 245330 (2005). [13] Y. Tserkovnyak, B. I. Halperin, A. A. Kovalev, and A. Brataas, Phys. Rev. B 76, 085319 (2007). [14] O. Bleibaum, Phys. Rev. B 74, 113309 (2006). [15] I. Adagideli and G. E.W. Bauer, Phys. Rev. Lett. 95, 256602 (2005). [16] B. K. Nikoli´c, S. Souma, L. P. Zarbo, and J. Sinova, Phys. Rev. Lett. 95, 046601 (2005). [17] G. Usaj and C. A. Balseiro, Europhys. Lett. 72, 631 (2005). [18] V. A. Zyuzin, P. G. Silvestrov, and E. G. Mishchenko, Phys. Rev. Lett. 99, 106601 (2007). [19] P. G. Silvestrov, V. A. Zyuzin, and E. G. Mishchenko, Phys. Rev. Lett. 102, 196802 (2009). [20] A. Khaetskii and E. Sukhorukov, Phys. Rev. B 87, 075303 (2013). [21] A. Khaetskii, Phys. Rev. B 89, 195408 (2014). [22] F. G. G. Hernandez, L. M. Nunes, G. M. Gusev, and A. K. Bakarov, Phys. Rev. B 88, 161305(R) (2013). [23] One-subband or two-subband configurations were real- ized experimentally using symmetric quantum wells of different widths. A 14 nm wide sample contained only one subband but a 45 nm wide sample contained two (even and odd) subbands. Due to the Coulomb repulsion of the electrons in the wide quantum well, the charge dis- tribution resulted in a bilayer electron system with a soft barrier inside the well. Tunneling through this barrier leads to the symmetric-antisymmetric splitting ∆SAS for which we use the notation ∆. [24] F. T. Vasko, JETP Letters 30, 541 (1979); Yu. Bychkov and E. I. Rashba, JETP Letters 39, 78 (1984). [25] M. I. Dyakonov and V. Kachorovskii, Sov. Phys. Semi- cond. 20, 110 (1986). [26] E. Bernardes, J. Schliemann, M. Lee, J. C. Egues, and D. Loss, Phys. Rev. Lett. 99, 076603 (2007). [27] R. S. Calsaverini, E. S. Bernardes, J. C. Egues, and D. Loss, Phys. Rev. B 78, 155313 (2008). [28] A. Khaetskii, Phys. Rev. B 73, 115323 (2006). [29] A.V. Shytov, E.G.Mishchenko, H.-A. Engel, and B. I. Halperin, Phys. Rev. B 73, 075316 (2006). [30] For the two-subband wide-well sample of Ref. [22] the total electron density n = 9.2 × 1011 cm−2 and the Fermi wave vectors kF,1 = kF,2 = kF = p2πn/2) = 1.7 × 106 cm−1. The Fermi energy is εF = 2k2 F /2m = 16.4 meV (assuming m = 0.067m0 for a GaAs well). Note that εF ≫ ∆ = 1.4 meV in Ref. [22]. [31] A fast contribution to the edge spin density which os- cillates as function of x with 2kF wave vector, gives parametrically smaller contribution to the total spin R ∞ 0 dxhSz(x)i, and we omit it. [21] [32] We consider a high mobility 2D structure, where scat- tering is due to the long-range disorder caused by the donor layer located at distance d from the 2D gas. Since kF d ≫ 1, the characteristic scattering angle is ≃ 1/kF d ≪ 1. Therefore the transport scattering time calculated in the Born approximation at a given momen- tum p is τtr(p) = Ap3, where A is some constant, see SM. [33] Jiyong Fu and J. Carlos Egues, Phys. Rev. B 91, 075408 (2015). 6
1607.04160	1	1607	2016-07-14T15:05:19	Control of the non-stationary spin-polarized tunneling currents by applied bias changing	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	We reveal that for the single Anderson impurity localized between non-magnetic leads of the tunneling contact $"$magnetic$"$ state can be distinguished from the $"$paramagnetic$"$ one only by the analysis of the non-stationary system characteristics or the behavior of the second order correlation functions for the localized electrons occupation numbers. We investigate the response of the system to the sudden shift of the applied bias and to the switching $"$on$"$ the coupling to the second lead of the tunneling contact. We demonstrate that in addition to the changes of the relaxation regimes and typical relaxation time scales, non-stationary spin-polarized currents flowing in the both leads are present in the system. Spin polarization and direction of the non-stationary currents in each lead can be simultaneously inverted by the sudden changing of the applied bias voltage.	cond-mat.mes-hall	cond-mat	Control of the non-stationary spin-polarized tunneling currents by applied bias changing N. S. Maslova1, P. I. Arseyev2, and V. N. Mantsevich1∗ 1Moscow State University, 119991 Moscow, Russia, 2 P.N. Lebedev Physical Institute RAS, 119991 Moscow, Russia (Dated: November 10, 2018) We reveal that for the single Anderson impurity localized between non-magnetic leads of the tunneling contact "magnetic" state can be distinguished from the "paramagnetic" one only by the analysis of the non-stationary system characteristics or the behavior of the second order correlation functions for the localized electrons occupation numbers. We investigate the response of the system to the sudden shift of the applied bias and to the switching "on" the coupling to the second lead of the tunneling contact. We demonstrate that in addition to the changes of the relaxation regimes and typical relaxation time scales, non-stationary spin-polarized currents flowing in the both leads are present in the system. Spin polarization and direction of the non-stationary currents in each lead can be simultaneously inverted by the sudden changing of the applied bias voltage. PACS numbers: 72.25.-b, 72.15.Lh, 73.63.-b I. INTRODUCTION Physics of spin-polarized electron transport in semi- conductor nanostructures is among the most rapidly de- veloping topics now a days1. Significant progress has been achieved in experimental and theoretical investi- gation of spin-polarized stationary transport in mag- netic tunneling junctions2,3,4,5,6. Different magnetic ma- terials, such as ferromagnetic metals7 or diluted mag- netic semiconductors8 have been applied as a spin in- jection sources and drains. Nevertheless spin-polarized current sources, that use nonmagnetic materials are at- tractable as they enable to avoid the presence of acci- dental magnetic field that may cause undesirable effects on the spin currents. It was demonstrated recently, that electron tunneling could be spin dependent even in the case of nonmagnetic leads9,10. Moreover, spin-filter de- vices, which can generate a spin-polarized current with- out using magnetic properties of materials were proposed in11,12. To the best of our knowledge stationary spin-polarized currents are usually under investigation. However, cre- ation, diagnostics and controlled manipulation of charge and spin states of the impurity atoms or quantum dots (QDs), applicable for ultra small size electronic devices design requires careful analysis of non-stationary effects and transient processes13,14,15,16,17,18. Consequently, time dependent dynamics of initial spin and charge con- figurations of correlated impurity or QD is an area of great interest both from fundamental and technological point of view. Moreover, characteristics of the station- ary state of single impurity interacting with the reservoir in the presence of strong Coulomb correlations are not completely understood19,20,21,22,23. In this paper we analyze non-stationary spin polar- ized currents through the single-impurity state local- ized in the tunnel junction in the presence of Coulomb correlations and applied bias voltage. We demonstrate that "magnetic" state can be distinguished from the "paramagnetic" one only by analyzing time evolution of opposite spin electron occupation numbers. We re- veal that non-stationary spin-polarized currents can flow in the both leads and their direction and polarization depend on the value of applied bias. Moreover, non- stationary spin-polarized currents simultaneously change their polarization and sign with the applied bias voltage variations. II. THEORETICAL MODEL We consider non-stationary processes in the system of single-level impurity placed between two non-magnetic electronic reservoirs (tunneling junction) with Coulomb correlations of localized electrons. The Hamiltonian of the system H = Himp + Hres + Htun (1) is written as a sum of the single-level Anderson impu- rity Hamiltonian Himp = (cid:88) σ ε1 n1σ + +U n1σ n1−σ, (2) non-magnetic electronic reservoirs Hamiltonian (cid:88) (cid:88) (εp − eV )c+ pσcpσ (3) (cid:88) pσ tp(c+ pσc1σ + c+ 1σcpσ). (4) Hres = εkc+ kσckσ + kσ pσ and the tunneling part Htun = tk(c+ kσc1σ + c+ 1σckσ) + (cid:88) kσ 6 1 0 2 l u J 4 1 ] l l a h - s e m . t a m - d n o c [ 1 v 0 6 1 4 0 . 7 0 6 1 : v i X r a Here index k(p) labels continuous spectrum states in the leads, tk(p) is the tunneling transfer amplitude be- tween continuous spectrum states and localized state with the energy ε1 which is considered to be indepen- dent on the momentum and spin. Operators c+ k(p)/ck(p) are the creation/annihilation operators for the electrons n1σ(−σ) = in the continuous spectrum states k(p). c+ 1σ(−σ)c1σ(−σ)-localized state electron occupation num- bers, where operator c1σ(−σ) destroys electron with the spin σ(−σ) on the energy level ε1. U is the on-site Coulomb repulsion for the double occupation of the local- ized state. Our analysis deals with the low temperature regime when the Fermi level is well defined and the tem- perature is much lower than all the typical energy scales in the system. Consequently the distribution function of the electrons in the leads (band electrons) is close to the Fermi step. III. NON-STATIONARY ELECTRONIC TRANSPORT: FORMALISM AND RESULTS and Eq. (7) correspondingly by the indexes substitution k ↔ p and k (cid:48) ↔ p (cid:48) . One can easily obtain: 2 (cid:88) (cid:48)(cid:54)=k k (cid:90) t (cid:88) c+ (cid:48) k σ ckσtk (cid:48) = i dt1ei·(εk−ε (cid:48) k)·(t−t1) × k (cid:48) (cid:48) · c+ × (t2 k 1σckσ − tktk (cid:48) · c+ (cid:48) k σ c1σ). (9) Similar expressions for the products c+ (cid:48) p ckσ and so on can be obtained from expression (9) by the indexes changing k ↔ p,k cpσ, c+ (cid:48) p (cid:48) ↔ p (cid:48) and so on. σ σ Substituting (9) in Eq. (6) we get Let us further consider = 1 and e = 1 elsewhere, so the motion equations for the electron operators products c+ 1σc1σ, c+ ck(p)σ can be written as: 1σck(p)σ and c+ (cid:48) k (p)σ ∂c+ 1σc1σ ∂t i = −(cid:88) −(cid:88) k,σ p,σ tk(c+ tp(c+ kσc1σ − c+ pσc1σ − c+ 1σckσ) − 1σcpσ), ∂c+ 1σckσ ∂t i = − (ε1 − εk)c+ 1σckσ − − U n1−σc+ 1σckσ + tk(n1σ − nkσ) − (cid:48) c+ pσckσ, (cid:48) k ckσ −(cid:88) − (cid:88) tpc+ tk σ (cid:48)(cid:54)=k p ∂c+ 1σckσ ∂t i + (ε1 − εk + iΓ)c+ + U n1−σc+ 1σckσ + (cid:90) t 1σckσ = tk(n1σ − nkσ) + (cid:90) t dt1tktk (cid:48) c+ (cid:48) k σ (cid:88) (cid:88) k (cid:48) + i + i dt1tktpc+ pσ(t1)c1σ(t1)ei·(εk−εp)·(t−t1), (t1)c1σ(t1)ei·(εk−ε (cid:48) k)·(t−t1) + (5) (6) p (10) where Γ = Γk + Γp and Γk(p) = νk(p)0t2 k(p), νk(p)0 - are the unperturbed densities of states in the left and right leads of the tunneling contact. 1σ cpσ ∂t Equation for ∂c+ can be obtained from Eq.(10) by the indexes k ↔ p and k changing. Multiplying Eq. (10) by the electron operators 1− n1−σ and n1−σ we obtain the following expressions: (cid:48) ↔ p (cid:48) k ckσ ∂c+ (cid:48) k σ ∂t i and ∂c+ pσckσ ∂t i (cid:48) − εk) · c+ = − (εk σ − tk (cid:48) c+ 1σckσ + tkc+ (cid:48) k ckσ − c1σ σ k (cid:48) (7) (1 − n1−σ) · i = ( εk − εp) · c+ pσckσ + pσc1σ − tp · c+ + tk · c+ 1σckσ, + i + i (8) where nk = c+ kσckσ is an occupation operator for the electrons in the reservoir. Equations of motion for the electron operators prod- cpσ can be obtained from Eq.(6) 1σcpσ and c+ (cid:48) p ucts c+ σ p and ∂c+ 1σckσ ∂t (cid:90) t (cid:90) t (cid:88) (cid:88) k (cid:48) + (ε1 − εk + iΓ)(1 − n1−σ)c+ 1σckσ = = (1 − n1−σ) · [tk · (n1σ − nkσ) + dt1tktk (cid:48) c+ (cid:48) k σ (t1)c1σ(t1)ei·(εk−ε (cid:48) k)·(t−t1) + dt1tktpc+ pσ(t1)c1σ(t1)ei·(εk−εp)·(t−t1)] (11) n1−σ · i ∂c+ 1σckσ ∂t + (ε1 − εk + U + iΓ)n1−σc+ 1σckσ = = n1−σ · [tk · (n1σ − nkσ) + (cid:90) t (cid:90) t (cid:88) (cid:88) k (cid:48) p + i + i dt1tktk (cid:48) c+ (cid:48) k σ (t1)c1σ(t1)ei·(εk−ε (cid:48) k)·(t−t1) + dt1tktpc+ pσ(t1)c1σ(t1)ei·(εk−εp)·(t−t1)]. ×[ (12) 3 1 2 i · dεk(p)f σ (cid:90) k(p)(εk(p)) × − 1 − e−i(ε1−i(cid:101)Γ−εk(p))t ε1 − i(cid:101)Γ − εk(p) (cid:90) k(p)(εk(p)) × − 1 − e−i(ε1+U−i(cid:101)Γ−εk(p))t ε1 + U − i(cid:101)Γ − εk(p) dεk(p)f σ i · ], ]. Nk(p)ε = ×[ 1 − ei(ε1+i(cid:101)Γ−εk(p))t ε1 + i(cid:101)Γ − εk(p) 1 − ei(ε1+U +i(cid:101)Γ−εk(p))t ε1 + U + i(cid:101)Γ − εk(p) Nk(p)ε+U (t) = 1 2 (16) rameter (cid:101)Γ (cid:39) Γ, because after applying approximation In Eq.(16) parameter Γ is replaced by the effective pa- (13) to Eq. (12) the omitted terms are of the order of Γ. We can obtain equations for the occupation numbers of localized electrons n1±σ by averaging Eqs. (14)-(16) for the operators and by decoupling electrons occupation numbers in the reservoir. Such decoupling procedure is reasonable if one considers that electrons in the macro- scopic reservoir are in the thermal equilibrium. After decoupling one has to replace electron occupation num- bers operators in the reservoir nσ k in Eqs. (15)-(16) by the Fermi distribution functions f σ k . If one is interested in the situation when the second lead and non-zero bias voltage are switched "on" at the time moment t = t0 > 0, Eqs. (14) can be easily gener- alized: ∂n1σ ∂t ×[n1σ − (1 − n1−σ)N T = −2Θ(t0 − t)Γ × ×[n1σ − (1 − n1−σ)Nkε(t) − n1−σNkε+U (t)] − −2Θ(t − t0)Γ × kε+U (t)], = −2Θ(t0 − t)Γ × ×[n1−σ − (1 − n1σ)Nkε(t) − n1σNkε+U (t)] − −2Θ(t − t0)Γ × kε+U (t)], ×[n1−σ − (1 − n1σ)N T kε(t) − n1−σN T kε(t) − n1σN T ∂n1−σ ∂t (17) If condition ε1−εF slowly varying quantity in comparison with the c+ ∂t c+ ( ∂ ∂t n1−σ << ∂ to consider that: >> 1 is fulfilled, n1−σ is a 1σck(p)σ 1σck(p)σ). Consequently, it is reasonable Γ ∂ ∂t n1−σc+ ∂ ∂t c+ 1σck(p)σ. (13) 1σck(p)σ ∼ n1−σ get expressions for (1−(cid:98)n1−σ)c+ So, the terms ( ∂ ∂t n1−σ)c+ 1σck(p)σ and(cid:98)n1−σc+ Substituting expressions for the c+ 1σck(p)σ are omitted. One can 1σck(p)σ [using the procedure similar to the one which was used to obtain Eq.(9) from Eq.(7)] and then for the c+ 1σck(p)σ. 1σck(p)σ and c+ k(p)σc1σ to Eq. (5) we obtain equations, which determine time evolution of the electron occupation numbers n1σ. It is necessary to note, that the last term in Eq. (10) af- ter summation over the index k(p) doesn't contribute to the non-stationary equations for the electron occupation numbers n1σ. So, time evolution of the electron occupa- tion numbers operators in the situation when the second lead and the non-zero bias voltage are present can be analyzed by means of the system of equations: n1σ ∂t n1−σ ∂t = −2Γ[n1σ − (1 − n1−σ) N T = −2Γ[n1−σ − (1 − n1σ) N T ε (t) − n1−σ N T ε (t) − n1σ N T ε+U (t)], ε+U (t)], where N T ε (t) = N T ε+U (t) = Γk Γ Γk Γ · Nkε(t) + Γp Γ · Nkε+U (t) + · Npε(t), Γp Γ · Npε+U (t) and (14) (15) where Γ = Γk for t < t0 and Γ = Γk + Γp for t > t0. Solution can be easily obtained by the numerical simu- lations of the system of equations. Time dependent dy- namics of the electron occupation numbers and their cor- relation functions can be analyzed for the different initial conditions: 1) the non-zero localized magnetic moment exists on the impurity (n1σ − n1−σ ∼ 1). Such state can be prepared due to the applied external magnetic field µB >> ε1, which is switched "off" at the initial time moment t = 0; 2) the initial state close to the highly oc- cupied paramagnetic one (1 − n1±σ << 1) can be pre- pared by the applied bias voltage eV > ε1 + U switch- ing "off" or "on" at the initial time moment t = 0; 3) the initial state close to the low occupied paramagnetic one (n1±σ << 1) can be prepared by the applied bias voltage eV < ε1 switching "off" or "on" at the initial time moment t = 0. It will be shown that relaxation time scale strongly depends on the properties of the initially prepared state. If one is interested in system time evolution for the time scales t >> 1 , fast oscillating terms, which contain ε1 time dependent exponents can be neglected and functions Nk(p)ε, Nk(p)ε+U [see Eq.(16)] become independent from t. So, the localized electrons occupation numbers n1σ(t), n1−σ(t) can be easily found for the arbitrary initial con- ditions: For 0 < t < t0 n1σ(0) − n1−σ(0) 2 n1−σ(0) − n1σ(0) 2 n1σ(t) = · eλ1t + n1−σ(t) = · eλ1t + n1σ(0) + n1−σ(0) Nkε 1 + ∆N 2 Nkε 1 + ∆N 2 · (1 − eλ2t) + · eλ2t, · (1 − eλ2t) + · eλ2t, n1−σ(0) + n1σ(0) + + where n1±σ(0) are the initial conditions. For t > t0 (18) n1σ(t) = + + n1−σ(t) = + + N T ε 1 + ∆N T · (1 − eλT n1σ(t0) − n1−σ(t0) n1σ(t0) + n1−σ(t0) N T kε 1 + ∆N T · (1 − eλT n1−σ(t0) − n1σ(t0) n1−σ(t0) + n1σ(t0) 2 (t−t0)) + · eλT · eλT 1 (t−t0) + 2 (t−t0), 2 (t−t0)) + · eλT · eλT 1 (t−t0) + 2 (t−t0). 2 2 2 2 4 FIG. 1. (Color online) Normalized relaxation rates behavior as a functions of the applied bias voltage for the parameters ε1/2Γ = −2.5 U/2Γ = 7.5 and Γ = Γk + Γp = 1 (Γk = Γp). voltage is shown in the Fig.1. For small values of applied bias eV < ε two time scales λT 1,2−1 strongly differ, while λT 1 2 → for the large absolute values of the applied bias λT 1/3. For the infinitely large times t → ∞ the stationary state is always "paramagnetic" one and electron occupa- tion numbers are: nstT 1σ = nstT 1−σ = N T ε 1 + ∆N T (22) The behavior of localized state electron occupation numbers for the different initial conditions and the set of system parameters in the case, when the second lead is switched on at the time moment t = t0 > 0 is depicted in the Fig.(2). Obtained results demonstrate, that switch- ing "on" of the second lead with the non-zero applied bias results in the increasing of the relaxation rate and con- sequently destroys the long-living "magnetic" moment. IV. NON-STATIONARY SPIN-POLARIZED CURRENTS: FORMALISM AND RESULTS (19) Eigenvalues λ1,2 and λT 1,2 have the following form: If the initial state is a "magnetic" one, non-stationary spin-polarized currents Ik(p)(t)± flow in the each lead: and λ1,2 = −2Γ · (1 ∓ ∆N ), 1,2 = −2Γ · (1 ∓ ∆N T ) λT ∆N = Nε − Nε+U , ∆N T = N T ε+U . ε − N T (20) (21) n1±σ(t0) is determined by expressions (18) for t = t0. Relaxation rates behavior as a function of applied bias k (t) = −2Γk[n1±σ − (1 − n1∓σ)Nkε(t) − n1∓σNkε+U (t)], I± p (t) = −2Γp[n1±σ − (1 − n1∓σ)Npε(t) − n1∓σNpε+U (t)], I± (23) where electron occupation numbers n1±σ are deter- mined from the system of equations (14) with the mag- netic initial conditions. Non-stationary spin-polarized currents can flow in the both leads and their direction and polarization depend on the value of applied bias. Non-stationary spin-polarized currents Ik(p)(t)± for the initially prepared "magnetic" 5 FIG. 2. (Color online) Changing of relaxation regimes of the electron occupation numbers when coupling to the second lead is switched "on" at 2Γt0 = 8 for different initial conditions. Black line demonstrates n1σ(t) and red line - n1−σ(t) in the presence of Coulomb correlations. Blue line demonstrates n1σ(t) and green line - n1−σ(t) in the absence of Coulomb correlations. a),b) n1σ = 1, n1−σ = 0; c),d) n1σ = 0.8, n1−σ = 0.7. a),c) eV /2Γ = −5.0 and b),d) eV /2Γ = 12.5. Parameters U/2Γ = 7.5, ε/2Γ = −2.5, and Γ = 1 (Γk = Γp) are the same for all the figures. state for the different constant values of applied bias are depicted in Fig.(3)- Fig.(4). Schemes of the spin- polarized currents directions are shown in Fig.(5). For the large negative values of applied bias voltage (see Fig.3a and Fig.5a) non-stationary spin-polarized currents p (t) and I− I + p (t) are flowing in the same direction in the tunneling contact lead with the Fermi level shifted by the applied bias voltage (lead p). In this case strong spin po- larization of the total current occurs at the initial stage of relaxation as the amplitude of current I + p (t) strongly exceeds the amplitude of current I− p (t). Non-stationary k (t) and I− spin-polarized currents I + k (t) in the lead with EF = 0 are also flowing in the same direction, but the dif- ference between currents amplitudes is small (see Fig.3b and Fig.5a). For the positive values of applied bias voltage (see Fig.4 and Fig.5b) direction of the non-stationary spin-polarized currents changes to the opposite one in comparison with the case, when large negative bias was applied to the tun- neling contact [see Fig.5a]. One can easily distinguish the presence of non-stationary spin-polarized currents I + p (t) and I− p (t) again flowing in the same direction in the tun- neling contact lead with the Fermi level shifted by the applied bias voltage (lead p) (see Fig.4a and Fig.5b). Non-stationary spin-polarized currents I + k (t) in the lead with EF = 0 are also flowing in the same di- rection, but the difference between currents amplitudes is k (t) and I− (Color online) Normalized non-stationary spin- FIG. 3. polarized tunneling currents I + k(p)(t)/2Γ (black line) and − k(p)(t)/2Γ (red line). a) Black line I + p (t); I − b) Black line I + k (t). Parameters U/2Γ = 7.5, ε/2Γ = −2.5, eV /2Γ = −5.0 and Γ = 1 are the same for all the figures. n1σ(0) = 1, n1−σ(0) = 0. p (t), red line I− k (t), red line I quite small (see Fig.4b and Fig.5b). Fig. (3)-(4) demon- strate equal amplitudes of non-stationary spin-polarized currents in the stationary state. For typical Γ ∼ 1 ÷ 10 meV and ε ∼ 50 meV24,25, corresponding to the situation depicted in Fig.3, Fig.4 the non-stationary spin-polarized current value is about 1 ÷ 10 nA (1nA (cid:39) 6 × 109e/sec). We revealed, that spin polarization and direction of the non-stationary currents in each lead can be simulta- neously inverted by the sudden changing of the applied bias voltage (see Fig.6 and Fig.7). Fig.6a demonstrates that initially spin-polarized non-stationary current with the dominant I + p (t) component changes direction and po- larization (component I− p (t) starts to prevail), when the applied bias changes the value from the large negative to the large positive one (system energy scheme changes from the one shown in Fig.5a to the one demonstrated in Fig.5b). Tunneling current in the another contact lead also changes polarization and direction (see Fig.6b), but the difference between the components with different spins is not so well pronounced. Opposite situation is depicted in Fig.7. In this case applied bias sign changing leads to the situation when initially spin-polarized non-stationary current with the dominant component I− p (t) changes direction and po- 6 (Color online) Normalized non-stationary spin- FIG. 4. polarized tunneling currents I + k(p)(t)/2Γ (black line) and − k(p)(t)/2Γ (red line). a) Black line I + p (t); I − b) Black line I + k (t). Parameters U/2Γ = 7.5, ε/2Γ = −2.5, eV /2Γ = 12.5 and Γ = 1 are the same for all the figures. n1σ(0) = 1, n1−σ(0) = 0. p (t), red line I− k (t), red line I FIG. 6. (Color online) Triggering of the normalized non- stationary spin-polarized tunneling currents in the case, when the value of the applied bias voltage changes at 2Γt0 = 2. a) Black line I + k (t), red line k (t). Parameters U/2Γ = 7.5, ε/2Γ = −2.5, eV /2Γ = −7.5 − I for 2Γt < 2Γt0 and eV /2Γ = 7.5 for 2Γt > 2Γt0. Parameter Γ = 1 is the same for all the figures. n1σ(0) = 1, n1−σ(0) = 0. p (t), red line I− p (t); b) Black line I + bers: (cid:48) (t − t (cid:48) K σσ ) =< n1σ(t)n1σ (cid:48) (t (cid:48) ) > . FIG. 5. (Color online) Schemes of the spin-polarized currents directions. Scheme a) corresponds to the results, shown in the Fig.3, b) in the Fig.4. Correlation functions K σσ tem of equations: (24) (cid:48) (τ = t − t (cid:48) ) satisfy the sys- larization to the opposite one and component I + p (t) be- comes the leading one (see Fig.6a) (system energy scheme changes from the one shown in Fig.5b to the one demon- strated in Fig.5a). Corresponding electron occupation numbers behavior is shown in Fig.8. Electron occupation numbers reveal non-monotonic behavior. V. STATIONARY CORRELATION FUNCTIONS: FORMALISM AND RESULTS The behavior of the local "magnetic" moments can be also analyzed from the time dependence of the stationary correlation functions for the electron occupation num- K +− = −2Γ · [K +− + ∆N T K−− − N T K−− = −2Γ · [K−− + ∆N T K +− − N T ε n1−σ], ∂ ∂t ∂ ∂t ε n1−σ].(25) Initial conditions are determined as: K +−(t, t) = K +−(0) = ε+U · N T N T 1 + ∆N T , ε K−−(0) = nst 1 = N T ε 1 + ∆N T . (26) Time evolution of the correlation functions can be ob- tained from Eq. (25): 7 FIG. 8. (Color online) Electron occupation numbers time evolution in the case, when the value of applied bias voltage changes at 2Γt0 = 2. Black line n1σ(t), red line n1−σ(t); a). eV /2Γ = −7.5 for t < t0 and eV /2Γ = 7.5 for 2Γt > 2Γt0 b). eV /2Γ = 7.5 for t < t0 and eV /2Γ = −7.5 for 2Γt > 2Γt0. Parameters U/2Γ = 7.5, ε/2Γ = −2.5 and Γ = 1 are the same for all the figures. n1σ(0) = 1, n1−σ(0) = 0. (Color online) Triggering of the normalized non- FIG. 7. stationary spin-polarized tunneling currents in the case, when the value of the applied bias voltage changes at 2Γt0 = 2. a) Black line I + k (t), red line k (t). Parameters U/2Γ = 7.5, ε/2Γ = −2.5, eV /2Γ = 7.5 − I for 2Γt < 2Γt0 and eV /2Γ = −7.5 for 2Γt > 2Γt0. Parameter Γ = 1 is the same for all the figures. n1σ(0) = 1, n1−σ(0) = 0. p (t), red line I− p (t); b) Black line I + N T K +−(τ = t − t (cid:48) ε+U − 1) ε (N T 2(1 + ∆N T ) K−−(τ = t − t (cid:48) + + N T ε (N T ε+U + 1) 2(1 + ∆N T ) 2 τ ) + 2 τ , ) = eλT ) = N T (N T ε )2 1 τ + (1 + ∆N T )2 (1 − eλT ε (N T kε+U + 1) eλT 2(1 + ∆N T ) ε )2 (1 + ∆N T )2 (1 − eλT ε (1 − N T N T ε+U ) eλT 2(1 + ∆N T ) 2 τ + (N T 2 τ ) + eλT 1 τ . (27) The behavior of the stationary correlation functions for the localized electrons occupation numbers with the different spin orientation is depicted in Fig.9. It is clearly evident, that for the deep energy levels correlation func- tions time evolution is much slower, than for the states with the shallow energy levels. For τ → ∞ correlation functions turn to the product of the decoupled electronic occupation numbers mean val- ues: K +−st = K−−st (cid:39) ( N T ε 1 + ∆N T )2. (28) So for t < 1λT 1 the "magnetic" correlations are still present in the system. VI. CONCLUSION We analyzed time evolution of the opposite spin elec- tron occupation for the single-localized state with the Coulomb interaction coupled to two reservoirs in the presence of applied bias voltage. We revealed that in the presence of the second reservoir with non-zero ap- plied bias, "magnetic" state can be distinguished from the "paramagnetic" one by analyzing time evolution of the electron occupation numbers. Typical time scales strongly depend on the value of applied bias and initial conditions. We revealed that non-stationary spin-polarized cur- rents can flow in the both leads and their direction and polarization depend on the value of applied bias. We revealed, that spin polarization and the direction of the non-stationary currents in each lead can be simultane- ously inverted by the sudden changing of the applied bias voltage. But in the stationary state occupation numbers for the electrons with the opposite spins have the same values. Spin polarized tunneling currents in each lead also become equal. 8 We also investigated the changes of the time evolution regimes when the second lead is switched on at the par- ticular time moment. We found out that switching on of the second lead with the non-zero applied bias destroys long-living "magnetic" moment. This work was supported by RFBR grant 16 − 32 − 60024 mol − a− dk and by RFBR grant 14− 02− 00434. FIG. 9. (Color online) Correlation functions time evolution for the different initial conditions. Black lines demonstrate . a) eV /2Γ = −5; b) eV /2Γ = 2.5. K +− parameters U/2Γ = 7.5, ε/2Γ = −2.5 and Γ = 1 are the same for all the figures. , red line - K−− τ τ ∗ [email protected] 1 Semiconductor Spintronics and Quantum Computation, edited by D.D. Awschalom, D. Loss, N. Samarth, Nanoscience and Technology (Springer, Berlin, 2002). 2 E.Y. Tsymbal, O. Mryasov, P.R. LeClair, J. Phys.: Con- dens. Matter 15, R109, (2003) 3 I. Zuti´c, J. Fabian, S. Das Sarma, Rev. Mod. Phys. 76, 323, (2004) 4 H.J. Zhu, M. Ramsteiner, H. Kostial, M. Wassermeier, H.- P. Schonherr, K.H. Ploog, Phys. Rev. Lett. 87, 116601, (2001) 5 Y. Ohno, D.K. Young, B. Beschoten, F. Matsukura, H. Ohno, D.D. Awschalom, Nature(London) 402, 790, (1999) 6 R. Fiederling, M. Keim, G. Reuscher, W. Ossau, G. Schmidt, A. Waag, L.W. Molenkamp, Nature(London) 402, 787, (1999) 7 H.B. Heersche, Th. Schapers, J. Nitta, H. Takayanagi, Phys. Rev. B 64, 161307, (2001) 8 J.C. Egues, Phys. Rev. Lett. 80, 4578, (1998) 9 V.I. Perel', S.A. Tarasenko, I.N. Yassievich, S.D. Ganichev, V.V. Bel'kov, W. Prettl, Phys. Rev. B 67, 201304, (2003) 10 M.M. Glazov, P.S. Alekseev, M.A. Odnoblyudov, V.M. Chistyakov, S.A. Tarasenko, I.N. Yassievich, Phys. Rev. B 71, 155313, (2005) 11 T. Koga, J. Nitta, H. Takayanagi, S. Datta, Phys. Rev. Lett. 88, 126601, (2002) 12 A. Voskoboynikov, S.S. Liu, C.P. Lee, Phys. Rev. B 58, 15397, (1998) 13 I. Bar-Joseph, S.A. Gurvitz, Phys.Rev B, 44, 3332, (1991). 14 S.A. Gurvitz, M.S. Marinov, Phys.Rev A, 40, 2166, (1989). 15 P.I. Arseyev, N.S. Maslova, V.N. Mantsevich, European Physical Journal B, 85(7), 249, (2012). 16 C.A. Stafford, N.S. Wingreen, Phys. Rev. Lett., 76, 1916, (1996). 17 B.L. Hazelzet, M.R. Wegewijs, T. H. Stoof, Y.V. Nazarov, Phys. Rev. B, 63, 165313, (2001). 18 E. Cota, R. Aguado, G. Platero, Phys. Rev. Lett., 94, 107202, (2005). 19 P.I. Arseyev, N.S. Maslova, V.N. Mantsevich, Solid State Comm., 152, 1545, (2012). 20 V.N. Mantsevich, N.S. Maslova, P.I. Arseyev, JETP, 118(1), 136, (2014). 21 L.D. Contreras-Pulido, J. Splettstoesser, M. Governale, J. Konig, M. Buttiker, Phys. Rev. B, 85, 075301, (2012). 22 Florian Elste, David R. Reichman, and Andrew J. Millis, Phys. Rev. B, 81, 205413, (2010). 23 D. M. Kennes, S. G. Jakobs, C. Karrasch, V. Meden, Phys. Rev. B, 85, 085113, (2012). 24 S. Amaha, W. Izumida, T. Hatano, S. Teraoka, S. Tarucha, J. A. Gupta, and D. G. Austing, Phys. Rev. Lett., 110, (2013), 016803. 25 J. Fransson Phys. Rev. B, 69, 201304, (2004). 9
1602.07735	1	1602	2016-02-24T22:26:11	Phonon Monte Carlo: Generating Random Variates for Thermal Transport Simulation	[ "cond-mat.mes-hall", "cond-mat.dis-nn", "cond-mat.mtrl-sci" ]	Phonon Monte Carlo (PMC) is a versatile stochasic technique for solving the Boltzmann transport equation for phonons. It is particularly well suited for analyzing thermal transport in structures that have real-space roughness or are too large to simulate directly using atomistic techniques. PMC hinges on the generation and use of \textit{random variates} -- specific values of the random variables that correspond to physical observables -- in a way that accurately and efficiently captures the appropriate distribution functions. We present the relative merits of the inversion and rejection techniques for generating random variates on several examples relevant in thermal transport: drawing phonons from a thermal distribution and with full or isotropic dispersion, randomizing outgoing momentum upon diffuse boundary scattering, implementing contacts (boundary and internal), and conserving energy in the simulation. We also identify common themes in phonon generation and scattering that are helpful for reusing code in the simulation (generating thermal-phonon attributes vs internal contacts; diffuse surface scattering vs boundary contacts). We hope these examples will inform the reader about the mechanics of random-variate generation and how to choose a good approach for whatever problem is at hand, and aid in the more widespread use of PMC for thermal transport simulation.	cond-mat.mes-hall	cond-mat	Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 6 1 0 2 b e F 4 2 ] l l a h - s e m . t a m - d n o c [ 1 v 5 3 7 7 0 . 2 0 6 1 : v i X r a Chapter 1 Phonon Monte Carlo: Generating Random Variates for Thermal Transport Simulation L. N. Maurer,a S. Mei,b and I. Knezevicc University of Wisconsin-Madison, Madison, WI 53706, USA [email protected], [email protected], [email protected] 1 Introduction Thermal transport in semiconductors is governed by phonons, the quanta of lattice waves [Ziman (1960); Nika and Balandin (2012)]. At temperatures above several degrees Kelvin, phonons experience considerable multiphonon interactions, in addition to scattering with boundary roughness, interfaces, or atoms of different mass that stem from doping, alloying, or natural isotope variation. Therefore, phonon transport in nanostructures is typically in the quasiballistic or diffusive regimes, and is described well by the phonon Boltmann transport equation (PBTE). The PBTE can be solved via deterministic [Broido et al. (2007); Aksamija and Knezevic (2010)] or stochastic techniques [Peterson (1994); Mazumder and Majumdar (2001); Lacroix et al. (2005)] . Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 2 Phonon Monte Carlo Phonon Monte Carlo (PMC) is an efficient stochastic technique for the solution to the PBTE [Peterson (1994); Mazumder and Majumdar (2001); Lacroix et al. (2005, 2006); Randrianalisoa and Baillis (2008); Ramayya et al. (2012); Bera (2012); Mei et al. (2014); P´eraud et al. (2014); Maurer et al. (2015); Ramiere et al. (2016)] which can incorporate real-space roughness and simulate nanostructures of experimentally relevant sizes. In PMC, a large ensemble of numerical phonons (typically of order 105 -- 106) is tracked over time as they fly freely and undergo scattering according to relevant scattering rates [Glassbrenner and Slack (1964)]. Modern transport simulations based on the PBTE involve phonon-phonon scattering rates obtained from first principles [Broido et al. (2007); Esfarjani et al. (2011)] and incorporate full phonon dispersions [Mei et al. (2014)]; the latter is very important in anisotropic systems, such as superlattices, nanowires, and nanoribbons [Mei et al. (2014); Mei and Knezevic (2015)]. A number of random variables with generally complicated distribution functions underscore the behavior of the phonon ensemble. Examples of random variables include the phonon energy or momentum for a bulk phonon system in equilibrium, or the outgoing-momentum direction after a scattering event. The PMC simulation hinges on the generation and use of random variates -- specific values of the random variables that correspond to physical observables -- in a way that accurately and efficiently captures the appropriate distribution functions. Accurate and efficient generation of random variates that numerically represent nonuniform distribution functions is not a simple matter, yet most articles on PMC do not show much detail on this aspect. In this chapter, we discuss numerical generation of random variates relevant for PMC, assuming that the uniformly distributed variates on the [0,1] interval are accessible. We discuss the relative merits of different approaches (direct inversion versus the rejection technique) from both theoretical and practical standpoints, which are sometimes at odds, and show several specific examples of nonuniform distributions relevant for phonon transport. We also identify common themes in phonon generation and scattering that are useful for reusing code in a simulation. We trust these examples will inform the reader about both the mechanics of random-variate Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Generating Random Variates 3 generation and how to choose a good approach for whatever problem is at hand. We review the two main methods for generating random variates in Sec. 2 and the PMC method in Sec. 3. Several applications are presented next: generating the attributes for phonons (Sec. 4.1) and isotropic dispersions (Sec. 4.2), randomizing outgoing momentum upon diffuse boundary scattering (Sec. 5), implementing contacts (Sec. 6), and conserving energy in the simulations (Sec. 7). in equilibrium with full 2 Generating Random Variates Here, we present a quick overview of the methods used to generate random variates with a given probability distribution function (PDF). We assume that the computer used can generate random variates that are uniformly distributed over the interval [0, 1]. There are two techniques for generating nonuniform random variates from uniform random variates: the inversion and rejection methods, which we explain below. (For more details on random- variate generation, see a book such as [Devroye (1986)]). Generally speaking, the inversion method requires more analytical manipulation of the PDF than the rejection method. When the analytical manipulations are possible, the inversion method is usually the simpler of the two. While the inversion method can be performed numerically, the rejection technique is generally simpler to implement in cases when analytical inversion is not possible. 2.1 Inversion Method Consider a PDF p (x). The first step in the inversion method is to integrate the PDF into the cumulative distribution function (CDF), (cid:90) x −∞ F (x) = dx(cid:48)f (x(cid:48)) , (1) Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 4 Phonon Monte Carlo where F (x) is the probability that a random variate will have a value less than or equal to x. Next, we generate a random variate r, which is uniformly distributed in [0, 1]. We solve r = F (x) for x, i.e., we invert the CDF to get the quantile function Q (r) = F −1 (r). Finally, we solve x = Q(r) for x.The resulting x is a random variate that follows our original PDF [Devroye (1986)]. The technique can be generalized to PDFs with multiple variables, but we will only consider PDFs that effectively only depend on a single variable. For example, say we want to generate a random variate from the distribution given by Lambert's cosine law1 in three dimensions (3D), p (θ, φ) = c cos θ for spherical coordinates θ ∈ [0, π/2] , φ ∈ [0, 2π) and a normalizing constant c [Modest (2003)]. Lambert's cosine law will prove important later (Secs. 5 and 6.2). First, we must properly normalize p (θ, φ): 1 = p (θ(cid:48), φ(cid:48)) sin θ(cid:48)dθ(cid:48)dφ(cid:48), (cid:90) π/2 (cid:90) 2π (cid:90) 2π (cid:90) θ 0 0 0 = sin2 θ. which yields c = π−1. The CDF for θ is then 0 F (θ) = p (θ(cid:48), φ(cid:48)) sin θ(cid:48)dθ(cid:48)dφ(cid:48) (2) (3) Note that our CDF only depends on θ. We can do this since p (θ, φ) does not depend on φ, but we keep the φ dependence to make clear that we still need to integrate over φ. Finally, invert the CDF for a random variate rθ that is uniformly distributed in [0, 1]: F (θ) = rθ . θ = arcsin ( √ rθ) . (4) Using the same method, we can find the unsurprising result that φ = 2πrφ, where rφ is uniformly distributed in [0, 1]. In this example, both the integration and inversion can be done analytically, which is the exception rather than the rule. It 1This is also known as Knudsen's cosine law in the context of gas molecules scattering from surfaces. Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Generating Random Variates 5 is possible to do both the integration and inversion numerically, but this reduces the accuracy and computational efficiency of the method. We will see cases where both the integration and inversion are done numerically (Sec. 4.1) and where the integration can be done analytically, but the inversion is done numerically (Sec. 5). 2.2 Rejection Method In contrast with the inversion method, the rejection method does not require any integration or inversion steps. The rejection method requires that we calculate the PDF p (x) for any x, there exists a bounding function g (x) such that ∀x, g (x) ≥ p (x), and that we can generate random variates from a PDF that is proportional to g (x). p (x) and g (x) do not have to be normalized; they must simply be proportional to probability distribution functions. The largest drawback of the rejection method is that, unlike the inversion method, the rejection method generally requires the computer to generate several random variates that are uniformly distributed in [0, 1]. Choosing a g (x) that closely resembles p (x) will reduce the number of random variates that the computer will have to generate. The rejection method to generate a random variate from p (x) follows. (1) Generate a random variate x(cid:48) from the PDF that is proportional to g (x). (2) Generate a random variate y that is uniformly distributed in [0, g (x(cid:48))] (3) If y < p (x(cid:48)), then x(cid:48) is the random variate generated from p (x). Otherwise, return to step (1). The third step ensures that the probability of choosing x(cid:48) is proportional to p (x(cid:48)), which is all that we require of a method to generate random variates from a distribution. This procedure is easy to visualize if p (x) is non-zero only on a finite interval [a, b], and p (x) ≤ c, where a, b, c are constants (Fig. 1). In this case, we can choose g (x) = c. Then the rejection method is equivalent to throwing a dart randomly and uniformly at the box defined by x(cid:48) ∈ [a, b] , y ∈ [0, c]. If y ≤ p (x(cid:48)), i.e., the dart Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 6 Phonon Monte Carlo Figure 1 Illustration of the rejection method for a distribution p (x) using a constant g (x). For a constant g (x), the rejection technique is the same as randomly throwing darts uniformly in the range x ∈ [a, b] , y ∈ [0, c]. If the dart lands above the curve p (x), then another dart is thrown. When a dart lands below the curve p (x), then its x value, x(cid:48), is the random variate that is accepted. The figure depicts possible dart throws (marked with bullseyes). The throws above p (x) are rejected, and x(cid:48) is taken from the throw that lands below p (x). falls below the curve p (x), then we take x(cid:48) as our random variate and repeat the procedure otherwise. The larger the area between the line y = c and the curve p (x), the more dart throws will be required for a dart to land below the curve p (x), which reduces the efficiency of the rejection method. For this reason, even if p (x) ≤ c, it may be wise to use a g (x) other than g (x) = c. We do this in Sec. 5 when we consider Soffer's model for momentum-dependent boundary scattering [Soffer (1967)]. We note that the rejection technique can work even if p (x) diverges, which is common in physics problems (e.g., Van Hove singularities). Take the case of a probability distribution that might arise from the Bose-Einstein distribution in two dimensions (2D) and in polar coordinates: p (r, θ) = 1 er − 1 . (5) p(x)g(x)abcx'Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Overview of Phonon Monte Carlo 7 (cid:82) R (cid:82) 2π Note that p (r, θ) is not normalized and diverges at r = 0. Also note that p (r, θ) ≤ r−1, so we choose g (r, θ) = r−1. Finally, suppose that we are considering the 2D domain θ ∈ [0, 2π) , r ∈ [0, R], where R is a constant. On that domain, g (r, θ) and p (r, θ) are normalizable because the singularity at r = 0 is integrable: 0 g (r, θ) rdrdθ = 2πR, which is finite. Because the integrand g (r, θ) r is constant, the random variate r(cid:48) is equally likely to take any value in [0, R]. So, it is quite simple to generate the random variates r(cid:48) and θ(cid:48): r(cid:48) is uniformly distributed in [0, R], and θ(cid:48) is uniformly distributed in [0, 2π). 0 Putting everything together, the rejection technique in this example works as follows: (1) Generate random variates r(cid:48) and θ(cid:48) that are uniformly distributed in [0, R] and [0, 2π), respectively. (2) Generate a random variate y that is uniformly distributed in (cid:104) 0, (r(cid:48)) −1(cid:105) (cid:16) (cid:17)−1 (3) If y < p (r(cid:48), θ(cid:48)) = er(cid:48) − 1 , then use r(cid:48) and θ(cid:48) as your random variates. Otherwise, return to step (1). 3 Overview of Phonon Monte Carlo the main carriers of heat Phonons are in semiconductor materials [Nika and Balandin (2012)]. Phonons can be treated as semiclassical particles on the spatial scales longer than the phonon coherence length and time scales longer than the phonon relaxation time. Phonon transport under these conditions is captured via the phonon Boltzmann transport equation (PBTE) [Ziman (1960)]: ∂nb(r, q, t) ∂t + vb,q · ∇rnb(r, q, t) = ∂nb(r, q, t) ∂t . (6) (cid:12)(cid:12)(cid:12)(cid:12)scat nb(r, q, t) is the time-dependent distribution of phonons with respect to position r and the phonon wave vector q for phonon branch b. vb,q = ∇qωb,q is the phonon group velocity, where ωb,q is the phonon angular frequency in branch b at wave vector q. In equilibrium, the average occupancy of a phonon state with Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 8 Phonon Monte Carlo energy ω at absolute temperature T is given by the Bose-Einstein distribution function , (7) (cid:104)nBE(ω, T )(cid:105) = 1 ω kBT − 1 e where kB is the Boltzmann constant. When addressing out-of- equilibrium phonon transport, we assume local equilibrium and employ the concept of a local and instantaneous temperature, T (r, t), then calculate the expectation number of phonons accordingly. To obtain the phonon group velocity vb,q, we require the phonon dispersion relations -- the relationships between the phonon angular frequency ωb,q and the phonon wave vector q. Figure 2 shows the full phonon dispersions along the high-symmetry directions in 3D silicon and 2D graphene. In acoustic branches, the phonon dispersion is relatively isotropic, therefore an analytical isotropic phonon dispersion (ωb,q depends only on the magnitude of q) is often adopted, as it enables a direct and simple evaluation of the phonon group velocity from the random phonon wave vector obtained after scattering (more on this in Sec. 2), which speeds up the simulation. A quadratic isotropic dispersion2 has been shown to be very accurate in simulating the thermal properties of silicon [Mazumder and Majumdar (2001); Lacroix et al. (2006); Ramayya et al. (2012)]. However, when the simulated structure is very small, such as in the case of thin nanowires or graphene nanoribbons (GNRs), anisotropy is prominent and adopting the full dispersion is necessary at the cost of slowing down the simulation. In the following sections, the isotropic approximation and the full dispersion relations in the context of PMC will be demonstrated on the examples of 3D silicon and 2D graphene, respectively. Phonon Monte Carlo is a widely used stochastic technique for solving the PBTE and obtaining thermal-transport properties of semiconductor materials [Mazumder and Majumdar (2001); Lacroix et al. (2006)]. PMC tracks phonon transport in real space, 2ωb(q) = vs,bq + cbq2, where q = q ∈ [0, qmax] is the allowed wave number, vs,b is the mode-dependent sound velocity, and cb is quadratic coefficient fitted from the full dispersion. The maximum allowed wave number qmax is calculated from 2π a0 , where a0 is the lattice constant of the material. Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Overview of Phonon Monte Carlo 9 Figure 2 Phonon dispersions for (a) 3D silicon and (b) 2D graphene along high-symmetry directions. thus allowing easy implementation of nontrivial geometries such as rough boundaries and real-space edge structures [Ramayya et al. (2012); Mei et al. (2014); Maurer et al. (2015)]. Figure 3 shows the flowchart of a typical PMC simulation. The simulation domain 00.511.522.53MKΓΓLXAngular Frequency [rad/s]00.20.40.60.81.01.21.41.6XKWΓΓAngular Frequency [rad/s](a)(b)Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 10 Phonon Monte Carlo Figure 3 Flowchart of a PMC simulation. The dashed box encloses the transport kernel. is a wire with a square cross-section (3D) or a rectangle (2D) divided into Nc cells of equal length along the heat-transport direction, as shown in Fig. 4. The two end cells are connected to heat reservoirs fixed at slightly different temperatures (Th and Tc). For structures much longer than the phonon mean free path (i.e., in the diffusive-transport limit) and in a steady state, a linear temperature profile between Th and Tc will naturally develop inside the simulation domain (Fig. 4). Therefore, we can achieve a steady state in the simulation faster if we initialize the cells according to a linear temperature profile, with the ith cell temperature set to Ti = Th − i−1 The total energy associated with the ith cell is then Db(ω)(cid:104)nBE(ω, Ti)(cid:105)ω dω, (8) Nc−1 (Th − Tc). (cid:90) (cid:88) Ei = Ωi b phonon driftphonon scatterreinitializationsteady state forlong enoughrecord cell energyrecord cell energyafter scatteringbefore scatteringexitEc, Edinitializationt=t+ΔtnoyesBook chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Overview of Phonon Monte Carlo 11 Figure 4 A typical simulation domain for a (a) 3D and (b) 2D PMC simulation. The color represents a typical steady-state temperature profile in each structure when connected to cold (blue, left end) and hot (red, right end) reservoirs. is the volume (3D) or area (2D) of the ith cell where Ωi in real space, Db(ω) is the phonon density of states (PDOS) in the material associated with branch b and energy level ω, and the sum is over all three acoustic branches. The expectation number of phonons in the ith cell is then (cid:90) (cid:88) Ni,exp = Ωi Db(ω)(cid:104)nBE(ω, Ti)(cid:105) dω. (9) b In simulations where the sample size is large (on the order of microns) or the temperature is not very low (a few hundred Kelvin), the expectation number of phonons Ni,exp can be very high (107 − 109), and it is computationally expensive to keep track of this many particles in the simulation. Instead, a weighting factor W is often introduced [Mazumder and Majumdar (2001); Lacroix et al. (2006)] to reduce the number of simulation particles to a tolerable range (typically 105 − 106). (a)(b)Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 12 Phonon Monte Carlo In 2D materials, like graphene, the number of flexural out- of-plane (ZA) phonons is overwhelmingly larger than that of transverse acoustic (TA) or longitudinal acoustic (LA) phonons, owing to the shape of the dispersion curves; therefore, a branch- dependent weighting factor Wb should be used. With the weighting factor taken into consideration, the total energy of the simulation particles in a cell becomes Ei = Ωi Db(ω) ω dω, (10) (cid:90) (cid:88) b (cid:104)nBE(ω, Ti)(cid:105) Wb During the initialization, we keep generating phonons following a desired distribution according to each cell's temperature and add them to random positions inside the cell until the cell has the desired energy, as expressed by Eq. (10). After initialization, we enter the transport kernel (enclosed in the dashed box in Fig. 3), where time is discretized in steps of ∆t. Each phonon is allowed to drift according to the group velocity obtained through the dispersion relation. A random number is drawn to decide whether the phonon will be scattered during its drift; the possibility of being scattered is captured through a phonon relaxation time. We record the heat flux along the wire (3D)/ribbon (2D) to monitor whether a steady state has been reached (i.e., whether the flux has become constant) and use ensemble averages to compute the thermal properties of the material. Special measures (e.g., reinitialization) are needed to make sure the energy in each cell is properly conserved without violating the distribution (Sec. 7). More details about the full-dispersion PMC simulation can be found in [Mei et al. (2014)]. 4 Generating Phonon Attributes in PMC 4.1 Thermal Phonons with Full Dispersion in 2D With the basic knowledge of the methods to generate random variates following certain distributions, this section gives examples of using these methods to randomly draw a phonon from the equilibrium distribution, following the full dispersion relation in 2D graphene. Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Generating Phonon Attributes in PMC 13 Here, we will present both the inversion technique with numerical integration and inversion and the rejection technique; the former is commonly used for thermal phonons with isotropic dispersion in 3D [Mazumder and Majumdar (2001); Lacroix et al. (2005); Ramayya et al. (2012)], as is the rejection technique [Peterson (1994); Maurer et al. (2015)]. The inversion technique is used to choose the phonon frequency and branch, and the rejection technique is used to choose a wave vector that matches the chosen frequency and branch. We note that we could use the rejection technique to choose the branch and wave vector directly (the method would be similar to the example at the end of Sec. 2.2), but we choose a hybrid approach to allow for code reuse: for internal scattering, we already need the code that can generate phonons of a specific frequency. So, we use the inversion technique to choose the phonon frequency and then use the preexisting code to draw a corresponding wave vector. As introduced in Sec. 3, we have assigned a temperature to the cell that we are generating a phonon in. The first step in generating the phonon is to find an angular frequency which follows the Bose- Einstein distribution according to the temperature. To do that, we use the CDF of ω at the given temperature T with the help of DOS: (cid:80) (cid:80) (cid:82) ω (cid:82) ωmax 0 dω(cid:104)nBE(ω, T )(cid:105)Db(ω)/Wb dω(cid:104)nBE(ω, T )(cid:105)Db(ω)/Wb b b 0 F (ω, T ) = . (11) With full dispersion, we do not have an analytical expression for Db(ω). Therefore, we divide the range of [0, ωmax] into Nint equal (cid:101) is the interval length and ωc,i = energy bins where ∆ω = (cid:100) ωmax 2i−1 2 ∆ω is the central frequency of the ith interval. We can obtain Nint Db(ωc,i) for i = 1, 2, . . . , Nint and evaluate the discrete CDF as (cid:80) (cid:80) b i(cid:80) Nint(cid:80) j=1 (cid:104)nBE(ωc,j, T )(cid:105)Db(ωc,j)/Wb Fi(T ) = (cid:104)nBE(ωc,j, T )(cid:105)Db(ωc,j)/Wb . (12) b j=1 To complete the table, set F0(T ) = 0, meaning that all phonons must have positive energy. The numerically evaluated CDF for phonons at 300 K is shown in Fig. 5. Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 14 Phonon Monte Carlo Figure 5 Cumulative distribution function of angular phonon frequency at 300 K. The frequency range ω ∈ [0, 2.5 × 1014] rad/s was divided into Nint = 2500 equal intervals in the numerical calculation. Reprinted with permission from S. Mei, L. N. Maurer, Z. Aksamija, and I. Knezevic, J. Appl. Phys. 116, 164307 (2014). Copyright 2014, American Institute of Physics. The next step is to numerically invert the the table: we draw a random number r1 and look for the interval i satisfying Fi−1 < r1 < Fi. We decide the frequency of this phonon falls in the ith interval and the actual frequency is determined with another random number r2, ω = ωc,i + (2r2 − 1) ∆ω 2 . When the ω is chosen, phonon branch b can be chosen in a similar fashion. Use index 1, 2, and 3 to represent the TA, LA, and ZA branches, respectively. Since there are only 3 branches, we can enumerate the CDFs as (cid:80) (cid:80) F1(ω) = F2(ω) = F3(ω) = 1, D1(ω)/W1 b(cid:48) Db(cid:48)(ω)/Wb(cid:48) , D1(ω)/W1 + D2(ω)/W2 b(cid:48) Db(cid:48)(ω)/Wb(cid:48) , (13a) (13b) (13c) 00.511.522.500.20.40.60.81Phonon angular frequency [1014 rad/s]Cumulative distribution functionBook chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Generating Phonon Attributes in PMC 15 Figure 6 Isoenergy curves in the first Brillouin zone for TA-branch phonons. Adjacent curves are offset by 2 × 1013 rad/s and the shaded region in the black triangle is the irreducible wedge in the first Brillouin zone. Reprinted with permission from S. Mei, L. N. Maurer, Z. Aksamija, and I. Knezevic, J. Appl. Phys. 116, 164307 (2014). Copyright 2014, American Institute of Physics. and a third random number r3 is used to choose the phonon branch b. The next step is generating the phonon wave vector, q, for the ω and b we already found. The distribution is 2D and complex, so it is hard to calculate a CDF. As a result, we employ the rejection technique. Figure 6 shows the isoenergy curves for the TA branch in the first Brillouin zone (1BZ), adjacent curves differing by 2 × 1013 rad/s. Because the isoenergy curves are close to circles, it is convenient to use the polar coordinate. Further, because of the symmetry, we can generate q in the shaded area and simply map it to the whole 1BZ. (More details can be found in [Mei et al. (2014)].) We use the rejection technique to choose an angle θ ∈ [0, π 6 ] and use −0.6−0.4−0.200.20.40.6−0.6−0.4−0.200.20.40.6qx [2/a]qy [2/a]Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 16 Phonon Monte Carlo a lookup table to get the corresponding q. The probability of a phonon with frequency ω in branch b having angle θ is represented by p(ω, θ) ∝ arc(θ − δθ, θ + δθ) vg(ω, θ) , (14) 2 where arc(θ − δθ, θ + δθ) is the arc length on the isoenergy curve between (θ − δθ, θ + δθ) and vg(ω, θ) is the magnitude of group velocity. When δθ is small, it is acceptable to assume the group velocity is constant along the arc. Since we do not have an analytical expression for the probability, we make a rejection table 6 ] where ∆θ = θmax of Na equally separated points between [0, π Na is the spacing and θc,i = (2i − 1) ∆θ is the central frequency in the ith interval. Note that energy is also discrete; we evaluate Eq. (14) only at (ωc,i, θc,j) and obtain an Nint × Na interpolation table. For any 0 < ω < ωmax and 0 < θ < θmax we can get the probability of having a phonon from interpolation. Na = 100 is enough for accurate interpolation. Upon getting the energy ω, we get the 1 × Na angle-distribution table fω(θc,i) and record fω,max = maxNa i=0 (fω(θc,i)). A pair of uniformly distributed random numbers (x, y) is drawn. Therefore, our angle in consideration is 6 · x. Then interpolate the distribution table to obtain fω(α). α = π If fω,max · y ≤ fω(α), the angle α is accepted, and we can proceed to look up q(ω, α) by interpolation. The final chosen wave vector is then (q(ω, α) cos α,q(ω, α) sin α). If fω,max · y > fω(α), the angle is rejected. We keep generating (x, y) pairs until an angle is accepted, which typically occurs within two iterations with the simple fω,max bound. 4.2 Thermal Phonons with an Isotropic Dispersion in 3D As we mentioned in the previous section, it is common to draw thermal phonons with isotropic dispersion relations in 3D using the inversion technique with numerical integration and inversion [Mazumder and Majumdar (2001); Lacroix et al. (2005); Ramayya et al. (2012)], although the rejection method is also sometimes used [Peterson (1994); Maurer et al. (2015)]. We believe that the rejection method is better suited to the task than the inversion Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Generating Phonon Attributes in PMC 17 method because the numerical integration and inversion can lead to a loss off accuracy or decreased computational performance [Mazumder and Majumdar (2001)]. Additionally, the rejection method does not require the density of states, which can be difficult to calculate. We will consider two branches, a TA branch and an LA branch, but the method can easily be generalized to more branches. We will also assume qmax ≥ q is an upper bound on the wave- vector magnitude. The rejection method works as follows. At temperature T , the number of phonons from branch b in each infinitesimal unit of reciprocal space is nb (q, T ) = d3q (2π)3(cid:104)nBE(ωb,q, T )(cid:105). (15) For an isotropic dispersion relation, nb can be reduced to a function of q alone by integrating over the polar and azimuthal angles θ and φ (cid:90) 2π (cid:90) π q2 sin θdqdθdφ (cid:104)nBE(ωb,q, T )(cid:105) (16) niso,b (q, T ) = 0 (2π)3 (cid:104)nBE(ωb,q, T )(cid:105)dq. 0 q2 2π = Although (cid:104)nBE(cid:105) diverges at q = 0, limq→0 nb (q) = 0 because of the q2 term. Thus, the maximum value of niso,b (q, T ) is finite, and we can use the bounding function g (q) = c as described in Sec. 2.2, where c is any number greater than the maximum value of niso,TA (q, T ) + niso,LA (q, T ). The maximum value of nb (q, T ) is temperature dependent, but instead of finding a new c whenever the temperature changes, simply find a c that works for a temperature higher than any conceivable temperature in your simulation. Once c is found, the rejection method follows the familiar pattern: (1) Generate a random variate q(cid:48) that is uniformly distributed in [0, qmax] (2) Generate a random variate y that is uniformly distributed in [0, c] Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 18 Phonon Monte Carlo (3) If y < niso,TA (q, T ), then generate a TA phonon with wave number q(cid:48). If niso,TA (q(cid:48), T ) < y < niso,TA (q(cid:48), T ) + niso,LA (q(cid:48), T ), then generate an LA phonon with wave number q(cid:48). Otherwise, return to step (1). Choosing between different branches in the last step is similar to the procedure used for choosing a branch in Eq. (13). Once we know the wave number q of the new phonon, we need to choose a direction. In equilibrium, all directions are equally likely [p (θ, φ) = d, where d is a constant], and we can use the inversion technique to choose a direction. First, we find d (cid:90) π (cid:90) 2π 1 = p (θ(cid:48), φ(cid:48)) sin θ(cid:48)dθ(cid:48)dφ(cid:48) 0 0 −1 . d = (4π) The CDF is3 P (θ) = = (cid:90) θ (cid:90) 2π p (θ(cid:48), φ(cid:48)) sin θ(cid:48)dθ(cid:48)dφ(cid:48) 0 0 1 − cos θ 2 . (17) (18) Inverting for a random variate rθ that is uniformly distributed in [0, 1] θ = arccos (1 − 2rθ). φ is uniformly distributed in [0, 2π]. All together  sin (θ) cos (φ) sin (θ) sin (φ) q = q(cid:48) (19) (20)  cos (θ) 3In the CDF, (1 − cos θ) /2 can be rewritten as sin2(θ/2), but making that change has no advantage in implementing the code, although it does make the equation look more like the results from Lambert's cosine law (Eq. (4)). Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Diffuse Boundary Scattering 19 5 Diffuse Boundary Scattering Nanostructure surfaces can have a large impact on thermal conductivity through increased boundary-surface scattering [Lim et al. (2012)], so implementing real-space phonon -- surface scat- tering is important for PMC simulations. A common class of phonon -- surface scattering models rely on diffuse scattering and a specularity parameter: a phonon that strikes a surface either reflects specularly or is scattered into a new phonon of the same frequency and branch, but with a random wave vector. The probability of specular scattering is controlled by a specularity parameter, which is either constant [Berman et al. (1953)] or momentum dependent [Berman et al. (1955); Ziman (1960); Soffer (1967)]. Specularity-parameter models fail when the surfaces are very rough, but appear to work well for small roughnesses [Maurer et al. (2015)]. Additionally, real phonons of one branch can be scattered into phonons of a different branch at a surface [Northrop and Wolfe (1984)]. This process, referred to as mode conversion, is often ignored, but can be included in phonon Monte Carlo simulations [Couchman et al. (1992)]. All phonon -- surface scattering models must respect detailed balance in equilibrium: the number of phonons scattered into any solid angle must be equal to the number of phonons incident on the surface from the same solid angle. Consider a flat surface with a surface normal vector n. In this section, θ will refer to the angle between a phonon's wave vector q and the surface's unit normal vector. (We will assume that the dispersion relation is isotropic, so that q is in the same direction as the velocity.) Phonons with a larger q · n = q cos θ will strike the surface more frequently, and in equilibrium this leads to Lambert's cosine law: the number of phonons incident on a surface at an angle θ must also be proportional to cos θ (Fig. 7). Because of detailed balance, the number of phonons being scattered into an angle θ is proportional to cos θ. For specular scattering, detailed balance in equilibrium holds automatically. For totally diffuse scattering, the inversion method can beused to generate random wave vectors that satisfy Lambert's Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 20 Phonon Monte Carlo Figure 7 Illustration of Lambert's cosine law. An equilibrium ensemble of phonons is depicted as a group of particles with uniform angular distribution (lower left). The number of phonons striking a small patch of surface dA at an angle θ (measured from the surface normal) is proportional to the angle subtended by the shaded wedges. This angle is proportional to cos θ (if dA is small), which is the basis for Lambert's cosine law. cosine law. Lambert's cosine law in 3D was an example in Sec. 2.1 with the result given in Eq. (4). In 2D, the procedure is very similar except that we work in polar rather than spherical coordinates. We use the distribution p (θ) = c cos θ, where c is a normalizing constant. The normalization condition is now (cid:90) π/2 1 = 0 p (θ(cid:48)) dθ(cid:48), (21) (22) which yields c = 1. The CDF is F (θ) = p (θ(cid:48)) dθ(cid:48) (cid:90) θ 0 = sin θ. Finally, we invert the CDF for a random variate rθ that is uniformly distributed in [0, 1]: dAdAθBook chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Diffuse Boundary Scattering 21 Figure 8 Polar plots of the angular distribution of the outgoing phonon momentum upon diffuse scattering from a surface with (left) a constant specularity parameter and (right) Soffer's momentum- dependent specularity parameter, Eq. (24) [Soffer (1967)]. Soffer's momentum-dependent specularity parameter decreases the chance of diffuse scattering as θ (the angle between the phonon wave vector and the surface normal) increases. To satisfy detailed balance, the distribution of outgoing phonons must match the probability of diffuse scattering. The result is a teardrop-shaped distribution, which suppresses scattering at large θ relative to the constant-specularity-parameter model. F (θ) = rθ, θ = arcsin θ, (23) which resembles the 3D result. For the constant-specularity-parameter model in both 2D and 3D, one simply chooses a probability p of specular scattering before running the simulation. Then, each time a phonon strikes the surface, the phonon is specularly scattered with probability p and is otherwise diffusely scattered into a randomly chosen direction using Eqs. (4) or Eq. (23) for 3D or 2D, respectively. The situation is somewhat more complicated for momentum- dependent specularity-parameter models. Take Soffer's model in 3D [Soffer (1967)]. The probability that a phonon scatters specularly is p (θ, φ) ∝ e−(2σq cos θ)2 , (24) where σ is the RMS roughness of the surface. Because the probability of diffuse scattering now depends on the angle of DiffuseSpecularIncidentBook chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 22 Phonon Monte Carlo incidence, the outgoing phonon distribution no longer follows a simple cosine law (Fig. 8). We will first attempt to use the inversion technique to find the correct distribution of scattered phonons; after encountering difficulties, we will turn to the rejection technique. The angular distribution of phonons incident on the surface is still proportional to cos θ, and the probability of diffuse scattering is 1− p (θ, φ), so the distribution of incoming phonons that will be diffusely scattered is f (θ, φ) = C cos θ (1 − p (θ, φ)) (cid:16) 1 − e−(2σq cos θ)2(cid:17) = C cos θ (25) , where C is a normalization constant given by 1 = C−1 = 2π (cid:33) (cid:90) π/2 (cid:32) 0 0 (cid:90) 2π 1 − 1 − e−4σ2q2 (cid:90) 2π 4σ2q2 (cid:90) θ f (θ(cid:48), φ(cid:48)) sin θ(cid:48)dθ(cid:48)dφ(cid:48), (26) . Now that C is known, we can find the CDF F (θ) = = f (θ(cid:48), φ(cid:48)) sin θ(cid:48)dθ(cid:48)dφ(cid:48) 0 0 e−4σ2q2−e−4σ2q2 cos2(θ) 1 − 1−e−4σ2q2 4σ2q2 4σ2q2 + sin2 (θ) (27) . We have been able to do the integration step analytically, but we cannot do the inversion step analytically. So, rθ = e−4σ2q2−e−4σ2 q2 cos2(θ) 1 − 1−e−4σ2 q2 4σ2q2 4σ2q2 + sin2 (θ) (28) will have to be solved numerically, for example with Newton's method (which requires computing dF dθ ) or the bisection method. The other option is to use the rejection technique. This is particularly appealing because we already have a good distribution Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Contacts 23 function g (θ, φ) ≥ f (θ, φ), namely g (θ, φ) = cos θ, which is the distribution we use for the constant-specularity parameter model.4 The rejection method to find the outgoing angle θ(cid:48) of a scattered phonon is then: (1) Generate a random variate θ(cid:48) = arcsin(cid:0)√ 1 − e−(2σq cos θ(cid:48))2(cid:17) (3) If y < f (θ(cid:48), φ) = cos θ(cid:48)(cid:16) uniformly distributed in [0, 1]. [0, cos θ(cid:48)] (cid:1) where rθ is rθ , then use θ(cid:48) as (2) Generate a random variate y that is uniformly distributed in your random variate. Otherwise, return to step (1). Because we have a good g (θ, φ), and because the inversion method will require an iterative solution, the rejection method is both faster and simpler to implement. 6 Contacts In the PMC transport simulation, we need contacts at the ends of the simulation domain to act as fixed-temperature phonon reservoirs. These reservoirs must inject new phonons into the simulation domain and absorb the phonons that leave it. There are two basic methods to implement contacts: either the boundaries of the simulation domain mimic reservoirs outside the simulation domain, or parts of the simulation domain get turned into reservoirs. We will call the former a boundary contact and the latter an internal contact. Both approaches can make use of existing code, because creating new phonons has algorithmically much in common with scattering existing phonons. Internal contacts can reuse the code for internal scattering mechanisms, and boundary contacts can reuse the code for diffuse surface scattering. All PMC simulations will implement internal scattering mechanisms, so it is always straightforward to implement internal contacts. In contrast, not all PMC simulations will implement diffuse surface scattering, so it 4We could also use a constant distribution function for g (θ, φ) as described in Sec. 2.2, but that would result in reduced performance. Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 24 Phonon Monte Carlo may take extra work to implement boundary contacts. We will see that boundary contacts require fewer new phonons to be generated each time step, which increases the performance of the simulation. 6.1 3D Internal Contacts Internal contacts are relatively simple to implement because the approach almost exclusively reuses code from other parts of the simulation. To implement internal contacts in 3D, one simply deletes all the phonons in a volume near at the end of the simulation domain and then fills the volume with new phonons drawn from the equilibrium distribution (Sec. 4.2). The only size requirement is that the volume must be large enough that no phonons can traverse the volume from end to end in one time step. There are two equivalent ways to fill the volume: either with a certain number of phonons ("fill by number"), or by adding phonons until the cell has the correct energy ("fill by energy"). It is simpler to implement code that fills by number, but the simulation will already need code to fill volumes by energy, in which case filling by energy can simply re-use existing code. We will discuss how to fill by energy in Sec. 7, so we explain how to fill by number here. (Boundary contacts will also use number rather than energy.) The average number of phonons in branch b in the volume at a temperature T is5 Nb(T ) = V (cid:90) d3q (2π)3(cid:104)nBE(ωb,q, T )(cid:105), (29) where the integral is over all allowed q (the Brillouin zone in the case of a full dispersion relation). The expression for the average energy in the volume is the same except that a factor of ωi (q) is included in the integral. The above expression can be simplified for an isotropic dispersion relation: 5Eq. (29) is similar to Eq. (9) except that that latter uses the DOS instead of an integral over reciprocal space. Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Contacts 25 Nb = V (cid:90) 2π (cid:90) π (cid:90) qmax 0 0 (cid:90) qmax 0 (cid:104)nBE(ωb,q, T )(cid:105) q2 sin θdqdθdφ (2π)3 , (30) 0 = (cid:104)nBE(ωb,q, T )(cid:105)q2dq, V 2π2 where q = q. Nb will not be an integer in general. So, add (cid:98)Nb(cid:99) phonons and then randomly, with a probability of Nb − (cid:98)Nb(cid:99), add one more phonon. For example, if Nb = 1192.63, then always add 1192 phonons and add an additional phonon 63% of the time. In this case, it might seem that the additional phonon probability can be done without, but incorrectly generating even one excess phonon per time step per cell may lead to a failure of the simulation (see Sec. 7). 6.2 3D Boundary Contacts The advantage of boundary contacts is that, instead of having to generate all the phonons inside a volume every time step, you only need to generate the phonons which would have drifted into the simulation domain from a reservoir outside the simulation domain. This effectively makes a boundary contact a blackbody that emits phonons into the simulation domain, so we can use results for the theory of blackbody radiation [Modest (2003)]. Boundary contacts have much in common with diffuse scattering (Sec. 5). The expectation number of phonons of branch b and wave vector q passing through a surface per unit area and time is nb (q, T ) = vg,b(q) cos θ(cid:104)nBE(ωb,q, T )(cid:105) d3q (2π)3 , (31) where θ is the angle between the group velocity vg,b(q) and the surface normal. Because we only are interested in the number of phonons entering the domain, we only consider θ ∈ [0, π/2]. We now turn to the case of an isotropic dispersion relation with qmax ≥ q as an upper bound on the wave-vector magnitude. Then nb (q, T ) can be simplified to [Randrianalisoa and Baillis (2008)] Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 26 Phonon Monte Carlo (cid:90) π/2 (cid:90) 2π 0 0 nb (q, T ) = vg,b (q) cos θ(cid:104)nBE(ωb,q, T )(cid:105) q2 sin θdqdθdφ (2π)3 , = vg,b (q)(cid:104)nBE(ωb,q, T )(cid:105) q2 8π2 dq. (32) Then the total number of phonons entering the simulation domain due to a blackbody of area A in a period of time ∆t is (cid:90) qmax (cid:88) vg,b (q) cos θ(cid:104)nBE(ωb,q, T )(cid:105) q2 8π2 dq. N 3D bc (T ) = A∆t 0 b (33) The cosine in Eq. (31) means that phonons entering the simulation will have the angular distribution from Lambert's cosine law, so we can generate random directions for the incoming phonons using Eq. (4). In principle, the phonons from the boundary contact will enter the simulation domains at different times, but in practice, all the phonons can be added to the simulation domain at the same time because ∆t is small. Then, the overall procedure is to calculate N 3D bc (Th) and N 3D bc (Tc) for the hot and cold contacts, which are at temperatures Th and Tc respectively. Then at each time step, any phonons that drift outside the simulation domain are deleted and N 3D bc (Th) and N 3D bc (Tc) equilibrium phonons are created (Sec. 4.2) at the two contacts, with angular distributions given by Lambert's cosine law, Eq. (4). 6.3 2D Contacts Two-dimensional internal contacts are essentially the same as their 3D counterpart, except we use the cell area instead of the volume and the integration is over a 2D wave vector q. When using internal contacts, it does not matter what kind of dispersion we use, because we just initialize the cell to the reservoir energy. Two-dimensional boundary contacts need more in implementation. In 2D, to mimic a reservoir from the outside, we need to delete phonons that drift outside of the end-cell boundary care Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Energy Conservation 27 (a line) and inject phonons into the cell as if they come from an outside reservoir fixed at temperature T . For isotropic dispersion, it is easy to work out the expectation number of phonon to be injected per unit length, per branch, per unit wave number, and per unit time as6 nb(q, T ) = 1 4π (cid:104)nBE(ωc,j, T )(cid:105) q 2π2 vg,b(q), (34) where vg,b(q) is the group velocity for phonon with wave number q and branch b. The total expectation number per time step is then N 2D bc = H∆t nb(q)dq, (35) (cid:90) qmax (cid:88) 0 b (cid:82) θ (cid:82) π 0 cos θ(cid:48)dθ(cid:48) 0 cos θ(cid:48)dθ(cid:48) = sin θ. where ∆t is the time step and H is the height of the simulation bc (cid:99) phonons each step domain. Like in 3D, we only generate (cid:98)N 2D bc (cid:99)). and add an additional phonon with a chance of (N 2D The CDF of angle in 2D is bc − (cid:98)N 2D F (θ) = (36) The cos θ(cid:48) term comes from the fact that phonons come in at a rate proportional to the group velocity perpendicular to the boundary. Then inversion method can be used to choose the angle as θ = arcsin(rθ), just like for 2D diffuse scattering, Eq. (23). When H is small, which is typically the case for PMC for quasi-one- dimensional graphene ribbons, the expectation number of injected particles is too small (< 10) to represent the distribution, even though the distribution would be correct if we drew a large number of phonons. As a result, the 2D boundary contacts have stability issues and the simulation may take a longer time to converge than the internal-contact implementation. 7 Energy Conservation If we use the rejection technique to generate a phonon in a cell for PMC simulation, the phonon energy is independent of how much energy is already in the cell. As a result, it is 6Compare with the 3D equivalent in Eq. (32). Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 28 Phonon Monte Carlo impossible to infuse the cell with exactly the expected energy. It is tempting to force the last phonon's energy to achieve exact energy conservation. However, by doing so, we are introducing excess phonons with energies smaller than average, i.e., more of them than the equilibrium distribution would predict. Owing to the low energies, these phonons have large group velocities and small scattering rates, so they carry their energy out of the simulation domain almost ballistically and lead to energy depletion inside the structure. Even generating a single phonon from the wrong distribution in every cell once a time step can add up and cause the simulation to fail dramatically. To respect the phonon energy distribution, we allow the energy inside a cell to be uncertain up to half the maximal energy of a single phonon: we consider the desired cell energy to be reached with satisfactory accuracy if the cell energy falls within (cid:3), where E is the exact desired cell energy (cid:2)E − ωmax and ωmax is the maximum energy carried by a phonon in the material. , E + ωmax 2 2 Furthermore, the treatment of inelastic scattering does not conserve energy precisely at each time step, but only in an average sense [Mei et al. (2014)]. To better conserve the energy after scattering, each cell is reinitialized after each time step by adding/deleting phonons from the equilibrium distribution. The accumulation of offset energy might cause a problem, so we record the offset energy Eoffset = E − Eactual at each step and add it to the desired energy at the next reinitialization. During drift, phonons might cross cell boundaries and therefore change the total energy in each cell. The energy of the phonons that drifted in (out) has to be added (subtracted) before entering the scattering routine. As a result, we record the cell energy just after the phonons finish drift in a prescattering array, Eprescat. Together with the array Ei,offset, which records the offset energy from our last attempt at enforcing energy conservation, we calculate the desired energy after scattering as Ed = Eprescat + Eoffset. Then we can scatter the phonons and calculate the actual cell energy after scattering, stored in an after-scattering array, Ei,afterscat. Again, we only enforce that the cell energy E get into the range [Ed − ωmax ], where Ed is the desired energy. The initial cell , Ed + ωmax 2 2 Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Conclusion 29 energy in the reinitialization process is Ei = Ei,afterscat. For each cell i, we compare Ei and Ei,d. (1) If Ei ∈ [Ei,d − ωmax (2) If Ei < Ei,d − ωmax and move on to the next one. ], we consider this cell good , Ei,d + ωmax 2 2 2 i = Eold , we generate a phonon from the equilibrium distribution and add it to a random place in the i + ω0, cell. The new cell energy after addition is Enew where ω0 is the energy carried by the added phonon. Keep adding phonons until the final Ei falls in the appropriate range. , we would randomly choose one phonon in the cell and delete it. 7 After the deletion, the new energy in i − ω0 where ω0 is the energy carried the cell is Enew by the deleted phonon. Then we compare Enew and Ei,d again i and keep this random deletion until Enew falls in the desired range. i = Eold ωmax 2 (3) If Ei > Ei,d + i After all cells have energies in the desired range, we record the new offset energy as Eoffset = Ed − Ei and use it in the next reinitialization process (after the next time step). 8 Conclusion PMC is a versatile stochasic technique for solving the Boltzmann equation for phonons in structures that can have real-space sizes. We presented roughness and experimentally relevant the relative merits of for generating random variates to represent the random variables with nonuniform distributions, which are relevant in thermal transport: generating the attributes for phonons in equilibrium with full and isotropic dispersions, randomizing outgoing momentum upon rejection techniques inversion vs 7It is very important that the phonon be chosen randomly. For example, simply deleting the oldest phonon will slowly skew the phonon distribution, because the oldest phonon is likely to be a phonon with a low scattering rate. Depending on the data structure used for storing phonons, deleting algorithms may vary. One can use a random number to choose the index of phonon to be deleted if the structure supports easy random access. For structures like the linked list, one can use the Fisher-Yates shuffle algorithm to generate a random permutation of the array and delete in sequence. Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 30 Phonon Monte Carlo implementing contacts (boundary diffuse boundary scattering, and internal), and conserving energy in the simulation. We also identified common themes in phonon generation and scattering that are helpful for reusing code in the simulation (generating thermal phonon attributes vs internal contacts, diffuse surface scattering vs boundary contacts). We hope these examples will inform the reader about both the mechanics of random variate generation and choosing a good approach for whatever problem is at hand, and aid in the more widespread use of PMC for thermal transport simulation. 9 Acknowledgement The authors gratefully acknowledge support by the U.S. Department of Energy, Office of Basic Energy Sciences, Division of Materials Sciences and Engineering under Award DE-SC0008712. This work was performed using the compute resources and assistance of the UW-Madison Center for High Throughput Computing (CHTC) in the Department of Computer Sciences. Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Bibliography Aksamija, Z. and Knezevic, I. (2010). Anisotropy and boundary scatter- ing in the lattice thermal conductivity of silicon nanomembranes, Phys. Rev. B 82, p. 045319. Bera, C. (2012). Monte Carlo simulation of thermal conductivity of Si nanowire: An investigation on the phonon confinement effect on the thermal transport, J. Appl. Phys. 112, p. 074323. Berman, R., Foster, E. L. and Ziman, J. M. (1955). Thermal Conduction in Artificial Sapphire Crystals at Low Temperatures. I. Nearly Perfect Crystals, Proc. R. Soc. A 231, 1184, pp. 130 -- 144, doi:10.1098/rspa.1955.0161, URL http://rspa. royalsocietypublishing.org/content/231/1184/130. Berman, R., Simon, F. E. and Ziman, J. M. (1953). The Thermal Conductivity of Diamond at Low Temperatures, Proc. R. Soc. A 220, 1141, pp. 171 -- 183, doi:10.1098/rspa.1953.0180, URL http: //rspa.royalsocietypublishing.org/content/220/1141/171. Broido, D. A., Malorny, M., Birner, G., Mingo, N. and Stewart, D. A. (2007). Intrinsic lattice thermal conductivity of semiconductors from first principles, Appl. Phys. Lett. 91, p. 231922. Couchman, L., Ott, E. and Antonsen, T. M. (1992). Quantum chaos in systems with ray splitting, Phys. Rev. A 46, pp. 6193 -- 6210, doi:10.1103/PhysRevA.46.6193, URL http://link.aps. org/doi/10.1103/PhysRevA.46.6193. Devroye, L. (1986). Non-Uniform Random Variate Generation (Springer-Verlag), ISBN 0387963057. Esfarjani, K., Chen, G. and Stokes, H. T. (2011). Heat transport in silicon from first-principles calculations, Phys. Rev. B 84, p. 085204. Glassbrenner, C. J. and Slack, G. A. (1964). Thermal conductivity of silicon and germanium from 3ok to the melting point, Phys. Rev. 134, p. A1058. Lacroix, D., Joulain, K. and Lemonnier, D. (2005). Monte Carlo transient phonon transport in silicon and germanium at nanoscales, Phys. Rev. B 72, p. 064305, doi:10.1103/PhysRevB. 72.064305, URL http://link.aps.org/doi/10.1103/PhysRevB. Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC 32 Bibliography 72.064305. Lacroix, D., Joulain, K., Terris, D. and Lemonnier, D. (2006). Monte Carlo simulation of phonon confinement in silicon nanostructures: Application to the determination of the thermal conductivity of silicon nanowires, Appl. Phys. Lett. 89, 10, p. 103104, doi: 10.1063/1.2345598, URL http://scitation.aip.org/content/ aip/journal/apl/89/10/10.1063/1.2345598. Lim, J., Hippalgaonkar, K., Andrews, S. C., Majumdar, A. and Yang, P. (2012). Quantifying Surface Roughness Effects on Phonon Transport in Silicon Nanowires, Nano Lett. 12, 5, pp. 2475 -- 2482, doi:10.1021/nl3005868, URL 10.1021/nl3005868. Maurer, L. N., Aksamija, Z., Ramayya, E. B., Davoody, A. H. and Knezevic, I. (2015). Universal features of phonon transport in nanowires with correlated surface roughness, Appl. Phys. Lett. 106, 13, 133108, doi:10.1063/1.4916962, URL http://scitation.aip. org/content/aip/journal/apl/106/13/10.1063/1.4916962. Mazumder, S. and Majumdar, A. (2001). Monte Carlo Study of Phonon Transport in Solid Thin Films Including Dispersion and Polarization, J. Heat Transfer 123, 4, p. 749, doi:10.1115/1. 1377018, URL http://heattransfer.asmedigitalcollection. asme.org/article.aspx?articleid=1445087. Mei, S. and Knezevic, I. (2015). Thermal conductivity of iii-v semicon- ductor superlattices, J. Appl. Phys. 118, 17, 175101, doi:http:// dx.doi.org/10.1063/1.4935142, URL http://scitation.aip.org/ content/aip/journal/jap/118/17/10.1063/1.4935142. Mei, S., Maurer, L. N., Aksamija, Z. and Knezevic, I. (2014). Full- dispersion Monte Carlo simulation of phonon transport in micron- sized graphene nanoribbons, Journal of Applied Physics 116, 16, 164307, doi:10.1063/1.4899235, URL http://scitation.aip. org/content/aip/journal/jap/116/16/10.1063/1.4899235. Modest, M. F. (2003). Radiative Heat Transfer (Academic Press), ISBN 0125031637. Nika, D. L. and Balandin, A. A. (2012). Two-dimensional phonon transport in graphene. J. Phys. Condens. Matter 24, 23, p. 233203, doi:10.1088/0953-8984/24/23/233203, URL http://www. ncbi.nlm.nih.gov/pubmed/22562955. Northrop, G. A. and Wolfe, J. P. (1984). Phonon Reflection Imaging: A Determination of Specular versus Diffuse Boundary Scattering, Phys. Rev. Lett. 52, pp. 2156 -- 2159, doi:10.1103/PhysRevLett.52. 2156, URL http://link.aps.org/doi/10.1103/PhysRevLett. 52.2156. P´eraud, J.-P. M., Landon, C. and Hadjiconstantinou, N. (2014). Monte Carlo methods for solving the Boltzmann transport equation, Book chapter, to appear in Nanophononics, Ed. Z. Aksamija, Pan Stanford Publishing, 2016 MaurerPMC Bibliography 33 Annual Review of Heat Transfer 17, pp. 205 -- 265. Peterson, R. B. (1994). Direct simulation of phonon-mediated heat transfer in a debye crystal, J. Heat Transfer 116, 4, pp. 815 -- 822, doi:10.1115/1.2911452, URL http://heattransfer.asmedigitalcollection.asme.org/ article.aspx?articleid=1441842. Ramayya, E. B., Maurer, L. N., Davoody, A. H. and Knezevic, I. (2012). Thermoelectric properties of ultrathin silicon nanowires, Phys. Rev. B 86, p. 115328, doi:10.1103/PhysRevB.86.115328, URL http://link.aps.org/doi/10.1103/PhysRevB.86.115328. Ramiere, A., Volz, S. and Amrit, J. (2016). Geometrical tuning of thermal phonon spectrum in nanoribbons, J. Phys. D: Appl. Phys. 49, 11, p. 115306, URL http://stacks.iop.org/0022-3727/49/ i=11/a=115306. Randrianalisoa, J. and Baillis, D. (2008). Monte Carlo simulation of steady-state microscale phonon heat transport, J. Heat Transfer 130, doi:10.1115/1.2897925, URL http://heattransfer.asmedigitalcollection.asme.org/ article.aspx?articleid=1449183. 072404, p. Soffer, S. B. (1967). Statistical Model for the Size Effect in Electrical Conduction, J. of Appl. Phys. 38, 4, pp. 1710 -- 1715. Ziman, J. M. (1960). Electrons and Phonons: the Theory of Transport Phenomena in Solids (Clarendon Press), ISBN 978-0198507796.
1807.03411	2	1807	2018-07-12T15:34:28	Probing a spin transfer controlled magnetic nanowire with a single nitrogen-vacancy spin in bulk diamond	[ "cond-mat.mes-hall" ]	The point-like nature and exquisite magnetic field sensitivity of the nitrogen vacancy (NV) center in diamond can provide information about the inner workings of magnetic nanocircuits in complement with traditional transport techniques. Here we use a single NV in bulk diamond to probe the stray field of a ferromagnetic nanowire controlled by spin transfer (ST) torques. We first report an unambiguous measurement of ST tuned, parametrically driven, large-amplitude magnetic oscillations. At the same time, we demonstrate that such magnetic oscillations alone can directly drive NV spin transitions, providing a potential new means of control. Finally, we use the NV as a local noise thermometer, observing strong ST damping of the stray field noise, consistent with magnetic cooling from room temperature to $\sim$150 K.	cond-mat.mes-hall	cond-mat	Probing a spin transfer controlled magnetic nanowire with a single nitrogen-vacancy spin in bulk diamond Adrian Solyom†,1 Zackary Flansberry†,1 Marta A. Tschudin,1 Nathaniel Leitao,1 Michel Pioro-Ladri`ere,2, 3 Jack C. Sankey,1 and Lilian I. Childress∗1 2Universit´e de Sherbrooke Institut Quantique and D´epartement de Physique 3Quantum Information Science Program, Canadian Institute for Advanced Research 1McGill University Department of Physics (Dated: July 13, 2018) The point-like nature and exquisite magnetic field sensitivity of the nitrogen vacancy (NV) center in diamond can provide information about the inner workings of magnetic nanocircuits in comple- ment with traditional transport techniques. Here we use a single NV in bulk diamond to probe the stray field of a ferromagnetic nanowire controlled by spin transfer (ST) torques. We first report an unambiguous measurement of ST tuned, parametrically driven, large-amplitude magnetic oscilla- tions. At the same time, we demonstrate that such magnetic oscillations alone can directly drive NV spin transitions, providing a potential new means of control. Finally, we use the NV as a local noise thermometer, observing strong ST damping of the stray field noise, consistent with magnetic cooling from room temperature to ∼150 K. In nanoscale magnetic circuits, spin transfer (ST) ef- fects provide an efficient means of all-electronic con- trol, prompting the development of ST-based non-volatile memories, microwave-frequency oscillators, filters, detec- tors, and amplifiers [1, 2]. These developments call for new tools to understand the interplay between spins, magnons, and the environment. In parallel, the nitrogen- vacancy (NV) center in diamond [3] has emerged as a versatile sensor for studying magnetic systems, with ex- cellent spatial and spectral resolution [4, 5]. The tiny magnetic moment of a single spin provides a non-invasive probe of local stray fields, and can be combined with scanned-probe [6] or subwavelength imaging [7] tech- niques to achieve nanoscale spatial resolution. Moreover, the NV offers a variety of sensing modalities appropri- ate for DC and AC magnetic fields [8, 9] or noise spec- troscopy [10–12]. Recently, NV centers have probed fer- romagnetic phenomena including vortex cores [13, 14], domain walls [15], ferromagnetic resonance (FMR) [16– 20], and magnetic thermal quantities in an extended YIG film [20] modified by ST effects. Here, we probe a metallic ST-controlled ferromagnetic nanowire with a single NV in bulk diamond. Differ- ent from microwave-frequency magnetoresistance read- out [21–23], the NV's point-like nature enables coupling to shorter-wavelength spin waves, thereby revealing qual- itatively new features. First, we provide transport-based evidence of parametrically-driven magnetic oscillations – tuned through threshold with spin transfer anti-damping – and use the resulting (phase-locked) stray field oscilla- tions alone to drive the NV spin resonance. This demon- †These authors contributed equally. ∗[email protected] Figure 1. Device geometry. (a) Cross-section of nanowire on a diamond substrate with implanted NVs. Nanowire current (red arrow along x) drives spins (green circle) polarized along y into the Py layer (red arrow along z). (b) Confocal PL map of the device. The nanowire (dotted box) appears as a shadow. Circle indicates the NV under study. stration removes any ambiguity from the interpretation of the transport measurements, and provides a potential new means of NV control. Additionally, transport read- out indicates a precession angle up to ∼55◦, suggesting the absence of Suhl-like (multi-magnon) instabilities [24], which we tentatively attribute to the reduced density of states in our confined geometry. The observed changes in field at the NV independently corroborate the estimated angle, and (more importantly) we observe no signs of NV spin flips in the PL spectra when directly driving FMR – a feature normally appearing in extended mag- netic films [16–20] – providing additional evidence of sup- pressed multi-magnon instabilities. This highlights the NV's potential use as a tool for understanding internal workings of magnetic nanosystems. Finally, we demon- strate strong ST control of the magnetic thermal fluc- tuations: adapting the NV noise relaxometry method of Ref. [20], we observe large ST damping of the stray field noise, providing evidence of magnetic cooling from room arXiv:1807.03411v2 [cond-mat.mes-hall] 12 Jul 2018 PL (kcounts/sec) 80 40 NVs 2 μm Contact Au Ti Pt Py NVs a b Contact 160 120 2 Figure 2. Transport FMR and parametric response. (a) Voltage ∆VMR spectra (vertically separated for clarity) for bias I0 = 0, RF amplitude IRF = 0.95 mA, and magnetic field µ0H0 = 16.5 mT applied along the NV axis (orange) or rotated about z by θ = 5◦ (blue). Red points (y-values divided by 10) correspond to I0 = 4.08 mA and θ = 0◦ (taken at µ0H0 = 23.5 mT), showing a large resonance near twice the FMR frequency νFMR. (b) and (c) show the frequency and field dependence of ∆VMR for θ = 5◦ and 0◦, with symbols denoting fit values of νFMR. (d) Same as (c), but with I0 = 4.08 mA. Vertical lines in (b)-(d) correspond to the traces plotted in (a). temperature to ∼150 K. The observed noise suppression is orders of magnitude larger than measured with an NV near a YIG film [20], and a factor of ∼2 larger than mea- sured with Brillouin light scattering from a related metal- lic structure [25]. An interesting open question is the fundamental limits of this cooling technique, in particu- lar if and from what temperature the ground state can be reached. Interestingly, our demonstrated NV sensitivity suggests it may be possible to resolve magnetic zero-point fluctuations, thereby providing an optically active "han- dle" on a macroscopic magnet in the quantum regime. These techniques, which are especially advantageous in systems having low magnetoresistance, pave the way to a deeper understanding of short-wavelength magnons, ST effects, and quantum emitter control. DEVICE GEOMETRY We fabricate [26] an 8-µm-long × 417-nm-wide Py (Ni81Fe19, 10 nm) / Pt (10 nm) multilayer nanowire on electronic grade diamond with a layer of NVs implanted 60±15 nm below the surface, as shown in Fig. 1(a). Fig- ure 1(b) shows a photoluminescence (PL) image of the device pumped (532 nm) and collected (>594 nm) from above. Bright spots are NVs, the nanowire appears dark, and the contacts (Au) exhibit typical background PL. We focus on the indicated spot, with PL primarily from sin- gle NV whose symmetry axis lies 35◦ from the y-axis in the yz plane [26]. Static magnetic fields are applied with a permanent neodymium magnet on a motorized stage, calibrated using the NV spin transitions [26]. For this device, a moderate field (10 − 40 mT) applied along the NV symmetry axis saturates the Py magnetization along y, canted at most ∼3◦ out of the xy plane (dictated by shape anisotropy [26]). All measurements are taken at room temperature. TRANSPORT CHARACTERIZATION OF MAGNETIC RESONANCES To verify the device's functionality, we perform ferro- magnetic resonance (FMR) by applying current I(t) = I0 + IRF cos 2πνNWt through the nanowire (bias I0, am- plitude IRF, frequency νNW), and reading the anisotropic magnetoresistance (AMR) response via a generated DC voltage ∆VMR [26, 27]. Due to the spin Hall effect, the electrical current I(t) drives a pure spin current (po- larized along y) into the Py layer, applying a torque ∂t m k y on the Py magnetization's unit vector m = mxx + my y + mzz. The field ~HI k y generated by this current applies a torque along z. A "typical" ST-FMR spectrum for field µ0H0 = 16.5 mT applied along the NV axis but rotated θ = 5◦ about z is shown in Fig. 2(a) (blue circles), with I0 = 0, and IRF = 0.95 mA. The resonant feature at 1.6 GHz is well fit by the expected Fano-like lineshape [26, 27], allowing us to extract the resonant frequency νFMR and linewidth ∆ν. Figure 2(b) shows the field dependence of these spectra (color scale) and the fit values of νFMR (points). The initial decrease in frequency corresponds to the equilibrium orientation h mi shifting from x (where shape anisotropy provides νFMR ∼ 1 GHz) and saturat- ing along ≈ y, where the anisotropy field opposes the applied field. When θ = 0◦ (Fig. 2(a), orange squares), this FMR signal vanishes (as expected above the satu- ration field), since both the drive torques and AMR re- sponse also vanish to lowest order in mx and mz [26]. Figure 2(c) shows the field dependence of these θ = 0◦ spectra, along with νFMR estimated from values at θ 6= 0 [26]. Above saturation, νFMR is well fit by the Kittel for- In-plane field (mT) c d ΔVMR (μV) ΔVMR (μV) ΔVMR (μV) b Frequency, νNW (GHz) Fit 5°, 0 mA 0°, 0 mA 0°, 4.08 mA (×0.1) a ΔVMR (μV) Frequency, νNW (GHz) Applied field, μ0H0 (mT) 3 Figure 3. NV detection of parametrically-driven FMR. (a) Top: Photoluminescence (PL) versus stripline (nanowire) frequency νSL (νNW) with field µ0H0 = 22.5 mT along the NV axis, and nanowire currents IRF = 1.15 ± 0.08 mA and I0 = 4.9 mA. PL decreases near the ESR frequency ν− = 2.19 GHz, and ν− shifts due to a reduced axial stray field. Data are normalized to off-resonant levels, which is why the contrast vanishes when νNW matches ESR frequencies (i.e., near 2.19 GHz, 3.59 GHz) or their harmonics (4.38 GHz). Bottom: Transport FMR readout ∆VMR under the same conditions. (b) Same PL measurement as (a), but with no stripline power, µ0H0 = 29.6 mT, and varied I0, driven near 2ν−. Solid curves represent Lorentzian fits used to determine the ESR contrast and linewidth. (c) ESR contrast (top) and linewidth (bottom) for a wider range of currents. Below I0 = 4.3 mA, data are consistent with zero contrast (see (b)). mula for spatially uniform m, providing effective in-plane (out-of-plane) coercive fields µ0Hyx = 7.57 ± 0.08 mT (µ0Hzx = 517 ± 4 mT) [26]. Figure 2(d) shows the same θ = 0◦ measurement with I0 = 4.08 mA. This applies a steady torque along −y that anti-damps the magnetization (HI = −6 mT also shifts saturation to higher fields). Importantly, a significantly larger, asymmetric peak appears near twice the expected FMR frequency (see Fig. 2(a), red symbols), suggesting parametrically driven, large-amplitude oscillations. Sim- ple macrospin simulations [26] reproduce the sign, mag- nitude, frequency, and line shape of this feature semi- quantitatively, and we interpret it as arising from a large- angle elliptical precession [21] of the spatially-averaged magnetization. Within this macrospin approximation, the largest-amplitude signal (120 µV) in Fig. 2(a) cor- responds to an in-plane (out-of-plane) precession angle ≈ 30◦ (6◦). NV RESPONSE TO PARAMETRIC OSCILLATIONS This picture is validated by the NV's PL spectrum shown in Fig. 3. First, while driving the nanowire at fre- quency νNW, we apply a π-pulse with a nearby stripline to determine the NV's lower electron spin resonance (ESR) frequency. Figure 3(a) shows the PL response (top) and ∆VMR (bottom) for field µ0H0 = 22.5 mT along the NV axis. Reductions in PL are associated with driving the NV spin from its ms = 0 spin projection to ms = ±1, occurring at resonant frequencies ν±. Notably, the fre- quency ν− of the stripline-driven transition shifts by up to 13 MHz when m is driven to large amplitude, and follows a path qualitatively similar to ∆VMR. The max- imum shift in ν− corresponds to a decrease in the NV- axial magnetic field of 0.5 mT, as expected for the re- duced average magnetization. This is a large fraction of the estimated 2.4 mT total stray field [26], corresponding to an in-plane precession angle ≈ 60◦ that is within 10% of the value (55◦) estimated from ∆VMR in Fig. 3(a). More compellingly, we can drive the NV transition us- ing only the Py layer's stray field. Figure 3(b) shows a similar set of measurements in the absence of stripline current, with µ0H0 tuned to 29.6 mT, such that νFMR ≈ ν−. Above I0 = 4.3 mA, a PL dip (corresponding to field-driven transitions) appears at twice ν−, as expected for parametric oscillations at half the drive frequency. Figure 3(c) shows the contrast and linewidth of this PL dip for a wider bias range, illustrating a sharp threshold at I0 = 4.3 mA. This measured PL response provides unambiguous evidence of ST anti-damping and large- amplitude, parametrically driven oscillations. We note that details such as the precise threshold current and shapes of the bias dependences in (c), depend somewhat on ~HI, which slightly tunes the NV and FMR frequen- cies. In Fig. 4, we map out the parametrically driven phase space of (a) ∆VMR, and (b) PL under the same condi- tions as Fig. 3. For this larger drive, new features ap- pear above the primary parametric ∆VMR peak, which we tentatively attribute to higher order spin-wave modes [29]. Of interest, the Py-driven PL feature near 2ν− from Fig. 3(b) extends over a wide range of fields, consistent with the presence of these higher-frequency transport fea- Nanowire frequency, νNW (GHz) Nanowire frequency, νNW (GHz) Current, I0 (mA) c Contrast (%) Linewidth (MHz) 4.03 mA 4.20 mA 4.36 mA 4.52 mA 4.69 mA b PL (normalized) PL (normalized) 1.0 0.8 0.9 a νSL (GHz) ΔVMR (μV) 4 25]. Stated briefly, the noise spectral density S⊥ of the stray field perpendicular to the NV axis (units of T2/Hz) should have the form S⊥(ν, H0) =Xk ¯ n(νk(H0))Pk(ν, H0)fk, (1) where ¯n(νk) is the thermal occupancy of the spin wave mode having wavenumber k and frequency νk, fk is a constant converting magnon number to field noise power, and Pk(ν, H0) is a density function describing how this power is spread over the frequency domain (peaked at νk). In the presence of S⊥, the NV spin relaxation rate Γ(ν) increases from its nominal value Γ0 ∼ 60 Hz [26] to a value Γ(ν) = Γ0 + γ2 NV 2 S⊥, (2) where γNV is the gyromagnetic ratio of the NV spin. Fig- ure 5(a) shows the measured [26] relaxation rates Γ± for varied field H0, along with ν± and νFMR below. Quali- tatively similarly to extended YIG films [20], the largest relaxation occurs when when ν± is just above νFMR, with peak value determined by the balance of ¯n, Pk, fk (which takes the largest value when 2π/k is comparable to the NV-wire distance), and the density of spin wave states. In contrast to Ref. [20], spin-transfer damping (and an- tidamping) in this confined metallic system should com- pletely dominate over Joule heating (the nanowire tem- perature changes at most by ∼5 K [26]), enabling strong modification of the thermal fluctuations. Figure 5(b) shows how Γ± varies with I0 (while simultaneously com- pensating ~HI to fix νFMR); the weaker I0-generated field at the NV leads to modest shifts of ν± in Fig. 5(d). Both relaxation rates significantly decrease for negative cur- rent (damping), indicating a strong suppression of stray field noise, and increase for positive current (antidamp- ing). Naively assuming these changes arise entirely from ¯ n(ν±), and that the action of damping is to modify the effective magnetic temperature Teff [20, 25], the observed change in Γ− corresponds to ST cooling to Teff ∼ 150 K at I0 = −4.08 mA. However, the probe frequencies ν± drift with I0 and the ST damping broadens the peak in S⊥. A model combining these effects with the observed field- dependence in Fig. 5(a) [26] produces the "spin transfer free" relaxation rate estimates Γ0 shown in Fig. 5(b) and ± the color scale in the lower panels of (a) and (b). Impor- tantly, the expected changes are much smaller (and have the opposite trend), suggesting that our estimate of Teff might represent an upper bound. We note that, in this geometry, we could not extract trustworthy parameters from FMR at maximum damping, highlighting the utility of the NV as an independent probe. Figure 4. Phase space for parametrically driven (a) ∆VMR and (b) PL at I0 = 4.9 mA and IRF = 1.15 mA. The four promi- nent PL features correspond to current-driven ESR transi- tions of the ground (ν±) and excited (ν±,ES) states. Inset of (b) shows the Py-driven feature at 2ν−. Red curves indicate the expected FMR frequency νFMR. tures. At this larger drive, we observe a residual directly- driven precession amplitude up to ∼ 14◦ at νFMR in the transport data (Fig. 4(a)). Despite this, we observe no evidence of spin flips at νFMR in the PL spectra of Fig. 4(b), contrasting what is ubiquitously observed near extended films [16–20]. We tentatively attribute this to the reduced density of states in our confined geometry, which acts to suppress multi-magnon up-conversion pro- cesses. Finally, as shown in the inset, the Py-driven spin reso- nance splits by up to ∼60 MHz at some fields. The nature of this splitting will be the subject of future work, but we suspect it is due to the presence of quasistable mag- netic modes [30, 31] having different average stray fields. Notably, information about these dynamics are not ap- parent in (a), highlighting this technique's potential to provide qualitatively new information. STRONGLY DAMPED MAGNETIZATION The associated NV spin relaxation rates Γ± can also probe the local field noise, providing some information about the thermal occupancy of spin wave modes [20, ΔVMR (μV) 00 01 300 200 -100 -200 PL (normalized) 0.9 0.8 1.0 In-plane field (mT) 2ν- ν+ ν- νFMR Applied field, μ0H0 (mT) In-plane field (mT) ν+ ν+,ES ν- ν-,ES 25 30 νFMR Applied field, μ0H0 (mT) a Frequency, νNW (GHz) b Frequency, νNW (GHz) 4.5 4.0 5 sions and related work. JS, LC, and M-PL acknowl- edge support from Fonds de Recherche - Nature et Tech- nologies (FRQNT) PR-181274. LC acknowledges fund- ing support from Canada Foundation for Innovation and Canada Research Chairs (CRC) project 229003, National Sciences and Engineering Research Council of Canada (NSERC) RGPIN 435554-13, and l'Institut Transdisci- plinaire d'Information Quantique (INTRIQ). JS acknowl- edges support from CRC project 228130 & 231509. LC and JS acknowledge Benjamin Childress and Calvin Chil- dress for all sleep-deprived errors in judgment, including this acknowledgment. Figure 5. Spin relaxometry under bias. (a) Measured re- laxation rates Γ± versus applied field, with bias I0=0. Also shown is the internal rates Γ0 ±. (b) Measured Γ± for varied I0 (solid lines), adjusting µ0H0 (nominally 16.5 mT at I0 = 0, white line in (c)) to ensure νFMR remains constant to within the shown error (shading on red line in (d)). Dashed lines are estimated rates Γ0 ± in the absence of ST effects, with absolute bounds shown as dotted lines. (c)-(d) Comparison of ν± and νFMR (curves) with reconstructed estimate of Γ0 (color plots) [26]. CONCLUSIONS The NV PL spectrum provides qualitatively dif- ferent information about magnetic nanocircuits than is provided by traditional transport measurements. We demonstrated parametric magnetic oscillations con- trolled by ST effects, and that ESR transitions can be driven entirely by a local magnetization. The latter tech- nique may potentially facilitate fast, low-power control of NV spins, and the nonlinear dynamics involved could, e.g., be used to amplify incident fields [14]. We also ob- serve large changes in the NV spin relaxation time when subjecting the magnetic element to ST damping, consis- tent with magnon cooling to well below ambient tem- peratures. Finally, we note that the demonstrated stray field coupling between the NV and spin wave modes is quite large. Of note, Fig. 5(a) exhibits a maximal Γ at 15.5 mT, where ν− corresponds to a room-temperature magnon occupancy ∼2500, meaning the noise S⊥,ZPF generated by the ferromagnetic zero-point fluctuations would set Γ ∼1 Hz. This value is comparable to the intrinsic relaxation rates of bulk NV's at low tempera- ture [32], suggesting it might be feasible to resolve the quantum fluctuations of a magnetic nanocircuit. ACKNOWLEDGMENTS We thank Luke Hacquebard, Quentin Pognan, Isabelle Racicot, Souvik Biswas, and Tsvetelin Totev for discus- [1] D. Ralph and M. Stiles, "Spin transfer torques," J. Mag. & Mag. Mat. 320, 1190 (2008). [2] A. Manchon, I. M. Miron, T. Jungwirth, J. Sinova, J. Zelezn´y, A. Thiaville, K. Garello, and P. Gambardella, "Current-induced spin-orbit torques in ferromagnetic and antiferromagnetic systems," arXiv:1801.09636 (2018). [3] M. W. Doherty, N. B. Manson, P. Delaney, F. Jelezko, J. Wrachtrup, and L. C. Hollenberg, "The nitrogen- vacancy colour centre in diamond," Phys. Rep. 528, 1 (2013). [4] L. Rondin, J.-P. Tetienne, T. Hingant, J.-F. Roch, and V. Jacques, "Magnetometry with in P. Maletinsky, nitrogen-vacancy defects in diamond," Rep. Prog. Phys. 77, 056503 (2014). [5] F. Casola, T. van der Sar, and A. Yacoby, "Probing condensed matter physics with magnetometry based on nitrogen-vacancy centres in diamond," Nature Rev. Mat. 3, 17088 (2018). [6] P. Maletinsky, S. Hong, M. S. Grinolds, B. Hausmann, M. D. Lukin, R. L. Walsworth, M. Loncar, and A. Ya- coby, "A robust scanning diamond sensor for nanoscale imaging with single nitrogen-vacancy centres," Nature Nano. 7, 320 (2012). [7] E. Rittweger, K. Han, S. Irvine, C. Eggeling, and S. W. Hell, "STED microscopy reveals crystal colour centres with nanometric resolution," Nature Phot. 3, 144 (2009). [8] C. L. Degen, "Scanning magnetic field microscope with a diamond single-spin sensor," Appl. Phys. Lett. 92, 243111 (2008). [9] J. M. Taylor, P. Cappellaro, L. Childress, L. Jiang, D. Budker, P. R. Hemmer, A. Yacoby, R. Walsworth, and M. D. Lukin, "High-sensitivity diamond magnetometer with nanoscale resolution," Nature Phys. 4, 810 (2008). [10] J.-P. Tetienne, T. Hingant, L. Rondin, A. Cavaill`es, L. Mayer, G. Dantelle, T. Gacoin, J. Wrachtrup, J.- F. Roch, and V. Jacques, "Spin relaxometry of sin- gle nitrogen-vacancy defects in diamond nanocrystals for magnetic noise sensing," Phys. Rev. B 87 (2013). [11] T. Rosskopf, A. Dussaux, K. Ohashi, M. Loretz, R. Schirhagl, H. Watanabe, S. Shikata, K. M. Itoh, and C. L. Degen, "Investigation of surface magnetic noise by shallow spins in diamond," Phys. Rev. Lett. 112, 147602 (2014). [12] B. A. Myers, A. Das, M. C. Dartiailh, K. Ohno, D. D. Awschalom, and A. C. Bleszynski Jayich, "Probing sur- face noise with depth-calibrated spins in diamond," Phys. Γ′ (kHz) 45 3 2 1 Γ- Γ+ Γ′- Γ′+ ν+ ν- νFMR b 20 10 Γ (kHz) 25 1 d Γ+ Γ- Γ0 - Γ0 + ν+ ν- νFMR 20 30 Applied field, μ0H0 (mT) -5 0 -2.5 2.5 Current, I0 (mA) 5 a 10 In-plane field (mT) 15 20 25 024 1234 Γ (kHz) c Frequency (GHz) [13] J.-P. Tetienne, T. Hingant, L. Rondin, S. Rohart, A. Thi- aville, J.-F. Roch, and V. Jacques, "Quantitative stray field imaging of a magnetic vortex core," Phys. Rev. B 88, 214408 (2013). [14] R. Badea, M. S. Wolf, and J. Berezovsky, "Exploiting bistable pinning of a ferromagnetic vortex for nitrogen- vacancy spin control," Appl. Phys. Lett. 109, 132403 (2016). [15] J.-P. Tetienne, T. Hingant, J.-V. Kim, L. H. Diez, J.- P. Adam, K. Garcia, J.-F. Roch, S. Rohart, A. Thiaville, D. Ravelosona, and V. Jacques, "Nanoscale imaging and control of domain-wall hopping with a nitrogen-vacancy center microscope," Science 344, 1366 (2014). [16] C. S. Wolfe, V. P. Bhallamudi, H. L. Wang, C. H. Du, S. Manuilov, R. M. Teeling-Smith, A. J. Berger, R. Adur, F. Y. Yang, and P. C. Hammel, "Off-resonant manipula- tion of spins in diamond via precessing magnetization of a proximal ferromagnet," Phys. Rev. B 89, 180406 (2014). [17] T. van der Sar, F. Casola, R. Walsworth, and A. Yacoby, "Nanometre-scale probing of spin waves using single elec- tron spins," Nature Comm. 6, 7886 (2015). [18] C. S. Wolfe, S. A. Manuilov, C. M. Purser, R. Teeling- Smith, C. Dubs, P. C. Hammel, and V. P. Bhalla- mudi, "Spatially resolved detection of complex ferromag- netic dynamics using optically detected nitrogen-vacancy spins," Appl. Phys. Lett. 108, 232409 (2016). [19] M. R. Page, F. Guo, C. M. Purser, J. G. Schulze, T. M. Nakatani, C. S. Wolfe, J. R. Childress, P. C. Hammel, G. D. Fuchs, and V. P. Bhallamudi, "Opti- cally Detected Ferromagnetic Resonance in Metallic Fer- romagnets via Nitrogen Vacancy Centers in Diamond," arXiv:1607.07485 (2016). Rev. Lett. 113, 027602 (2014). [20] C. Du, T. van der Sar, T. X. Zhou, P. Upadhyaya, F. Casola, H. Zhang, M. C. Onbasli, C. A. Ross, R. L. Walsworth, Y. Tserkovnyak, and A. Yacoby, "Control and local measurement of the spin chemical potential in a magnetic insulator," Science 357, 195 (2017). [21] S. I. Kiselev, J. C. Sankey, I. N. Krivorotov, N. C. Emley, R. J. Schoelkopf, R. A. Buhrman, and D. C. Ralph, "Mi- crowave oscillations of a nanomagnet driven by a spin- polarized current," Nature 425, 380 (2003). [22] A. A. Tulapurkar, Y. Suzuki, A. Fukushima, H. Kub- ota, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and S. Yuasa, "Spin-torque diode effect in magnetic tunnel junctions," Nature 438, 339 (2005). 6 [23] J. C. Sankey, P. M. Braganca, A. G. F. Garcia, I. N. Kriv- orotov, R. A. Buhrman, and D. C. Ralph, "Spin-transfer- driven ferromagnetic resonance of individual nanomag- nets," Phys. Rev. Lett. 96 (2006). [24] H. Suhl, "The theory of ferromagnetic resonance at high signal powers," J. Phys. & Chem. Solids 1, 209 (1957). [25] V. E. Demidov, S. Urazhdin, B. Divinskiy, V. D. Bessonov, A. B. Rinkevich, V. V. Ustinov, and S. O. Demokritov, "Chemical potential of quasi-equilibrium magnon gas driven by pure spin current," Nature Comm. 8 (2017). [26] See supplementary materials for more information. [27] 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). [28] A. Dreau, M. Lesik, L. Rondin, P. Spinicelli, O. Arcizet, J.-F. Roch, and V. Jacques, "Avoiding power broadening in optically detected magnetic resonance of single NV defects for enhanced dc magnetic field sensitivity," Phys. Rev. B 84, 195204 (2011). [29] Z. Duan, C. T. Boone, X. Cheng, I. N. Krivorotov, N. Reckers, S. Stienen, M. Farle, and J. Lindner, "Spin- wave modes in permalloy/platinum wires and tuning of the mode damping by spin Hall current," Phys. Rev. B 90 (2014). [30] J. Sankey, I. Krivorotov, S. Kiselev, P. Braganca, N. Em- ley, R. Buhrman, and D. Ralph, "Mechanisms limiting the coherence time of spontaneous magnetic oscillations driven by dc spin-polarized currents," Phys. Rev. B 72 (2005). [31] D. V. Berkov and N. L. Gorn, "Magnetization oscillations induced by a spin-polarized current in a point-contact ge- ometry: Mode hopping and nonlinear damping effects," Phys. Rev. B 76 (2007). [32] A. Jarmola, V. M. Acosta, K. Jensen, S. Chemerisov, and D. Budker, "Temperature- and magnetic-field- dependent spin relaxation in nitrogen- vacancy ensembles in diamond," Phys. Rev. Lett. 108 (2012). longitudinal [33] C. Lee, E. Gu, M. Dawson, I. Friel, and G. Scarsbrook, "Etching and micro-optics fabrication in diamond using chlorine-based inductively-coupled plasma," Diamond & Rel. Mat. 17, 1292 (2008). [34] Y. Chu, N. de Leon, B. Shields, B. Hausmann, R. Evans, E. Togan, M. J. Burek, M. Markham, A. Stacey, A. Zi- brov, A. Yacoby, D. Twitchen, M. Loncar, H. Park, P. Maletinsky, and M. Lukin, "Coherent optical transi- tions in implanted nitrogen vacancy centers," Nano Lett. 14, 1982 (2014). Supplementary Information July 13, 2018 Contents 1 Device fabrication 2 NV measurements 2.1 Single NV identification and isolation . . . . . . . . . . . . . . . . . . . 2.2 Extracting spin relaxation rates . . . . . . . . . . . . . . . . . . . . . . 3 Magnetic field calibration 4 Spin transfer control and magnetoresistive readout 5 Joule heating background, differential resistance, and RF current cal- ibration 5.1 Derivation of Joule heating mixdown voltage . . . . . . . . . . . . . . . 5.2 Differential resistance and estimation of RF current . . . . . . . . . . . 5.2.1 Traditional lock-in readout . . . . . . . . . . . . . . . . . . . . . Impulse readout . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 5.2.3 Maximum change in wire temperature . . . . . . . . . . . . . . 5.2.4 Estimating RF current from rectified voltage . . . . . . . . . . . 5.3 Modeling the thermal transfer function for our wire . . . . . . . . . . . 5.3.1 Parallel resistor model . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 Finite-element simulation . . . . . . . . . . . . . . . . . . . . . 6 Macrospin model 6.1 Equation of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Small-angle precession frequency . . . . . . . . . . . . . . . . . . . . . . 6.3 Linear susceptibility for magnetization approximately along the in-plane hard axis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.1 Mixdown voltage and FMR fit function . . . . . . . . . . . . . . 6.3.2 Expected spectrum of Brownian magnetization noise . . . . . . 6.4 Parametrically driven, large-amplitude oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4.1 Nanowire stray field along NV axis 3 3 3 6 7 8 9 9 11 12 12 13 15 16 17 19 22 22 23 24 25 26 26 28 1 arXiv:1807.03411v2 [cond-mat.mes-hall] 12 Jul 2018 6.4.2 Estimating the parametric precession angle . . . . . . . . . . . . 7 Ferromagnetic resonance frequencies 7.1 Macrospin approximation to resonant frequency . . . . . . . . . . . . . 7.2 Lowest-order spinwave approximation to resonant frequency . . . . . . 8 Magnon-induced NV spin relaxation rates 29 31 32 32 34 2 1 Device fabrication We fabricate an 8-µm-long × 417-nm-wide Py (Ni81Fe19) / Pt nanowire (layer thick- nesses tPy = 10 nm and tPt = 10 nm) on CVD-grown, electronic grade diamond (Ele- ment Six), as shown in Fig. 1 of the main text. Prior to depositing the nanowire, we relieve strain and smooth the diamond surface to ∼0.2 nm-rms over micron length scales (∼0.05 nm-rms over smaller scales) with Ar/Cl2 etching [1], then create a layer of NV centers 75±15 nm below the surface with 15N+ implantation (60 keV, Innovion Corp.) and high-temperature annealing [2]. A subsequent Ar/O2 plasma etch (20 mTorr, 35 sccm Ar, 10 sccm O2, 108 W RF for 45 sec; etches 20 nm/min) removes contaminants and reduces the NV depth to 60±15 nm. To remove any residual graphitized carbon and oxygen-terminate the surface, we perform a final clean in a boiling 1:1:1 mixture of nitric, perchloric, and sulfuric acid at 200◦ C for 1 hour. The Py/Pt nanowire is deposited with e-beam lithography, sputtering, and liftoff (we recommend a more di- rectional deposition, e.g., evaporation to improve liftoff yield). We then deposit Ti (10 nm) / Au (50 nm) macroscopic leads and a microwave stripline (10 µm wide, 17 µm center-to-center distance) using photolithography, and finally connect the leads to the nanowire with another e-beam liftoff (Ti (10 nm) / Au (70 nm) evaporated). The contacts leave an "active" region of length 5.2 µm exposed. 2 NV measurements 2.1 Single NV identification and isolation We focus on an NV center ("NVA") having the strongest stray field coupling. However, this NV is positioned near a second NV "NVB", that provides an additional small photoluminescence (PL) signal. As described below, we identify these two NVs using a combination of electron spin resonance (ESR) spectroscopy and photon correlation measurements, taking advantage of their different symmetry axes to isolate signals from NVA. The two NVs are both associated with the single bright emission spot shown in Fig. 1(b) of the main text. With the 532 nm excitation focused on this spot, we observe ESR spectra indicating the presence of NVs having two different orientations. Figure 1 shows two ESR spectra taken with a 6 mT magnetic field oriented 35◦ from the diamond surface in the xz and yz planes, such that it is aligned with one of the allowed <111> crystallographic orientations of the diamond bonds (i.e., along one of the allowed NV symmetry axes). Dips in PL are associated with driving the NV from its ms = 0 spin projection to ms = ±1, occurring at frequencies ν± (nominally ν+ = ν− = 2.87 GHz at zero field). The deepest photoluminescence dip in both spectra is associated with ms = 0 → −1 transition of the better-coupled NV center (NVA), mostly because the polarization of the 532 nm excitation is perpendicular to the NVA symmetry axis. This transition shows the largest magnetic field dependence when the field is in the yz plane, parallel to NVA (corresponding to the arrow in Fig. 1(a) of the main text), while the 3 Figure 1: ESR measurement of the probed spot in Fig. 1(b) of the main text, showing PL dips associated with two NV centers. shallower ms = 0 → −1 transition of NVB shows the largest deviation with the field aligned in the xz plane. Note that the ms = 0 → +1 transition of NVA is barely visible above the noise in each spectrum, due to imperfect power coupling to our stripline at the higher frequencies (this is also why we do not see the ms = 0 → +1 transition for NVB in these spectra). In principle, there could be multiple NVs creating the observed spectra, but the total photon count rate is approximately what we expect for two NVs. Moreover, because the emission spot is close to the magnetic nanowire, in a region of high magnetic field gradient, NVs with the same orientation would be expected to show distinct ESR lines. We see no evidence of unexpected dips under any conditions, lending additional credence to the identification of two NVs. To further verify our interpretation that only two NV centers are co-located within the emission spot, we perform photon correlation measurements to estimate the number of contributing emitters. The second order correlation function g(2)(τ ) was measured with a Hanbury-Brown-Twiss setup, using a fiber beam-splitter and two single photon counting modules. A time-correlated single photon counting system (PicoHarp 300) was used to measure the histogram of times between two consecutive photons. Such a normalized distribution can be interpreted as a measurement of g(2)(τ ), as long as τ is much smaller than the mean time between two detection events (for us, τ (cid:28) 1/count rate= (30 kcounts/s)−1 ≈ 33 µs) [3]. We measure a background count rate (on a dark area of the sample) of ≈ 3 kcounts/s, which we subtract from our data before analysis. 4 cos(35°) sin(35°) cos(35°) sin(35°) 6 6 μ0H0 μ0H0 Figure 2: g(2)(τ ) correlation measurement of the studied emission spot, with g(2)(0)=0.45±0.01. Our data shows a clear anti-bunching signature at short times (see Fig. 2), charac- teristic of a small number of quantum emitters, and slight bunching at longer times, associated with metastable states in the emitters. To quantify g(2)(0) requires that we appropriately normalize the data. Merely setting the long-time value to 1 is problem- atic because of the exponential decay associated with the fact that we are not directly measuring the correlation function but a histogram of times between photons. Indeed, we observed such an exponential decays for g(2)(τ ) measured with attenuated laser light. We thus compensate for the decay by fitting an exponential with a decay constant τ, e−t/τ to the data from 1 µs onward. Multiplying the full spectrum of data points by et/τ corrects for the exponential decay, and allows us to used the long-time value for our normalization. We then use a three level model to fit the curve to the function [4]: τ2(cid:17) , (1) g(2)(τ ) = c1(cid:16)1 + c2e− τ τ1 + c3e− τ where c1,2,3 are scaling factors and τ1,2 correspond to the lifetimes of the involved NV states. We find that g(2)(0) = 1 + c2 + c3 = 0.45±0.01, consistent with two co-located emitters, where one is slightly better coupled to our optical excitation/collection path than the other. In our measurements, we isolate NVA from NVB by taking advantage of their dif- ferent orientations. For ESR measurements (such as in Fig. 3 of the main text), the transitions of NVA and NVB are spectrally resolved. Moreover, we align the polarization of the 532 nm illumination to preferentially excite NVA, reducing photoluminescence from NVB; we also work in magnetic fields aligned with NVA, which reduces the ESR contrast of NVB. As a result, the resonances from NVB are only weakly visible in the spin resonance data shown in Fig. 4(b) of the main text. For spin relaxation measure- 5 1.5 1 0.5 0 0 200 400 600 800 g2 NV5 minus BG Data Fit τ (ns) g(2)(τ) 0.45 ± 0.01 ments, we use π pulses tuned to the transitions of NVA to isolate its relaxation rates from those of NVB, as described in further detail below. 2.2 Extracting spin relaxation rates We extract the two relaxation rates Γ± for NVA by initializing the spin into the three spin projections ms = −1, 0, 1, observing its relaxation via photoluminescence measure- ments, and fitting our data to a rate equation model. We prepare the spin projections using a pulse of 532 nm excitation, which optically pumps both NVA and NVB into ms = 0; to prepare ms = ±1, we then apply a π pulse of microwaves tuned to the NVA spin transitions for durations of 50-200 ns, chosen to maximize the contrast of a Rabi sequence. These π pulses are far off resonance with NVB, so it remains in ms = 0. After a variable delay of duration τ, we again apply a pulse of 532 nm excitation and observe photoluminescence counts, which are higher for ms = 0 states than for ms = ±1. In the end, we record three photoluminescence traces: f0(τ ), f+(τ ), and f−(τ ), where the subscript indicates the spin state into which NVA was initialized. Since these photoluminescence traces include emission from both NVA and NVB, we consider the differences in the photoluminescence traces d−(τ ) = f0(τ ) − f−(τ ) and d+(τ ) = f0(τ ) − f+(τ ). Because NVB is prepared in the same way for all three initializations, its (possibly τ-dependent) photoluminescence is eliminated in the data sets d±(τ ), as is any background photoluminescence not associated with NVA. laxation: We then simultaneously fit d−(τ ) and d+(τ ) to a rate equation model for spin re- dp−(t) dt dp+(t) dt dp0(t) dt = Γ− (p0(t) − p−(t)) = Γ+ (p0(t) − p+(t)) = −(Γ− + Γ+)p0(t) + Γ−p−(t) + Γ+p+(t), (2) where pm(t) is the population of NVA in spin state m and Γ± are the relaxation rates on the ms = 0 ↔ ms = ±1 transitions. We also considered double quantum relaxation between ms = ±1 states, but our fits revealed that this rate was insignificant. The fit functions for d±(τ ) are found by calculating the population in ms = 0 for three different spin initializations, p0(τ )(m) (where the superscript indicates the initial state), and subtracting to reconstruct a signal proportional to d±(τ ) ∝ p0(τ )(0) − p0(τ )(±). Note that this proportionality holds even for imperfect optical pumping and imperfect π pulses – low fidelity initialization and incomplete π pulses reduce the contrast of our signals, but not their time dependence. We can thus fit d±(τ ) to A(cid:0)p0(τ )(0) − p0(τ )(±)(cid:1), where A is an unknown fitting parameter, and extract the two relaxation rates Γ±. 6 3 Magnetic field calibration We apply static magnetic fields using a permanent magnet (K&J magnetics, D42-N52) mounted on a three-axis translation stage equipped with closed-loop motorized actua- tors (CONEX-TRB25CC). The 25mm range of motion permits us to apply magnetic fields up to 90 mT oriented along the NV symmetry axis. To calibrate the relationship between the actuator position and the applied magnetic field, we exploit the field-dependent photoluminescence of the NV center [5]. Essentially, NV photoluminescence is maximized when the magnetic field is parallel to its symmetry axis, with heightened sensitivity near the ground- and excited-state level anti-crossings near 103 mT and 51 mT, respectively. Following the procedure described in Ref. [6], we measure the photoluminescence of an NV as a function of the position of the permanent magnet. From this data, we can extract the controller coordinates that bring the NV center to the excited state level anti-crossing (51 mT) and align the magnetic field with the NV axis. We then model the stray field of the permanent magnet using RADIA [7] and determine the position (relative to the permanent magnet origin) where the stray field is 51 mT aligned with the NV axis. This provides a coordinate transformation between our controller position and the model. We can then use the RADIA model to roughly predict the field as a function of controller position. We design a trajectory for the magnet that should maintain the magnetic field along the NV axis or at a defined angle relative to that axis. This calibration procedure gives only an approximation of the magnetic field due to potential inaccuracy of the RADIA model and possible misalignment between the (nominally aligned) coordinate axes of our sample and the controller axes. We therefore use NV electron spin resonance (ESR) measurements to determine the actual magnetic field taken along the trajectories used in our experiments. Note that these measure- ments were taken after the device magnetization disappeared (see Sec. 6.4.1), so the NV experiences only the applied magnetic field. We fit the ESR spectra to extract the transition frequencies ν±. From the linear and quadratic Zeeman effects, we can determine the on- and off-axis components of the magnetic field [6], and thus calculate its magnitude. Figure 3(a) shows the difference between the extracted and expected field magnitudes as a function of the expected field predicted by the RADIA model. These values are found using an NV g-factor of 2.0030(3) and a zero field splitting of 2.870(2) GHz, consistent with our observed splittings and literature values [8, 9]. The solid red curve is a sixth order polynomial interpolating function used to rescale the magnitudes of the magnetic fields in the data presented in the main text. Figure 3(b) shows the extracted angle of the magnetic field, which is poorly constrained by these measurements; due to the good ESR contrast observed over the entire magnetic field scan at θ = 0◦ in Fig. 4(b) of the main text, and the strong suppression of directly- driven FMR at θ = 0◦, we believe the angular alignment is good to within a couple of degrees. 7 Figure 3: (a) Magnetic field magnitude (as determined by ESR measurements), relative to the expected magnetic field (as calculated by RADIA). Red curve is a sixth order polynomial fit used to rescale the magnetic field values in the main text. (b) Off-axis angle of the magnetic field as a function of magnitude. The (large) error bars represent systematic uncertainties in our system parameters. 4 Spin transfer control and magnetoresistive readout We control the magnetization of the Py layer with current I(t) in the nanowire, which provides two torques. First, as indicated in Fig. 1(a) of the main text, the current density JPt in the Pt layer produces a spin current density Js = θSHJPt (where θSH is Pt's spin Hall angle), polarized along y and travelling along z [10]. A fraction η of these spins are absorbed by the Py layer (ηθSH = 0.055 [11]), thereby applying an areal spin transfer torque (STT) density at the Py interface, or an effective per-volume torque density (N·m/m3) (3) ~ TSTT = m × ( m × y) , ηθSHJPt 2etPy where the unit vector m = mx x + my y + mz z represents the orientation of the local magnetization. Second, the current in the wire generates a magnetic field ~HI in the Py layer, with a torque (volume) density ~ THI = µ0Ms ~HI × m, (4) where Ms is the Py saturation magnetization. Changes in the magnetization can then be read out via the device's resistance, which varies approximately1 as (5) where R0 = 237 Ω is the wire's unbiased resistance at mx = 0, and R0δAMR ≈ 0.2 mΩ is the change due to anisotropic magnetoresistance (AMR) from the fraction of current flowing through the Py layer (measured by monitoring R while sweeping the field). R = R0(1 + δAMRm2 x), 1This assumes mz (cid:28) mx, which is justified by the calculations and simulations in Sec. 6. 8 10 20 30 40 50 Field magnitude (mT) 30 20 10 0 -10 0 AnglefromNVAaxis(degrees) b 10 30 20 Nominal field (mT) 40 50 012 - 1 Fieldmagnitude-nominalfield(mT) a (7) (8) (9) ∆V = I0∆R0 + IRF∆RRF cos ψ (6) 1 2 To verify device functionality, we perform spin transfer ferromagnetic resonance (ST-FMR) similar to that of Ref. [11]. To briefly summarize, a current I(t) = I0 + IRF cos(2πνNWt) (with DC bias I0, "radio frequency" (RF) amplitude IRF, and frequency νNW) driving coherent resistance oscillations R(t) = R0 + ∆R0 + ∆RRF cos(νNWt + ψ) (for constant offset ∆R0, amplitude ∆RRF, and phase ψ) will generate a time-averaged voltage change that can be read out with a lock-in technique (see Sec. 5.2). This voltage comprises a resonant magnetoresistance signal ∆VMR (estimated for a macrospin in Sec. 6) and a broad background due to Joule heating (useful for estimating IRF, as discussed in Sec. 5) that is subtracted from all presented spectra. 5 Joule heating background, differential resistance, and RF current calibration When a time-dependent current passes through a resistive wire, it dissipates a time- dependent power, thereby heating the sample and causing a time-dependent change in resistance. In general, this nonlinear response can generate a static voltage hV i across the resistor. Here we develop a framework for quantifying this effect, and show how it can be used to provide a reasonable estimate of the RF current flowing through our nanowires. This calibration considers RF temperature oscillations not included in (e.g.) Tshitoyan et al [12], which are especially relevant when the substrate is highly thermally conductive. 5.1 Derivation of Joule heating mixdown voltage In general, the instantaneous voltage V will depend on current I and the change in temperature ∆T . For small I and ∆T , we can Taylor expand to 3rd order: T V(cid:1) ∆T 3, (cid:0)∂3 1 6 1 2 1 2 T V(cid:1) ∆T 2 I V(cid:1) I 2 + (∂T ∂IV ) I∆T + (cid:0)∂2 (cid:0)∂2 I V(cid:1) I 2∆T + I V(cid:1) I 3 + T ∂IV(cid:1) I∆T 2 + (cid:0)∂3 (cid:0)∂2 (cid:0)∂T ∂2 1 2 1 2 1 6 + + V ≈ (∂IV ) I + (∂T V ) ∆T where all of the parenthetical factors are constants of the system determined by the re- sistor's geometry, materials, and thermal anchoring. We simplify this with the following assumptions: 1. In the absence of temperature change, the resistor responds linearly. This means I V = 0; as a result, the terms (red) having no temperature dependence ∂2 ∂T ∂2 I V = 0 as well. I V = ∂3 9 2. Changing the temperature does not on its own generate a voltage. This means the current-independent terms ∂T V = ∂2 T V = ∂3 T V = 0 (teal). 3. The temperature change ∆T ∼ I 2 to lowest order, meaning the penultimate term (blue) is of order I 5 and can be dropped.2 Under these assumptions, V → (∂IV ) I + (∂T ∂IV ) I∆T. (10) The first term is the heat-free linear response of the resistor, with the constant ∂IV being the resistance near I = 0. We emphasize that ∂IV is a constant that does not depend on I or ∆T (all differentials in Eq. 10 are evaluated at I = ∆T = 0) and is qualitatively different from laboratory measurements commonly referred to as "differential resistance", wherein the current is slowly modulated and the resulting voltage modulations are recorded (see Section 5.2). The second term is the Joule heating nonlinearity of interest; an "extra" voltage can be generated from this term only if there is both a current and a temperature change. For example, if I = I0 is some constant value, this will heat the sample, causing ∆T > 0, which raises the resistance by (∂T ∂IV ) ∆T , at which point I0 produces an additional voltage. If I changes slowly enough with time that the system remains in steady state, it is this term that is responsible for a bias dependence in a low-frequency differential resistance measurement (see Sec. 5.2.1). We eliminate temperature ∆T by assuming small enough changes that ∆T responds linearly to the applied power P . In this limit, an oscillatory component in the power P1 cos ωt of amplitude P1 and frequency ω will induce a temperature change ∆T = [X(ω) cos ωt + Y (ω) sin ωt] P1, (11) where X(ω) and Y (ω) (units of K/W) represent the thermal transfer function of the wire, capturing the magnitude and phase shift of the thermal response.3 In our exper- iments, we apply a current of the form I = I0 + I1 cos ωt, (12) for constants I0, I1 and frequency ω, such that the instantaneous power is, to leading order (again assuming negligible Peltier effects) (13) I 2 1 cos 2ωt(cid:19) (∂IV ) . 1 2 P (t) ≈ (I0 + I1 cos ωt)2 (∂IV ) 0 + I 2 1 + 2I0I1 cos ωt + 1 2 =(cid:18)I 2 2In systems having a more significant Peltier effect (e.g., some spin valves), this should not be dropped. 3We choose this quadrature formulation of the transfer function over the "usual" complex one due to the nonlinear operations in the rest of the analysis. The quadrature basis is also very convenient for calculating the mixdown voltage, which requires only the in-phase component X(ω). 10 This comprises a thermal drive at three frequencies (zero, ω, and 2ω) which, in this linear-response limit, can be treated separately. As such, the temperature ∆T ≈ P0X0 + P1X(ω) cos ωt + P1Y (ω) sin ωt + P2X(2ω) cos 2ωt + P2Y (2ω) sin 2ωt (14) with 1(cid:19) (∂IV ) 1(cid:19) (∂IV ) . 1 2 I 2 I 2 P0 ≡(cid:18)I 2 P2 ≡(cid:18)1 2 P1 ≡ (2I0I1) (∂IV ) 0 + Plugging Eq. 14 into the voltage expansion of Eq. 10, V ≈ (∂IV ) (I0 + I1 cos ωt) + (∂T ∂IV ) (I0 + I1 cos ωt) (P0X(0)) + (∂T ∂IV ) (I0 + I1 cos ωt) (P1X(ω) cos ωt + P1Y (ω) sin ωt) + (∂T ∂IV ) (I0 + I1 cos ωt) (P2X(2ω) cos 2ωt + P2Y (2ω) sin 2ωt) . Taking a time-average yields (15) (16) (17) (18) (19) (20) (21) (22) [χ(0) + 2χ(ω)] I 2 I1P1X(ω)(cid:19) 1(cid:19) 1 2 1 2 hV i = (∂IV ) I0 + (∂T ∂IV )(cid:18)I0P0X(0) + = I0 (∂IV )(cid:18)1 + χ(0)I 2 0 + where we have defined a fractional resistance change transfer function χ(ω) ≡ (∂T ∂IV ) X(ω) (23) for brevity (units of 1/mA2). The first term is Ohm's law and the second is the steady state heating due to DC current. The third term is the response to an RF drive, and is responsible for large backgrounds observed in FMR (see Sec. 5.2.2). Importantly, this comprises two terms, the first (χ(0)) arising from the time-averaged power absorbed by the wire, and the second (χ(ω)) arising from the heat-induced resistance oscillations at ω mixing with the drive current. 5.2 Differential resistance and estimation of RF current We now discuss a means of interpreting the signal from a lock-in-based differential resistance measurement in the presence of DC and RF current. Section 5.2.1 discusses a "traditional" continuous-wave lock-in approach, and Sec. 5.2.2 discusses an impulse method that happens to be easier with our apparatus. Section 5.2.4 then discusses a means of using the observed voltages in these measurements to estimate the RF current flowing through the device. 11 5.2.1 Traditional lock-in readout When "differential resistance" is measured using a lock-in technique, the bias is modu- lated, such that the current (24) where Im is the lock-in's modulation amplitude, and ωm (cid:28) ω is the modulation fre- quency (typically chosen to be low enough that the device remains in steady state). In this case, the voltage (Eq. 22) becomes I0 → I0 + Im cos ωmt, [χ(0) + 2χ(ω)] I 2 1 2 1(cid:19) 1(cid:19) cos ωmt (26) (25) χ(0) + χ(ω)(cid:21) I 2 2 I 3 3 4 hV i → (∂IV ) (I0 + Im cos ωmt)(cid:18)1 + χ(0) (I0 + Im cos ωmt)2 + = ... + (∂IV )(cid:18)ILI + χ(0)(cid:20)3I 2 LI(cid:21) + Im(cid:20)1 0 Im + χ(0) + χ(ω)(cid:21) I 2 1(cid:19) . (28) I 2 m(cid:21) +(cid:20)1 2 3 4 Rdiff ≡ (∂IV )(cid:18)1 + χ(0)(cid:20)3I 2 0 + where we have lumped all terms not contributing to the measured amplitude at ωm into "..." for brevity, and used the identity (27) cos 3ωmt. 1 4 cos ωmt + 3 4 cos3 ωmt = The lock-in measurement demodulates at ωm to record the response amplitude, which is then divided by Im to define a "differential resistance" Note this result agrees with Eq. 22 in the low-frequency limit χ(ω) → χ(0), but here we can see how the presence of RF current will alter this signal. Impulse readout 5.2.2 In our experiment, we employ an impulse readout of Rdiff wherein the amplitude of the modulation is slowly increased to its maximum value and then decreased to zero as (29) (30) (31) (cid:19) t(cid:21)(cid:19) , 1 N (cid:19) t(cid:21) + sin(cid:20)ωm(cid:18)1 + 1 N I0 → I0 + Is(t) 2N Is(t) ≡ Im sin2(cid:16) ωm t(cid:17) sin (ωmt) = Im 4 (cid:18)2 sin (ωmt) − sin(cid:20)ωm(cid:18)1 + where N is a large integer. We have expanded Is in the last line to highlight that this waveform includes only 3 frequencies near ωm, which helps minimize artifacts associated with abrupt changes in current. Similar to the continuous-wave measurement, the 12 device responds adiabatically for sufficiently small ωm, and the low-frequency voltage becomes hV i → V0 + Vs(t) = (∂IV ) (I0 + Is(t))(cid:18)1 + χ(0) (I0 + Is(t))2 +(cid:20)1 χ(0) + χ(ω)(cid:21) I 2 = (∂IV )(cid:18)DC +(cid:18)1 + 3χ(0)I 2 0 +(cid:20)1 χ(0) + χ(ω)(cid:21) I 2 1(cid:19) 1(cid:19) Is + 3χ(0)I0I 2 2 2 s + χ(0)I 3 (32) s(cid:19) , (33) where "DC" contains all time-independent terms. We then extract a similar "differential resistance" Rdiff,s by taking an overlap with the injected impulse Is(t) to extract a (normalized) "in-phase" response amplitude Rdiff,s = = 0 1 Im ×R 2πN/ω R 2πN/ω mZ 2πN/ω 3πN I 3 8 0 [V0 + Vs(t)] Is(t)dt 0 I 2 s (t)dt Vs(t)Is(t)dt. By symmetry, the even powers of Is in Eq. 33 vanish, leaving behind Rdiff,s = (∂IV )(cid:18)1 + χ(0)(cid:18)3I 2 0 + 35 64 I 2 m(cid:19) +(cid:20)1 2 1(cid:19) . χ(0) + χ(ω)(cid:21) I 2 (34) (35) (36) In practice, the measured value of Rdiff,i does not depend on frequency below ∼ 300 Hz (limited in our case by the low-pass action of our bias-T), which allows us to estimate χ(0) from the bias-dependence of Rdiff,i in the absence of RF current (I1 = 0). Figure 4 shows a measurement of Rdiff,i(I0), with ωm = 2π×265 Hz. The data is well- fit by Eq. 36 (with a small offset bias Ioff = −15 ± 1 µA), from which we extract ∂IV = 237.3 ± 0.3 Ω, χ(0) = 5.3085 ± 0.004 × 10−4 mA−2. The small offset current is likely due to a combination of non-ideal electronics and Peltier effects associated with asymmetric contacts to our nanowire on chip. In the present experiment, this offset leads to at most a few-percent error in our calibration of I1 (discussed below), along with a small, smoothly-varying background signal in our ferromagnetic resonance measurements of up to ∼5 µV at low frequencies. The associated correction, which does not affect the central conclusions of the present work, will be the subject future work. 5.2.3 Maximum change in wire temperature In the presence of only DC bias, Eq. 22 simplifies to hV i → (∂IV ) (I0)(cid:0)1 + χ(0)I 2 0(cid:1) , 13 (37) Figure 4: Measured impulse differential resistance Rdiff,i versus DC bias I0 with modulation amplitude Im = 81.5 µA, modulation frequency ωm = 2π × 265 Hz, N = 53 periods per impulse, and no RF current (I1 = 0). Solid red curve m + 3χ(0) (I0 − Ioff)2(cid:1), with fit is a fit to the form Rdiff,s = (∂IV )(cid:0)1 + χ(0) 35 parameters ∂IV = 237.3±0.3 Ω, χ(0) = 5.3085±0.004×10−4 mA−2, and offset current Ioff = −15± 1 µA. Uncertainty on χ(0) and Ioff are statistical, and the uncertainty on ∂IV is systematic, due to uncertainty in series resistors leading to the sample. Note this measured ∂IV includes the resistances of everything after the bias-T, including a circuit board, wire bonds, and on-chip leads / contacts, so this is an upper bound on the wire resistance. 64I 2 such that the DC resistance hV i I0 = (∂IV ) + (∂IV ) χ(0)I 2 0 = R0 + R0α∆T, where α is the temperature coefficient of resistance. From this, we identify R0 = ∂IV ∆T = χ(0)I 2 0 α (38) (39) (40) (41) Using the above fit values and using the Pt coefficient α & 0.003 K−1 as a lower bound, we place an upper bound on the maximum Joule heating temperature change ∆T . 5 K for our bias range (I0 < 5 mA). 14 ∆hV i = hV i − [hV i]I1=0 (∂IV ) [χ(0) + 2χ(ω)] I0I 2 1 (43) 15 1 2 = (a) "Typical" rectified voltage hV i versus Figure 5: Joule heating spectra. RF frequency for bias I0 = −2.04 mA, nominal power -5 dBm (∼ 0.5 mA into the nanowire), and field µ0H0 =20.6 mT applied a in-plane angle θ = 5◦ (orange) and 0◦ (blue) from the wire axis and 35◦ out of plane. The (heavily damped) magnetic resonance near 2.4 GHz should only appear in the off-axis (θ = 5◦) case, and evidently represents at most a sub-percent contribution to this signal. (b) Frequency-dependent output power used to compensate for losses in the electronic components between the RF source and nanowire. (c) Ratio of the two curves in (b), showing at most a few-percent difference between the compensation spectra. (42) [χ(0) + 2χ(ω)] I 2 1(cid:19) 1 2 hV i = I0 (∂IV )(cid:18)1 + χ(0)I 2 0 + 5.2.4 Estimating RF current from rectified voltage Knowing χ(0), we can, in principle, use the measured rectified voltage (Eq. 22 rewritten) to estimate the RF current I1. Figure 5(a) shows a "typical" spectrum of just the RF-induced rectified voltage 5o 0o a b c 5o 0o as recorded with a chopped lock-in technique at 265 Hz [13]4, with a bias I0 = −2.04 mA, a nominal power -5 dBm (∼ 0.5 mA into the nanowire), and field µ0H0 =20.6 mT applied along an in-plane angle θ = 5◦ (orange) and 0◦ (blue) from the wire's long axis and 35◦ out of plane. Under these conditions, the magnetization dynamics are heavily damped, and the magnetization is saturated approximately along the in-plane short axis (y) of the nanowire (canted a few degrees out of plane). Importantly, this data was recorded after adjusting the output power at each fre- quency (Fig. 5(b)) to ensure a frequency-independent ∆hV i at I0 = −4.08 mA, which allows us to roughly compensate for the strongly frequency-dependent input coupling of our microwave circuitry. Figure 5(c) shows the ratio of the powers in b, highlighting that the damped ferromagnetic resonance feature, which should appear near 2.4 GHz in the θ = 5◦ spectrum, represents at most a ∼1% correction. The high-frequency peak in Fig. 5(a) is a noise feature owing to poor coupling near 4.7 GHz. The low-frequency tail is a bias-independent feature that we suspect arises from electronics nonidealities and / or a small Peltier effect. As such, we subtract it from all spectra in the main text. Its presence or absence from the data does not quantitatively or qualitatively affect the conclusions of this work. Having no reliable means of measuring the loss spectrum of our circuit board, wire bonds (as connected to the sample), and microfabricated leads, we cannot use the measurement in Fig. 5(a) to estimate χ(ω) directly. We can, however, still solve Eq. 43 for the RF amplitude 2∆hV i I0 (∂IV ) χ(0)(cid:16)1 + 2 χ(ω) χ(0)(cid:17), (44) I1 =vuut s 2∆hV i 3I0 (∂IV ) χ(0) and, since we expect 0 < χ(ω) < χ(0) (i.e., χ is largest at ω = 0), we can constrain the RF current I1 to the range < I1 <s 2∆hV i I0 (∂IV ) χ(0) , (45) representing a maximum systematic error of ±37%. However, the frequency dependence of χ(ω) will impose a frequency-dependent systematic error, which can in principle skew observed ferromagnetic resonance curves. In the following section, we improve upon this estimate by simulating the thermal response of our wire on a diamond substrate. 5.3 Modeling the thermal transfer function for our wire The goal of this section is to simulate the thermal response of our nanowire so that we may more accurately estimate how the Joule-rectified voltage depends on the RF 4The only difference from Ref. [13] is that we employ a Wheatstone bridge on the low-frequency port of our bias-T to eliminate common-mode noise from our current source, and use the "single-shot" readout discussed in Sec. 5.2.2. 16 current and DC bias. Inspecting Eq. 44, the only quantity of which we do not have an independent measure is the ratio χ(ω)/χ(0),5 which the following sections address. 5.3.1 Parallel resistor model To model the thermal response of our multilayer nanowire to an oscillatory current, we first must estimate the current carried by each layer. We model the Pt and Py layers of the wire as two resistors connected in parallel, with values RPt and RPy, such that our total current through the wire (46) (47) (48) (49) (50) (51) is the total wire resistance. The dissipated power in each layer is then I = I0 + I1 cos ωt, (I0 + I1 cos ωt) j R R Ij = R ≡ RPyRPt RPy + RPt is divided into layer currents for j ∈ {Pt, Py}, where Pj = 0 + 2I0I1 cos ωt + I 2 R2 Rj(cid:0)I 2 1 cos2 ωt(cid:1) . and the first harmonic has amplitude Pj1 = 2I0I1R2 Rj , Importantly, Pj1 ∝ I1, so the ratio of first-harmonic amplitudes is PPy1 PPt1 = RPt RPy . With this oscillatory drive, we expect a steady state solution to cause a (time de- pendent) temperature change ∆Tj, such that the layer resistances Rj = Rj0 (1 + αj∆Tj) (52) where Ij is the layer current, Rj0 is the zero-heat layer resistance, and αj is the layer's temperature coefficient of resistance. The resistance of the combined nanowire is then R = R0 (1 + αPt∆TPt) (1 + αPy∆TPy) 1 + R0 αPy∆TPy RPy αPt∆TPt + R0 RPt (53) 5Recall the Fig. 5 provides the Joule-rectified voltage ∆hV i, I0 is measured with a series resistor, and (∂I V ) χ(0) can be estimated as in Fig. 4. 17 with zero-heat total wire resistance R0 ≡ RPy0RPt0 RPy0 + RPt0 . (54) Assuming the resistance changes due to Joule heating are a small fraction of the zero- heating values (αj∆Tj (cid:28) 1; see Fig. 4), R ≈ R0 + R2 0 RPt αPt∆TPt + R2 0 RPy αPy∆TPy, (55) and the power amplitudes in each layer (keeping only terms of order I 2 j ∼ ∆Tj) become Pj ≈ R2 0 Rj0(cid:0)I 2 0 + 2I0I1 cos ωt + I 2 1 cos2 ωt(cid:1) . (56) (57) (58) (59) (60) ∆Tj = (Xj(0)Pj0 + Pj1 (Xj(ω) cos ωt + Yj(ω) sin ωt) + Xj(2ω) cos 2ωt) I 2 1(cid:19) R2 0 Rj 1 2 Pj0 ≡(cid:18)I 2 0 + , I 2 1 R2 0 Pj1 ≡ 2I0I1 Rj R2 0 Rj Pj2 ≡ 1 2 In the presence of these static and oscillatory powers, the layer temperature change where Xj(ω) and Yj(ω) are the quadratures of the transfer function for each layer. The instantaneous total voltage is then (61) (62) (63) (64) (65) V = IR = R0 (I0 + I1 cos ωt) 1 +Xj + R0 (I0 + I1 cos ωt) Xj + R0 (I0 + I1 cos ωt) Xj R0 Rj R0 Rj R0 Rj αjXj(0)Pj0! αjPj1 (Xj(ω) cos ωt + Yj(ω) sin ωt)! αjXj(2ω) cos 2ωt! , 18 [χ(0) + 2χ(ω)] I 2 1(cid:19) 1 2 hV i = I0R0(cid:18)1 + χ(0)I 2 0 + and the time-averaged voltage can be written as with the total in-phase transfer function χ(ω) ≡Xj = R3 0 R2 Pt R3 0 R2 j αjXj(ω) αPtXPt(ω) + R3 0 R2 Py αPyXPy(ω). (66) This formula is identical to that of the single-resistor case (Eq. 22 of Sec. 5.1, identi- fying R0 = ∂IV ), except that χ(ω) is now a weighted average of the layers' individual responses. Not surprisingly, the weighting factors scale as αj, and increase as the layer resistance Rj is reduced (when a larger fraction of the current flows through layer j). Additionally, for the case of a single layer (of resistance R0, thermal coefficient α, and transfer function X), this expression simplifies to χ(ω) → RαX(ω) and we can identify ∂T ∂IV = R0α from Eq. 22, as expected. As mentioned above (see also Eq. 44), we are interested in the ratio χ(ω) χ(0) =(cid:18) XPt(ω) XPt(0)(cid:19) 1 + ηXPy(ω)/XPt(ω) 1 + ηXPy(0)/XPt(0) η = R2 R2 PtαPy PyαPt . (67) Assuming RPt/RPy = 0.325 [14] and αPy/αPt ∼ 1.33 for our system, η ∼ 0.15. As we will show in the following section, XPy/XPt < 1 over the frequency range of interest, so we can make a further approximation χ(ω) χ(0) ≈(cid:18) XPt(ω) XPt(0)(cid:19)(cid:18)1 + η(cid:20)XPy(ω) XPt(ω) − XPy(0) XPt(0)(cid:21)(cid:19) (68) which illustrates the correction due to the presence of the Py layer is small (as discussed below, it should be . 3%). Nonetheless, we simulate both layers because there is not much additional overhead. 5.3.2 Finite-element simulation We perform finite-element simulations (COMSOL) of the nanowire's active region (be- tween the contacts, where the current is concentrated) and the diamond substrate. Figure 6(a) shows the simulated geometry, comprising a L× W × H diamond substrate (dimensions varied with frequency as discussed below), upon which a 10 nm Py (bot- tom) / 10 nm Pt (top) nanowire spanning 5.2 × 0.417 µm2 is positioned at the center. We assume the heat is primarily dissipated in the constriction (the nanowire), and that the large-area connection to diamond serves as the dominant heat sink in this system; as such we do not bother to include the leads in this study. Including a "standard" convective heat loss of ∼25 W/m2K boundary condition on all free surfaces does not 19 Figure 6: Simulated thermal response of Py/Pt nanowire on diamond, assum- ing a wire length l=5.2 µm, width w=417 nm, Pt and Py thicknesses h=10 nm, resistivities ρPt = 21.9 Ω-cm and ρPy = 65.2 Ω-cm. (a) Simulation volume for the low-frequency response (left) and nanowire mesh (right). (b) In-phase temperature oscillation amplitude at the drive frequency for the Pt (blue) and Py (orange) layer, for a drive power amplitude PPt1 = 1017 W/m3 and PPy1 = 0.336 × 1017 W/m3 in the Py layer. (b) Out-of-phase oscillation am- plitude, showing steady-state behavior at low frequencies (where the thermal penetration depth in the diamond is much larger than the wire) and the onset of a lagged response at higher frequencies (where the heat is confined to very near the wire). Inset shows χ(ω)/χ(0) (black) as well as XPt(ω)/XPt(0) (red), XPy(ω)/XPt(ω) (blue), XPy(ω) XPt(0) (magenta) from Eqs. 67 and 68. XPt(ω) − XPy(0) significantly alter the results, nor does reducing the mesh density along each axis by a factor of 2. As per Eq. 51, we introduce an oscillatory power of amplitude PPt1=1017 W/m3 and PPy1 = (RPy/RPt) PPt1 and frequency ω to the Pt and Py layers. We let the simulation run until it converges to a steady state, then extract the in-phase (Xj, Fig. 6(b)) and 20 Pt Diamond Py 0.8 mm Frequency (GHz) Ratios a b c out-of-phase (Yj, Fig. 6(c)) amplitudes from the time-dependent temperatures Tj(t).6 At lower frequencies, the temperature change penetrates further into the diamond, and a larger diamond volume is required for the results to converge. At the lowest frequency simulated (ω = 2π×265 Hz), for example, we employed the 0.8 mm × 0.8 mm × 0.4 mm volume shown in Fig. 6(a), but a small deviation from adiabatic response is still evidenced by a sub-percent out-of-phase response due to integrator-like behavior as heat reaches the boundaries of the diamond. This phase lag returns at higher frequencies (ω > 2π × 106 Hz) as the heat ceases to efficiently escape the nanowire. Figure 7: RF current amplitude I1 calibrated from the data in Figs. 4-5 (Fig. 2 of the main text) and the simulated thermal response in Fig. 6 (black). Also shown is the model-independent calibration range (shaded region) bounded by the case χ(ω)/χ(0) = 0 (red) and χ(ω)/χ(0) = 1 (blue). The inset of Fig. 6(c) shows the quantities of interest in Eqs. 67-68, as well as χ(ω)/χ(0) for the total wire assuming η = 0.15. We can now plug this into our expres- sion for the RF current amplitude (Eq. 44 rewritten) to convert the observed rectified voltage ∆hV i in Fig. 5(a) (at I0 = −2.04 mA) to RF current, using the values ∂IV = 237.3± 0.3 Ω, χ(0) = 5.3085± 0.004× 10−4 mA−2 from the fit in Fig. 4. Figure 7 shows the frequency dependence of I1 (black curve) along with the absolute calibration range of Eq. 45 (shaded region), highlighting the importance of the frequency-dependent thermal response in our system. If we ignored χ(ω) as in Ref. [12] (e.g.) we would overestimate I1 by up to 65%, and miss a systematic variation of ∼ 10% over the probed frequency range. We suspect this correction is smaller in more thermally-isolated systems such as wires on oxide or spin valves, though we note that the thermal transfer function falls off very gradually at high frequencies. 6Note this single-frequency method is significantly more efficient than performing an impulse re- sponse when covering this many decades of frequency. 21 I1 =vuut 2∆hV i I0 (∂IV ) χ(0)(cid:16)1 + 2 χ(ω) χ(0)(cid:17) (69) 6 Macrospin model In this section we develop a useful toy model in which the magnetization of the Py layer is assumed to be spatially uniform. Section 6.1 presents the equations of motion, Sec. 6.2 derives the natural frequency for small-angle precession about equilibrium, and Sec. 6.3 derives the magnetization's linear susceptibility to general oscillatory torques. Finally, we perform simulations to roughly describe the parametrically driven, large-amplitude dynamics in Figs. 3-4 of the main text in Sec. 6.4. 6.1 Equation of motion We employ the Landau-Lifshitz-Gilbert equation in the low-damping limit, with the spin Hall torque from the Pt layer [11]: ∂t m = γ0µ0 ~Heff × m + α m × ∂t m + gµBηθSHJPt 2etP yMs ≈ γ0µ0(cid:16) ~Heff × m + α m ×(cid:16) ~Heff × m(cid:17)(cid:17) + m × ( m × y) gµBηθSHJPt 2etPyMs m × ( m × y) . (70) Here, m = mx x+my y+mz z is a unit vector describing the orientation of the magnetiza- tion, γ0 is the magnitude of the gyromagnetic ratio, µ0 is the magnetic permeability of free space, ~Heff is the effective field (discussed below), α is the damping parameter (ap- proximately equal to the Gilbert damping for weak damping and spin transfer torques), µB is the Bohr magneton, η is the fraction of incident spins that are absorbed by the Py layer, θSH is the spin Hall angle of Pt, JPt is the charge current density in the Pt layer (determined by the parallel resistor model of Sec. 5.3.1), e is the electron charge, tPy is the thickness of the Py layer, and Ms is its effective saturation magnetization. The effective field Heff = ~H0 + ~Han + ~HI, ~ (71) where ~H0 = Hx x + Hy y + Hz z is the applied field, the shape anisotropy field ~ Han = −Ms (Nyy − Nxx) my y − Ms (Nzz − Nxx) mz z ≡ −Hyxmy y − Hzxmz z, (72) (73) with Nij being the elements of a demagnetization tensor (assumed diagonal for simplic- ity, with Nxx + Nyy + Nzz = 1), and ~ HI = −aI(t)y (74) is the spatially averaged field generated by the instantaneous current I(t) flowing through the wire, with proportionality constant a. For our long wire (aligned along x), we assume Nxx (cid:28) Nyy (cid:28) Nzz ∼ 1, such that Hyx + Hzx ≈ Ms is the effective magnetization. 22 6.2 Small-angle precession frequency Our first goal is to estimate the resonant frequency for for small-angle precession when the applied field is sufficient to saturate my. To do so, we ignore dissipation and current in Eq. 70, and include an applied field with Hx = 0 (as in the experiment): ∂t m = γ0µ0 ~Heff × m ∂t m γ0µ0 = ((Hy − Hyxmy) y + (Hz − Hzxmz) z) × m (Hy − Hyxmy) mz − (Hz − Hzxmz) my (Hz − Hzxmz) mx − (Hy − Hyxmy) mx =   . (75) We find the equilibrium values mx0 and my0 of mx and my by setting ∂t m = 0, which gives three equations and three unknowns. Assuming Hy is large enough that mx = 0 leaves only the first equation, which can be written in terms of mz0 as = Hz mz0 + Hyx − Hzx. (76) Hyp1 − m2 z0 We note that Hz ∼ 0.5 T is required to saturate the magnetization along z meaning our ∼ 10 mT fields will only slightly raise the magnetization out of the plane. As such, we assume mz0 (cid:28) 1 so that my0 =p1 − m2 z0. To first order in mz0, this simplifies to z ≈ 1 − 1 2m2 mz0 ≈ Hz Hy + Hzx − Hyx . (77) For our system's effective saturation fields µ0Hyx = 7.57 mT and µ0Hzx = 517 (see Sec. 7), a 35-mT field applied along the NV axis (35◦ out of plane), mz0 ≈ 0.04 (2.3◦ out of plane). To find the natural precession frequency νFMR, we therefore apply the limit mx, mz (cid:28) 1 and my ≈ 1, to Eq. 75, which yields coupled differential equations for mx and mz: ∂tmx γ0µ0 ≈ (Hy + Hzy) mz − Hz ∂tmz γ0µ0 ≈ − (Hy − Hyx) mx, with Hzy = Hzx − Hyx. Using the trial solution mx = X0 cos(2πνFMRt) mz = mz0 − Z0 sin(2πνFMRt) 23 (78) (79) (80) (81) with real-valued amplitudes X0 and Z0 yields from which the amplitude ratio 2πνFMRX0 ≈ γ0µ0 (Hy + Hzy) Z0 2πνFMRZ0 ≈ γ0µ0 (Hy − Hyx) X0, X0 Z0 =s Hy + Hzy Hy − Hyx (82) (83) (84) (85) and resonant frequency νFMR = γ0µ0 2π q(Hy − Hyx) (Hy + Hzx − Hyx). Importantly, this is the same (Kittel) formula one would arrive at with a purely in-plane field, which makes sense in this small-mz limit. Also, even for our maximal in-plane field µ0Hy = 33 mT, X0/Z0 ∼ 4.7, and this ratio increases at lower fields, diverging at Hy = Hyx, as expected. As such, the mz-generated stray field power at the NV (i.e., the quantity responsible for the spin relaxation rates) is at least (X0/Z0)2 ∼ 22 times smaller than that of mx (and much smaller at the fields of interest). 6.3 Linear susceptibility for magnetization approximately along the in-plane hard axis We now derive the susceptibility of mx to an oscillatory drive at frequency ν. Since we know the magnetization is saturated along y to good approximation (and behaves as though Hz = 0 for our parameter range; see Sec. 6.2), we consider an applied field ~ H0 k y for simplicity, and include a general infinitesimal drive torque (86) of amplitudes δx and δz. Equation 70 can then be written (replacing the current-induced terms with this torque) as ∂t m = γ0µ0 (δx x + δz z) ei2πνt (87) (88) (89) (90) δx 0 δ z mx my mz ∂t m γ0µ0 ∂t m = (cid:16) ~H0 + ~Han − α(cid:16) ~H0 + ~Han(cid:17) × m(cid:17) × m + (δx x + δz z) ei2πνt.   ei2πνt In the limit mx, mz (cid:28) 1 and my ≈ 1 to first order, so this becomes  +  γ0µ0 ≈  −Hzxmz + α (Hy − Hyx) mx  ×  −α (Hy + Hzy) mz Hy − Hyx or ∂tmx γ0µ0 ≈ (Hy + Hzy) mz − α (Hy − Hyx) mx + δxei2πνt ∂tmz γ0µ0 ≈ −α (Hy + Hzy) mz − (Hy − Hyx) mx + δzei2πνt. 24 (91) (92) (93) (94) (95) (96) (97) Using the trial solution mx = X0ei2πνt mz = Z0ei2πνt with complex amplitudes X0 and Z0 yields an in-plane steady state amplitude 1 2 2π (cid:17)(cid:18) νzyδz + iνδx FMR − ν2 + iν∆ν(cid:19) Hzx(cid:19) 2π (cid:18)Hy − Hyx + γ0µ0 ν2 X0(ν) ≈ (cid:16) γ0µ0 ∆ν ≡ 2α γ0µ0 νzy ≡ 2π γ0µ0 νyx ≡ 2π νFMR ≡ √νzyνyx. (Hy + Hzy) (Hy − Hyx) with The linewidth nominally depends on Hy, but Hzx is the dominant effect, and so ∆ν should remain approximately constant over our (small) field range. Also, torques along x and z produce qualitatively different lineshapes, in principle enabling a torque vector measurement [11, 15, 16]. 6.3.1 Mixdown voltage and FMR fit function If the applied field is tilted slightly toward x, the equilibrium magnetization will gain a small component mx0. As long as mx0 (cid:28) 1, the response X0(ν) should not change to first order (by symmetry7). However such a tilt does provide access to an in-phase anisotropic magnetoresistance (AMR) oscillation with phase delay ψ, amplitude ∆RRF, and in-phase component ∆RRF cos ψ (see Sec. 4) proportional to the real part of X0, which can be written (see Eq. 93) Reh X0(ν)i ≈(cid:16)γ0µ0 2π (cid:17) δzνzy (ν2 FMR − ν2) + δxν2∆ν FMR − ν2)2 + ν2∆ν2 (ν2 . (98) For small-angle precession, the static change in resistance ∆R0 should contribute very little to the FMR signal, but would scale as FMR − ν2)2 + ν2∆ν2 (ν2 7The resonant frequency νFMR increases ∝ m2 x0 to lowest order, since the anisotropy field maximally opposes the applied field when aligned with y. ν2 zyδ2 z + ν2δ2 x , (99) 2 (cid:12) (cid:12) X0(ν)(cid:12)(cid:12)(cid:12) (cid:12) ∝ 25 which is close in form to Reh X0(ν)i. As such, FMR spectra are well fit by VMR(ν) = az (ν2 FMR − ν2) + axν2∆ν FMR − ν2)2 + ν2∆ν2 (ν2 , (100) with free parameters az, ax, νFMR, and ∆ν. 6.3.2 Expected spectrum of Brownian magnetization noise A thermal (Langevin) field [17] is typically assumed isotropic, exerting stochastic, un- correlated torques with a white noise spectrum in all three dimensions. If the torque power spectral densities of each component are ST (units of rad2 sec−2 Hz−1), then we expect the power spectral density of mx to be scaled by the magnitude of the suscep- tibility squared: Smx ∝ (cid:12)(cid:12)(cid:12)(cid:12) ∝ (ν2 ν2 2 νzy FMR − ν2 + iν∆ν(cid:12)(cid:12)(cid:12)(cid:12) zy + ν2 ν2 FMR − ν2)2 + (ν∆ν)2 . ST +(cid:12)(cid:12)(cid:12)(cid:12) ν2 iν FMR − ν2 + iν∆ν(cid:12)(cid:12)(cid:12)(cid:12) 2 ST (101) 6.4 Parametrically driven, large-amplitude oscillations The transport technique used to measure the signals shown in Fig. 2 of the main text are most sensitive to the most spatially uniform magnetic oscillations. To gain some qualitative intuition about the observed large-amplitude, parametrically driven dynamics, we numerically integrate Eq. 70 with γ0 = 2π × 29.25 GHz/T (electron g-factor g = 2.09 for 10-nm-thick Py [18]), ηθSH = 0.055 [11], layer resistivity ratio ρPy/ρPt = 2.977, and tPy = 10 nm. From our fits (Sec. 7) we use µ0Hyx = 7.57 mT and µ0Hzx = 517 mT and effective magnetization µ0Ms = 525 mT for consistency. The proportionality constant for current-induced field a = 1.72 mT/mA is estimated from the required compensation field at 4.08 mA in Fig. 2 of the main text. In order to achieve a similar threshold for parametric oscillations (i.e., not needing unreasonably large currents), we choose α = 0.02, which is approximately half the value estimated from our FMR fits. This discrepancy is under study, but we nominally think it is related to the actual device's nonuniform magnetization and / or other nonidealities such as material contamination (e.g., oxidization) and roughness.8 To mimic the red data in Fig. 2(a) of the text, we apply a field µ0H0 = 23.5 mT ×(cid:16)q 2 3 y +q 1 3 z(cid:17) along the NV axis, and current I(t) = I0 + I1 cos(2πν1t) (102) 8Macrospin models are often surprisingly accurate in describing actual nanostructured systems, but their results should always be considered with caution. 26 Figure 8: Simulated magnetoresistance signal ∆VMR (orange) and the con- tribution ∆Vmix from mixing with RF current I1 alone (blue). Simulation parameters are listed in the main text. At this value of DC bias I0 = 4.08 mA, the signal is dominated by I0 h∆R(t)i, which has a different line shape at the fundamental frequency. (i) Trajectory of m at ν1 = 3.1875 GHz, where ∆VMR = 120 µV. (ii) Time domain showing the relative phase of the resistance (∝ m2 x) and current oscillations. with I0 = 4.08 mA, I1 = 1 mA, and ν1 stepped from 1-3.5 GHz. At each frequency, the magnetization is initialized to within 0.1◦ of equilibrium (along y and canted 1.4◦ out of plane), and evolved for 100 ns to ensure steady state. The time-averaged change in voltage ∆VMR due to magnetoresistance is then calculated as ∆VMR = hI(t)R(t) − I0R0i = h(I0 + I1 cos(2πν1t)) (R0 + ∆R(t)) − I0R0i = hI(t)∆R(t)i, where R0 is the undriven resistance, and ∆R(t) = R0δAMRm2 x (103) (104) (105) (106) is the time-dependent resistance change due to precession, with R0δAM R = 0.2 Ω. Figure 8 shows ∆VMR calculated using an integer number of oscillations in the last 10 ns of each simulation. A large-amplitude parametrically driven peak occurs near the second harmonic of the ferromagnetic resonance above 3 GHz, with a skew toward higher frequencies similar to the data in Fig. 2(a) of the main text. The magnitude is also quantitatively similar, corresponding to in-plane precession about the equilibrium offset (mz = 0.025) amplitude of 30◦ when ∆VMR = 120 µV, the trajectory of which is shown in inset (i). The peak is positive because the antidamping spin transfer torque 27 (i) (ii) (pointing away from the inset plot's origin) is, on average, larger when the angle is maximal, though the phase of the resistance oscillations lags behind the drive as shown in Fig. 8(ii). Finally, we note the presence of a much smaller directly-driven oscillation at νFMR, arising from the small equilibrium value hmzi and the oscillatory current- induced field ~HI. This feature is visible in Fig. 4(a) of the main text. 6.4.1 Nanowire stray field along NV axis In this section, we use the NV's field-dependent electron spin resonance (ESR) to es- timate the strength of the stray field along the NV axis when the device is statically magnetized along y. Figure 9 shows the photoluminescence (PL) spectra with field µ0H0 = 22.5 mT applied along the NVA axis for (blue data) NVA, (green data) a refer- ence NVref having the same orientation as NVA but positioned 5 µm from the device, where the stray field is negligible, and (orange data) NVA after the nanowire's demise, which removed all evidence of ferromagnetism (we suspect due to oxidization or de- struction the Py layer). From the difference in fit ESR frequencies (see Table 1), we estimate the axial stray field at NVA most likely lies between 2.1 and 2.6 mT, and we use the "dead device" value 2.5 ± 0.1 mT as our best estimate. Figure 9: Nanowire stray field as measured by ESR. Measurements of the ms = 0 → ±1 transitions of NVA while the device was magnetized, after the device had oxidized, and of NVref are fit to Lorentzian profiles in the solid, dotted and dash-dotted lines respectively. Note the differences in the linewidths and contrasts are due in part to imperfect power coupling to the stripline, and inhomogeneity in the external field can only account for a maximum of 2 MHz of deviation between NVA and NVref. 28 NVA, device magnetized NVA, no magnetization NVref 2.2 2.3 2.4 0.2 0.4 0.6 Frequency, SL (GHz) 3.4 3.5 3.6 0.8 3.7 1.0 1.0 1.00 0.8 0.95 0.6 0.90 0.85 0.4 0.80 0.2 0.75 PL (normalized) 0.70 0.0 2.1 0.0 Table 1: Fit results from ESR resonances in Fig. 9. The net NV axial field is estimated assuming a the free electron gyromagnetic ratio γNV = 2π × 28.0 GHz/T. NVA, magnetized device NVA, oxidized device NVref ∆ν (MHz) 1362.0 ± 0.2 1222 ± 4 1230 ± 10 Bk (mT) 24.409 ± 0.003 21.9 ± 0.1 22.1 ± 0.2 6.4.2 Estimating the parametric precession angle Knowing the strength of the nanowire's stray field at NVA (Sec. 6.4.1 above) allows us to estimate the parametric oscillation angle ∆θ from the reduction in stray field shown in Fig. 3(a) of the main text, using the macrospin approximation. As discussed above, the steady-state trajectory is roughly sinusoidal and highly confined to the xy plane, so we approximate for simplicity mx ≈ sin θ(t), my ≈ cos θ(t), mz ≈ 0, θ(t) ≈ ∆θ cos(πνNWt + ψ), (107) (108) (109) (110) where θ(t) is the time-dependent in-plane angle from y, ∆θ is the steady-state am- plitude, and νNW nanowire's drive frequency, equal to twice the parametric response frequency, and ψ is a steady-state phase shift. In this limit, we can calculate the time- averaged magnetization along y, which will reduce the stray field experienced by NVA as hmyi ≈ hcos (∆θ cos(πνNWt + ψ))i = J0(∆θ) (111) (112) 4∆θ2) is the zeroth order Bessel function. Figure 3(a) in the main where J0(∆θ) (≈ 1− 1 text shows an increase in the ms = 0 → −1 transition frequency of ∆ν− = 13 MHz, which corresponds to a decrease in stray field of ∆Bstray ≈ 0.5 mT. Assuming the precession is sufficiently symmetric that the stray field orientation remains the same, the fractional change ∆Bstray Bstray ≈ 1 − hmyi. (113) Using Bstray = 2.5 mT from Sec. 6.4.1 at the same external field, we estimate ∆θ ≈ 60◦. Under the same approximations, we can independently estimate ∆θ from the mag- netoresistance signal ∆R(t) = R0δAMR sin2 (∆θ cos(ωrt + ϕ)) . (114) 29 (115) 1 2 When current I(t) = I0 + IRF cos(ωt) is sent through the device, with I0 & 4 mA and IRF ∼ 1, time-averaging the precession-induced voltage yields R0δAMRI0 (1 − J0(2∆θ)) , ∆VMR ≈ where we have dropped the comparatively small "mixdown" term involving IRF for simplicity (the IRF term contributes . 10% to the total signal for this range of pa- rameters, as shown in supplementary Sec. 6.4). In this limit, the peak measurement of ∆VMR = 360 µV in Fig. 3(a) of the main text (I0 = 4.9 mA, I1 = 1.15 mA, R0δAMR = 0.2 Ω) corresponds to ∆θ ≈ 55◦. Similarly, the 120 µV parametric peak in Fig. 2(a) of the main text (I0 = 4.08 mA) corresponds to ∆θ ≈ 32◦, in reasonable agreement with the macrospin simulation (Sec. 6.4). 30 Figure 10: Estimating the ferromagnetic resonance (FMR) frequency νFMR at in-plane angle θ = 0. (a) Spin-transfer-driven FMR spectra for θ = ±4◦ and ±8◦ for static field µ0H0 = 16.5 mT applied 35◦ out of plane, as in the main text. Traces are offset for clarity. Dark lines show Fano fits, and gray lines extend these fits outside the chosen range of frequencies. (b) Fitted frequencies (blue points) over the full range of angles, fit to a symmetric quartic function. (c) Summary of so-estimated νFMR(0) 0◦, with the range 14-35 mT fit with a simple Kittel model (blue curve), as well as a lowest-order spinwave model assuming the "extreme" values of fixed effective widths 50 nm (orange curve) and 1.6 µm (green curve). 7 Ferromagnetic resonance frequencies Because the spin transfer drive and magnetoresistance signal both approximately vanish when the equilibrium value of m is (nearly) parallel to y (i.e., the field's in-plane angle θ = 0, we cannot measure the ferromagnetic resonance (FMR) frequency νFMR directly. Instead, we measure νFMR(θ) for a set of 8 to 9 small angles spanning ±10◦ (maintaining the out-of-plane angle 35◦ as in the main text), and then fit the resulting frequencies to a symmetric polynomial to estimate νFMR(0) (and a misalignment angle θ0). Figure 10(a) shows a "typical" set of spin-transfer-driven FMR spectra with applied field µ0H0 = 16.5 mT, taken as discussed in Sec. 5.2.2, with the Joule heating back- ground (Sec. 5) subtracted. Due to the frequency-dependence of the drive current (see Sec. 5.2.4), we fit only the data near the resonant feature to Eq. 100 to extract the frequency νFMR and width ∆ν. As expected, the signal increases with θ. Also, as shown in Fig. 10(b), the frequency decreases as θ approaches zero, consistent with the shape anisotropy maximally opposing the applied field at θ = 0. Exploiting the mirror symmetry of our geometry, we fit the observed angular dependence in Fig. 10(b) to a low-order symmetric polynomial of the form (116) with fit constants C0, C2, C4, and offset angle θ0. The offset angle (θ0 = 0.4◦ ± 0.1◦ for the shown data set) takes on values within ±0.5◦ of 0◦ over the usable range νFMR(θ) = C0 + C2(θ − θ0)2 + C4(θ − θ0)4, 31 a b c ip of applied fields (14-35 mT). Completing the same analysis at each value of applied field produces the zero-angle frequency data νFMR(0) shown in Fig. 10(c). Clear fit systematics preclude the trustworthiness of frequencies so estimated below 14 mT. The "reliable" region (dark symbols) can then be fit to a variety of models to estimate material parameters of the permalloy (Py) layer. 7.1 Macrospin approximation to resonant frequency To gain immediate insight, we first assume the magnetization simply behaves as a uniformly-magnetized ellipsoid with equilibrium magnetization m approximately par- allel to y. As derived in Sec. 6.2, the resonant frequency for our geometry (Eq. 85) νFMR = γ0µ0 2π q(Hy − Hyx)(Hy + Hzx), (117) where Hy = H0 cos(35◦) is the in-plane component of the applied field. Fitting the "reliable" range of data (blue curve in Fig. 10(c)) yields effective saturation fields µ0Hyx = 7.57 ± 0.08 mT and µ0Hzx = 517 ± 4 mT. The low value of Hzx (nomi- nally close to the saturation magnetization) suggests non-uniform magnetic dynamics and / or other nonidealities of the Py layer. 7.2 Lowest-order spinwave approximation to resonant frequency To get a sense of scale for the potential impact of nonuniform dynamics, we can also perform a fit to an approximate lowest-order (most uniform) spinwave resonance, which has frequency [19, 20] νk = γ0µ0 2π q(Hy + Msλexk2 − Hd) (Hy + Msλexk2 − Hd + MsFk), (118) where Hy = H0 cos(35◦) is the in-plane component of the applied field, Hd is an effective demagnetizing (dipole) field, Ms is Py's saturation magnetization, λex = 2Aex/(µ0M 2 s ), with exchange constant Aex = 1.05 × 10−11 J/m [21], k is the magnitude of the spin wave vector k = kx x + ky y + kz z, and Fk = 1 + gk(cid:0)sin2 θk − 1(cid:1) + Msgk(1 − gk) sin2 θk Hy − Hd + Msλexk2 , (119) where θk is the angle between the equilibrium magnetization orientation h mi and k, and gk = 1 −(cid:0)1 − e−ktPy(cid:1) /(ktPy) with device thickness tPy. For our thin film, we expect m to be approximately uniform along z [19, 22] (we also ignore the small offset in the equilibrium out-of-plane component hmzi). The nanowire further constrains the longitudinal wavenumber as kx = nxπ/Leff, with nx = 1, 2, 3, . . . and Leff ≈ 8.05 µm deviating slightly from the geometrical length due to effective dipolar boundary conditions on the wire [22, 23]. The remaining relevant wave number is often written ky = nyπ/weff in terms of the transverse mode number ny = 1, 2, 3, . . . 32 x + π weff y. and an effective width weff, which we expect to be smaller than the actual wire width when m k y [20, 22, 24]. The lowest frequency (and most spatially homogeneous) mode should have nx = ny = 1, so we treat the resonances measured in transport as having a wave vector kFMR = π Leff Equation 118 effectively contains two fit parameters, Msλexk2−Hd and MsFk, which themselves are composed of three unknown quantities, weff, Ms, and Hd. Table 2 shows the fit values of Hd for a wide range of assumed weff, with the corresponding fits for weff = 50 nm (orange) and weff = 1.6 µm (green) plotted in Fig. 10(c) showing negligible deviation from the macrospin model. As expected, as weff becomes large, the Hd → Hyx and Ms → Hyx + Hzx from the macrospin approximation (Sec. 7.1). We presume weff should not be smaller than 50 nm, and the low value of µ0Ms still suggests some nonidealities in the Py layer, which will be the subject of future investigations and higher quality materials deposition. Note that modifying the parameters Ms, Hd, weff or Aex by factors of order unity does not affect the key results – observed spin transfer threshold for parametric oscillations, observed stray fields (or lack thereof) at the NV, and observed spin transfer damping – of the main text. Table 2: Fit results of lowest-ordered magnon modes with fixed weff to data in Fig. 10(c). Fixed weff (nm) µ0Hd (mT) 50 100 200 400 800 1600 155 ± 1 50.1 ± 0.3 19.17 ± 0.04 10.81 ± 0.06 8.79 ± 0.08 8.58 ± 0.09 µ0Ms (mT) 709 ± 6 613 ± 5 569 ± 5 548 ± 5 537 ± 5 532 ± 5 33 8 Magnon-induced NV spin relaxation rates Figure 11: Reconstructing the field and frequency dependence of the NV spin relaxation rate Γ0 in the absence of spin transfer effects. (a) Spectrum of spin wave mode frequencies for the first transverse mode (ny = 1, nx = 1, 2, 3, ..., gray lines), and NV resonance frequencies ν+ (orange) and ν− (blue) when vary- ing only the field (filled circles) or varying the DC bias while compensating the field at the Py layer (hollow squares). The fundamental mode (nx = ny = 1) is highlighted in red, and the green lines correspond to the second fundamental transverse mode (nx = 1, ny = 2) for different assumed effective widths weff (labeled); all modes follow the same family of curves to good approximation (deviating from each other by much less than a spin wave linewidth) over the studied range. The vertical line at µ0H0 = 16.5 mT indicates the field at which the spin transfer effects were probed in Fig. 5(right) of the main text. Dashed lines highlight which NV measurements (filled symbols) are used to estimate Γ0 at which frequencies along the vertical line. (b) Reconstructed spin-transfer- free relaxation rates (solid line) along the vertical line cut in (a) with error bars from the bias-free measured values (solid points in (a)). The dashed lines represent the absolute (and quite extreme) bounds of the analysis. Orange (blue) squares correspond to Γ0+ (Γ0 −) in Fig. 5(b) of the main text. In this section, we use the spin-transfer-free relaxation rates Γ(ν, H) measured at a variety of NV probe frequencies ν and applied fields H to estimate the rates at other values of ν and H (i.e., the color scales in Fig. 5 of the main text). The basic idea can be understood by inspecting the phase space of spin wave modes shown in Fig. 11(a). Here, many spin wave mode frequencies νk (where k = kx x + ky y + kz z is the mode wavenumber; see Sec. 7.2) are plotted over our range of applied fields. Importantly, all modes follow the same family of curves to good approximation, meaning one can 34 a = 2 0 0 n m w e f f b (ii) f f = 4 0 w e 0 n m 0 n m f f = 8 0 w e (i) I0 < 0 (damping) I0 > 0 (antidamping) use a measurement of Γ(ν, H) (taken at the filled symbols) to estimate the relaxation rate Γ0(ν0, H0) at another location along the nearest νk curve; the quantities that vary the most along these curves are the field, frequency, density of states, and thermal occupancy, all of which are known or can be approximated, as discussed below. Figure 12: Spin wave dispersion and broadened density of modes assuming an effective width weff = 200 nm. (a) Noise distribution Pk versus frequency ν and longitudinal wavenumber kx, showing the first (ny = 1) and second (ny = 2) transverse modes at an applied field of µ0H0 = 16.5 mT along the NV axis. (b) Broadened density of modes g(ν), and the contributions from the first (gny=1(ν)), second (gny=2(ν)), and third (gny=3(ν)) modes. First, we note that the total NV spin relaxation rate [25] Γ(ν, H) = Γ0 + γ2 NV 2 S⊥(ν, H) (120) ≈ (121) S⊥(ν, H) comprises the sum of the NV's (small) internal rate Γ0 −(+) = 64± 7 Hz (54± 12 Hz) (as measured with NVA at µ0H0 = 22.5 mT after the device magnetization disappeared) NVS⊥/2 (∼ kHz) driven by the magnetization's stray field noise power and the rate γ2 spectral density S⊥(ν, H) (units of T2/Hz), where γNV is the magnitude of the NV spin's gyromagnetic ratio and the subscript ⊥ reminds us that it is the fields perpendicular to the NV axis that drive the transitions. Each spin wave mode contributes noise power in proportion to its thermal occupancy ¯n(νk), and so we can write where fk is a mode-dependent geometrical constant converting occupancy to noise power at the NV (units of T2/magnon), and Pk(ν, H) is a unity-normalized density γ2 NV 2 35 S⊥(ν, H) =Xk ¯ n(νk)fkPk(ν, H), (122) b a ny = 2 ny = 1 function (units of 1/Hz) describing how this power is distributed over the frequency domain (i.e., a normalized version of the mode's power susceptibility, such as Eq. 101). To simplify the analysis, we assume the mode profiles do not change much over our range of applied fields, so that fk is approximately independent of field. We do expect fk to depend on k and the exact location of the NV relative to the nanowire, taking on the largest values when 1/k is comparable to the wire-NV distance d [6, 26]. To simplify further, we note that the observed linewidth ∆ν ∼ 600 MHz of the fundamental mode is approximately constant over our field range, consistent with the behavior predicted by the macrospin approximation (Sec. 6.3). We therefore assume the distributions Pk(ν, H) depend only on νk(H) and the probe frequency ν. Figure 12(a) shows an example Pk for applied field µ0H0 = 16.5 mT along the NV axis. Summing across kx, we then define a total "linewidth-broadened density of modes" g(ν, H), which is shown in Fig. 12(b). If we imagine following one of the νk(H) curves in Fig. 11(a), we notice two impor- tant quantities change – the mode frequencies νk(H) and the density of states g(ν, H), suggesting that, if we factor these trends from Eq. 122, we can write (123) (124) in terms of a "weighting factor" S⊥(ν, H) = ¯n(ν)g(ν, H)Xk wk(ν, H)fk, wk(ν, H) ≡ ¯ n(νk)Pk(ν, H) ¯ n(ν)g(ν, H) that should be fairly insensitive to the distance traveled along a given νk curve. Of particular relevance to the cooling argument of the main text, Fig. 13(a) shows these weights for the direct measurement at point (i) in Fig. 11(a) (cyan circles) as well as the location of maximum cooling (ii) (magenta markers). As we have engineered, both peaks occur at the same value of kx (7 rad/µm), which is a restatement of the fact that we have moved along a νk curve. Importantly, the distributions at these two extremes look very similar, with individual weights differing by at most 34%, as shown in Fig. 13(b). We remind ourselves that these weights are multiplied by values of fk expected to oscillate with kx underneath a smooth envelope, such that the summed effect is more likely of order the wk-weighted average of the ratio in Fig. 13(b) (i.e., ∼3%). The presence of additional transverse modes having comparable values of will not have an impact beyond that of Fig. 13(b) unless their mode frequencies deviate from the fundamental mode's family of νk(H) curves (gray lines in Fig. 11(a)) by an amount comparable to ∆ν over the length of the dotted blue line. Even including a more complicated spin wave model or micromagnetic simulations, we do not expect to find modes so strongly deviating from these trends over so small a field range. We can now convert the directly-measured values of Γ at frequencies ν and fields H (filled circles in Fig. 11(a)) into estimates at other values ν0 and H0 using the ratio Γ0(ν0, H0) Γ(ν, H) = S⊥(ν0, H0) S⊥(ν, H) = n(ν0) ¯ ¯ n(ν) 36 g(ν0, H0) g(ν, H) Pk wk(ν0, H0)fk Pk wk(ν, H)fk . (125) Figure 13: Magnon transfer function weights at points (i) and (ii) in Fig. 11(a). (a) Relative magnon weights wk probed at frequencies νk = 2.45 GHz (13.5 mT) and ν0k = 2.67 GHz (16.5 mT) versus kx, which peaks at kx = 7 rad/µm (as indicated in Fig. 12 for µ0H00 = 16.5 mT). (b) Ratio of weights for these two fields. The first ratio ¯n(ν0)/¯n(ν) = (ehν/kBT − 1)/(ehν0/kBT − 1), with Planck constant h, Boltzmann constant kB, and temperature T . The second ratio can be calculated as in Fig. 12, and the final ratio should be comparable to 1, as discussed above. The resulting reconstruction is plotted (solid line) for µ0H0 = 16.5 mT, i.e., the field at which the spin transfer measurements were made in Fig. 11(b). Similar curves can be generated at other values of field, as is plotted on the color scales in Fig. 5 of the main text. Also plotted in Fig. 11(b) are upper and lower bounds (dashed curves) corresponding to worst-case-scenario systematic errors in the final ratio, calculated by assuming the only contributing mode is the one having the largest and smallest values of wk(H0)/wk(H) in Fig. 13(b). Importantly, the expected bias-dependent changes in Γ0 ± for Fig. 5(b) of the main text (the probe frequencies of which are indicated by orange and blue squares in Fig. 13) is much smaller than what is observed, and has the opposite trend with applied current. Figure 14(a) shows the same calculations for effective widths spanning a wide range. As expected, the presence of different transverse mode structure has little effect on these estimates. Note that, below the fundamental resonance, there is no obvious choice for the blue dashed curves, but there are still modes whose tails contribute to S⊥. As such, we have chosen to fix the difference from the fundamental mode frequency νFMR (where the density of modes is highest). In this region, the approximation that the final ratio in Eq. 125 is ≈ 1 becomes increasingly incorrect – mainly because the occupation at the probing frequency no longer matches the occupation of the nearest magnon modes – as evidenced by Γ0 exceeding the upper bounds at frequencies below νFMR = 1.64 GHz. This "spin-transfer-free" approximation is valid to the specified tolerances provided the linewidth ∆ν is constant. When damped by spin transfer, however, we expect the linewidth to broaden, which can redistribute noise away from νk and potentially reduce 37 a b (i) (ii) Figure 14: Reconstructed relaxation rates Γ0 for a range of effective widths weff. (a) Γ0 at µ0H0 = 16.5 mT for different assumed values of weff, illustrat- ing a minimal impact from changes in mode structure, in particular near the spin-transfer probe frequencies of ν±,ST. (b) Ratio of the linewidth-broadened density of states at the same field for a factor of 2 increase in ∆ν, illustrating that the total density of states increases at the probe frequency ν− = 2.58 GHz of maximal cooling (vertical line). Γ without cooling. However, as shown in Fig. 14(b), the broadened density of modes actually increases under the conditions of maximal spin transfer cooling (vertical line), where the linewidth changes by at most a factor of ∼2. In the worst shown scenario, where the effective width weff = 100 nm, such that the second transverse mode is well above the probe frequency, the broadened density of modes at the probe frequency ν− = 2.58 GHz still increases due to the tails of the modes away from ν−. As weff increases, the higher-order transverse mode frequencies approach ν−, and g is found to increase by as much as ∼ 25%. Therefore, if we assume that fk does not vary significantly over the resonance linewidth (and / or varies linearly), then the sum in Eq. 123 should remain roughly the same or increase (in opposition to the observed 38 I0 > 0 (antidamping) I0 < 0 (damping) a b trend) as a result of the increased ∆ν. Combined with the fact that the maximum temperature change due to Joule heating ∆T . 5 K (see Sec. 5.2.3), we expect this is not a dominant issue. References [1] C. Lee, E. Gu, M. Dawson, I. Friel, and G. Scarsbrook, "Etching and micro-optics fabrication in diamond using chlorine-based inductively-coupled plasma," Diamond & Rel. Mat. 17, 1292 (2008). [2] Y. Chu, N. de Leon, B. Shields, B. Hausmann, R. Evans, E. Togan, M. J. Burek, M. Markham, A. Stacey, A. Zibrov, A. Yacoby, D. Twitchen, M. Loncar, H. Park, P. Maletinsky, and M. Lukin, "Coherent optical transitions in implanted nitrogen vacancy centers," Nano Lett. 14, 1982 (2014). [3] C. Kurtsiefer, S. Mayer, P. Zarda, and H. Weinfurter, "Stable solid-state source of single photons," Phys. Rev. Lett. 85, 290 (2000). [4] R. Verberk and M. Orrit, "Photon statistics in the fluorescence of single molecules and nanocrystals: Correlation functions versus distributions of on- and off-times," J. Chem. Phys. 119, 2214 (2003). [5] J.-P. Tetienne, L. Rondin, P. Spinicelli, M. Chipaux, T. Debuisschert, J.-F. Roch, and V. Jacques, "Magnetic-field-dependent photodynamics of single NV defects in diamond: an application to qualitative all-optical magnetic imaging," New J. Phys. 14, 103033 (2012). [6] T. van der Sar, F. Casola, R. Walsworth, and A. Yacoby, "Nanometre-scale probing of spin waves using single electron spins," Nature Comm. 6, 7886 (2015). [7] P. Elleaume, O. Chubar, and J. Chavanne, "Computing 3d magnetic fields from in- sertion devices," in Proceedings of the 1997 Particle Accelerator Conference (IEEE, 1997). [8] M. W. Doherty, N. B. Manson, P. Delaney, F. Jelezko, J. Wrachtrup, and L. C. Hollenberg, "The nitrogen-vacancy colour centre in diamond," Phys. Rep. 528, 1 (2013). [9] S. Felton, A. M. Edmonds, M. E. Newton, P. M. Martineau, D. Fisher, D. J. Twitchen, and J. M. Baker, "Hyperfine interaction in the ground state of the negatively charged nitrogen vacancy center in diamond," Phys. Rev. B 79 (2009). [10] S. O. Valenzuela and M. Tinkham, "Direct electronic measurement of the spin Hall effect," Nature 442, 176 (2006). 39 [11] L. Liu, T. Moriyama, D. C. Ralph, and R. A. Buhrman, "Spin-torque ferromag- netic resonance induced by the spin Hall effect." Phys. Rev. Lett. 106, 036601 (2011). [12] V. Tshitoyan, C. Ciccarelli, A. P. Mihai, M. Ali, A. C. Irvine, T. A. Moore, T. Jung- wirth, and A. J. Ferguson, "Electrical manipulation of ferromagnetic NiFe by antiferromagnetic IrMn," Phys. Rev. B 92, 214406 (2015). [13] J. C. Sankey, P. M. Braganca, A. G. F. Garcia, I. N. Krivorotov, R. A. Buhrman, and D. C. Ralph, "Spin-transfer-driven ferromagnetic resonance of individual nano- magnets," Phys. Rev. Lett. 96 (2006). [14] Z. Duan, C. T. Boone, X. Cheng, I. N. Krivorotov, N. Reckers, S. Stienen, M. Farle, and J. Lindner, "Spin-wave modes in permalloy/platinum wires and tuning of the mode damping by spin Hall current," Phys. Rev. B 90 (2014). [15] J. C. Sankey, Y.-T. Cui, J. Z. Sun, J. C. Slonczewski, R. A. Buhrman, and D. C. Ralph, "Measurement of the spin-transfer-torque vector in magnetic tunnel junctions," Nature Phys. 4, 67 (2008). [16] H. Kubota, A. Fukushima, K. Yakushiji, T. Nagahama, S. Yuasa, K. Ando, H. Maehara, Y. Nagamine, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and Y. Suzuki, "Quantitative measurement of voltage dependence of spin-transfer torque in MgO-based magnetic tunnel junctions," Nature Phys. 4, 37 (2008). [17] W. F. Brown, "Thermal fluctuations of a single-domain particle," Phys. Rev. 130, 1677 (1963). [18] J. P. Nibarger, R. Lopusnik, Z. Celinski, and T. J. Silva, "Variation of magnetiza- tion and the Landé g factor with thickness in Ni-Fe films," Appl. Phys. Lett. 83, 93 (2003). [19] B. A. Kalinikos and A. N. Slavin, "Theory of dipole-exchange spin wave spectrum for ferromagnetic films with mixed exchange boundary conditions," J. Phys. C: Solid State Phys. 19, 7013 (1986). [20] T. Brächer, P. Pirro, and B. Hillebrands, "Parallel pumping for magnon spintron- ics: Amplification and manipulation of magnon spin currents on the micron-scale," Phys. Rep. 699, 1 (2017). [21] N. Smith, D. Markham, and D. LaTourette, "Magnetoresistive measurement of the exchange constant in varied-thickness permalloy films," J. Appl. Phys. 65, 4362 (1989). [22] C. Bayer, J. Jorzick, B. Hillebrands, S. O. Demokritov, R. Kouba, R. Bozinoski, A. N. Slavin, K. Y. Guslienko, D. V. Berkov, N. L. Gorn, and M. P. Kostylev, 40 "Spin-wave excitations in finite rectangular elements of Ni80Fe20," Phys. Rev. B 72 (2005). [23] K. Y. Guslienko, S. Demokritov, B. Hillebrands, and A. Slavin, "Effective dipolar boundary conditions for dynamic magnetization in thin magnetic stripes," Phys. Rev. B 66 (2002). [24] T. Brächer, O. Boulle, G. Gaudin, and P. Pirro, "Creation of unidirectional spin- wave emitters by utilizing interfacial Dzyaloshinskii-Moriya interaction," Phys. Rev. B 95 (2017). [25] C. P. Slichter, Principles of Magnetic Resonance (Springer Berlin Heidelberg, 1990). [26] C. Du, T. van der Sar, T. X. Zhou, P. Upadhyaya, F. Casola, H. Zhang, M. C. Onbasli, C. A. Ross, R. L. Walsworth, Y. Tserkovnyak, and A. Yacoby, "Control and local measurement of the spin chemical potential in a magnetic insulator," Science 357, 195 (2017). 41
1703.04295	1	1703	2017-03-13T08:46:22	Phonon induced optical gain in a current carrying two-level quantum dot	[ "cond-mat.mes-hall" ]	In this work we consider a current carrying two level quantum dot(QD) that is coupled to a single mode phonon bath. Using self-consistent Hartree-Fock approximation, we obtain the I-V curve of QD. By considering the linear response of our system to an incoming classical light, we see that depending on the parametric regime, the system could have weak or strong light absorption or may even show lasing. This lasing occurs at high enough bias voltages and is explained by a population inversion considering side bands, while the total electron population in the higher level is less than the lower one. The frequency at which we have the most significant lasing depends on the level spacing and phonon frequency and not on the electron-phonon coupling strength.	cond-mat.mes-hall	cond-mat	a Phonon induced optical gain in a current carrying two-level quantum dot Amir Eskandari-asla,b, aDepartment of physics, Shahid Beheshti University, G. C. Evin, Tehran 1983963113, Iran bSchool of Nano Science, Institute for Research in Fundamental Sciences (IPM),P.O.Box: 19395-5531, Tehran, Iran Abstract In this work we consider a current carrying two level quantum dot(QD) that is coupled to a single mode phonon bath. Using self-consistent Hartree-Fock approximation, we obtain the I-V curve of QD. By considering the linear re- sponse of our system to an incoming classical light, we see that depending on the parametric regime , the system could have weak or strong light absorption or may even show lasing. This lasing occurs at high enough bias voltages and is explained by a population inversion considering side bands, while the total electron population in the higher level is less than the lower one. The frequency at which we have the most significant lasing depends on the level spacing and phonon frequency and not on the electron-phonon coupling strength. Keywords: optical gain, quantum dot, phonon, Hartree-Fock approximation 1. Introduction One of the most interesting issues in applied physics is the light matter interaction. Recent advances in nano technology made it possible to investigate the interaction of nano-meter sized objects with light. Quantum dot systems have been studied both experimentally[1, 2, 3, 4, 5, 6] and theoretically[7, 8, 9, 10] and it has been shown that these systems could be used as light sources or amplifiers. The electron-phonon interaction is shown to assist the optical gain. In these theoretical studies, a multi-level system is considered that is con- nected to electronic sources and drains and couples to a continues phonon bath and cavity photon fields. They usually use equations of motion in the Lindblad form to consider the effect of all environments on the system. However, if we want to manipulate molecular sized QDs, it is the case to consider a single mode phonon bath, as it is done in investigating molecular junctions[11]. Moreover, we could apply the methods of non-equilibrium Green's functions(NEGF)[12] to make it possible to study the effect of bias voltage more accurately. Email address: [email protected] (Amir Eskandari-asl) Preprint submitted to Physica B November 11, 2018 In this work, we theoretically consider a two-level QD connected to two leads and coupled to a single mode phonon bath. Using Hartree-Fock(HF) approx- imation we obtain the I-V curve of our system for different electron coupling strength. At a fixed bias voltage, by investigating linear response of the QD to an incoming classical light, we show that depending on the coupling strength and applied bias voltage, we could have weak or strong light absorption and at high enough bias voltages the system shows lasing at some special frequencies. Even though for the optical gain we don't have total population inversion and the higher level is less populated than the lower one, some kind of population inversion occurs between the side-bands, which explains the lasing in our model. In sec.1 we introduce our Hamiltonian and describe the HF approximation. In sec.2 we consider linear response to a classical electric field and obtain the formula for one photon absorption(OPA) cross section. In sec.3 we present our numerical results and sec.4 concludes our work. The paper is organized as follow. 2. Hamiltonian model and the method Our system consists of a QD with two orbitals with different parities, con- nected to two leads. Each orbital is spin degenerate and coupled to a single mode phonon bath. The Hamiltonian of the model in the presence of external electromagnetic field is HT = H 0 dot + Hleads + Htun + H 0 ph + He−ph + He−l. (1) The first two terms are, respectively, the effective Hamiltonian of a non-interacting two-orbital QD and isolated leads, given by † ic iσciσ dot = H 0 (2) (cid:88) (cid:88) i,σ and Hleads = † ka kσakσ, (3) k∈{R,L},σ † † where σ is the spin index, ciσ(c iσ) and akσ(a kσ) are, respectively, the annihila- tion (creation) operators of electron with spin σ of the i-th orbital of QD and k-th level of leads and i is the energy of i-th orbital. The third term in Eq.1 describes the hybridization between the orbitals of QD and the two leads. For spin independent hybridization it has the form † (−tika kσciσ + h.c.), (4) (cid:88) Htun = k∈{R,L},i,σ where tiks are hopping integrals where we consider them in the wide-band limit. 2 The forth term represents the free Hamiltonian of the single mode phonon bath and the fifth term is the electron-phonon interaction Hamiltonian, that is, (cid:18) Hph = Ω b†b + , 1 2 (cid:19) (cid:16)b† + b (cid:17) and (cid:88) i,j,σ † γc iσcjσ He−ph = (5) (6) , where b (b†) is the annihilation (creation) operator of phonons, Ω is phonon frequency and γ is the strength of electron-phonon coupling. Finally, the last term in the Hamiltonian represents the interaction of exter- nal classical electromagnetic fields with the QD in the electric-dipole approxi- mation. In terms of creation and annihilation operators of the QD, it is given by He−l = −→ −−→µ . † 1σc2σ + h.c.), E (t)(c (7) (cid:88) σ where −→µ is the electric-dipole matrix element between the two orbitals of QD −→ E (t) is the external electric field. and The bias voltage, V , is applied symmetrically to the right and left leads, thus their Fermi energies are, respectively, µL = eV /2 and µR = −eV /2. Moreover, the system is considered at zero temperature, so that the Fermi distribution of the leads are Heaviside theta functions (θ(µα − ω), α = R, L), and we choose the system of units that e = = 1. 2.1. The HF Approximation In order to consider the electron-phonon interaction and determine the op- tical properties of our model, we use the Keldysh formalism of non-equilibrium (contour-ordered) Greens functions (NEGF) which is an extremely useful method for studying the non-equilibrium properties of many-body systems. In our model, we deal with both electron and phonon GFs. The free Phonon GF on the Keldysh time contour is defined as (cid:68) (cid:110)b(τ ) + b†(τ ) (cid:111)(cid:110)b(τ (cid:48) ) + b†(τ (cid:48) ) (cid:111)(cid:69) D0c(τ, τ (cid:48) ) = −i Tc (8) It is a straight forward calculation to obtain the retarded, advanced and lesser GFs as functions of frequency (the greater GF is not independent of these tree): Dr 0(ω) = [Da ∗ 0 (ω)] = 1 ω − Ω + i0+ − 1 ω + Ω + i0+ , and 0 (ω) = −2πi [nphδ (ω − Ω) + (1 + nph) δ (ω + Ω)] , D< 3 (9) (10) where nph is the expectation value of phonon number in free phonon bath, which is identically zero at zero temperature. The electron GF on the Keldysh time contour is defined as (cid:68) (cid:16) (cid:17)(cid:69) Gc,ij(τ, τ (cid:48) ) = −i Tc † cH,iσ(τ )c H,jσ(τ (cid:48) ) , i, j = 1, 2 (11) † H,iσ(τ )) is the annihilation (creation) operator in the Heisen- where cH,iσ(τ ) (c berg representation on the Keldysh time contour. In our model the electron GFs are spin independent, so we show them by two by two matrices and drop their spin indices. The non-interacting electron GFs, that is, the GFs in the absence of electron- phonon and electron-light interactions, are obtained by doing a straight forward calculation to be (cid:104) (cid:16) G0r(ω) (cid:17)† , G0r(ω) = G0a(ω) = (ω + i0+) I − hdot − Σr lead (cid:105)−1 , (12) (13) (14) and G0<(ω) = G0r(ω)Σ< leads(ω) G0a(ω), where I is the unity matrix, hdot is the single particle non-interacting Hamilto- nian of the QD, and the lead self-energies are and Γ, leads = − i Σr 2 (cid:88) Γ i 2 α∈{R,L} Σ< leads(ω) = θ(µα − ω), (15) (16) in which, Γ is a two by two matrix that determines the level broadening of QD due to coupling to leads and its elements are given by kjδ( − k), (cid:88) Γij() = 2π tikt∗ (17) k which is independent of in wide-band approximation. In order to consider the effect of electron-phonon interaction in electron GFs, we use the self consistent Hartree-Fock approximation to find an appropriate self-energy, ΣHF , and insert it into the Dyson equation, which on the Keldysh contour reads as Gc = G0 c + G0 c ΣHF,cGc, (18) 4 where contour integrations are implicitly understood. The Hartree-Fock self- energy is the sum of Hartree and Fock terms. The Hartree self-energy on the Keldysh contour is (cid:90) c ij (τ, τ (cid:48) ) = −iγ2δijδc(τ − τ (cid:48) ) dτ1D0(τ, τ1) 2G< ll (τ1, τ1), (19) ΣH (cid:88) l where δc stands for the contour Dirac delta function. Because of this delta func- tion, the lesser Hartree self-energy vanishes. Using the Langreth rules for ana- lytical continuation and Fourier transforming the results, the retarded Hartree self energy in frequency domain is obtained as ij = −iγ2δijD0(ω = 0) ΣH,r 2G< ll (ω (cid:48) ), (20) (cid:90) ∞ −∞ (cid:48) dω 2π (cid:88) l which is frequency independent. The electron population in the i-th orbital of the QD is (cid:90) ∞ −∞ ni = −i (cid:48) dω 2π 2G< ii (ω (cid:48) ), where the factor 2 is for spin. Combining Eqs.9,20 and 21, we arrive at ΣH,r ij = −2γ2 Ω δij (cid:88) l nl. The Fock self-energy on the Keldysh contour is (cid:48) ΣF ij(τ, τ ) = iγ2D0(τ, τ(cid:48))Gij(τ, τ(cid:48)), Using the Langreth rules and Fourier transformation, the lesser and retarded self-energies are obtained to be (cid:48) −∞ dω 2π (cid:90) ∞ (cid:90) ∞ −∞ 0(ω − ω G< ij(ω (cid:48) )D< 0 (ω − ω (cid:48) ), (cid:48) {Gr dω 2π ) + G< (cid:48) ij(ω (cid:48) (cid:48) (cid:48) 0 (ω − ω )D< ) 0(ω − ω )}. (cid:48) ij(ω )Dr ΣF,< ij (ω) = iγ2 and ΣF,r ij (ω) = iγ2 (cid:48) +Gr ij(ω Using Eq.10, we arrive at )Dr (21) (22) (23) (24) (25) (26) ΣF,< ij (ω) = γ2G< ij(ω + Ω), 5 and ΣF,r (cid:90) ∞ ij (ω) = γ2Gr {Gr (cid:48) iγ2 dω 2π −∞ ij(ω + Ω) + (cid:48) ) + G< ij(ω (cid:48)}Dr 0(ω − ω (cid:48) ). (27) ij(ω Adding the Hartree and Fock terms, we obtain the HF self-energy and conse- quently, the electron GFs as (cid:105)−1 , (cid:105) Ga(ω). (28) (29) (30) ( G0r(ω))−1 − Σr HF Gr(ω) = Ga(ω) = Gr(ω)†, (cid:104) (cid:104) Σ< and G<(ω) = Gr(ω) leads(ω) + ΣF,<(ω) In numerical calculations, we have to find the electron GFs, HF self-energies and electron populations self consistently. After doing these self-consistent calculations and obtaining the electron GFs, the left to right current could be computed from the formula (cid:90) µL µR I = ie dω 2π T r (cid:104)Γ (cid:16) Gr(ω) − Ga(ω) (cid:17)(cid:105) , (31) where the trace is taken over the orbital degrees of freedom of the QD and a factor of 2 is already taken into account for spin. 3. linear response to classical light The polarization of QD is defined by −→ P (t) = −→µ (cid:88) (cid:68) σ (cid:69) † † H,1σ(t) H,2σ(t) + cH,2σ(t)c cH,1σ(t)c . (32) Using equal time anti-commutation properties of annihilation and creation op- erators, Eq.32 can be written as[13] (cid:104)−→µ G< (cid:105) −→ P (t) = 4Im 12(t, t) , (33) where G< taken into account. 12(t, t) is the lesser GF between the two orbitals of QD when He−l is Since, we are interested in the OPA cross section, we determine −→ E (t). The contour-ordered GF to first order in He−l is order in −→ P (t) to first (cid:73) (1) G (τ, τ (cid:48) ) = dτ1 G(τ, τ1)he−l(τ1) G(τ1, τ (cid:48) ), (34) 6 where all the τ s lie on the Keldysh time contour, G represents the HF GFs of the QD, and he−l is (cid:32) he−l(τ ) = 0 −→ −−→µ . E (τ ) −→ −−→µ . E (τ ) 0 (cid:33) . (35) Using the Langreth rules and doing the Fourier transformations, the first −→ E in the direction of −→µ could be written as order polarization for P (1)(ω) = 2iµ2E(ω)(γ(−ω, ω) − γ∗(ω,−ω)), (cid:90) dω (cid:48) 2π where E(ω) is the Fourier transform of the electric field and γ(−ω, ω) = (cid:48) [Gr 12(ω + ω (cid:48) (cid:48) )Ga (cid:48) (cid:48) )G< 12(ω ) + Gr 11(ω + ω (cid:48) (cid:48) (cid:48) ) × G< ) + G< )]. In terms of γ(−ω, ω), the frequency dependent OPA cross section is 12(ω + ω 11(ω + ω 22(ω 12(ω )Ga 22(ω ) + G< σ(ω) = 8πωµ2 c Re[γ(−ω, ω) − γ∗(ω,−ω)]. If we have emission in our system, this OPA takes negative values. 4. Numerical Results (36) (37) (38) In this section we present our numerical results. For our numerical calcula- tions, we set 1 = −2 = δ so that the level spacing is 2δ. Moreover, we take all Γijs to be equal to a same value of Γ which is our energy unit. In Figs.1 a, b and c, we show, respectively, the electron populations of QD levels, n1 and n2, and left to right current I, as functions of applied bias voltage for δ/Γ = 3.0, Ω/Γ = 1.8 , γ/Γ = 0.6, 0.8 and 1.0. As we see in Fig.1a, for small electron- phonon couplings and at low bias voltages, the upper orbital of QD is almost empty, that is, n1 is small. But by increasing the strength of coupling, the upper orbital gets more populated even at small bias voltages. This behavior could be understood by noticing that the Hartree self energy, Eq.22, is of minus sign and acts effectively as a gate voltage that lowers the onsite energies of QD, and by increasing the electron-phonon coupling strength, its value increases. By increasing the bias voltage, the electron populations of the two orbitals of QD change, until both n1 and n2 approach 1, which means the QD gets half filled. At bias voltages where n1 is substantially greater than n2, the electrons could absorb light and make inter dot transition. In Fig.2a, we show this behavior by plotting OPA for δ/Γ = 3.0, Ω/Γ = 1.8 , γ/Γ = 0.8 and eV /Γ = 5.0. In Fig.2b, we show the density of states, A(ω) = −2 ii(ω)), and in order to see what states are filled, we plot the filling functions Fi = −2i ii (ω), whose integrals give the electron populations. Please note that the factor of i Im(Gr (cid:80) 2π G< π 7 Figure 1: (a), (b) and (c) are, respectively, the electron populations of the two orbitals of QD and the current for δ/Γ = 3.0, Ω/Γ = 1.8 , γ/Γ = 0.6, 0.8 and 1.0, as functions of the bias voltage. 2 is considered for spin. Fig.2b reveals that the main peaks in OPA, Fig.2a, are caused by electron transitions between the states of QD. The two peaks of OPA at frequencies ω/Γ = 5.72 and 7.51, are due to electron transitions from state with energy ω/Γ = −4.60 to states with energies ω/Γ = 0.97 and 2.85, respectively. At high bias voltages we have optical gain at some frequencies. In Fig.3a, we show OPA for δ/Γ = 3.0, Ω/Γ = 1.8 , γ/Γ = 0.8 and eV /Γ = 15.0. Even though n1 is smaller than n2, we have lasing with most strength at frequency ω/Γ = 4.31. Fig.3b shows the density of states and filling functions of QD. This figure reveals that the lasing takes place because one of the side-bands whose peak is at frequency ω/Γ = −2.54 is almost empty and electrons could make transition from the higher orbital, whose main peak is at frequency ω/Γ = 1.52, to it. For the cases where level spacing is substantially more than the phonon frequency, since the side-bands are approximately at frequency distances of Ω 8 Figure 2: (a)The OPA cross section as a function of frequency and (b) the density of states and filling functions for δ/Γ = 3.0, Ω/Γ = 1.8 , γ/Γ = 0.8 and eV /Γ = 5.0. The arrows indicate the important peaks with their frequencies. The absorption peaks are explained by electron transitions between the states of QD. c is the speed of light. Figure 3: (a)The OPA cross section as a function of frequency and (b) the density of states and filling functions for δ/Γ = 3.0, Ω/Γ = 1.8 , γ/Γ = 0.8 and eV /Γ = 15.0. The arrows indicate the important peaks with their frequencies. The emission peak is explained by electron transmission from the higher main peak of density state to the side band at the right of the lower main peak. In addition, there is an absorption peak because electrons make transition from the lower main peak to the side band at right to the higher main peak. c is the speed of light. from the main peaks, we expect that the main optical gain occurs at frequency of about 2δ−Ω, which is indeed the case. On the other hand, since the side-band 9 right to the main higher peak of density of states is almost empty, electrons could absorb light and makes transition from lower main peak of density of states to this side-band. The distance between these two peaks is approximately 2δ + Ω, which explains an absorption peak at almost this frequency. Figure 4: OPA cross sections as functions of frequency for δ/Γ = 3.0, Ω/Γ = 1.8 ,eV /Γ = 15.0, γ/Γ = 0.0, 0.6, 0.8 and 1.0. The electron-phonon coupling strength is written on top of each curve. We see that the emission peak and the absorption resonance next to it, are completely due to electron-phonon interaction. c is the speed of light. The emission and absorption peaks at approximately the frequencies of 2δ∓Ω are completely due to electron-phonon coupling. In order to illustrate that, in Fig.4, we show OPA for δ/Γ = 3.0, Ω/Γ = 1.8 ,eV /Γ = 15.0, γ/Γ = 0.0, 0.6, 0.8 and 1.0. We see that increasing the electron-phonon coupling strength, the optical gain and the absorption peak next to it, both increase. 5. Conclusions In conclusion, we considered a two-level QD connected to two biased leads and coupled to a single-mode phonon bath. We obtained the I-V curve of our system and showed that depending on the used parameters, the populations of QD levels could be close to each other or differ substantially. By considering the linear response of the system to an incoming classical light, we concluded that at low bias voltages the QD absorbs light and we don't have optical gain. Moreover, at these low biases if the populations of the QD levels differ substantially, the light absorption is much stronger, that is, it would be easier for electrons to absorb light and make transition from the lower level to the higher one. On the other hand, at high enough voltages where both levels of the QD are almost half-filled, the system shows lasing. 10 This lasing is seen because one of the lower side-bands is almost empty and electrons could make transition from the higher levels to it. On the other hand, one of the higher side-bands is also empty which explains an absorption peak. The frequency at which we have strong lasing is determined by the level spacing of QD, 2δ, and the phonon frequency, Ω. For Ω considerably less than 2δ, this frequency is approximately 2δ − Ω, while we have an absorption peak at the approximate frequency of 2δ + Ω. Additionally, the strength of the lasing increases with increasing the electron-phonon coupling strength, provided that the QD remains almost half-filled. References [1] C. Weber, A. Fuhrer, C. Fasth, G. Lindwall, L. Samuelson, A. Wacker, Probing confined phonon modes by transport through a nanowire double quantum dot, Physical review letters 104 (2010) 036801. [2] T. Frey, P. Leek, M. Beck, A. Blais, T. Ihn, K. Ensslin, A. Wallraff, Dipole coupling of a double quantum dot to a microwave resonator, Physical Review Letters 108 (2012) 046807. [3] J. Basset, D.-D. Jarausch, A. Stockklauser, T. Frey, C. Reichl, W. Wegscheider, T. M. Ihn, K. Ensslin, A. Wallraff, Single-electron double quantum dot dipole-coupled to a single photonic mode, Physical Review B 88 (2013) 125312. [4] Y.-Y. Liu, K. Petersson, J. Stehlik, J. M. Taylor, J. Petta, Photon emission from a cavity-coupled double quantum dot, Physical review letters 113 (2014) 036801. [5] Y.-Y. Liu, J. Stehlik, C. Eichler, M. Gullans, J. M. Taylor, J. Petta, Semi- conductor double quantum dot micromaser, Science 347 (2015) 285 -- 287. [6] A. Stockklauser, V. F. Maisi, J. Basset, K. Cujia, C. Reichl, W. Wegschei- der, T. Ihn, A. Wallraff, K. Ensslin, Microwave emission from hybridized states in a semiconductor charge qubit, Physical review letters 115 (2015) 046802. [7] P.-Q. Jin, M. Marthaler, J. H. Cole, A. Shnirman, G. Schon, Lasing and transport in a quantum-dot resonator circuit, Physical Review B 84 (2011) 035322. [8] M. Kulkarni, O. Cotlet, H. E. Tureci, Cavity-coupled double-quantum dot at finite bias: Analogy with lasers and beyond, Physical Review B 90 (2014) 125402. [9] M. Gullans, Y.-Y. Liu, J. Stehlik, J. Petta, J. M. Taylor, Phonon-assisted gain in a semiconductor double quantum dot maser, Physical review letters 114 (2015) 196802. 11 [10] C. Karlewski, A. Heimes, G. Schon, Lasing and transport in a multilevel double quantum dot system coupled to a microwave oscillator, Physical Review B 93 (2016) 045314. [11] M. Galperin, M. A. Ratner, A. Nitzan, A. Troisi, Nuclear coupling and po- larization in molecular transport junctions: beyond tunneling to function, Science 319 (2008) 1056 -- 1060. [12] G. Stefanucci, R. van Leeuwen, Nonequilibrium Many-Body Theory of Quantum Systems: A Modern Introduction, Cambridge University Press, 2013. [13] A. J. White, M. Sukharev, M. Galperin, Molecular nanoplasmonics: Self- consistent electrodynamics in current-carrying junctions, Phys. Rev. B 86 (2012) 205324. 12
1811.09963	1	1811	2018-11-25T06:40:26	Temperature-Dependent Magnetization Reversal in Exchange Bias NiFe/IrMn/NiFe Structures	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	We demonstrate the magnetization reversal features in NiFe/IrMn/NiFe thin-film structures with 40% and 75% relative content of Ni in Permalloy in the temperature range from 80 K to 300 K. At the descending branches of the hysteresis loops, the magnetization reversal sequence of the two ferromagnetic layers is found to depend on the type of NiFe alloy. In the samples with 75% relative content of Ni, the bottom ferromagnetic layer reverses prior to the top one. On the contrary, in the samples with 40% of Ni, the top ferromagnetic layer reverses prior to the bottom one. These tendencies of magnetization reversal are preserved in the entire range of temperatures. These distinctions can be explained by the morphological and structural differences of interfaces in the samples based on two types of Permalloy.	cond-mat.mes-hall	cond-mat	Journal of Magnetism and Magnetic Materials 00 (2017) 000--000 TEMPERATURE-DEPENDENT MAGNETIZATION REVERSAL IN EXCHANGE BIAS NiFe/IrMn/NiFe STRUCTURES Ch. Gritsenko1, I. Dzhun2, M. Volochaev3, M. Gorshenkov4, G. Babaytsev2, N. Chechenin2, 5, A. Sokolov3, Oleg A. Tretiakov6, 7, and V. Rodionova1, 4 1Immanuel Kant Baltic Federal University, A. Nevskogo 14, 236041 Kaliningrad, Russia 2Skobeltsyn Institute of Nuclear Physics, Lomonosov Moscow State University, Leninskie Gory 3Kirensky Institute of Physics, Federal Research Center KSC SB RAS, Akademgorodok 50/38, 1/2, 119991 Moscow, Russia Krasnoyarsk, 660036, Russia 119049, Russia Moscow, Russia 4National University of Science and Technology MISIS, Leninsky Prospect 4, Moscow, 5Faculty of Physics, Lomonosov Moscow State University, Leninskie Gory 1/2, 119991 6Institute for Materials Research and Center for Science and Innovation in Spintronics, 7School of Physics, The University of New South Wales, Sydney 2052, Australia Tohoku University, Sendai 980-8577, Japan Elsevier use only: Received date here; revised date here; accepted date here Abstract We demonstrate the magnetization reversal features in NiFe/IrMn/NiFe thin-film structures with 40% and 75% relative content of Ni in Permalloy in the temperature range from 80 K to 300 K. At the descending branches of the hysteresis loops, the magnetization reversal sequence of the two ferromagnetic layers is found to depend on the type of NiFe alloy. In the samples with 75% relative content of Ni, the bottom ferromagnetic layer reverses prior to the top one. On the contrary, in the samples with 40% of Ni, the top ferromagnetic layer reverses prior to the bottom one. These tendencies of magnetization reversal are preserved in the entire range of temperatures. These distinctions can be explained by the morphological and structural differences of interfaces in the samples based on two types of Permalloy. PACS: 75.80.+q, 96.12.Hg, 75.50.Bb Keywords:magnetization reversal; exchange bias; Permalloy 2 / Journal of Magnetism and Magnetic Materials 00 (11/25/18) 000 -- 0001 a in with interaction system of 1. Introduction The exchange bias phenomenon can be observed adjacent antiferromagnetic (AFM) and ferromagnetic (FM) layers under the condition of an induced uniaxial magnetic anisotropy. It leads to a shift of the hysteresis loop along the field axis. Along a fixed direction, the ferromagnetic layer becomes harder for magnetization reversal due to the exchange coupling the antiferromagnetic layer. As a result, the FM- layer is considered as a pinned layer, which is widely used in spin valves [1 -- 4], magnetic sensors [5], and MRAM [6]. Depending on the application, the most important features of magnetization process can be either the magnitude of the exchange bias or sequence of ferromagnetic layers magnetization in peculiarities of hysteresis loops shape [7]. The AFM-layer thickness plays a crucial role for the exchange bias effect. Therefore, it is important to study how the material parameters and characteristic properties of the exchange bias systems influence the aforementioned features. switching, reflected The exchange bias phenomenon has been studied for more than a half of a century [8], and it has been confirmed to have a strong dependence on the types of FM and AFM materials [9]. Commonly used in exchange bias systems ferromagnetic materials are Ni, Co, Fe, their alloys, and the alloys doped with impurities of elements, such as Pt. Depending on magnetization, coercive force, and magnetic anisotropy of these FM materials, one can find different exchange bias values and coercivity for a required application [10 -- 13]. NiFe alloys have small coercivity, high initial and maximum as magnetic permeabilities, as well types of Permalloy: corrosion resistance that can be useful for digital memory devices [14,15]. There are two 'High-nickel' Permalloy [16,17], which contains 72%- 80% of Ni, and 'low-nickel' Permalloy [17], which contains 40%-50% of Ni. The small 'High-nickel' Permalloy has a crystalline initial permeability, and in traditional exchange bias systems [18 -- 20]. The 'Low-nickel' Permalloy has higher crystalline anisotropy and larger saturation magnetization in comparison with the 'high- nickel' Permalloy. The 'Low-nickel' one is usually used in write heads [15]. is usually used anisotropy, large One of the ways to improve the required properties of the exchange biased systems is to use the trilayer structures instead of bilayers [9,21], where the two ferromagnetic layers are separated by the antiferromagnetic one. Such structures produce step-wise hysteresis loops [22,23] due to the two exchange-coupled interfaces with different energies. systems temperature dependences Exchange bias effect depends strongly on temperature [24,25]. The thicknesses of both ferromagnetic and antiferromagnetic layers also affect the temperature dependence of the exchange bias. In particular, for the NiFe/FeMn/NiFe a peak-like behavior of the exchange bias was observed in for thicknesses of the seed NiFe layer from 10 nm to 50 nm, while for the 5 nm thickness the temperature dependence had been shown to exhibit a different behavior [26]. Such a behavior has been found to depend on the bottom FM-layer thickness of FM/AFM/FM trilayer systems and is related to the amount of FM-layer spins that are strongly coupled with the AFM. Concerning the blocking (the temperature exchange bias the of and this both NiFe/IrMn temperature at which the exchange bias becomes nonzero), it was reported in [27] that temperature decreases with decreasing antiferromagnetic layer thickness in IrMn/NiFe structures, as well as the same dependence was observed for different CoO layer thicknesses in CoO/Co/Ge films [28]. In [25], it has been shown that as temperature changes, the domain structure of NiFe/IrMn, in particular the size of domain walls, also changes affecting the exchange bias. In this work, the FM/AFM/FM trilayer compositions with either low- or high- nickel Permalloy have been studied. The magnetization reversal features as well as exchange bias and coercivity have been found to depend on temperature in the range from 80 K to 300 K. 2. Experimental details The NiFe/IrMn/NiFe thin-film structures were fabricated by magnetron sputtering at an ambient temperature in Ar atmosphere with pressure of 3 mTorr. Magnetic field of 500 Oe was applied in plane of the substrate during the deposition process to induce the unidirectional anisotropy in samples. The substrate was Si/SiO2 (100). The buffer Ta layer with a thickness of 30 nm was deposited onto the substrate to improve the growth of further layers. For each structure, the FM-layer that was deposited prior to the other one (i.e. onto the Ta buffer layer) is denoted as "bottom". Accordingly, the FM layer that was deposited on top of the IrMn layer is denoted as "top". The last layer of 30 nm of Ta was deposited on top of each sample to prevent them from the oxidation. We prepared two series of samples, one using 'low-nickel' Permalloy that is Ni40Fe60 (LNiPy), and the other using 'high-nickel' out The layer. Permalloy that is Ni75Fe25 (HNiPy). Two targets of separated Ni and Fe were used for co-deposition of NiFe alloys. The target Ir45Mn55 alloy was used for deposition of an antiferromagnetic layers thicknesses were set by the deposition time with the deposition rates estimated from the measurements of the the Rutherford calibration samples by backscattering method. The NiFe layers were fabricated to have thickness of 10 nm, while the thickness of IrMn layers was varied to be 2, 4 or 10 nm. thickness of the samples structural The study of properties was carried using transmission electron microscopy (TEM) using Hitachi HT7700 microscope at the accelerating voltage of 100 kV. Cross- section pieces of samples were prepared using a Hitachi FB2100 (FIB) single-beam focused ion beam system. The magnetic properties of the samples were investigated using a Vibrating Samples Magnetometer (VSM, Lake Shore, Model 7400). The hysteresis loops for each sample were measured for in-plane geometry with the magnetic field of the VSM oriented along the induced unidirectional anisotropy, in the temperature range from 80 K to 300 K. 3. Results and discussion The TEM images, presented in Fig. 1, show that the interfaces between the NiFe layers and IrMn are the smooth except the interface between the bottom LNiPy and IrMn. In the case of the LNiPy/IrMn partial intermixing of layers is observed. It can occur because the LNiPy grows with a large grain size [29,30] due to the stoichiometric ratio of elements in the alloy [16]. The large grain size of LNiPy causes the enhanced layer surface. roughness of cross-sectional the LNiPy interface, the the (i.e. 4 Assuming the IrMn layer grains to be smaller than LNiPy grains [31,32], it can possibly fill the gaps between the LNiPy grains. shows the Figure 2 (a) shows the hysteresis loops for the samples NiFe/IrMn(10 nm)/NiFe with LNiPy (red curve) and HNiPy (black curve). As it can be seen, the loops exhibit the step-like shape that corresponds to the separate magnetization reversal of the two ferromagnetic layers. This is more visible in Fig. 2 (b), (c), which the differential susceptibility first derivative of magnetic moment by magnetic field) distributions for the descending and ascending branches of the hysteresis loops for the aforementioned samples. The steps in the black hysteresis loop are of equal height because the two FM layers of HNiPy are of the same thickness 10 nm, as it was confirmed by the TEM (Fig. 1). However, the red hysteresis loop has non-equal heights of the steps in the descending branch: the bottom step is of smaller height than the top one. This may be caused by the partial intermixing of the bottom LNiPy layer with IrMn layer demonstrated earlier [29,33]. It should be mentioned that in the HNiPy sample in the descending branch of the hysteresis the bottom FM-layer reverses prior to the top one. This can be confirmed by the coercivities for the top and bottom subloops that correlate perfectly with the coercivities of the separate HNiPy/IrMn and IrMn/HNiPy structures [34]. However, for the LNiPy sample in the descending branch of the hysteresis loop, the top FM-layer reverses prior to the bottom one. This can be revealed from the observed difference in the heights of the steps in the hysteresis loops as it was described above. in magnetization Thus, the difference loop, / Journal of Magnetism and Magnetic Materials 00 (11/25/18) 000 -- 0001 the reversal for the HNiPy and LNiPy samples is observed. The typical hysteresis loops for the LNiPy and HNiPy NiFe/IrMn/NiFe samples at temperatures 90 K, 200 K, and 300 K, with thicknesses of AFM-layer of 2 nm and 4 nm, are presented in Fig. 3. The remnant magnetization of the samples at 2 nm of IrMn layer (Fig. (a), (b)) increases with the decrease of temperature. For all samples, when the coercive force increases. These facts can be explained in terms of thermal fluctuations model [35], according to which the spin structure at the interfaces becomes more stable, because when is decreased it reduces the thermal-fluctuations energy of AFM atoms and therefore of the AFM-grains. temperature decreases, temperature susceptibility For the LNiPy samples with 2 nm of AFM-layer the kinks appear in the sections of the loops preceding the saturation state (Fig. 3 (a)). This is more visible in the differential distribution (Fig. 4 (a)). The second peak is observed in the descending branch. This may mean that due to an increase in the anisotropy energy of the sample layers, some of the magnetic moments of the FM-layer reverse later, after the majority of the moments is already reversed. At the same time, an increase in the slope of the hysteresis loops for the LNiPy samples indicates a change in the mechanism of the magnetization reversal of the samples. Thus, if at room temperature, the contribution of magnetization reversal through the motion of the domain wall was greater than through the rotation of the magnetic moments, then with a decrease of temperature, the contribution to the process through of magnetization the motion of the domain wall decreases. reversal is At 4 nm of IrMn layer thickness the hysteresis loops for the LNiPy samples are asymmetric, since in the descending branch the two ferromagnetic layers reverse at different values of the magnetic field, whereas in the ascending branch it occurs at the same value of the field. The sequence of the magnetization reversal for two LNiPy layers is the same as at 10 nm of AFM-layer thickness at room temperature (Fig. 2 (a)). For the HNiPy samples, the magnetization- reversal sequence of the top and bottom FM- layers the HNiPy/IrMn/HNiPy sample with 10 nm of AFM layer at room temperature (Fig. 2 (a)). That is, in the descending branch the bottom FM-layer reverses prior to the top one, whereas in the ascending branch the top FM-layer reverses prior to the bottom one. Moreover, in the case of LNiPy a complete separation of the loops into the top and bottom subloops does not occur. the same as found for As it can be seen (Fig. 3 (a), (b)), the slope of hysteresis loops for both LNiPy and HNiPy samples at 2 nm of AFM-layer is smaller than the one at 4 nm. Considering the peaks of the differential susceptibility distribution for descending and ascending branches (Fig. 4), it should be mentioned that the increase of the AFM-layer thickness. This means that the contribution to the reversible switching of i.e. magnetic moments rotation, increases with increasing the AFM-layer thickness. the peaks width widens with the magnetization, The dependences of estimated values of the exchange bias and coercivity on the thickness of the antiferromagnetic layer for the top and bottom FM-layers are presented in Fig. 5. A decrease in temperature leads to an increase in the exchange bias. It can be explained by the decreasing of the energy of thermal fluctuations in the system. This leads to the spin structure at the interfaces to be more stable, and hence the magnetic moments of the ferromagnetic layer need more energy to overcome this barrier. As the temperature decreases to 80 K, the coercive force of the samples increases. For the samples with IrMn-layer thickness of 2 nm, the blocking temperature below which the exchange bias was found is 250 K for the LNiPy and 200 K for the HNiPy sample. For the samples with the antiferromagnetic layer thickness of 4 nm, the exchange bias was observed at temperatures below 290 K. The tendency to an increase of the blocking temperature with increasing AFM-layer thickness can be explained by an increase of the anisotropy the AFM layer [6,24,27,36]. energy of with has Thus, it was shown that the mechanism of magnetization reversal is maintained for the Permalloy of each composition with decreasing temperature. The sequence of magnetization reversals of ferromagnetic layers for structures based on LNiPy in the descending branch is the same as in the ascending branch, whereas it is different for the structures based on HNiPy. structures Conclusions The studies of magnetic properties performed in the temperature range from 80 K to 300 K allowed us to determine the blocking temperatures of NiFe/IrMn/NiFe thin-film the antiferromagnetic layer thicknesses of 2 nm and 4 nm, which do not exhibit the exchange bias effect at temperature. The sequence of magnetization reversal for the two been determined to depend on the AFM-layer thickness and the temperature. Thus, at 4 nm thickness of IrMn layers for the LNiPy samples the top FM-layer reverses prior to the bottom one. On the contrary, for the ferromagnetic layers room / Journal of Magnetism and Magnetic Materials 00 (11/25/18) 000 -- 0001 samples 6 the bottom FM-layer HNiPy reverses prior top one. The aforementioned differences are supposedly caused by the structural qualities of the systems based on the LNiPy and HNiPy. the to of [7] [4] B.G. Park, J. Wunderlich, X. Martí, V. Holý, Y. Kurosaki, M. Yamada, H. Yamamoto, A. Nishide, J. Hayakawa, H. Takahashi, A.B. Shick, T. Jungwirth, spin-valve-like A magnetoresistance an antiferromagnet-based tunnel junction, Nat. Mater. 10 (2011) 347 -- 351. doi:10.1038/nmat2983. [5] Y.-T. Chen, The Effect of Interface in Texture on Exchange Biasing System, Ni80Fe20/Ir20Mn80 Nanoscale Res. Lett. 4 (2009) 90 -- 93. doi:10.1007/s11671-008-9207-4. for exchange bias [6] K. O'Grady, L.E. Fernandez-Outon, new G. Vallejo-Fernandez, A paradigm in polycrystalline thin films, J. Magn. Magn. Mater. 322 (2010) 883 -- 899. doi:10.1016/j.jmmm.2009.12.011. J. Sort, K.S. Buchanan, V. Novosad, A. Hoffmann, G. Salazar-Alvarez, A. Bollero, M.D. Baró, B. Dieny, J. Nogués, into antiferromagnets, Phys. Rev. Lett. 97 (2006) 1 -- 4. doi:10.1103/PhysRevLett.97.067201. [8] W. H. Meiklejohn and C. P. Bean, New Magnetic Anisotropy, Phys. Rev. 102 (1956) 1413 -- 1414. J. Nogués, I.K. Schuller, Exchange bias, J. Magn. Magn. Mater. 192 (1999) 203 -- 232. doi:10.1016/S0304- 8853(98)00266-2. Imprinting vortices [10] L. Del Bianco, F. Spizzo, M. Tamisari, S. Laureti, Dependence of exchange bias on the field-cooled remanent magnetic state in Ni/NiO nanogranular samples, Solid State Commun. 151 (2011) 351 -- 353. doi:10.1016/j.ssc.2010.12.027. [11] A. Maitre, D. Ledue, R. Patte, [9] and support by Acknowledgments M. Gorshenkov Ch. Gritsenko acknowledge the Russian Foundation for Basic Research (RFFI grant № 17-32-50170). O.A.T. acknowledges support by the Grants-in-Aid for Scientific Research (Grant Nos. 17K05511 and 17H05173) from MEXT, Japan, by the grant of the Center for Science and Innovation in Spintronics (Core Research Cluster), Tohoku University, and by JSPS and RFBR under Research Cooperative Program. Japan-Russian the [2] References [1] Ching Tsang, R.E. Fontana, Tsann Lin, D.E. Heim, V.S. Speriosu, B.A. Gurney, M.L. Williams, Design, fabrication and testing of spin-valve read heads for high density recording, IEEE Trans. Magn. 30 (1994) 3801 -- 3806. doi:10.1109/20.333909. Z. Wei, A. Sharma, A.S. Nunez, P.M. Haney, R.A. Duine, J. Bass, A.H. MacDonald, M. Tsoi, Changing exchange bias in spin valves with an electric current, Phys. Rev. Lett. 98 (2007). doi:10.1103/PhysRevLett.98.116603. [3] Y. Jiang, T. Nozaki, S. Abe, T. Ochiai, A. Hirohata, N. Tezuka, K. Inomata, Substantial reduction of critical current for magnetization switching in an exchange-biased spin valve, Nat. Mater. 3 (2004) 361 -- 364. doi:10.1038/nmat1120. Interfacial roughness and temperature effects on exchange bias properties in coupled ferromagnetic / antiferromagnetic bilayers, J. Magn. Magn. Mater. 324 (2012) 403 -- 409. doi:10.1016/j.jmmm.2011.07.049. [13] [12] S.P. Hao, Y.X. Sui, R. Shan, L. Sun, S.M. Zhou, Exchange biasing in as- prepared Co/FeMn bilayers and magnetic properties of ultrathin single layer films, Thin Solid Films. 485 (2005) 212 -- 217. doi:10.1016/j.tsf.2005.03.019. I. Suzuki, Y. Hamasaki, M. Itoh, T. Taniyama, Controllable exchange in Fe/metamagnetic FeRh bias bilayers, Appl. Phys. Lett. 105 (2014). doi:10.1063/1.4900619. [14] M.Y. S. X. Wang, N. X. Sun, S. Yabukami†, Properties of a new soft magnetic material, Nature. 407 (2000). [15] C. Tannous, R.L. Comstock, Springer Handbook of Electronic and Photonic Materials, 2017. doi:10.1007/978-3- 319-48933-9. [16] E.S. Borovik, V.V. Ereyomenko, A.S. Milner, Lectures on magnetism, FIZMATLIT, Moscow, 2005. I.O. Dzhun, G. V. Babaytsev, N.G. Chechenin, C.A. Gritsenko, V. V. Rodionova, FMR investigations of exchange biased NiFe/IrMn/NiFe trilayers with high and low Ni relative content, J. Magn. Magn. Mater. (2017) 1 -- 5. doi:10.1016/j.jmmm.2017.11.028. [18] C. Liu, C. Yu, H. Jiang, L. Shen, C. Alexander, G.J. Mankey, Effect of interface roughness on the exchange bias for NiFe/FeMn, J. Appl. Phys. 87 (2000) 6644 -- 6646. doi:10.1063/1.372797. [17] S. of 115 bias [19] V.P. Nascimento, E.C. Passamani, A.D. Alvarenga, F. Pelegrini, A. Biondo, E. Baggio Saitovitch, Influence of the roughness on the exchange of NiFe/FeMn/NiFe trilayers, J. Magn. Magn. Mater. 320 (2008) 272 -- 274. doi:10.1016/j.jmmm.2008.02.059. effect [20] F. Xu, Z. Liao, Q. Huang, N.N. Phuoc, Li, Characteristics of NiFe / IrMn / NiFe Trilayers, 47 (2011) 3486 -- 3489. C.K. Ong, [21] P.N. Lapa, I. V. Roshchin, J. Ding, J.E. Pearson, V. Novosad, J.S. Jiang, A. Hoffmann, Magnetoresistive detection pinned uncompensated magnetization in antiferromagnetic FeMn, Phys. Rev. B. 95 (2017) 1 -- 6. doi:10.1103/PhysRevB.95.020409. strongly [22] K.W. Lin, T.C. Lan, C. Shueh, E. Lierop, Skoropata, J. Modification of the ferromagnetic anisotropy and exchange bias field of NiFe/CoO/Co trilayers through the CoO spacer thicknesses, J. Appl. Phys. 1 -- 5. doi:10.1063/1.4861216. (2014) [23] G. Malinowski, M. Hehn, S. Robert, O. Lenoble, A. Schuhl, P. Panissod, Magnetic origin of enhanced top exchange biasing in Py/IrMn/Py multilayers, Phys. Rev. B - Condens. Matter Mater. Phys. 68 (2003) 1 -- 8. doi:10.1103/PhysRevB.68.184404. [24] M. Ali, C.H. Marrows, M. Al-Jawad, B.J. Hickey, A. Misra, U. Nowak, K.D. Usadel, Antiferromagnetic layer thickness dependence of the IrMn/Co exchange-bias system, Phys. Rev. B. 68 (2003) 214420. doi:10.1103/PhysRevB.68.214420. [25] J.P. King, J.N. Chapman, M.F. Van 8 / Journal of Magnetism and Magnetic Materials 00 (11/25/18) 000 -- 0001 [26] K.M. Chui, D. Tripathy, Gillies, J.C.S. Kools, Magnetization reversal of NiFe films exchange- biased by IrMn and FeMn, J. Phys. D. Appl. Phys. 34 (2001) 528 -- 538. doi:10.1088/0022-3727/34/4/315. A.O. Adeyeye, Temperature dependence of magnetotransport properties of Ni80Fe20⁄Fe50Mn50⁄Ni80Fe20 trilayers, J. Appl. Phys. 101 (2007) 09E512. doi:10.1063/1.2709741. [27] X.Y. Lang, W.T. Zheng, Q. Jiang, blocking exchange biased the of in Dependence temperature ferromagnetic/antiferromagnetic bilayers on thickness of the antiferromagnetic Nanotechnology. 18 (2007). doi:10.1088/0957- 4484/18/15/155701. the layer, [28] C.H.T. Chang, S.C. Chang, J.S. Tsay, Y. Der Yao, Variation of blocking temperatures for exchange biased CoO/Co/Ge(100) films, AIP Adv. 6 (2016) 1 -- 6. doi:10.1063/1.4942553. [29] L.X. Phua, N.N. Phuoc, C.K. Ong, Investigation of the microstructure, magnetic and microwave properties of electrodeposited NixFe1−x(x=0.2 -- 0.76) films, J. Alloys Compd. 520 (2012) 132 -- 139. doi:10.1016/j.jallcom.2011.12.164. [30] X. Chen, H. Qiu, H. Qian, P. Wu, F. Wang, Tian, Characteristics of Ni x Fe 100Àx films deposited on SiO 2 /Si(1 0 0) by co-sputtering, DC Vacuum. 217 -- 223. doi:10.1016/j.vacuum.2004.03.001. magnetron (2004) [31] K.-M.H. Lenssen, A.E.M. De Veirman, J.J.T.M. Donkers, Inverted Pan, 75 Y. L. spin valves for magnetic heads and sensors, J. Appl. Phys. 81 (1997) 4915 -- 4917. doi:10.1063/1.364818. a [33] [32] H.N. Fuke, K. Saito, M. Yoshikawa, H. Iwasaki, M. Sahashi, Influence of crystal structure and oxygen content on exchange-coupling properties of IrMn/CoFe spin-valve films, Appl. Phys. Lett. 75 (1999) 3680 -- 3682. doi:10.1063/1.125427. I. Dzhun, N. Chechenin, K. Chichay, V. Rodionova, Dependence of exchange on antiferromagnetic layer thickness in NiFe / CoO bilayers, ACTA Phys. Pol. A. 83 (1998) 6822 -- 6824. doi:10.1063/1.367863. coupling [34] C. Gritsenko, I. Dzhun, G. Babaytsev, N. Chechenin, V. Rodionova, Exchange Bias and Coercivity Fields as the Antiferromagnetic Layer Thickness in bi- and tri- layered thin-films Based on IrMn and NiFe, Phys. Procedia. 82 (2016) 51 -- 55. doi:10.1016/j.phpro.2016.05.010. Function [35] D. Ledue, A. Maitre, F. Barbe, L. Lechevallier, Temperature dependence of the exchange bias properties of ferromagnetic/ antiferromagnetic polycrystalline bilayers, J. Magn. Magn. Mater. 372 (2014) 134 -- 140. doi:10.1016/j.jmmm.2014.07.021. [36] F. Radu, H. Zabel, Exchange bias effect ferro-/antiferromagnetic heterostructures, in: S. Bader, H. Zabel (Eds.), Magn. Heterostruct., 1st ed., Springer, Berlin, 2008: pp. 97 -- 184. doi:10.1007/978-3-540-73462-8. of of Fig. 1. TEM-images for the samples LNiPy(10 nm)/IrMn(10 nm)/LNiPy(10 nm) and HNiPy(10 nm)/IrMn(5 nm)/HNiPy(10 nm). 1.0 (a) S M M / 0.5 0.0 -0.5 -1.0 1.0 ) (b) s t i n u . b r a ( H ∂ Μ / ∂ ) s t i n u . b r a ( H ∂ Μ / ∂ (с) 0.5 0.0 -0.5 -1.0 1.0 0.5 0.0 -0.5 -1.0 -200 -100 LNiPy HNiPy ascending descending ascending descending 0 H, Oe 100 200 10 / Journal of Magnetism and Magnetic Materials 00 (11/25/18) 000 -- 0001 Fig. 2. (a) Hysteresis loops for the LNiPy and HNiPy samples NiFe/IrMn(10 nm)/NiFe, obtained along the unidirectional anisotropy, and corresponding differential susceptibility distribution for the LNiFe (b) and HNiPy (c) samples. S M M / 1.0 0.5 0.0 -0.5 -1.0 S M M / 1.0 0.5 0.0 -0.5 -1.0 90 К 200 К LNiPy 300 К HNiPy (a) (b) -300 -150 0 H, Oe 150 300 -100 -50 50 100 0 H, Oe (c) (d) m n 2 = ) n M r I ( t m n 4 = ) n M r I ( t -350 -175 0 H, Oe 175 -350 -175 0 H, Oe 175 Fig. 3. Hysteresis loops obtained along the unidirectional anisotropy for the LNiPy (a), (c) and HNiPy(b), (d) NiFe/IrMn(t)/NiFe samples. LNiPy Ascending Descending (b) HNiPy (a) 1.0 0.5 0.0 (c) (d) ) s t i n u . b r a ( H ∂ Μ / ∂ ) s t i n u . b r a ( H ∂ Μ / ∂ -0.5 -1.0 1.0 0.5 0.0 -0.5 -1.0 m n 2 = ) n M r I ( t m n 4 = ) n M -400 -200 0 200 400 -400 -200 0 200 400 H, Oe H, Oe r I ( t Fig. 4. The differential susceptibility distribution for LNiPy (a), (c) and HNiPy(b), (d) NiFe/IrMn(t)/NiFe samples at 90 K. (a) LNiPy-top LNiPy-bot HNiPy-top HNiPy-bot e O , C H 100 200 T(K) 300 (c) (b) LNiPy-top LNiPy-bot HNiPy-top HNiPy-bot m n 2 = ) n M r I ( t 100 (d) 200 T(K) 300 150 120 90 60 30 0 100 e O , C H 80 60 40 100 200 T(K) 300 100 200 T(K) 300 m n 4 = ) n M r I ( t , e O X E H 60 45 30 15 0 , e O X E H 300 200 100 0 Fig. 5. Exchange bias and coercivity dependences on the temperature for the top and bottom FM-layers in samples NiFe/IrMn(t)/NiFe with LNiPy (a), (c) and HNiPy (b), (d).
1501.04872	1	1501	2015-01-20T16:35:55	Competing scanning tunneling microscope tip-interlayer interactions for twisted multilayer graphene on the a-plane SiC surface	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	Scanning tunneling microscopy (STM) images are obtained for the first time on few layer and twisted multilayer epitaxial graphene states synthesized on n+ 6H-SiC a-plane non-polar surface. The twisted graphene is determined to have a rotation angle of 5.4{\deg} between the top two layers, by comparing moir\'e patterns from stick and ball models of bilayer graphene to experimentally obtained images. Furthermore, the experimental moir\'e pattern shows dynamic behavior, continuously shuffling between two stable surface arrangements one bond length apart. The moir\'e pattern shifts by more than 1 nm, making it easy to observe with STM. Explanation of this dynamic behavior is attributed to electrostatic interactions between the STM tip and the graphene sample.	cond-mat.mes-hall	cond-mat	Competing scanning tunneling microscope tip-interlayer interactions for twisted multilayer graphene on the a-plane SiC surface P. Xu a, M.L. Ackerman a, S.D. Barber a, J.K. Schoelz a, P.M. Thibado a,, V.D. Wheeler b, L.O. Nyakiti b, R.L. Myers-Ward b, C.R. Eddy, Jr. b, D.K. Gaskill b a Department of Physics, University of Arkansas, Fayetteville, AR 72701, USA b U.S. Naval Research Laboratory, Washington, DC 20375, USA ABSTRACT Scanning tunneling microscopy (STM) images are obtained for the first time on few layer and twisted multilayer epitaxial graphene states synthesized on n+ 6H-SiC a-plane non-polar surface. The twisted graphene is determined to have a rotation angle of 5.4° between the top two layers, by comparing moiré patterns from stick and ball models of bilayer graphene to experimentally obtained images. Furthermore, the experimental moiré pattern shows dynamic behavior, continuously shuffling between two stable surface arrangements one bond length apart. The moiré pattern shifts by more than 1 nm, making it easy to observe with STM. Explanation of this dynamic behavior is attributed to electrostatic interactions between the STM tip and the graphene sample. Corresponding author. E-mail address: [email protected] (P.M. Thibado). 1 1. Introduction The novel electronic properties of graphene have been spurring intense research interest ever since this two-dimensional (2D) material was first successfully isolated [1]. These intriguing properties, such as ballistic transport [2], the quantum Hall effect [3], and ultra-high mobility [4], mark graphene as a potentially crucial material in post-silicon electronics. In particular, epitaxial graphene grown on SiC via thermal decomposition has been identified as one of the most likely avenues to graphene-based electronics [5]. As a highly resistive material, semi- insulating SiC has a minimal effect on current flow in graphene, and it is already available in the form of large-diameter wafers compatible with current industrial technology [6-11]. This enormous promise has, as a result, triggered a much closer investigation of microscopic structural properties of epitaxial graphene on SiC [12-14]. For example, large-scale atomic force microscopy images have revealed 1-2 nm high ridges running parallel and perpendicular to steps in the substrate [15; 16]. These have typically been attributed to stress induced in the graphene layers during post-growth cooling as graphene expands and the substrate contracts [17-20]. Atomic-scale scanning tunneling microscopy (STM) experiments performed on the ridges established them as buckled regions of graphene and showed that they occasionally rearrange under the influence of the STM tip [21; 22]. However, the lack of an intrinsic band gap in graphene has proved a major deterrent to its application in digital electronic devices. One possible solution to this problem is graphene nanoribbons, which have a band gap inversely proportional to their width due to quantum confinement [23]. Initial methods for their manufacture involved cutting ribbons from larger sheets, resulting in disordered edges detrimental to transport [24]. Recently, though, graphene nanoribbons as narrow as 40 nm have been grown directly on nanofacets in SiC surfaces, 2 advantageously preserving their structural integrity [25]. This discovery has spurred the interest in studying other surface orientations of SiC. Another approach to introducing a band gap is by using bilayer graphene, which has a band gap in proportion to the strength of an applied perpendicular electric field [26; 27]. Bilayer graphene may also be grown directly on SiC surfaces [28], but its electronic properties will depend strongly on the angle of rotation between the two layers [29]. For instance, a reduction in the carrier velocity as a function of twist angle has been predicted [30], although this effect has recently been contested for small (< 3°) twist angles [31]. It has also been predicted that twisted bilayer graphene will have no band gap even in the presence of a perpendicular field [32]. It is therefore imperative to have more experimental studies of the properties of bilayer graphene [33]. SiC(112(cid:3364)0) or a-plane surface and show that there are regions of few layer and twisted multilayer In this paper, we study epitaxial graphene grown on the non-traditional, non-polar 6H- graphene. Experimental STM images are presented which display the characteristic moiré pattern corresponding to a 5.4° twist angle of the top layer relative to the layer underneath. In addition, we show that interactions between the tip and sample cause the top layer to undergo small horizontal displacements, with a ~10x larger displacement in the moiré pattern. 2. Experimental The epitaxial graphene sample used in this study was grown on the non-polar face of a n+ 6H-SiC a-plane nominally-oriented sample (Aymont Technology) measuring 16 mm × 16 mm. Growth was carried out in a commercial hot-wall Aixtron VP508 chemical vapor deposition reactor. Prior to graphene growth, the SiC substrate was etched in situ in a 100 mbar H2 ambient 3 at 1520 °C for 50 min. This etching produces a controlled starting surface that is smooth, specular and has an atomic force microscopy RMS roughness values of 0.12 nm (10 μm x10 μm height scan) that are within instrument resolution limit (Veeco D3100 operated in tapping mode), as shown in Fig. 1(a). After the H2 etching step, the ambient was switched to Ar with a transition time of 2 minutes during which pressures varied by ±50% around 100 mbar, this was followed by a temperature ramp to 1620 °C. The subsequent 60 minute graphene synthesis process was conducted under a flowing Ar environment of 20 standard L/min at 100 mbar, with a substrate temperature of 1620 °C. After growth, the sample was cooled in Ar to room temperature, diced into 5 mm × 5 mm samples, and diamond scribed labels were added to the back side. The post- growth morphology of the sample has terraces and steps throughout the surface as shown in Fig. 1(b). Next, Raman data was collected for the sample using a Thermo DXR system. A 532 nm, 8 mW laser was used as the pump probe with a spot size 0.7 m. After characterization with Raman, the sample was secured, sealed, and sent to the STM facility. Constant-current filled-state STM images were obtained using an Omicron ultrahigh- vacuum (base pressure is 10–10 mbar), low-temperature model STM operated at room temperature. The sample was mounted with silver paint onto a flat tantalum sample plate and transferred through a load-lock into the STM chamber where it was electrically grounded. STM tips were electrochemically etched from 0.25 mm diameter polycrystalline tungsten wire via a custom double lamella method with an automatic gravity-switch cutoff. After etching, the tips were gently rinsed with distilled water, briefly dipped in a concentrated hydrofluoric acid solution to remove surface oxides, and then transferred into the STM chamber. All images were acquired using a positive tip bias of 0.100 V and a tunneling current setpoint of 1.00 nA. 4 3. Results and discussion The average roughness of the surface after graphene formation was determined from AFM to be 1.06 nm and different graphene regions bounded by wrinkles/ridges were found as shown in Fig. 1(c). The average size of the regions is around 1 micron. Confocal Raman spectrometry confirms the presence and layer thickness of epitaxial graphene on the surface as shown in Fig. 1(d). The intensity of major Raman modes (D, G and 2D) as a function of Raman shift after subtracting the SiC background are shown. The small D peak intensity indicates the absence of a significant number of defects. The 2D peak is symmetric with a full width at half maximum (FWHM) of ~50 cm-1, which indicates that there are 2-4 layers of graphene on the surface of our sample. Note, a symmetric 2D peak having a FWHM of ~25 cm-1 would indicate a monolayer [34-37]. Three STM images of the graphene/SiC a-plane surface are shown in Fig. 2. A large- scale (200 nm × 200 nm) image is first given in Fig. 2(a), and it is characterized by a series of terraces with the surface height increasing from left to right. A horizontal line profile spanning the STM image and extracted from near the bottom is displayed beneath it, showing the step height and width of the substrate terraces. The steps average approximately 2.5 nm in height, each followed by a plateau around 50 nm in width, which gives an overall nominal wafer miscut of 2-3º, which is also confirmed with larger scale AFM images. Atomically resolved STM images were obtainable virtually everywhere on the terraces, and two such small-scale images, measuring 6 nm × 4 nm, are presented in Figs. 1(b) and 1(c). The triangular atomic symmetry indicative of highly coupled, few layer Bernal stacked graphene[38], appears throughout Fig. 2(b), yet a honeycomb structure, characteristic of graphene that has been largely electronically decoupled [39] appears throughout Fig. 2(c). Note, however, that the topography 5 of the latter also exhibits long-wavelength undulations. These two atomic-scale STM images show that few graphene layers were formed, which is not uncommon for thermal decomposition of SiC [40-42]. Further analysis of the modulated hexagonal pattern detected in Fig. 2(c) is performed in Fig. 3. First, a similar 6 nm × 6 nm atomic-scale STM image of the graphene/SiC surface displaying the honeycomb lattice with large alternating bright and dark spots is shown in Fig. 3(a). Then to highlight the topographic structure of the surface, two line profiles were extracted from the image and plotted in Fig. 3(b). The starting heights have been offset in order to see the detail in both curves. The upper (blue) curve was taken from top to bottom along the nearly vertical arrow in Fig. 3(a), and the lower (green) curve was taken from left to right along the nearly horizontal arrow. These directions were chosen to align with and bisect the bright spots, resulting in curves with peak-to-peak amplitude of 0.08 nm. The upper line has two peaks roughly 3 nm apart, whereas the two peaks of the lower line are separated by 4.5 nm. Using larger scale STM images with ten repeats the distance between the features is on average 2.8 nm and 4.7 nm, respectively. In an attempt to replicate the peak separation distances found in the experimental image, two computer-generated 2D hexagonal lattices were stacked on top of one another, as depicted in Fig. 3(c). The top layer was rotated clockwise relative to the lower layer in increments of 0.1° around a normal axis running through a carbon atom approximately in the middle of Fig. 3(c) until the spatial periodicity matched that of Fig. 3(a). An angular displacement of 5.4° was found to best match the experimental results. This displacement is illustrated more clearly in Fig. 3(d), which shows a close-up of hexagons from the top and bottom layers around the point of rotation as cropped from Fig. 3(c). 6 The simulated model suggests that the STM image of Fig. 3(a) portrays a moiré pattern of twisted graphene with a twist angle of 5.4°. The twist angle between the top two layers causes carbon atoms in the top layer to precisely overlap with carbon atoms in the lower layer (i.e., AA stacked) in certain sections. It is known that AA stacking of graphene layers is less energetically stable than AB stacking [43]. This produces a local displacement between the two layers and subsequent height increase of the top layer, which is measured in the line profiles of Fig. 3(b). In fact, at this twist angle Luican et al. indicate significant decoupling compared to untwisted bilayer graphene [30], consistent with our experimental results showing that the honeycomb structure of monolayer graphene is visible everywhere in Fig. 3(a) [39]. Interestingly, the experimental moiré pattern in many cases exhibited unusual systematic time dependence. To illustrate this effect, a series of 6 nm × 6 nm STM images was continuously collected without delay between scans at a single position on the graphene/SiC surface, where each scan took ~2 min. These images are presented in sequential order in Fig. 4, and overlaid on each is a diamond-shaped box outlining the unit cell of the moiré periodicity. Between Fig. 4(a) and Fig. 4(b), the cell has shifted up and to the right by about 1 nm. In the image taken immediately afterward [Fig. 4(c)], the cell has returned to its original position, yet in Fig. 4(d) it has shifted again as in Fig. 4(b). Finally, the initial configuration is once more observed in Fig. 4(e). Comparing the positions of the moiré pattern’s unit cell in this series of images implies that there are two distinct configurations occurring between the top and bottom graphene layers, with movement between them. Previous STM experiments on graphene and graphite have established that an electrostatic interaction between the tip and the sample can cause small horizontal shifts in the top graphitic layer [43], and here the moiré pattern images serve to magnify the fidelity of the movement. Since its spatial periodicity is approximately 10 times that 7 of monolayer graphene, the displacement in the moiré pattern is an order of magnitude larger than the actual top layer displacement. In order to fully characterize the bi-stable graphene configurations, as seen in Fig. 4, the 5.4° twisted graphene model was again employed. First, the simulated graphene bilayer was cropped to match the dimensions and moiré pattern of the STM image in Fig. 4(a). The resulting image with the moiré unit cell outlined is shown in Fig. 5(a). Then the top layer in the model was shifted by various amounts and in various directions. The resulting images were next compared to Fig. 4(b), and it was found that a shift along the carbon-carbon bond axis by one bond length best replicated the experimental results. This configuration is shown in Fig. 5(b). In order to more clearly see the atomic-scale shift, the two model images were cropped down to a few graphene unit cells. The small-scale picture before the shift is shown in Fig. 5(c), below the corresponding large-scale picture. Near the lower left corner the two layers exhibit AB stacking. The small-scale image after the horizontal shift is shown in Fig. 5(d), below its corresponding large-scale image. Near the lower left corner the two layers now exhibit AA stacking, indicative of a shift by one full bond length. The chosen magnitude and direction of the shift given above corresponds to transitions between the stacking patterns of graphite layers, as well [44]. Interestingly, shifting the top layer of trilayer graphite by this amount and in this direction shifts the stack from ABA stacking to ABC stacking. After this analysis of the local structural changes that occur between the two different moiré pattern configurations, we developed a proposed mechanism for the process. As the STM tip scans over the sample, the tip is disproportionately electrostatically attracted to the nearby AA stacked regions, which have a larger charge density and height. The STM tip, when in the correct position naturally exerts a small horizontal force, pulling the AA stacked graphene into 8 the energetically favored AB stacking position and, thereby leading to a decrease in the energy of that part of the system. However, the areas of AB stacked graphene are consequently pushed into AA stacking, increasing the energy in those parts of the system. So the total energy of the top two layers remains the same, even as the moiré pattern has shifted by a nanometer, as in Fig. 4. Thus, the combination of unfavorable energetics near AA stacked twisted graphene, along with electrostatic interactions between the tip and the sample, results in the formation of a dynamic moiré pattern with two stable states. 4. Conclusions Atomic-scale STM images of epitaxial graphene grown via thermal decomposition on the non-traditional, non-polar 6H-SiC a-plane surface were acquired for the first time. Multiple arrangements of few layer graphene were observed on the surface, including twisted multilayer graphene. Using the spatial periodicity of the associated moiré pattern, an angle of 5.4° between the top two layers was determined for the twisted graphene. Dynamic atomic-scale horizontal displacements in the twisted graphene were concluded based on magnified bimodal behavior in the moiré pattern, likely facilitated by partial decoupling between the graphene layers and the electrostatic influence of the STM tip. Acknowledgments P.X. and P.M.T. gratefully acknowledge the financial support of ONR under grant N00014-10-1-0181 and NSF under grant DMR-0855358. Work at the U.S. Naval Research Laboratory is supported by the Office of Naval Research. L.O.N. gratefully acknowledges postdoctoral fellowship support through the ASEE. 9 10 K.S. Novoselov, A.K. Geim, S.V. Morozov, D. Jiang, Y. Zhang, S.V. Dubonos, I.V. [1] Grigorieva, A.A. Firsov, Science 306 (2004) 666-669. [2] [3] X. Du, I. Skachko, A. Barker, E.Y. Andrei, Nature Nanotech. 3 (2008) 491-495. Y. Zhang, Y.-W. Tan, H.L. Stormer, P. Kim, Nature 438 (2005) 201-204. K.I. Bolotin, K.J. Sikes, Z. Jiang, M. Klima, G. Fudenberg, J. Hone, P. Kim, H.L. [4] Stormer, Solid State Commun. 146 (2008) 351-355. [5] [6] [7] [8] C. Berger, et al., Science 312 (2006) 1191-1196. T. Ohta, A. Bostwick, T. Seyller, K. Horn, E. Rotenberg, Science 313 (2006) 951-954. C. Riedl, U. Starke, J. Bernhardt, M. Franke, K. Heinz, Phys. Rev. B 76 (2007) 245406. J.B. Casady, R.W. Johnson, Solid-State Electronics 39 (1996) 1409-1422. L.O. Nyakiti, V.D. Wheeler, N.Y. Garces, R.L. Myers-Ward, C.R. Eddy, D.K. Gaskill, [9] Mrs Bull. 37 (2012) 1149-1157. [10] N. Camara, B. Jouault, A. Caboni, B. Jabakhanji, W. Desrat, E. Pausas, C. Consejo, N. Mestres, P. Godignon, J. Camassel, Appl. Phys. Lett. 97 (2010). [11] S. Tanaka, K. Morita, H. Hibino, Physical Review B 81 (2010). [12] M.K. Yakes, D. Gunlycke, J.L. Tedesco, P.M. Campbell, R.L. Myers-Ward, C.R. Eddy, Jr., D.K. Gaskill, P.E. Sheehan, A.R. Laracuente, Nano Lett. 10 (2010) 1559-1562. [13] N. Camara, B. Jouault, A. Caboni, A. Tiberj, P. Godignon, J. Camassel, Nanosci. Nanotech. Lett. 3 (2011) 49-54. [14] C. Vecchio, S. Sonde, C. Bongiorno, M. Rambach, R. Yakimova, V. Raineri, F. Giannazzo, Nano. Res. Lett. 6 (2011). [15] G.G. Jernigan, B.L. VanMil, J.L. Tedesco, J.G. Tischler, E.R. Glaser, A. Davidson, III, P.M. Campbell, D.K. Gaskill, Nano Lett. 9 (2009) 2605-2609. [16] Luxmi, P.J. Fisher, N. Srivastava, R.M. Feenstra, Y. Sun, J. Kedzierski, P. Healey, G. Gu, Appl. Phys. Lett. 95 (2009) 073101. [17] N. Ferralis, R. Maboudian, C. Carraro, Phys. Rev. Lett. 101 (2008) 156801. [18] N. Ferralis, J. Kawasaki, R. Maboudian, C. Carraro, Appl. Phys. Lett. 93 (2008) 191916. [19] Luxmi, S. Nie, P.J. Fisher, R.M. Feenstra, G. Gu, Y. Sun, J. Electron. Mater. 38 (2009) 718-724. 11 J. Röhrl, M. Hundhausen, K.V. Emtsev, T. Seyller, R. Graupner, L. Ley, Appl. Phys. [20] Lett. 92 (2008) 201918. [21] G.F. Sun, J.F. Jia, Q.K. Xue, L. Li, Nanotech. 20 (2009) 355701. [22] P. Xu, et al., Jap. J. Appl. Phys. 2013, 035104 [23] M.Y. Han, J.C. Brant, P. Kim, Phys. Rev. Lett. 104 (2010) 056801. [24] L. Tapasztó, G. Dobrik, P. Lambin, L.P. Biró, Nature Nanotech. 3 (2008) 397-401. [25] M. Sprinkle, M. Ruan, Y. Hu, J. Hankinson, M. Rubio-Roy, B. Zhang, X. Wu, C. Berger, W.A. de Heer, Nature Nanotech. 5 (2010) 727-731. [26] Y. Zhang, T.-T. Tang, C. Girit, Z. Hao, M.C. Martin, A. Zettl, M.F. Crommie, Y.R. Shen, F. Wang, Nature 459 (2009). [27] T. Ohta, A. Bostwick, T. Seyller, K. Horn, E. Rotenberg, Science 313 (2006) 951-954. [28] H. Kageshima, H. Hibino, S. Tanabe, J. Phys.: Condens. Matter 24 (2012) 314215. [29] G. Li, A. Luican, J.M.B. Lopes dos Santos, A.H. Castro Neto, A. Reina, J. Kong, E.Y. Andrei, Nature Phys. 6 (2010) 109-113. [30] A. Luican, G. Li, A. Reina, J. Kong, R.R. Nair, K.S. Novoselov, A.K. Geim, E.Y. Andrei, Phys. Rev. Lett. 106 (2011) 126802. [31] W. Yan, M. Liu, R.-F. Dou, L. Meng, L. Feng, Z.-D. Chu, Y. Zhang, Z. Liu, J.-C. Nie, L. He, Phys. Rev. Lett. 109 (2012) 126801. J.M.B. Lopes dos Santos, N.M.R. Peres, A.H. Castro Neto, Phys. Rev. Lett. 99 (2007) [32] 256802. [33] W.-T. Pong, C. Durkan, J. Phys. D: Appl. Phys. 38 (2005) R329-R355. [34] L.G. Cancado, K. Takai, T. Enoki, M. Endo, Y.A. Kim, H. Mizusaki, N.L. Speziali, A. Jorio, M.A. Pimenta, Carbon 46 (2008) 272-275. [35] L.G. Cancado, M.A. Pimenta, R. Saito, A. Jorio, L.O. Ladeira, A. Grueneis, A.G. Souza, G. Dresselhaus, M.S. Dresselhaus, Phys. Rev. B 66 (2002). [36] D.S. Lee, C. Riedl, B. Krauss, K. von Klitzing, U. Starke, J.H. Smet, Nano Letters 8 (2008) 4320-4325. [37] L.O. Nyakiti, et al., Nano Letters 12 (2012) 1749-1756. [38] D. Tománek, S.G. Louie, H.J. Mamin, D.W. Abraham, R.E. Thomson, E. Ganz, J. Clarke, Phys. Rev. B 35 (1987) 7790-7793. 12 [39] P. Xu, Y.R. Yang, D. Qi, S.D. Barber, J.K. Schoelz, M.L. Ackerman, L. Bellaiche, P.M. Thibado, Phys. Rev. B 86 (2012). [40] P.N. First, W.A. de Heer, T. Seyller, C. Berger, J.A. Stroscio, J.-S. Moon, MRS Bull. 35 (2010) 296-305. [41] B.K. Daas, S.U. Omar, S. Shetu, K.M. Daniels, S. Ma, T.S. Sudarshan, M.V.S. Chandrashekhar, Crys. Grow. Des. 12 (2012) 3379-3387. [42] K.V. Emtsev, F. Speck, T. Seyller, L. Ley, J.D. Riley, Phys. Rev. B 77 (2008). [43] P. Xu, Y. Yang, D. Qi, S.D. Barber, J.K. Schoelz, M.L. Ackerman, L. Bellaiche, P.M. Thibado, Phys. Rev. B 86 (2012) 085428. [44] P. Xu, Y. Yang, D. Qi, S.D. Barber, M.L. Ackerman, J.K. Schoelz, T.B. Bothwell, S. Barraza-Lopez, L. Bellaiche, P.M. Thibado, Appl. Phys. Lett. 100 (2012) 201601. 13 Figure captions Fig. 1 (a) An AFM image of the SiC(112(cid:3364)0) or a-plane surface after it was etched in situ in a 100 mbar H2 ambient at 1520 °C for 50 min. (b) An AFM image of the sample surface after multilayer epitaxial graphene was formed. (c) A zoomed-in view AFM image of multilayer epitaxial graphene. (d) Raman spectra of the surface obtained using confocal Raman spectrometry with an excitation wavelength of 532 nm and a laser spot size of 0.7 m. Fig. 2 (a) Large-scale (200 nm × 200 nm) STM image of epitaxial graphene on 6H-SiC a-plane surface showing a terraced structure. Line profile beneath was extracted horizontally across the image from near the bottom. (b) Atomic-scale (6 nm × 4 nm) STM image taken from a terrace and having the appearance of few layer graphene. (c) Atomic-scale (6 nm × 4 nm) STM image taken from a different region and having the appearance of graphene. Fig. 3 (a) Experimental 6 nm × 6 nm STM image of epitaxial graphene on 6H-SiC a-plane surface displaying a moiré pattern. (b) Line profiles taken along the arrows shown in (a). (c) Twisted bilayer graphene model. The top layer has been rotated 5.4° relative to the bottom layer. (d) Small-scale image of the model shown in (c). The axis of rotation is in the lower left corner. Fig. 4 Chronological sequence of atomic-scale 6 nm × 6 nm STM images of epitaxial graphene on 6H-SiC a-plane surface taken at a single location and revealing the horizontal movement of the top layer of graphene. A diamond representing the unit cell of the moiré pattern has been drawn on each image, and two configurations are found to exist. Fig. 5 (a) Twisted bilayer graphene model. The top layer is rotated 5.4° relative to the bottom layer. The image is cropped so that the regions of high carbon-carbon overlap match the positions of the bright spots in Fig. 4(a). (b) The same model after the top layer has been shifted 14 by one bond length. The regions of high carbon-carbon overlap now match the positions of the bright spots in Fig. 4(b). (c,d) Small-scale, zoomed-in images of the models shown in (a,b), respectively. 15 Fig. 1 16 Fig. 2 17 Fig. 3 18 Fig. 4 19 Fig. 5 20
1712.09773	2	1712	2018-03-13T08:30:18	Nodal surface semimetals: Theory and material realization	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	We theoretically study the three-dimensional topological semimetals with nodal surfaces protected by crystalline symmetries. Different from the well-known nodal-point and nodal-line semimetals, in these materials, the conduction and valence bands cross on closed nodal surfaces in the Brillouin zone. We propose different classes of nodal surfaces, both in the absence and in the presence of spin-orbit coupling (SOC). In the absence of SOC, a class of nodal surfaces can be protected by spacetime inversion symmetry and sublattice symmetry and characterized by a $\mathbb{Z}_2$ index, while another class of nodal surfaces are guaranteed by a combination of nonsymmorphic two-fold screw-rotational symmetry and time-reversal symmetry. We show that the inclusion of SOC will destroy the former class of nodal surfaces but may preserve the latter provided that the inversion symmetry is broken. We further generalize the result to magnetically ordered systems and show that protected nodal surfaces can also exist in magnetic materials without and with SOC, given that certain magnetic group symmetry requirements are satisfied. Several concrete nodal-surface material examples are predicted via the first-principles calculations. The possibility of multi-nodal-surface materials is discussed.	cond-mat.mes-hall	cond-mat	a Nodal surface semimetals: Theory and material realization Weikang Wu,1 Ying Liu,1 Si Li,1, 2 Chengyong Zhong,3 Zhi-Ming Yu,1 Xian-Lei Sheng,4, 1, ∗ Y. X. Zhao,5, 6, † and Shengyuan A. Yang1 1Research Laboratory for Quantum Materials, Singapore University of Technology and Design, Singapore 487372, Singapore 2Beijing Key Laboratory of Nanophotonics and Ultrafine Optoelectronic Systems, School of Physics, Beijing Institute of Technology, Beijing 100081, China 3Institute for Quantum Information and Spintronics, School of Science, Chongqing University of Posts and Telecommunications, Chongqing 400065, China 4Department of Physics, Key Laboratory of Micro-nano Measurement-Manipulation and Physics (Ministry of Education), 5National Laboratory of Solid State Microstructures and Department of Physics, Nanjing University, Nanjing 210093, China 6Collaborative Innovation Center of Advanced Microstructures, Nanjing University, Nanjing 210093, China Beihang University, Beijing 100191, China We theoretically study the three-dimensional topological semimetals with nodal surfaces protected by crystalline symmetries. Different from the well-known nodal-point and nodal-line semimetals, in these materials, the conduction and valence bands cross on closed nodal surfaces in the Brillouin zone. We propose different classes of nodal surfaces, both in the absence and in the presence of spin- orbit coupling (SOC). In the absence of SOC, a class of nodal surfaces can be protected by spacetime inversion symmetry and sublattice symmetry and characterized by a Z2 index, while another class of nodal surfaces are guaranteed by a combination of nonsymmorphic two-fold screw-rotational symmetry and time-reversal symmetry. We show that the inclusion of SOC will destroy the former class of nodal surfaces but may preserve the latter provided that the inversion symmetry is broken. We further generalize the result to magnetically ordered systems and show that protected nodal surfaces can also exist in magnetic materials without and with SOC, given that certain magnetic group symmetry requirements are satisfied. Several concrete nodal-surface material examples are predicted via the first-principles calculations. The possibility of multi-nodal-surface materials is discussed. I. INTRODUCTION Topological states of matter have been attracting great interest in recent physics research. Inspired by the pre- vious studies on topological insulators [1 -- 3], the research focus is now shifted towards topological semimetals [4 -- 7]. In these materials, the electronic band structures possess nontrivial topology and/or symmetry-protected band crossings close to the Fermi level, such that the low-energy electrons behave drastically different from the usual Schrodinger-type fermions. For example, in Weyl/Dirac semimetals [7 -- 10], the conduction and va- lence bands cross linearly at isolated k points, around which the electrons resemble the massless Weyl/Dirac fermions from the standard model, making it possible to simulate intriguing high-energy and relativistic physics phenomena in table-top experiments [11 -- 13]. The nontrivial band crossings can be classified based on their dimensionalities. The crossings in the afore- mentioned Dirac/Weyl semimetals are isolated zero- dimensional (0D) points. Materials with 1D band cross- ings, known as nodal-line semimetals, have also been proposed and intensively studied in recent works [14 -- 21]. For 3D materials, there is one remaining possibil- ity: The band crossings may form a 2D nodal surface, namely each point on the surface is a crossing point be- ∗Electronic address: [email protected] †Electronic address: [email protected] tween two bands whose dispersions are linear along the surface normal direction. Although both are 2D mani- folds in the 3D Brillouin zone (BZ), we stress that the nodal surface is distinct from the ordinary Fermi surface, because the coarse-grained quasiparticles excited from a nodal surface have an intrinsic pseudospin degrees of free- dom (representing the two crossing bands) [22], behav- ing effectively as 1D massless Dirac fermions along the surface normal direction, and therefore may have inter- esting physical properties. In addition, in contrast to the ordinary Fermi surface as a sphere, each nodal surface identified in this article is a torus with opposite edges attached on the periodic boundaries of the BZ. To date, such nodal-surface semimetals (NSSMs) have only appeared in a few scattered theoretical studies. In Ref. [22], Zhong et al. proposed a family of sta- ble graphene network materials, each exhibiting a pair of nodal surfaces close to the Fermi energy. Liang et al. [23] studied the hexagonal BaM X3 (M = V, Nb, or Ta; X = S or Se) and found a single nodal surface in these materials when the spin-orbit coupling (SOC) can be neglected. From symmetry and topology analysis, Bzdusek and Sigrist discussed the possibility to stabilize nodal surfaces in centrosymmetric systems [24]. More recently, the nodal surface similar to Ref. [23] was pre- dicted for a proposed acoustic metamaterial [25], and the stability of nodal surfaces against perturbations was theoretically investigated [26]. Admittedly, our current understanding of NSSMs is still at the primitive stage with several important issues to be addressed. First, the nodal surfaces in the carbon allotropes were found through first-principles calculations, but its topological origin has not been clearly elucidated. Second, given that the nodal surfaces in the aforementioned works are all vulnerable against SOC, is it possible to have nodal surfaces robust under SOC? This important question is still waiting to be answered. Third, the carbon allotropes in Ref. [22] have yet to be synthesized, while the nodal surfaces in BaVS3 [23] can only be maintained in the room-temperature structural phase. It is thus important to search for more realistic material candidates to facili- tate the experimental investigation on NSSMs. Motivated by the above questions and issues, in this work, we present a theoretical study of the 3D NSSMs both in the absence and in the presence of SOC. In the absence of SOC, we propose two different classes of symmetry-protected nodal surfaces corresponding to the two materials studied in Refs. [22, 23]. The first class of the nodal surfaces (as those in the graphene networks) are protected by spacetime inversion symmetry and sublat- tice symmetry, and are characterized by a Z2 topological index. The second class of nodal surfaces are guaranteed by a combination of two-fold screw-rotational symmetry and time-reversal symmetry. After the inclusion of SOC, the former class of nodal surfaces is destroyed generically, but the latter can still be protected with the requirement that the space-time inversion symmetry is violated. Fur- thermore, it is found that both in the absence and in the presence of SOC, the protection in the latter case persists even in the presence of magnetic orders perpen- dicular to the screw axis, where, although time reversal symmetry is violated, the system is still invariant un- der the combination of screw rotation and time reversal. Via first-principles calculations, we identify several can- didate NSSM materials with relatively clean low-energy band structures, which support the theoretical analysis in this paper and will facilitate experimental studies on the novel physical properties associated with nodal surfaces. It is noteworthy that the band structures of the two classes are essentially different due to their qualitatively distinct origins. Nodal surfaces in the first classes have the nontrivial Z2 topological charge, and therefore ap- pear in pairs in the BZ conforming the Nielsen-Ninomiya no-go theorem [27, 28], while those in the second class mainly due to the two-fold nonsymmorphic symmetry exist alone in the BZ, for which the topological essence was revealed in Ref. [29]. Particularly, each degenerate point on such a nodal surface belong to a twisted band structure analogous to the Mobius strip [29], and the band structure as a whole is equivalent to a collection of Mobius strip parametrized by a torus (a 2D sub-BZ perpendicular to the screw axis). This paper is organized as follows. In Sec. II, we dis- cuss the two classes of nodal surfaces in the absence of SOC. In Sec. III, we propose nodal surfaces that can be protected in the presence of SOC. In Sec. IV, we gen- eralize the discussion to magnetically ordered systems, predicting nodal surfaces in magnetic materials. In each section, we present corresponding material examples with 2 band structures obtained from first-principles calcula- tions. The details of our first-principles methods are pre- sented in the Appendix. Discussion and conclusion are presented in Sec. V. II. NODAL SURFACE IN THE ABSENCE OF SOC A. Nodal surface with Z2 topological charge Let us start with the first class (class I) of nodal sur- faces, which require spacetime inversion symmetry PT with the relation (PT )2 = 1 and sublattice symmetry S. Here P and T denote space inversion and time rever- sal, respectively, and the sublattice symmetry S satis- fies the anti-commutation relation with the Hamiltonian: {S,H(k)} = 0. It is required that PT and S are mu- tually independent, namely [PT ,S] = 0. The relation of (PT )2 = 1 can be realized by particles with integer spin, such as photonic crystals or phonons of mechani- cal systems and certain cold-atom systems and by elec- tronic systems with negligible SOC, such as the carbon allotropes studied in Ref. [22]. In Ref. [22], the symme- try/topology aspect was not fully exposed, and will be our focus in the following discussion. We now address the Z2 topological classification of the class-I nodal surfaces under the constraint of PT and S symmetries [24, 30]. Since both P and T in- verse k, PT operates trivially in momentum space. As P is unitary but T anti-unitary, we can choose the rep- resentation in momentum space that PT = K, which is unique up to a unitary transformation. Here K is the complex conjugation. Therefore, a PT -symmetric Hamiltonian H(k) is real in this representation. On the other hand, if S is represented by S = σz ⊗ 1N , where σz is the third Pauli matrix and 1N the N × N iden- tity matrix with N being the number of valence bands, then the Hamiltonian H(k) takes the block anti-diagonal form, i.e., H(k) = antidiag[Q(k), Q†(k)] [30]. Because of [PT ,S] = 0, Q(k) can always be converted to a real ma- trix by a unitary transformation. Thus, for any gapped momentum k, the Hamiltonian H(k) (after being flat- tened) can be topologically regarded as a point in the space O(N ) (with the assumption N ≥ 2), which is ex- actly the Hamiltonian space for the class BDI among the ten Altland-Zirnbauer (AZ) symmetry classes [31, 32]. The space of O(N ) has two disconnected components, namely π0[O(N )] = Z2, and the sign of the determinant of Q(k) specifies which component H(k) belongs to. In the framework of topological classification of nodal surfaces, the spatial codimension of nodal surfaces in three dimensions is zero, namely a 0D sphere S0 consist- ing of two points is chosen to surround a nodal surface in 3D BZ from its transverse dimensions [12, 33]. We now consider two gapped points k1,2 with sgn[DetQ(k1,2)] = ±1, respectively. Then, any path connecting the two points in momentum space has to pass through at least a band-crossing point k0 with DetQ(k0) = 0 as a result of the mean value theorem, and the band-crossing point has a two-fold degeneracy, which is of topological stabil- ity once PT and S are preserved. These two-fold degen- erate band-crossing points generically spread out a com- pact nodal surface in momentum space, namely a class-I nodal surface, which accordingly has a non-trivially Z2 topological charge. 3 at odd sites to holes at their even nearest neighbours and vice versa (see the detailed discussion below), given that the chemical potential is appropriately tuned for each orbital. The band structure of QGN(1,2) obtained from the first-principles calculations is plotted in Fig. 1(d). The band structures with and without SOC have little dif- ference, due to the negligible SOC strength of carbon atoms. One observes the linear-type band crossing along the Γ-Z and the M-A paths. As mentioned in Ref. [22], these crossing points form nodal surfaces and there is a pair of such surfaces related by T or P in the BZ, as illustrated in Fig. 1(c). Each point on the surfaces is a linear crossing point between two bands along the surface normal direction (approximately the kz direction). And either nodal surface is quite flat with very small energy variation, and is close to the Fermi level. To reveal the topological nature of the class-I nodal surface of QGN(1,2), we first work out the tight-binding (TB) model corresponding to its low-energy physics. At low energies, the two crossing bands are dominated by the π orbitals from the eight sp2 carbon atoms in a unit cell, which are numbered clockwisely from 1 to 8 as shown in Fig. 1(b), and the eight-band TB model H(k) is ex- plicitly given in Appendix A. The sublattice symmetry † S relates ci to c i+1 with i = 1, 3, 5, 7, and therefore is represented as S = 14 ⊗ σz. Here and hereafter, we express the 8 × 8 matrices using the Kronecker prod- ucts of the Pauli matrices (σi, τi) and the identity ma- trices. The chiral symmetry operator can be converted to (cid:101)S = σz ⊗ 14 by the unitary transformation UΓ that exchanges c2 and c7 as well as c4 and c5, so that the trans- formed Hamiltonian is block diagonalized with the upper right block denoted by A(k). The operators are now or- dered as c = (c1, c7, c3, c5, c4, c6, c2, c8)T . From Fig. 1(b), P maps c1 to c5, c3 to c7 for odd orbitals, and c2 to c6, c4 to c8 for even orbitals, namely P = 12⊗ (σx⊗ τx) with τx being the first Pauli matrix. Since T is simply com- plex conjugation, it is found that PT = 12 ⊗ σx ⊗ τxK , there exists a unitary transformation converting PT to 4 σx⊗τx . After this unitary transformation, the upper right block A(k) becomes a real matrix Q(k). Now the Z2 topologi- cal charge of a nodal surface can be defined as which commutes with (cid:101)S. The relation (PT )2 = 1 implies (cid:103)PT = K, which turns out to be UP T = 12 ⊗ ei π ν = (sgnDet[Q(k1)] − sgnDet[Q(k2)])/2 mod 2, (1) where k1,2 are two points on two sides of the nodal sur- face, respectively. It is verified in Appendix A that both nodal surfaces illustrated in Fig. 1(c) have the nontrivial topological charge. We have two remarks before proceeding. First, in Ref. [24], Bzdusek and Sigrist developed a theory for the centrosymmetric extension of the ten AZ classes. The example discussed here would fit into the BDI class dis- cussed in their work. Second, the similar S symmetry (known as the chiral symmetry) also naturally emerge FIG. 1: (a) Lattice structure and (b) unit cell of the carbon al- lotrope QGN(1,2). (a) shows a 2×2×2 supercell, in which the unit cell is marked by the black box. In (b), the red-colored carbon atoms have a sp2 hybridization character (which are labeled from site 1 to site 8), whereas the blue-colored carbon atoms have a sp3 hybridization character. (c) Corresponding Brillouin zone, where the two nodal surfaces are schematically shown by the shaded surfaces. (d) Calculated band structure for QGN(1,2) without SOC. The class-I nodal surface can be realized by the carbon allotropes, which has been studied in Ref. [22] with other emphases. This family of materials share both P and T symmetry with negligible SOC, and therefore have the desired PT symmetry with (PT )2 = 1. The chiral sym- metry S also emerges in the low-energy effective theory by appropriately tuning the chemical potential of each site. To be concrete, we take one particular example in this family, namely QGN(1,2), where QGN stands for the quadrilateral graphene network, as shown in Fig. 1. The crystal structure of QGN(1,2) has the space group sym- metry of P 42/mmc (No. 131). Its unit cell contains two kinds of graphene nanorbbon-like structural motifs: The edge atoms show a strong sp3 hybridization character, whereas the remaining atoms exhibit sp2 hybridization character and form four zigzag chains running along the c-axis. From atomic orbital projection, one finds that the low-energy states are dominated by the π orbitals on the eight sp2 carbon atomic sites which are marked in red color in Fig. 1(b). Thus, it is emerged a sublattice symmetry in low-energy regime, which relates electrons (cid:66)(cid:67)(cid:24)(cid:25)(cid:18)(cid:19)(cid:22)(cid:21)(cid:20)(cid:23)(cid:66)(cid:67)(cid:68)(cid:9)(cid:66)(cid:10)(cid:9)(cid:68)(cid:10)(cid:9)(cid:69)(cid:10)(cid:9)(cid:67)(cid:10)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:1013)(cid:59)(cid:34)(cid:46)(cid:57)(cid:51)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:57)(cid:46)(cid:59)(cid:51)(cid:34)(cid:1013)in the low-energy excitation spectrum for superconduc- tors. Hence, nodal surfaces may also appear in certain centrosymmetric superconductors, like that discussed in Refs. [24, 34]. B. Essential nodal surface dictated by nonsymmorphic symmetry Next, we consider the second class (class-II) of nodal surfaces existing in the absence of SOC. Such surface is a result of the combination of a two-fold screw symmetry and time reversal symmetry. It is essential in the sense that its presence and location are solely dictated by the symmetries, as we discuss below. Consider the two-fold screw rotation S2z : (x, y, z) → (−x,−y, z + 1 2 ), which is a nonsymmorphic symmetry involving a half translation along the rotation axis. In momentum space S2z inverses kx and ky, and preserves kz. Without SOC, one finds that (S2z)2 = T001 = e−ikz , where T001 is the translation along the z-direction by a lattice constant. T is anti-unitary and inverses k with the relation T 2 = 1. Thus, the combination T S2z is anti-unitary and only inverses kz. Since [T , S2z] = 0, T S2z satisfies (T S2z)2 = e−ikz . (2) We now regard the system under consideration as a col- lection of 1D kz-subsystems parametrized by (kx, ky). Each 1D subsystem with given (kx, ky) effectively has the same non-symmorphic magnetic symmetry T S2z with the relation (2). Hence for a two-band theory, it has a twisted band structure of Mobius strip with a single band-crossing at the boundary of the 1D sub-BZ, which is enforced by the topological nature of the two-fold non- symmorphic symmetry, as shown in Refs. [29, 35, 36], where the momentum-dependent symmetry operator is also associated with a unit winding number. The Mobius- strip type band structure has a nontrivial Z2 topological charge, which means that, given the symmetry being pre- served, a single band-crossing is topologically stable, and can only be gapped in pair in four-band theories [29]. Since a linear crossing point exists for every 1D subsys- tem (kx, ky) at the sub-BZ boundary, the collection of crossings form a nodal surface on the boundary plane of the 3D BZ with kz = π. The crossings on the surface can also be understood as a result of the Kramers degeneracy. One notes that any point on the kz = π plane is invariant under T S2z, but from Eq. (2), the anti-unitary symmetry satisfies (T S2z)2 = −1 (3) on the whole plane. Thus, the two-fold Kramers degen- eracy arises at every point on the plane. This degeneracy is generically lifted away from this plane due to the loss of symmetry protection, so that a nodal surface is formed at the kz = π plane. The above argument was presented 4 in Ref. [23], where BaVS3 was identified as a candidate material with this type of nodal surfaces. As we have mentioned, the condition for the presence of the class-I nodal surface is quite stringent: Besides the symmetries PT and S, its appearance also requires regions in BZ with inverted band orderings (different Z2 indices). In contrast, the presence of the class-II nodal surface discussed here is solely guaranteed by symmetry (hence can be regarded as an essential band-crossing), and its location is fixed at kz = π (if the screw axis is along z). This makes it easier for search for candidate materials by analyzing the space groups. We emphasize that although the existence of class-II nodal surface is determined by symmetry, its energy is not determined and depends on the specific material. In addition, there could be strong energy variation from point to point on the nodal surface. In order for the nodal surface to manifest in physical properties, a "good" can- didate NSSM needs to satisfy the following requirements: (i) the nodal surface should have relatively small energy variation; (ii) its energy should be close to Fermi level; and (iii) it is desired that no other extraneous band is present at low energy. FIG. 2: (a) Crystal structure of K6YO4 and (b) the corre- sponding Brillouin zone. In (b), R and E are the mid-points of the paths A-L and Γ-M, respectively. (c) Band structure of K6YO4 without SOC. The red arrows indicate the band degeneracy along the paths on the nodal surface. Here we identify example materials which possess the class-II nodal surfaces. The first example is the com- pound K6YO4 with the structure in the space group P 63mc (No. 186), as illustrated in Fig. 2(a). This mate- rial is predicted in Materials Project [37]. In the crystal structure, each yttrium atom is surrounded by a tetra- hedron of O atoms and is located at the 2b positions: 2 , 2 ( 1 2 ) with u = 0.239, while the K atoms fill the space between the tetrahedra, forming two types of triangles with different sizes. The material is 3 , u) and ( 2 3 , 1 3 , u + 1 (cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:1013)(cid:34)(cid:46)(cid:44)(cid:41)(cid:45)(cid:38)(cid:51)(cid:67)(cid:68)(cid:66)(cid:48)(cid:44)(cid:58)(cid:9)(cid:66)(cid:10)(cid:9)(cid:67)(cid:10)(cid:9)(cid:68)(cid:10)nonmagnetic and has a two-fold screw axis along the z- direction exists, hence the condition for a class-II nodal surface is satisfied. To be noted, the material does not have the inversion symmetry. The calculated band structure of K6YO4 in the ab- sence of SOC is displayed in Fig. 2(c). Since the material is composed of light elements, the SOC has negligible ef- fect on the band structure, which has been checked by our density functional theory (DFT) calculation. From the result, one observes that a nodal surface is indeed present in the kz = π plane, where the two low-energy bands cross linearly. In Fig. 2(c), we purposely show the dispersion along a generic k-path E-R, which is not a high-symmetry path [see Fig. 2(b)] and on which the lin- ear crossing can be clearly observed. This is in agreement with the above argument that the anti-unitary symmetry T S2z guarantees a nodal surface at the kz = π plane. In addition, the energy variation of the nodal surface at the kz = π plane is small (< 0.2 eV), the linear dispersion range is relatively large (above 0.5 eV), and there is no other extraneous band close to the Fermi level. These features are desired for a NSSM. FIG. 3: (a) Perspective view and (b) top view of the crystal structure of the AMo3X3 family materials, where A = (Na, K, Rb, In, Tl) and X = (S, Se, Te). The insert in (b) shows the face-sharing Mo6 octahedra chains. (c) Band structure of RbMo3S3 without SOC. The red arrows indicate the band de- generacy along the paths on the nodal surface. The Brillouin zone here has the same shape as that in Fig. 2(b). The second example is a family of materials with the formula AMo3X3, where A = (Na, K, Rb, In, Tl) and X = (S, Se, Te). These AMo3X3 compounds have been synthesized in experiment [38, 39]. They share the same crystalline structure with the space group P 63/m (No. 176), as shown in Figs. 3(a) and 3(b). The structure features close-packed one-dimensional [Mo3X3] columns, consisting of face-sharing Mo6 octahedra surrounded by X atoms. The [Mo3X3] columns are oriented along the z- 5 direction, arranged in a trigonal lattice in the x-y plane, and related to each other by a screw rotation. The A atoms are intercalated in the large holes between the columns. The space group contains the S2z and T sym- metries. And it should be noted that the structure pre- serves the inversion symmetry P. This point will be im- portant for the later discussion in Sec. III when we deal with the case with SOC included. Let us consider one specific example RbMo3S3 in this family, for which the effect of SOC on the low- energy bands is small. The calculated band structure for RbMo3S3 in the absence of SOC is shown in Fig. 3(c). Similarly to K6YO4, one observes a nodal surface located at the kz = π plane, guaranteed by the anti-unitary sym- metry T S2z. The linear dispersion range is above 1 eV, even larger than that of K6YO4. The nodal surface is very flat in energy, and it sits almost exactly at the Fermi level. Such a band structure would be ideal for studying the class-II nodal surfaces. Before proceeding, we mention that the structure of the AMo3X3 family compounds have strong quasi-1D character, which may induce structural instabilities to- wards Peierls distortion. First-principles calculations in- dicated that except for TlMo3Te3 and RbMo3Te3, other members in this family would be prone to a Peierls distor- tion that breaks the S2z symmetry [40], hence destroying the nodal surface. However, such distortion has not been detected in experiment [41, 42], and future studies are needed to clarify this issue. III. NODAL SURFACE IN THE PRESENCE OF SOC Compared to the case without SOC, the inclusion of SOC makes at least two important differences in terms of symmetry properties: (i) The time reversal operator now satisfies T 2 = −1 because the time reversal operation reverses the electron spin and (ii) all rotation operations need to operate on the spin in addition to the spatial degree of freedom. For class-I nodal surfaces, the protection mechanism discussed in Sec. II A no longer holds, because of point (i). Hence introducing SOC will generally destroy class-I nodal surfaces. We explicitly verify this point by arti- ficially increasing the SOC strength in QGN(1,2), and indeed find that the nodal surface is gapped out by SOC (see Appendix C). For class-II nodal surfaces, a more involved analysis is needed. It turns out that the inversion symmetry P plays an important role in this case. In order to ensure the boundary Kramers degeneracies due to T S2z are pro- tecting a nodal surface, PT symmetry has to be violated. First, Eq. (3) still holds with SOC. Although T 2 = −1, performing S2z twice also rotates spins, and we have (S2z)2 = T001 ¯E = −e−ikz , (4) where ¯E denotes the 2π-rotation on spin, contributing (cid:67)(cid:68)(cid:66)(cid:9500)(cid:46)(cid:80)(cid:9477)(cid:66)(cid:67)(cid:9)(cid:66)(cid:10)(cid:9)(cid:67)(cid:10)(cid:9)(cid:68)(cid:10)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)a factor of −1. As [T , S2z] = 0, Eq. (2) still holds, so does Eq. (3). However, as long as inversion symmetry is present, the Kramers degeneracies on the BZ bound- ary with kz = π can extend to the whole bulk of BZ (along any generic path) because of the spacetime inver- sion symmetry PT . In contrast to the case without SOC, (PT )2 = −1, hence there is actually the Kramers degen- eracy at every point in the BZ due to PT , which acts locally in momentum space. Consequently, the degener- acy at the kz = π plane in this case does not represent a nodal surface according to our definition. 6 generacy by T S2z in the kz = π plane does not lead to a nodal surface. This demonstrates that when P is preserved, T S2z can no longer guarantee a nodal surface in the presence of SOC. Meanwhile, it is noted that in Fig. 4(b), the crossing is maintained along Γ-A path at the A point, which is enabled by the six-fold rotational symmetry on this path. A recent study shows that the crossing point at A is actually a cubic Dirac point with linear dispersion along kz and cubic dispersion in the kz = π plane [40]. FIG. 5: (a) Side view and (b) top view of the crystal structure for Ta3TeI7. The inset in (b) shows the cluster unit of the triangular lattice. (c) Calculated band structure for Ta3TeI7 in the presence of SOC. The right panel shows the enlarged view of the band structure near the Fermi level along a generic path E-R-H, where E-R is perpendicular to the kz = π plane and R-H is in the kz = π plane. The Brillouin zone and the labeled k points are the same as in Fig. 2(b). Hence, from the above argument, a necessary condition for a T S2z-protected nodal surface would be that the P symmetry needs to be broken (here we consider nonmag- netic materials where T is preserved). This will generally lift the two-fold degeneracy away from the kz = π plane, thus pairs of non-degenerate bands along a generic path have to cross at the kz = π plane, making the doubly- degenerate kz = π plane a nodal surface. Here, it should be mentioned that for systems without SOC (those in Sec. II B), spin is a dummy degree of freedom. Hence, if the spin degeneracy is counted, then the nodal sur- face there would be four-fold degenerate. In comparison, in the presence of SOC, the obtained nodal surfaces are two-fold degenerate essentially due to a Kramers-like de- generacy, and there is no protection for a four-fold de- generate surface (at least by the symmetries discussed in this work). This picture is indeed confirmed by our calculation on an example material Ta3TeI7. Its crystal structure has the space group P 63mc (No. 186), which does not possess an inversion center. Ta3TeI7 is a member of the M3QX7 material family (M = Nb, Ta; Q = S, Se, Te; and X = Cl, Br, I) [43, 44]. The crystal structure consists of or- dered, close-packed layers of I and Te atoms, interleaved FIG. 4: Calculated band structures of TlMo3Te3 (a) with- out SOC and (b) with SOC. Panel (c) shows the Brillouin zone. R and E are the mid-points of the paths A-L and Γ-M, respectively. (d) Enlarged view of the low-energy band dis- persion around the R and H points, showing that the original band-crossing at the nodal surface is gapped by SOC. To explicitly demonstrate the above point, we perform the first-principles calculation on the material TlMo3Te3 in the AMo3X3 family with SOC included. Its crystal structure has been shown in Fig. 3(a) and 3(b). As men- tioned before, the lattice structure of TlMo3Te3 preserves the inversion symmetry P, and unlike RbMo3S3, the SOC in TlMo3Te3 has appreciable effects on the band struc- ture and needs to be taken into account. In Figs. 4(a) and 4(b), we plot the band structures of TlMo3Te3 without and with SOC. In the absence of SOC, a nodal surface is present at the kz = π plane, similar to that for RbMo3S3. After including SOC, band splitting is observed at the original nodal surface. One notes that the band disper- sion along the K-H and E-R no longer shows the band crossing [see Figs. 4(b) and 4(d)], hence the two-fold de- (cid:9)(cid:66)(cid:10)(cid:9)(cid:67)(cid:10)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:14)(cid:17)(cid:15)(cid:19)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:19)(cid:1)(cid:17)(cid:15)(cid:21)(cid:51)(cid:41)(cid:44)(cid:14)(cid:17)(cid:15)(cid:17)(cid:25)(cid:14)(cid:17)(cid:15)(cid:17)(cid:21)(cid:1)(cid:17)(cid:15)(cid:17)(cid:17)(cid:38)(cid:51)(cid:41)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:9)(cid:68)(cid:10)(cid:1013)(cid:34)(cid:46)(cid:44)(cid:41)(cid:45)(cid:38)(cid:51)(cid:9)(cid:69)(cid:10)(cid:79)(cid:80)(cid:1)(cid:52)(cid:48)(cid:36)(cid:52)(cid:48)(cid:36)(cid:67)(cid:68)(cid:42)(cid:53)(cid:66)(cid:53)(cid:70)(cid:66)(cid:67)(cid:9)(cid:66)(cid:10)(cid:9)(cid:67)(cid:10)(cid:9)(cid:68)(cid:10)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:1062)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:14)(cid:17)(cid:15)(cid:18)(cid:19)(cid:14)(cid:17)(cid:15)(cid:17)(cid:25)(cid:14)(cid:17)(cid:15)(cid:17)(cid:21)(cid:38)(cid:51)(cid:41)by Ta atoms, as shown in Fig. 5(a) and 5(b). Figure 5(b) displays the top view of a single Ta3TeI7 layer, which is similar to the structure of 1T'-MoS2. This layered ma- terial have several types of stacking. Here we consider the bulk structure with the ABAB layer stacking pat- tern. The calculated band structure with SOC included is shown in Fig. 5(c). The material is a semiconduc- tor with a bandgap about 0.48 eV. Although it is not a semimetal, by examining the band dispersion around the kz = π plane, one finds that there is indeed a nodal surface in that plane consisting of linear band-crossing points. This is more easily observed in the right panel of Fig. 5(c), which shows a zoom-in image for the dispersion along the generic path E-R. Along the path, the bands are non-degenerate (due to the broken P), and a pair of bands must cross at R in the kz = π plane, hence a nodal surface is formed in that plane. IV. NODAL SURFACE IN MAGNETIC MATERIALS The discussion in the previous sections are about non- magnetic materials, such that T is preserved. We have seen that T plays an important role in stabilizing the nodal surfaces. Then the question arises: Is it possible to realize protected nodal surfaces also in magnetic materi- als? Here we show that this is indeed possible, and the re- sult also depends on whether the SOC in the system can be neglected or not. First, consider the case in the absence of SOC. Without SOC, the spin and the orbital degrees of freedom are in- dependent and can be considered as different subspaces. The spins can be oriented in any direction without affect- ing the orbital part of the wave-function. With a chosen spin polarization axis, the two spin channels are decou- pled, and hence the bands for each spin species can be effectively regarded as for a spinless system. Therefore, for the states of one spin, all the crystalline symmetries are preserved [45, 46]. Consequently, the analysis can be reduced to those in Sec. II, which means that it is in principle possible to have class-I as well as class-II nodal surfaces in a magnetic system, if the symmetry requirements presented in Sec. II are satisfied. Particu- larly, if the crystal structure possesses a two-fold screw axis S2z, then there must exist a nodal surface (for each spin species) at kz = π. We demonstrate the above point using a concrete ma- terial example. We consider the material CsCrI3, which is a member of the material family CsCrX3 (X = Cl, Br, I) [47]. The material takes the BaVS3-type struc- ture, with space group P 63/mmc (No. 194), as shown in Fig. 6(a) and 6(b). The Cr atoms are surrounded by octahedra of X atoms, forming one-dimensional chains along the z-axis, with octahedra sharing common faces. Those chains are arranged into a trigonal lattice in the x-y plane, with the Cs atoms intercalated between the 7 FIG. 6: (a) Side view and (b) top view of the crystal struc- ture for CsCrI3. (c) Calculated band structure for CsCrI3 in the absence of SOC, showing a ferromagnetic state. The red and blue lines represent spin-up and spin-down bands, respec- tively. In the absence of SOC, the nodal surface is preserved, and the band structure does not depend on the direction of the magnetic moments. chains. Without considering the spin polarization, the lattice structure has a two-fold screw axis S2z. The ma- terial assumes a ferromagnetic ground state, for which the magnetic moments are provided by the Cr atoms. The band structure for CsCrI3 without SOC is plotted in Fig. 6(c). One observes that the system is a half metal: The Fermi level crosses only the spin-up bands, whereas the spin-down bands are high in energy. Focusing on the spin-up bands that are near the Fermi level, one finds that there is indeed a nodal surface in the kz = π plane, as consistent with our above argument for the case without SOC. Due to the absence of SOC, orienting the magnetic moments along different directions will not af- fect the band structure, and hence will not affect the nodal surface. This is confirmed by our DFT results: We have repeated the calculation with different magnetic mo- ment orientations (e.g., along the x-direction and the z- direction), and the obtained results are exactly the same as in Fig. 6(c). Next, we consider the case in the presence of SOC. The spin and orbital degrees of freedom are tied with each other by SOC. It is generally not possible to label the bands as spin-up/spin-down. Hence, the symmetry op- erations need to act on both orbital and spin as a whole. Nevertheless, we note that the arguments in Sec. III rely on the existence of the composite symmetry T S2z, which can still be preserved although the individual T or S2z symmetry may be broken. If so, then the previous argu- ments in Sec. III still apply and a nodal surface can be guaranteed. Explicitly, this means that: in the presence (cid:66)(cid:67)(cid:68)(cid:42)(cid:36)(cid:83)(cid:36)(cid:84)(cid:66)(cid:67)(cid:9)(cid:66)(cid:10)(cid:9)(cid:67)(cid:10)(cid:9)(cid:68)(cid:10)(cid:14)(cid:19)(cid:15)(cid:17)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:1)(cid:19)(cid:15)(cid:17)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:79)(cid:80)(cid:1)(cid:52)(cid:48)(cid:36)of SOC, a nodal surface in the kz = π plane can be pro- tected by the combined symmetry T S2z in conjunction with the absence of the PT symmetry in the system (note that PT may be preserved in certain antiferromagnet). FIG. 7: Calculated band structures for CsCrI3 in the presence of SOC, with magnetic moments along (a) the x direction and (b) the z direction. The nodal surface is preserved in (a) but not in (b). We still take CsCrI3 as an example and now include SOC in the DFT calculation. It is important to note that: (i) if the magnetic moments are aligned in a direction perpendicular to the z direction, then the combined T S2z symmetry is preserved (since the spins aligning in the x-y plane are flipped by both T and S2z, and therefore are intact by T S2z); and (ii) for other directions (e.g., along the z direction), the T S2z symmetry is broken. The PT symmetry is broken for both cases. Then according to our previous analysis, the nodal surface in the kz = π plane should be preserved in case (i) but not in case (ii). This is indeed confirmed by our DFT results. Figure 7(a) shows that result with the moments aligned along the x- direction, where a nodal surface is observed in the kz = π plane. Figure 7(b) shows that result when the moments are aligned along the z-direction, and one can see that the nodal surface is destroyed. V. DISCUSSION AND CONCLUSION We have discussed two classes of nodal surfaces in the absence of SOC. They are essentially different, namely the class-I surfaces are topologically charged, and two patches on the surfaces related by P have oppo- site charges to satisfy the Nielsen-Ninomiya no-go the- 8 orem [27, 28]; however, the class-II ones are mainly orig- inated from the topological features of screw symmetry, and therefore each degenerate point belongs to a Mobius- strip band structure, all of which spread out as a torus (a kx-ky sub-BZ). The class-I surfaces are not essential, requiring band inversions in part of the BZ; whereas the class-II surfaces are essential and guaranteed to appear solely by the symmetry. In addition, the shape and loca- tion of the class-I surfaces can vary in the BZ, and can be tuned, e.g. by lattice strain; whereas the the location of the class-II surface is fixed at the BZ boundary plane as a result of the combination of time reversal and screw rotation. The nodal surfaces in the examples discussed in Sec. III and IV are generalizations of the class-II sur- faces, so they share similar characteristic features of the class-II surfaces listed above. We mention that both classes do not lead to any spe- cial boundary gapless modes. Although the first class has the nontrivial Z2 topological charge, similar to Weyl semimetals, the codimensionality of nodal surfaces or the dimensionality of sub topological insulators is zero, im- possible to generate any boundary state. It may be note- worthy that each nodal surface considered here is a torus in the BZ, which cannot be continuously deformed to be a point, and hence should be distinguished from a nodal surface that is topologically a sphere deformable to a sin- gle point in the BZ [26]. (Note that for class-I surfaces, a pair of surfaces may merge into a single sphere when the system is strongly deformed.) We also mention that although the 0D Z2 charge does not produce any topo- logical surface state, it is possible that additional 1D or 2D topological charges (having codimensions 1 and 2) may exist and lead to nontrivial surface states in these systems. As we have mentioned, the nodal surfaces are distinct from the usual Fermi surfaces because it is formed by the crossing of two bands. As a result, the low-energy elec- trons acquire an intrinsic pseudo-spin degree of freedom, different from the usual Schrodinger fermions. Near a rel- atively flat nodal surface, the low-energy effective model may be written as Heff = vzqzσz, (5) where vz is the Fermi velocity, qz measures the deviation from the nodal surface along the surface normal direc- tion, and the Pauli matrix σz represents the pseudo-spin, corresponding to the two crossing bands. Equation (5) effectively describes 1D massless Dirac fermions, where the pseudospin is locked to the momentum. For transport along the z-direction, we may expect interesting proper- ties such as the enhanced mobility due to the suppression of back-scattering (for states around a single nodal sur- face) from scatterers that preserve the pseudospin. Regarding the class-II surfaces, if there are multiple two-fold screw axis, then it is possible to have multiple nodal surfaces on the BZ boundary planes. One exam- ple is shown in Fig. 8 for the material Cu3Se2 known as the mineral umangite, with the space group of P 421m (cid:9)(cid:66)(cid:10)(cid:9)(cid:67)(cid:10)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:52)(cid:48)(cid:36)(cid:13)(cid:1)(cid:46)(cid:1)(cid:16)(cid:16)(cid:1)(cid:89)(cid:52)(cid:48)(cid:36)(cid:13)(cid:1)(cid:46)(cid:1)(cid:16)(cid:16)(cid:1)(cid:91)9 candidate materials that satisfy the condition. However, for some cases, like the case in the presence of SOC, the examples we show are not ideal. For example, Ta3TeI7 discussed in Sec. III is actually a semiconductor. Nev- ertheless, it serves the purpose to illustrate that a nodal surface can indeed appear when the proposed symmetry requirements are satisfied. Based on our theory, we ex- pect better NSSM materials to be discovered in future studies. In conclusion, we have theoretically investigated the NSSMs which host robust nodal surfaces formed by the crossing of two bands close to the Fermi level. We clar- ify the symmetry/topology-protection of the nodal sur- faces. In the absence of SOC, we identify two classes of nodal surfaces which are protected by different mecha- nisms. The class-I surfaces are protected by a combina- tion of inversion, sublattice, and time reversal symme- tries and are characterized by a Z2 index. The class-II surfaces are protected by a two-fold screw axis and the time reversal symmetry. The two classes differ in sev- eral aspects including the surface shape, location, and robustness against perturbation. The inclusion of SOC generally gaps the class-I surface, and we show that the class-II surface may be preserved provided that the inver- sion symmetry is broken. Furthermore, we generalize the analysis to magnetic materials. We find that class-II sur- faces can exist in magnetic materials both without and with SOC, given that certain magnetic group symme- try requirements are satisfied. We have identified several concrete NSSM material examples, which will facilitate the experimental exploration of the intriguing properties of NSSMs. Acknowledgments The authors thank T. Bzdusek, X. C. Wu, and D. L. Deng for valuable discussions and comments on the manuscript. This work was supported by the Singapore Ministry of Education Academic Research Fund Tier 1 (SUTD-T1-2015004) and Tier 2 (MOE2015-T2-2-144). We acknowledge computational support from the Texas Advanced Computing Center and the National Super- computing Centre Singapore. FIG. 8: (a) Crystal structure of Cu3Se2 and (b) the corre- sponding Brillouin zone. In (b), Γ1, X1, Y1, and M1 label the four k-points that are above the points Γ, X, Y, and M, and are located in the kz = π/2c plane. (c) Band structure of Cu3Se2 (with SOC) along some generic paths. The red arrows indicate the band degeneracy along the paths on the two orthogonal nodal surfaces. 2 −x, z) with x = 0.145 and z = 0.763. (No. 113) [48, 49]. As a typical intermetallic compounds, Cu atoms are bonded closely both with one another and with Se atoms. Two kinds of Cu atoms exist in the struc- ture, with Cu(I) at the 2a position (0, 0, 0) and Cu(II) at the 4e position (x, 1 The material experiences a transition to an antiferromag- netic state at about 50 K [50]. Here, we focus on its paramagnetic phase above 50 K. The band structure of Cu3Se2 is plotted in Fig. 8(c). The space group P 421m contains two perpendicular screw axis S2x = {C2x 1 2 0} and S2y = {C2y 1 2 0}. With preserved T , two mutually orthogonal nodal surfaces can be found in the kx = π and ky = π planes [as illustrated in Fig. 8(b)]. This is indeed observed in the band structure in Fig. 8(c), which shows that the non-degenerate bands along the generic paths Γ1-X1 and Γ1-Y1 cross at the two nodal surfaces. A ma- terial example with three nodal surfaces is also presented in Appendix C. 2 1 1 2 Finally, we mention that as a good NSSM material, the nodal surface should be close to the Fermi level and formed by the crossing between conduction and valence bands. This condition imposes constraints regarding the electron filling of the bands. The recent work by Watan- abe et al. [51] has studied the detailed filling constraints for realizing semimetal states for nonsymmorphic space groups, which will offer useful guidelines for searching NSSMs. In this work, we have identified several good (cid:67)(cid:68)(cid:66)(cid:36)(cid:86)(cid:52)(cid:70)(cid:57)(cid:46)(cid:1062)(cid:9385)(cid:57)(cid:9385)(cid:46)(cid:9385)(cid:1013)(cid:58)(cid:58)(cid:9385)(cid:9)(cid:66)(cid:10)(cid:9)(cid:67)(cid:10)(cid:9)(cid:68)(cid:10)(cid:14)(cid:17)(cid:15)(cid:22)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:22)(cid:1013)(cid:1062)(cid:9385)(cid:57)(cid:9385)(cid:46)(cid:9385)(cid:58)(cid:9385)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:1062)(cid:9385)Appendix A: Tight-binding model and Z2 invariant for QGN(1,2) sites, and therefore the Hamiltonian of the tight-binding model, H =(cid:82) d3k ψ As discussed in the main text, the low-energy physics mainly comes from electrons hopping among the eight sp2 † kH(k)ψk with ψk = (c1, c2,··· , c8)T , describes an eight-band theory. After a gauge transformation to make the Hamiltonian periodic in kx and ky, we have  H(k) = 2t1 cos( 1 2 kz) 0 0 0 0 0 0 t2e−ikx 2 kz) 2t1 cos( 1 t2e−iky 0 0 t2eiky 0 0 0 0 0 0 2t1 cos( 1 2 kz) 0 0 0 0 0 0 2 kz) 2t1 cos( 1 t2e−ikx 0 0 0 0 t2eikx 0 0 0 0 2t1 cos( 1 2 kz) 0 0 0 0 0 0 2t1 cos( 1 2 kz) 0 t2eiky 0 0 0 0 0 0 t2e−iky 0 2t1 cos( 1 2 kz) 2t1 cos( 1 2 kz) 0 10 . (A1) t2eikx 0 0 0 0 0 Here the wave-vectors are measured in the respective in- verse lattice constants, and t1 and t2 are the real hopping amplitudes from site 1 to site 2 and from site 2 to site 3, respectively [see Fig. 1(b)]. Performing the combined unitary transformation  UP T UΓ with UΓ = and  1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1  1 0 0 i 0 0 0 0 0 1 i 0 0 0 0 0 0 i 1 0 0 0 0 0 i 0 0 1 0 0 0 0 0 0 0 0 1 0 0 i 0 0 0 0 0 1 i 0 0 0 0 0 0 i 1 0 0 0 0 0 i 0 0 1  , UP T = 1√ 2 [i.e., H(k) = the Hamiltonian becomes anti-diagonal antidiag(Q(k), Q†(k))] with the upper right lock being  t2 sin kx 0 0 t2 cos ky 2 −t2 sin ky 2t1 cos kz t2 cos kx 2t1 cos kz 2 Q(k) = 2t1 cos kz 2 t2 sin ky t2 cos ky t2 cos kx 2t1 cos kz 2 −t2 sin kx 0 1(cos kz +1)2−t4 0 2/2t2 Then it is found that DetQ(k) = 4t4 (A4) 2. If t2/2t1 < 1, then there are two nodal surfaces with kz = ± arccos(t2 1 − 1) related by T or P symmetry. It is easy to check that DetQ(kz = 0) > 0 because t2/2t1 < 1 (t1 = 2.95 eV and t2 = 1.30 eV, as obtained from the fitting of DFT result in Fig. A1); whereas DetQ(kz = π) < 0. Thus, both nodal surfaces have the nontrivial topological charge according to the formula in Eq. (1). (A2) (A3)  . FIG. A1: Comparison of band structures of the eight-band tight binding model in Eq. A1 (red dashed lines) and the DFT result (blue solid lines). In the tight-binding model, the values t1 = 2.95 eV and t2 = 1.30 eV are obtained from fitting the low-energy DFT band structure. Appendix B: First-principles calculation method The first-principle calculations are based on the DFT, as implemented in the Vienna ab initio simulation pack- age [52, 53]. The projector augmented wave method was adopted [54]. The generalized gradient approximation (GGA) with the Perdew-Burke-Ernzerhof (PBE) realiza- tion [55] was adopted for the exchange-correlation po- tential. For all calculations, the energy and force conver- gence criteria were set to be 10−5 eV and 10−2 eV/A, respectively. The BZ sampling was performed by us- ing k grids with a spacing of 2π × 0.02 A−1 within a Γ-centered sampling scheme. As the transition metal d orbitals may have notable correlation effects, we have val- idated our DFT results by the GGA+U method [56] (see Appendix E). The key features are found to be qualita- tively the same as the GGA results. Hence in the main text, we focus on the GGA results. For K6YO4, the optimized lattice parameters (a = 9.59 A, c = 6.69 A) were used for the band structure calculation. For the (cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:1013)(cid:59)(cid:34)(cid:46)(cid:57)(cid:51)other materials discussed here, the experimental values of their respective lattice parameters were used in the calculation. are shown here). We find that the material is a semicon- ductor with a bandgap about 0.75 eV. Nevertheless, band crossings appear in both valence and conduction bands. 11 Appendix C: SOC effect on class-I surfaces Here we consider the SOC effect on the class-I nodal surfaces. Take the QGN(1,2) discussed in Sec. II A as an example. Since the carbon is a light element with very weak SOC. The band structure with SOC shows negligible difference from that in Fig. 1(d) without SOC. In order to more clearly demonstrate that SOC in fact gaps the nodal surface, we artificially enhance the SOC strength in the DFT calculation by 30 times. The ob- tained band structure is plotted in Fig. A2. One clearly observes that a gap is opened by SOC at the original band-crossing point, destroying the class-I nodal surface. FIG. A2: (a) Band structure for QGN(1,2) with SOC. The SOC strength is artificially enhanced by 30 times, in order to show that the SOC gaps the original nodal surfaces. (b) En- larged view around the original band crossing, corresponding to the region as indicated by the red arrow in (a). Appendix D: Material example with three nodal surfaces In Sec. V, we have shown that the compound Cu3Se2 with two perpendicular screw axis can host two orthog- onal nodal surfaces on the BZ boundary planes. Here we consider another example material Rb2Se5 with three perpendicular nodal surfaces on the BZ boundary planes [57]. Its crystal structure has the space group P 212121 (No. 19), which has three two-fold screw axes perpendic- 2}, ular to each other: S2x = {C2x 1 and S2z = {C2z 1 2}. The time reversal symmetry is 2 0 1 preserved due to the absence of magnetic ordering. There are four helical Se2− chains in the unit cell, and the Rb atoms are surrounded by the Se chains and related by the screw axis. 2 0}, S2y = {C2y0 1 2 1 1 2 5 The calculated band structure for Rb2Se5 with SOC included is shown in Fig. A3(c) (only some generic paths 5 chain. FIG. A3: (a) Crystal structure of Rb2Se5 and (b) the corre- sponding Brillouin zone. The insert in (a) shows the helical Se2− In (b), Γ1 is the body center of 1/8 Brillouin zone; X1, Y1, and Z1 are the face centers of the 1/8 Brillouin zone; U1, T1, and S1 are the mid-points of the paths U-R, T- R, and S-R, respectively. (c) Band structure of Rb2Se5 with SOC. By examining the band dispersion on the BZ boundary planes, one can find that the bands cross at these planes in pairs. For example, the non-degenerate bands along the three generic paths Γ1-X1, Γ1-Y1, Γ1-Z1 all cross at the BZ boundary planes. Hence there are three mutually orthogonal nodal surfaces as illustrated in Fig. A3(b). Appendix E: Hubbard U correction To test the correlation effects of the transition metal d -orbitals, we performed the GGA+U calculations, by taking into account the on-site Coulomb interaction. The U values proper for each transition metal element have been tested. Figure A4 shows the representative results for the band structures of K6YO4, RbMo3S3, TlMo3Te3, Ta3TeI7, CsCrI3, and Cu3Se2. One can observe that the results for all these compounds show little change in com- parison with the GGA results. Hence, we focus on the GGA results in the main text. (cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:1013)(cid:59)(cid:34)(cid:46)(cid:57)(cid:51)(cid:52)(cid:48)(cid:36)(cid:695)(cid:20)(cid:17)(cid:14)(cid:17)(cid:15)(cid:20)(cid:14)(cid:17)(cid:15)(cid:19)(cid:14)(cid:17)(cid:15)(cid:18)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:18)(cid:1062)(cid:59)(cid:9)(cid:66)(cid:10)(cid:9)(cid:67)(cid:10)(cid:67)(cid:68)(cid:66)(cid:51)(cid:67)(cid:52)(cid:70)(cid:58)(cid:9385)(cid:59)(cid:9385)(cid:54)(cid:9385)(cid:53)(cid:9385)(cid:57)(cid:9385)(cid:1013)(cid:52)(cid:9385)(cid:9385)(cid:1013)(cid:57)(cid:52)(cid:54)(cid:59)(cid:53)(cid:58)(cid:51)(cid:9)(cid:66)(cid:10)(cid:9)(cid:67)(cid:10)(cid:9)(cid:68)(cid:10)(cid:14)(cid:17)(cid:15)(cid:21)(cid:14)(cid:17)(cid:15)(cid:19)(cid:1)(cid:17)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:1013)(cid:59)(cid:9385)(cid:53)(cid:9385)(cid:58)(cid:9385)(cid:57)(cid:9385)(cid:52)(cid:9385)(cid:58)(cid:9385)(cid:9385)(cid:1013)(cid:9385)12 FIG. A4: Effects of the Hubbard U correction on band structures of (a) K6YO4 (without SOC) with U=4.0 eV, (b) RbMo3S3 (without SOC) with U=4.0 eV, (c) TlMo3Te3 (SOC included) with U=4.0 eV, (d) Ta3TeI7 (SOC included) with U=2.0 eV, (e) CsCrI3 (without SOC) with U=5.0 eV, and (f) Cu3Se2 (SOC included) with U=5.0 eV. [1] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 [20] C. Fang, H. Weng, X. Dai, and Z. Fang, Chinese Physics (2010). [2] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [3] A. Bansil, H. Lin, and T. Das, Rev. Mod. Phys. 88, 021004 (2016). B 25, 117106 (2016). [21] S. Li, Z.-M. Yu, Y. Liu, S. Guan, S.-S. Wang, X. Zhang, Y. Yao, and S. A. Yang, Physical Review B 96, 081106 (2017). [22] C. Zhong, Y. Chen, Y. Xie, S. A. Yang, M. L. Cohen, [4] C.-K. Chiu, J. C. Y. Teo, A. P. Schnyder, and S. Ryu, and S. Zhang, Nanoscale 8, 7232 (2016). Rev. Mod. Phys. 88, 035005 (2016). [23] Q.-F. Liang, J. Zhou, R. Yu, Z. Wang, and H. Weng, [5] A. A. Burkov, Nature Materials 15, 1145 (2016). [6] B. Yan and C. Felser, Annu. Rev. Condens. Matter Phys. Phys. Rev. B 93, 085427 (2016). [24] T. Bzdusek and M. Sigrist, Phys. Rev. B 96, 155105 8, 337 (2017), ISSN 1947-5454. (2017). [7] N. Armitage, E. Mele, and A. Vishwanath, Reviews of Modern Physics 90, 015001 (2018). [25] M. Xiao and S. Fan, arXiv:1709.02363. [26] O. Turker and S. Moroz, Physical Review B 97, 075120 [8] X. Wan, A. M. Turner, A. Vishwanath, and S. Y. (2018). Savrasov, Phys. Rev. B 83, 205101 (2011). [27] H. Nielsen and M. Ninomiya, Nuclear Physics B 185, 20 [9] S. Murakami, New Journal of Physics 9, 356 (2007). (1981), ISSN 0550-3213. [10] S. M. Young, S. Zaheer, J. C. Y. Teo, C. L. Kane, E. J. Mele, and A. M. Rappe, Phys. Rev. Lett. 108, 140405 (2012). [28] H. Nielsen and M. Ninomiya, Nuclear Physics B 193, 173 (1981), ISSN 0550-3213. [29] Y. X. Zhao and A. P. Schnyder, Phys. Rev. B 94, 195109 [11] H. Nielsen and M. Ninomiya, Physics Letters B 130, 389 (2016). (1983), ISSN 0370-2693. [30] Y. X. Zhao, A. P. Schnyder, and Z. D. Wang, Phys. Rev. [12] G. E. Volovik, The Universe in a Helium Droplet Lett. 116, 156402 (2016). (Clarendon Press, Oxford, 2003). [31] A. Altland and M. R. Zirnbauer, Phys. Rev. B 55, 1142 [13] S. Guan, Z.-M. Yu, Y. Liu, G.-B. Liu, L. Dong, Y. Lu, Y. Yao, and S. A. Yang, npj Quantum Materials 2, 23 (2017), ISSN 2397-4648. (1997). [32] A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Lud- wig, Phys. Rev. B 78, 195125 (2008). [14] H. Weng, Y. Liang, Q. Xu, R. Yu, Z. Fang, X. Dai, and [33] Y. X. Zhao and Z. D. Wang, Phys. Rev. Lett. 110, 240404 Y. Kawazoe, Phys. Rev. B 92, 045108 (2015). (2013). [15] S. A. Yang, H. Pan, and F. Zhang, Phys. Rev. Lett. 113, [34] D. F. Agterberg, P. M. R. Brydon, and C. Timm, Phys. 046401 (2014). [16] K. Mullen, B. Uchoa, and D. T. Glatzhofer, Phys. Rev. Lett. 115, 026403 (2015). Rev. Lett. 118, 127001 (2017). [35] L. Michel and J. Zak, Phys. Rev. B 59, 5998 (1999). [36] S. M. Young and C. L. Kane, Phys. Rev. Lett. 115, [17] R. Yu, H. Weng, Z. Fang, X. Dai, and X. Hu, Phys. Rev. 126803 (2015). Lett. 115, 036807 (2015). [18] Y. Kim, B. J. Wieder, C. L. Kane, and A. M. Rappe, Phys. Rev. Lett. 115, 036806 (2015). [19] Y. Chen, Y. Xie, S. A. Yang, H. Pan, F. Zhang, M. L. [37] A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek, S. Cholia, D. Gunter, D. Skinner, G. Ceder, et al., APL Materials 1, 011002 (2013), ISSN 2166532X. [38] W. Honle, H. Von Schnering, A. Lipka, and K. Yvon, Cohen, and S. Zhang, Nano Lett. 15, 6974 (2015). Journal of the Less Common Metals 71, 135 (1980). (cid:9)(cid:69)(cid:10)(cid:9)(cid:70)(cid:10)(cid:9)(cid:68)(cid:10)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:53)(cid:77)(cid:46)(cid:80)(cid:9387)(cid:53)(cid:70)(cid:9387)(cid:13)(cid:1)(cid:54)(cid:1)(cid:30)(cid:1)(cid:21)(cid:15)(cid:17)(cid:1)(cid:70)(cid:55)(cid:13)(cid:1)(cid:52)(cid:48)(cid:36)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:9)(cid:67)(cid:10)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:51)(cid:67)(cid:46)(cid:80)(cid:9387)(cid:52)(cid:9387)(cid:13)(cid:1)(cid:54)(cid:1)(cid:30)(cid:1)(cid:21)(cid:15)(cid:17)(cid:1)(cid:70)(cid:55)(cid:13)(cid:1)(cid:79)(cid:80)(cid:1)(cid:52)(cid:48)(cid:36)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:9)(cid:66)(cid:10)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:44)(cid:9390)(cid:58)(cid:48)(cid:9388)(cid:13)(cid:1)(cid:54)(cid:1)(cid:30)(cid:1)(cid:21)(cid:15)(cid:17)(cid:1)(cid:70)(cid:55)(cid:13)(cid:1)(cid:79)(cid:80)(cid:1)(cid:52)(cid:48)(cid:36)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:1062)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:53)(cid:66)(cid:9387)(cid:53)(cid:70)(cid:42)(cid:9391)(cid:13)(cid:1)(cid:54)(cid:1)(cid:30)(cid:1)(cid:19)(cid:15)(cid:17)(cid:1)(cid:70)(cid:55)(cid:13)(cid:1)(cid:52)(cid:48)(cid:36)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:1013)(cid:34)(cid:41)(cid:45)(cid:46)(cid:38)(cid:51)(cid:41)(cid:44)(cid:36)(cid:84)(cid:36)(cid:83)(cid:42)(cid:9387)(cid:13)(cid:1)(cid:54)(cid:1)(cid:30)(cid:1)(cid:22)(cid:15)(cid:17)(cid:1)(cid:70)(cid:55)(cid:13)(cid:1)(cid:79)(cid:80)(cid:1)(cid:52)(cid:48)(cid:36)(cid:14)(cid:19)(cid:15)(cid:17)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:1)(cid:19)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:1013)(cid:1062)(cid:9385)(cid:57)(cid:9385)(cid:46)(cid:9385)(cid:58)(cid:9385)(cid:1062)(cid:9385)(cid:14)(cid:18)(cid:15)(cid:17)(cid:1)(cid:17)(cid:15)(cid:17)(cid:1)(cid:18)(cid:15)(cid:17)(cid:38)(cid:79)(cid:70)(cid:83)(cid:72)(cid:90)(cid:1)(cid:9)(cid:70)(cid:55)(cid:10)(cid:36)(cid:86)(cid:9387)(cid:52)(cid:70)(cid:9386)(cid:13)(cid:1)(cid:54)(cid:1)(cid:30)(cid:1)(cid:22)(cid:15)(cid:17)(cid:1)(cid:70)(cid:55)(cid:13)(cid:1)(cid:52)(cid:48)(cid:36)(cid:9)(cid:71)(cid:10)13 [39] J. Huster, G. Schippers, and W. Bronger, Journal of the Less Common Metals 91, 333 (1983). [40] Q. Liu and A. Zunger, Phys. Rev. X 7, 021019 (2017). [41] M. Potel, R. Chevrel, M. Sergent, J. Armici, M. Decroux, and . Fischer, Journal of Solid State Chemistry 35, 286 (1980), ISSN 0022-4596. [42] R. Chevrel and M. Sergent, Crystal Chemistry and Prop- erties of Materials with Quasi-One-Dimensional Struc- tures: A Chemical and Physical Synthetic Approach (Springer Netherlands, Dordrecht, 1986). [43] M. D. Smith and G. J. Miller, Journal of the American Crystallographica Section B: Structural Crystallography and Crystal Chemistry 35, 1554 (1979). [48] N. Morimoto and K. Koto, Science 152, 345 (1966). [49] D. Chakrabarti and D. Laughlin, Bulletin of Alloy Phase Diagrams 2, 305 (1981). [50] K. Okamoto, S. Kawai, and R. Kiriyama, Japanese Jour- nal of Applied Physics 8, 718 (1969). [51] H. Watanabe, H. C. Po, M. P. Zaletel, and A. Vish- wanath, Phys. Rev. Lett. 117, 096404 (2016). [52] G. Kresse and J. Hafner, Phys. Rev. B 47, 558 (1993). [53] G. Kresse and J. Furthmuller, Phys. Rev. B 54, 11169 Chemical Society 118, 12238 (1996). (1996). [44] M. Smith and G. J. Miller, Journal of Solid State Chem- istry 140, 226 (1998). [45] Z. Wang, M. G. Vergniory, S. Kushwaha, M. Hirschberger, E. V. Chulkov, A. Ernst, N. P. Ong, R. J. Cava, and B. A. Bernevig, Phys. Rev. Lett. 117, 236401 (2016). [46] G. Chang, S.-Y. Xu, X. Zhou, S.-M. Huang, B. Singh, B. Wang, I. Belopolski, J. Yin, S. Zhang, A. Bansil, et al., Phys. Rev. Lett. 119, 156401 (2017). [47] L. Guen, R. Marchand, N. Jouini, and A. Verbaere, Acta [54] P. E. Blochl, Phys. Rev. B 50, 17953 (1994). [55] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). [56] S. L. Dudarev, G. A. Botton, S. Y. Savrasov, C. J. Humphreys, and A. P. Sutton, Phys. Rev. B 57, 1505 (1998). [57] P. Boettcher, Zeitschrift fuer Kristallographie (149,1979- ) 150, 65 (1979).
1104.3325	7	1104	2011-08-19T18:40:17	Numerical studies of the interaction of an atomic sample with the electromagnetic field in two dimensions	[ "cond-mat.mes-hall", "quant-ph" ]	We consider the interaction of electromagnetic radiation of arbitrary polarization with multi-level atoms in a self-consistent manner, taking into account both spatial and temporal dependencies of local fields. This is done by numerically solving the corresponding system of coupled Maxwell-Liouville equations for various geometries. In particular, we scrutinize linear optical properties of nanoscale atomic clusters, demonstrating the significant role played by collective effects and dephasing. It is shown that subwavelength atomic clusters exhibit two resonant modes, one of which is localized slightly below the atomic transition frequency of an individual atom, while the other is positioned considerably above it. As an initial exploration of future applications of this approach, the optical response of core-shell nanostructures, with a core consisting of silver and shell composed of resonant atoms, is examined.	cond-mat.mes-hall	cond-mat	Numerical studies of the interaction of an atomic sample with the electromagnetic field in two dimensions Maxim Sukharev1, ∗ and Abraham Nitzan2, † 1Department of Applied Sciences and Mathematics, Arizona State University, Mesa, Arizona 85212, USA 2School of Chemistry, Tel Aviv University, Tel Aviv, 69978, Israel (Dated: November 7, 2018) We consider the interaction of electromagnetic radiation of arbitrary polarization with multi-level atoms in a self-consistent manner, taking into account both spatial and temporal dependencies of local fields. This is done by numerically solving the corresponding system of coupled Maxwell-Liouville equations for various geometries. In particular, we scrutinize linear optical properties of nanoscale atomic clusters, demonstrating the significant role played by collective effects and dephasing. It is shown that subwavelength atomic clusters exhibit two resonant modes, one of which is localized slightly below the atomic transition frequency of an individual atom, while the other is positioned considerably above it. As an initial exploration of future applications of this approach, the optical response of core-shell nanostructures, with a core consisting of silver and shell composed of resonant atoms, is examined. PACS numbers: 42.50.Ct, 78.67.-n I. INTRODUCTION Nanoscale optical materials have long been attracting considerable attention due to many important applications ranging from optical nanodevices [1], plasmonic circuitry [2], nanoscale sources of coherent radiation [3], single atom/molecule manipulation [4], bio- medical applications [5, 6], and many others. Among such exciting applications lies the ∗Electronic address: [email protected] †Electronic address: [email protected] 2 yet-to-be explored sub-field of nanoscale optical atomic and molecular physics that deals with ensembles of atoms or molecules interacting with dielectric nanoparticles (NP) and their assemblies. Such systems are characterized by a significant spatial dependence of evanescent electromagnetic (EM) fields on the dielectric environment, providing the means to control the behavior of molecular systems by the combination of large EM fields and large field gradients, exemplified by recent work [7 -- 12] on the EM field associated with metal NP dimers and their dependence on particle sizes and interparticle distance. At their core, these phenomena rely mostly on the excitations of surface plasmon-polariton (SPP) resonances [13] in systems comprising metal NPs and their arrays, [14] as well as other nanoscale metal surfaces, such as subwavelength diffraction gratings, whose optical properties depend sen- sitively on their surface topology and material parameters [15]. Their studies have lead to many applications such as coherent EM energy transport in space [16], surface enhanced Raman spectroscopy (SERS) [17] and tip-enhanced microscopy [18]. Recent attention has focused on optical control scenarios, ranging from coupled exciton-plasmon dynamics in semiconductor nanodots [19 -- 27] and in molecular aggregates [28 -- 36] where metal NPs af- fect excitation energy transfer between molecules, to optical trapping of single atoms or molecules [37 -- 40]. Such applications are facilitated by the possibility to control the geom- etry of nano-materials (NPs sizes, their relative arrangement, etc.) with an outstanding precision [41]. While theoretical and computational methodologies for studying these phenomena have advanced considerably, the consequences of mutual feedback between molecular excitations and metallic SPP resonances are not well understood, especially when one probes systems comprising both metallic nanostructures and semiconductor or molecular particles or lay- ers. An often used simple description is based on assigning a dielectric response function to the semiconductor/molecular component and solve the electromagnetic problem for the corresponding composite dielectric. While such an approach can be useful for describing the effect of the molecular environment on the metallic plasmon, it cannot be used to de- scribe energy-transfer, relaxation, and spontaneous emission in the excitonic (molecular or semiconductor) system. It is hence important to develop a self-consistent description of the electromagnetic response of such systems. Such approach has to take into account the electrodynamics of the radiation field and the quantum dynamics of the molecular system in a self-consistent manner. This can be accomplished by solving simultaneously Maxwell(cid:48)s 3 equations for the radiation field and the Liouville equation for the molecular density matrix, including the molecular polarization current in the former and the molecules-field interaction in the latter. First attempts to consider numerically coupled Maxwell-Bloch equations have been ini- tiated by Ziolkowski et al. [42, 43] for simple two-level atoms in one and two dimensions utilizing finite-difference time-domain (FDTD) technique. Later on this approach has been extended to three dimensions [44]. Although these works contain interesting and important physics, they are limited to ensembles of two-level systems. Consideration of multilevel systems is critical for modeling of nano-lasing, which has to include at least three lev- els. Moreover, the proposed numerical implementation results in noticeably, long execution times. The scheme we propose is more efficient as discussed below. Similarly Neuhauser et al. proposed another approach [45] where the authors coupled Maxwell(cid:48)s equations to the Schrodinger equation describing a molecule located in the closed proximity of a metal NP. These works, however, cannot in their present form include relaxation and dephasing effects, which, as we demonstrate, are very important. In this paper we describe a numerical implementation of such a model, using the method- ology developed by Ziolkowski et al. [42, 43] as our starting point. This model captures collective effects that play pivotal role in electrodynamics of nano-systems, as well as the counterbalancing effect of dephasing processes. The questions to be addressed are: 1. How does a size of the system affect scattering/absorption of EM radiation? 2. What is a role of dephasing and relaxation effects? 3. When does one observe collective response of atoms to external EM excitation? These and other closely related questions are not only important from the fundamental point of view (how optically induced interatomic or intermolecular interactions depend on structural/material parameters), but also essential for general understanding of optics of many-body systems. In this regard it should be pointed out that although the technique we propose in this paper is utilized to capture collective effects of quantum particles in the linear response regime, it can easily be applied to nonlinear systems. The paper is organized as follows: Section II discusses our computational approach, based on coupled Maxwell-Liouville equations in the mean-filed approximation. In Section III we 4 provide details of the numerical implementation; Section IV describes and discusses the results of our numerical studies. Our main conclusions along with future research outlook are presented in Section V. II. MODEL We consider a general problem of a system of quantum particles (further referred to as atoms) interacting with EM radiation. We start from the time domain Maxwell(cid:48)s equations for the dynamics of the EM fields, (cid:126)E and (cid:126)H µ0 ∂ (cid:126)H ∂t = −∇ × (cid:126)E, ∂ (cid:126)E ∂t ε0 = ∇ × (cid:126)H − (cid:126)J, (1a) (1b) where µ0 and ε0 are magnetic permeability and dielectric permittivity of free space, respec- tively. In spatial regions occupied by a metal nanostructure (such as metal NP, for instance) the equation (1b) is evaluated in the standard way from the metal dielectric dispersion [46]. In the present study the dispersion of dielectric constant of metal, ε(ω), is taken in the form of the Drude model ε(ω) = εr − ω2 p ω2 − iγω (2) with numerical parameters describing silver for the wavelengths of interest εr = 8.26, ωp = 1.76 × 1016 rad/sec, γ = 3.08 × 1014 rad/sec. The time evolution of the current density (cid:126)J in metal regions is then ∂ (cid:126)J ∂t = a (cid:126)J + b (cid:126)E, (3) where a = −γ and b = ε0ω2 p. In the spatial regions occupied by atoms, the mutual interaction between the atomic system and the EM field is accounted for in a self-consistent manner as follows: first, the current density in Eq. (1b) is expressed in terms of the macroscopic polarization of the atomic system, (cid:126)P ((cid:126)r, t) The latter is given by (cid:126)J = ∂ (cid:126)P ∂t . (cid:126)P = na(cid:104)(cid:126)µ(cid:105), (4) (5) where (cid:104)(cid:126)µ(cid:105) = T r(ρ(cid:126)µ) 5 (6) is the expectation value of the atomic dipole moment and na is the atomic density. Eqs. (4) - (6) constitutes the main approximation of the present approach, whereupon the local polarizability is expressed in terms of the local atomic density multiplied by the local aver- aged single atomic dipole. The time evolution of the latter is obtained from the evolution of the single atom density matrix (described below) in the presence of the EM field, thus providing a self-consistent description of the field-matter dynamics. Next consider the atomic subsystem. While our ultimate goal is to study realistic 3- dimensional systems, the present study focuses on nanoscale atomic clusters in two dimen- sions, taken to lie in the XY plane [48]. The incident radiation field is represented by a transverse-electric (TE) mode with respect to z-axis. It is characterized by two in-plane electric field components, Ex and Ey, and one out-of-plane magnetic field component, Hz. To account for the (two-dimensional) spherical symmetry of the atomic polarization response, the atoms are described as 3-level systems: an s-type ground state and two degenerate p-type excited states of px and py character (as depicted in Fig. 1A). In anticipation of pos- sible generalizations to more complex models involving multilevel systems we use, in what follows, a basis of angular momentum wavefunctions with quantization axis in the z direc- √ tion, 1(cid:105) = s(cid:105), 2(cid:105) = (px(cid:105) + ipy(cid:105)) / 2, with optical transitions corresponding to ∆J = ±1 and ∆M = ±1 selection rules. The corresponding Hamiltonian is √ 2, 3(cid:105) = (px(cid:105) − ipy(cid:105)) /  0 Ω+ (t) −Ω− (t)  , Ω− (t) −Ω+ (t) ωa 0 0 ωa H = H0 − (cid:126)µ · (cid:126)E(t) = (7) √ where ωa is the atomic energy transition, Ω± = µsp (Ex (t) ± iEy (t)) / 6, and µsp is s − p the matrix element of the dipole moment operator, that, in principle, can be taken from experiment or calculated using standard quantum chemistry packages. Its Cartesian com- ponents are µx = − ∂ H ∂Ex = µsp√ 6 µy = − ∂ H ∂Ey = µsp√ 6  ,  . 6 (8a) (8b) −1 0 0 0 0 1  0 −1 1  0 i i −i 0 0 −i 0 0 Note that the dipole moment operator (8) differs from the one used in [43] by the factor of √ 3. The mean field approximation Eq. (5) make it possible to describe the atomic system in terms of the single atom density matrix ρ, which satisfies the Liouville equation i dρ dt = [ H, ρ] − iΓρ. (9) Eq. (9) describes the time evolution of an atom interacting with the radiation field and subject to (assumed Markovian) relaxation processes described by the Γ operator, which is taken in the Lindblad form [47]. The diagonal elements of this operator correspond to excited states lifetimes, while nondiagonal elements account for dephasing effects. Eqs. (1) - (9) describe the time evolution of the atomic system and radiation field in a self-consistent way. Note that in the single-atom Hamiltonian H given by Eq. (7) interatomic interactions are absent. Still, the dynamics described by Eqs. (1) - (9) is not that of independent atoms: the self-consistent scheme accounts for all interactions between atoms that are associated with their mutual interaction with the radiation field, including excitonic (energy transfer) interactions, which are all important for elucidating the overall system response. Note that, assuming that the EM field does not significantly vary within a volume occupied by a single atom, the spatial dependence of the density matrix in Eq. (9) depends parametrically on position via the EM field variables. Eqs. (9)-(8) lead to the following equations for the atomic density matrix elements dρ11 dt = iω+ (ρ12 + ρ∗ 13) − iω− (ρ13 + ρ∗ 12) + γ1 (ρ22 + ρ33) , dρ12 dt = iωaρ12 − iω− (ρ22 − ρ11) + iω+ρ∗ 23 − γ2ρ12, (10a) (10b) dρ13 dt = iωaρ13 + iω+ (ρ33 − ρ11) − iω−ρ23 − γ2ρ13, dρ22 dt dρ23 dt dρ33 dt = iω−ρ∗ = −iω+ (ρ13 + ρ∗ = iω−ρ13 − iω+ρ∗ 12 − iω+ρ12 − γ1ρ22, 12) − 2γ2ρ23, 13 − γ1ρ33, 7 (10c) (10d) (10e) (10f) where ω± = Ω±/, γ2 = γp + γ1/2 with γp denoting the pure dephasing rate due to envi- ronmentally induced random fluctuations in the atomic energy spacing. As noted above, in Eq. (10) we denoted the ground state as 1(cid:105) and the excited states J = 1, M = −1(cid:105) and J = 1, M = 1(cid:105) as 2(cid:105) and 3(cid:105), respectively. Finally, using Eqs. (8) and (10) we obtain the macroscopic polarization current (time derivative of Eq. (5)), which enters Ampere's law (1b) ∂ (cid:126)P ∂t = na ∂(cid:104)(cid:126)µ(cid:105) ∂t , where, back in cartesian coordinates, = ∂(cid:104)µx(cid:105) ∂t ∂(cid:104)µy(cid:105) ∂t Eyµ2 sp 3 (ρ22 − ρ33) − i 3 (ρ22 − ρ33) + sp = −Exµ2 [(ωa + iγ2) (ρ12 − ρ13) − (ωa − iγ2) (ρ∗ 12 − ρ∗ 13)] , µsp√ 6 [(ωa + iγ2) (ρ12 + ρ13) + (ωa − iγ2) (ρ∗ 12 + ρ∗ 13)] , µsp√ 6 (11a) (11b) (11c) We end this section with two comments. First, as already pointed out, Eqs. (10) a mean field description of a system of atoms interacting with the EM field. In this approximation a single atom interacts with other atoms through the electromagnetic field associated with their mean local density. Obviously, such an approach cannot account for specific atom-atom correlations, but it can describe collective effects in a system of atoms resulting from their interaction with the EM field. Second, we note that this procedure can be easily generalized to yield the analogous 3-dimensional coupled Maxwell-Liouville equations. To maintain spherical symmetry we would need to include an additional atomic level J = 1, M = 0(cid:105), which is coupled to the ground atomic state by Ez [44]. Obviously, it is also possible to expand the atomic basis and consider additional excited manifold starting with J = 2, M(cid:105). Although the number of equations similar to Eqs. (10) grows significantly, modern analytical computer packages such as Mathematica [49] can easily handle necessary algebra and subsequent computer 8 coding. For a molecule without rotational symmetry, average over angular distribution in the calculation of (cid:126)P from Eq. (5) may be used when relevant. III. NUMERICAL APPROACH To solve the system (1), (10), (11) of coupled Maxwell-Liouville equations we employ a generalized FDTD technique [50]. Within the FDTD, both the electric, (cid:126)E, and magnetic, (cid:126)H, fields are propagated in time and space by directly discretizing Maxwell(cid:48)s equations (1). This approach has several attractive technical features, including its numerical stability and the explicit description of the magnetic field. The latter is especially important if one considers structures with sharp corners, at which the tangential components of (cid:126)H have singularities. For the our purposes, the main advantage of the FDTD approach is that the boundary conditions (i.e. the continuity of the tangential (cid:126)H and (cid:126)E components) are automatically maintained at all grid points owing to the use of the Yee cell [51]. This allows straightforward programming of the complex geometries. For simulations of open systems, one needs to impose artificial absorbing boundaries in order to avoid numerical reflection of outgoing EM waves back to the simulation domain. Among the various approaches that address this numerical issue, the perfectly matched layers (PML) technique is considered to be the most powerful [52]. It reduces the reflection coefficient of outgoing waves at the simulation region boundary to ∼ 10−8. In essence, the PML approach surrounds the simulation domain by thin layers of non-physical material that efficiently absorbs outgoing waves incident at any angle. We have implemented the most efficient and least memory variant of the method, the convolution perfectly matched layers (CPML) absorbing boundaries [53]. Through extensive numerical experimentation, we have empirically determined optimal parameters for the CPML boundaries that lead to almost no reflection of the outgoing EM waves at all incident angles. In the calculations reported below we consider structures with a characteristic size much smaller than the incident wavelength. Hence it is a good approximation to excite such systems using a plane wave. The latter is accomplished via implementation of total field/scattered field approach [50] within the FDTD. We partition the FDTD scheme onto an array of parallel grid slices by dividing the cubic simulation cell into M xy slices, where M is the number of available processors. Point- 9 to-point message passing interface (MPI) communication subroutines [54] are implemented at the boundaries between slices. The number of xy planes in each slice usually varies in the range from 15 to 20. All simulations are performed on the home-build 128-core AMD Opteron based cluster at Arizona State University [55]. The numerical implementation of the proposed scheme is as follows 1. In the spatial regions occupied by atoms the Maxwell equations are solved utilizing the standard FDTD algorithm. First, magnetic field is updated according to Faraday(cid:48)s law, Eq. (1a). Next, using Ampere(cid:48)s law, Eq. (1b), we update the electric field with the macroscopic polarization current density, Eqs. (11), which is calculated using the density matrix of the previous time step. The EM fields in the regions occupied by metal are updated according to the auxiliary differential equation method, [50] Eq. (3). 2. With the knowledge of local electric field components (stored in memory at two pre- vious time steps) we update the density matrix at each spatial point on the grid according to Eq. (10) using the fourth order Runge-Kutta scheme [56]. 3. Finally, with the knowledge of the electric field components and the updated den- sity matrix we calculate the macroscopic polarization current, ∂ (cid:126)P ∂t , at each grid point according to (11). We have verified this scheme using several test cases. A most important test of numerical stability is to check that the condition Tr(ρ) = 1 is maintained at each time step. In all simulations this condition was perfectly satisfied with almost no dependence on incident field amplitude and other physical parameters. Another interesting test was to demonstrate the absence of self-interactions in our calculation. Such interactions often appear spuriously in mean field calculations, whereupon a particle interacts with its own contribution to the mean density. In the present situation, however, the field produced by the oscillating dipole of a given atom propagates away from this atom and can affect it only through the polarization induced in other atoms or (in different setups) through reflection from the boundaries, both physically valid phenomena. We have verified that direct self-interaction is indeed absent in our calculation by solving Eqs. (1) and (10) for the case where the system occupies a single grid point: the same solution is obtained whether or not the polarization source term is included in Eq. (1b). Finally, we have compared our results and execution time of the proposed integration scheme with those obtained by Ziolkowski et al. [43]. We have implemented the numerical approach based on the predict-corrector method and atomic 10 basis as used in [43] and compared it with ours (keeping in mind that µsp has to be re- normalized by the factor of 3). The simulation data obtained using both approaches were √ in excellent agreement. However execution times for the codes employing our approach were noticeably smaller. With the solution of Eqs. (1), (10), (11) obtained in this way, the following observables can be calculated: (1) The scattered radiation can be computed as the difference between the total and the incident EM fields. At any detection point, e.g. that depicted as a red diamond in Fig. 1B, we can calculate the Poynting vector components associated with the scattered EM field. This may be integrated over a spherical boundary surrounding the atomic system to yield the total scattered radiation. These calculations can be accomplished in a transient mode to give the time dependent response to an incident EM pulse, or in a steady state mode that yields the long time steady state response to CW incident radiation of a given frequency, ωin. In the latter case we need to propagate the Maxwell-Liouville equations under the incident CW radiation until steady-state is reached. (2) Generally, for a given incident frequency ωin the outgoing field may exhibit different frequencies ωout (with amplitudes obtained by Fourier-transforming the scattered sig- nal), making it possible to obtain the outgoing steady state flux (Poynting vector) in a given direction at any ωout for a given incident frequency. Integrating the outgoing flux over ωout and displaying the result as a function of ωin yields the absorption spectrum of the atomic cluster. (3) A much easier way to calculate the steady state absorption lineshape is by calculating the steady state relaxation flux of the excited state populations according to (cid:90) − dE dt ≈ ωaγ1 d3r (ρ22 ((cid:126)r) + ρ33 ((cid:126)r)) , (12) where the integral is taken over cluster volume [57]. Eq. (12) expresses for any given incoming frequency the rate of energy dissipation by the molecular system, which at 11 steady state should be equal to the rate of energy absorption at that frequency. (4) For some applications the short pulse method (SPM) [46] can save a substantial comput- ing effort. In this approach we use an ultra-short incident pulse with a wide bandwidth, which is almost flat in the spectral region of interest (in our simulations the incident pulse duration was set at 0.36 fs, which corresponds to flat spectrum throughout the frequency domain considered in this paper). Such pulse can be represented as a co- herent linear superposition of CW plane waves with different ωin. We then propagate Maxwell-Liouville equations for several picoseconds (the total propagation time has to be significantly longer than the lifetime of excited states of an atom, 1/γ1), and take the Fourier transform of the calculated field. Under conditions of linear response and elastic scattering (namely, when ωout = ωin), the Fourier component at frequency ω contains all the information relevant to a CW process at frequency ω. When applicable, the SPM can save substantially on computation time, since it yields the system response at many frequencies from one short time computation. It is important to understand its shortcomings. Two limitations, mentioned above, are obvious: this method is applicable only in linear response and only when the light scattering process is elastic. Both limitations are associated with the requirement that a given incident frequency can give rise to response only at the same frequency. A third limitation is important in the present context, because Eqs. (10b) and (10c) are inherently nonlinear. Linearity is obtained when ρ22 and ρ33 may be neglected, and ρ11 taken constant in these equations. In this case, the steady state solutions for ρ12 and ρ13, and therefore the polarization (Eq. (5)), oscillate with the incident frequency as required. Care has to be exercised if one attempts to calculate the steady state excitation, ρ22 + ρ33, in this method. An approximation may be obtained if the incident frequency is close to atomic transition frequency, ωa, by taking the Fourier transform at ω = 0, i.e. the time average, of the resulting time dependent signal. The method, as described, cannot describe spontaneous emission, i.e. fluorescence, since the latter is a quantum effect associated with the quantum nature of the radiation field. It has been demonstrated [58] that the effect of spontaneous emission can be accounted for partially by imposing a classical stochastic field on the system. Obviously, a classical EM noise cannot mimic vacuum fluctuations; in particular it can induce excitation of ground state molecules while vacuum fluctuations can lead only to radiative damping of excited 12 molecules. One can use this trick to study time evolution in a system which is initially inverted, that is, all molecules in the excited state. In this case, the induction of emission by the EM noise will soon lead to a dominant signal of induced emission that does not depend much on the nature of that noise, provided the latter is weak enough. This method has been succesfully used to simulate superradiance[58, 59] and gain [60] within the FDTD approach. However, this method cannot be used to generate fluorescence in a system of mostly ground state molecules (ρ22 + ρ33 (cid:28) ρ11), where the main effect of such noise will be to induce unphysical molecular excitation. IV. RESULTS AND DISCUSSION The simulation setup is shown in Fig. 1B: an atomic cluster is excited with an x-polarized low intense plane wave propagating in the y direction. We use low intensity incoming fields (our incident electric field was fixed at 1 V/m) in order to insure linearity of the system response [61]. Eqs. (1), (10), (11) are then evolved to yield the electric and magnetic fields as functions of time and position. In the current studies we focus on the y -component of the Poynting vector, Sy ∼ ExHz, as the observable of interest. Fig. 2 presents direct comparison of two methods: SPM approach and CW scheme. The scattering signals obtained from the two methods are in excellent agreement. Also shown is the absorption spectrum from the CW calculation using Eq. (12). The absorption lineshape (normalized so that it matches the scattering signals at the peak) also leads to nearly identical lineshape, except that it exhibits a more pronounced resonance near the atomic transition frequency. It is not surprising, but still providing a consistency check, that at the very low densities (na < 1024 m−3) and zero dephasing our simulations are in the perfect agreement with the Clausius-Mossotti approximation described above. Moreover, calculations, in which the polarization current term in (1b) was neglected (i.e. the atoms are not coupled to each other through their mutual interaction with the radiation field but rather driven only by external incident radiation), were in a perfect agreement with the data produced by the full self-consistent computations in the limit of low densities. Fig. 3 summarizes main results of our SPM calculations (note that we performed direct comparison of SPM data with that obtained via CW scheme for every set of parameters 13 discussed below). First the dependence of the scattering intensity on the density of atoms in the cluster is depicted in Fig. 3A. The atomic transition frequency is fixed at ωa = 3.1 eV (see figure caption for the rest of simulation parameters). The scattering radiation clearly exhibits two resonances; one, a relatively weak response, close to (slightly below) the atomic transition frequency. The other is a strong and broad peak at a higher frequency that moves to the blue at larger atomic densities. Additional simulations presented in Fig. 4 show that the intensity of the high frequency mode (unlike the low frequency one) scales as n2 a for the low density of atoms suggesting a possible collective nature of the peak. It should be noted that this collective mode is noticeably wider than the atomic transition resonance. The dependence of the scattering intensity on cluster's size is shown in Fig. 3B. The low energy resonance exhibits a red shift, when a radius of the cluster increases, while the resonant frequency of the high energy mode does not change with cluster's size. It, however, becomes significantly wider for larger clusters, which has been observed experimentally [62]. One of the advantages of the present calculations over the standard approach based on a dielectric model is the ability to examine the influence of the dephasing rate on optical properties. Fig. 3C shows simulation results obtained at three pure dephasing rates, γp, including the case without pure dephasing. We should note that at small γp numerical simulations tend to become hard to converge at the frequencies near the collective resonance (in our case this occurs for γp < 4 × 1012 s−1). While for relatively high dephasing rates regular simulations require spatial steps, δx, on the order of 1 nm, the case without pure dephasing should be explored at δx < 0.2 nm. It is interesting to note that the collective mode, while decreasing its width with the decrease of the dephasing rate, is still significantly wider than the atomic transition peak. In contrast, the low frequency peak becomes narrower with decreasing dephasing rate, with its width approaching γ1. Note that the scattering actually shows a dip at the atomic frequency, with the low frequency peak slightly below it. Fig. 3D explores how the scattering is affected by the matrix element of the atomic dipole moment, µsp. Not surprisingly, the result for increasing µsp is qualitatively similar to that obtained with increasing atomic density. It is seen that larger µsp results in blue shift of the higher frequency collective mode, and in a red shift of the lower frequency mode. The former shows a quadratic dependence of the resonant frequency on the dipole moment, which has been theoretically discussed in the case of a sphere with uniformly distributed linear quantum dipoles [63]. It is useful, for the sake of comparison, to consider the simplest theoretical description for the optical response of our system, by modeling it as a dielectric particle with a dielectric response function taken from the Clausius-Mossotti expression ε = 1 + 2x 1 − x , (13) 14 where x ≡ 4π 3 nα, n is the number density of atoms, and α is the atomic polarizability. The absorption lineshape can be calculated [64] as the ratio between the dissipated power, Pdiss = (1/2)(cid:82) d3rσ E2 and the incident flux, Jin = c(cid:12)(cid:12)Ein (cid:12)(cid:12)2 / (8π), where σ =(cid:0)ωin/ (4π)(cid:1) Im (ε) is the conductivity, ε is the dielectric response function, E is the electric field in the particle, Ein is the incident electric field and c is the speed of light. For a small spherical particle [65] E = 3 ε + 2 Ein. (14) Using these expressions we obtain the absorption cross-section of a small spherical particle (cid:12)(cid:12)(cid:12)(cid:12) 3 ε + 2 (cid:12)(cid:12)(cid:12)(cid:12)2 of volume Ω in the form σa = Pdiss Jin = Ω ωin c Im (ε) = 16π 3 nΩ Im (α) . (15) The imaginary part of the molecular optical polarizability is essentially a Lorentzian reso- nance peaked at the atomic transition frequency ωa. We see that the absorption cross-section in the Clausius-Mossotti approximation is proportional to the number of particles and to the absorption of a single particle, as would be predicted for a system of non-interacting particles. On the same level of theory, the dipole induced on the particle is [65] ε − 1 ε + 2 3Ω 4π (cid:126)µ = (cid:126)Ein = nΩα (cid:126)Ein. (16) And, since the scattered light is proportional to µ2, it is predicted to go like the square of the number of particles, and to have a similar resonant behavior as a function of the incident frequency. Eqs. (15) and (16) describe essentially a system of non-interacting particles, occupying a volume with linear dimensions much smaller than the radiation wavelength, that respond coherently to the incident radiation. This approximation is valid at low atomic density. We will see below how dephasing and through-field interatomic interactions at higher densities affect this behavior. While the calculation procedure applied here provides a route to explore the effect of dephasing on the optical response of atomic and molecular clusters, further studies will be 15 needed in order to determine how much of this effect was indeed captured in our simulations. The destruction of phase is affected within our calculations on the mean field evolution of a single atom, feeling the effects of others through their mutual interaction with the radiation field. It is not obvious that this mean field implementation can capture the full physics of atoms going out of phase from each other. We leave this important technical question to a future study. It is informative to explore the spatial dependence of EM intensity, I ∼ E2 y , at reso- nant conditions. Fig. 5 shows intensity distributions calculated using steady-state solutions x + E2 of Maxwell-Liouville equations for the two resonance modes. Clearly the EM intensity at the lower resonant frequency is mainly localized on the surface of the cluster exhibiting dipole radiation pattern similar to EM intensity distributions seen at the plasmon resonance for a single metal nanoparticle [13]. The collective high frequency mode is distributed in the entire volume of the particle, where all atoms coherently participate in the radiation pro- cess. This suggests that the high frequency mode is more collective in nature than the low frequency one, consistent with its large density dependent shift from the atomic frequency and its n2 a scaling at low densities. The low frequency mode, involving fewer surface atoms, may be more atomic in nature. It is important to note however that this is not a single atom response since it clearly shifts with increasing atomic density. Optical properties of molecular aggregates resonantly coupled to plasmonic materials have been a subject of extensive research for the past several years. Sugawara et al. [66] demonstrated significant modification of transmission and reflection spectra of a gold film with deposited J-aggregates. It has been shown experimentally that SPP resonances no- tably affect molecular electronic structure leading to resonance splitting [33]. The latter was proposed to be used for controlling optics of such hybrid material using femtosecond laser pulses [67]. Moreover core-shell metal NPs with a shell comprised of optically active molecules have been recently studied experimentally [68]. To demonstrate the generality of our approach we present simulations of the core-shell particle schematically depicted in the inset of Fig. 6. Here a silver nanoparticle is shelled by a resonant atomic layer, with atomic transition frequency equaled to the SPP resonance of silver, ωa = 3.61 eV. The optical prop- erties of silver are described within the Drude model with parameters as in [18]. The hollow atomic shell exhibits doubled collective mode (red dashed line in Fig. 6), which corresponds to the symmetry of the problem and can be understood within the plasmon hybridization 16 model proposed for noble metal core-shell particles [69]. The important observation is a clear splitting of the SPP mode with additional strong peak centered near atomic transition frequency (blue dash-dotted line in Fig. 6). The observed splitting as indicated in [33] is due to the strong optical coupling of atoms with the SPP mode. V. CONCLUSION We have presented a self-consistent electrodynamical model based on coupled Maxwell- Liouville equations that takes into account arbitrary polarization of the incident field. The proposed model is applied to investigate linear optical response of nanoscale atomic clusters in two dimensions merging classical electrodynamics with quantum mechanical description of atoms. The calculations can capture collective effects that play pivotal role in electrodynam- ics of nano-systems and, with limitations discussed below, includes the effect of dephasing on the optical response of these systems. We have found that spherical atomic clusters exhibit two well-distinguished resonances. The low energy resonance is close to the atomic transition frequency of individual atoms. The EM intensity distribution at this resonance is localized near the surface of a cluster. The high energy mode, where all atoms in the cluster coherently participate in the scattering, has clear collective nature. The dependence of the scattering intensity on various parameters was considered. It was demonstrated that the pure dephasing plays an important role in the scattering and absorption. Moreover we successfully applied our formalism to more complex systems, which comprise a resonant atomic shell and a silver core. Applications of the proposed scheme are many and vast. They range from complete three-dimensional description of nanoparticles resonantly coupled to ensembles of quantum particles, to nonlinear optical phenomena at the nanoscale. We note that our scheme can be extended to molecular systems, where one may investigate Raman processes. At the same time, we have indicated physically significant open technical issues. One is the limited ability of an approach based on classical electrodynamics to describe spontaneous emission and hence fluorescence. To account for such phenomena we need to modify the Maxwell- Liouville equations, Eqs. (1), (10), (11), so as to take into account the quantum nature of the radiation field, possibly using the quantization schemes described in Refs. [70 -- 72] or [73, 74] (which were shown to be equivalent [75]). Another is the need to examine the 17 adequacy of mean field calculations of dephasing. All these will be subjects of our continuing studies. Acknowledgments The authors would like to acknowledge fruitful discussions with Dr. Maxim Efremov. M.S. is grateful to the financial and technical support from Arizona State University via startup funds. The research of A.N. is supported by the Israel Science Foundation, the Israel-US Binational Science Fundation, the European Science Council (FP7 /ERC grant no. 226628) and the Israel-Niedersachsen Research Fund. [1] M. Li, W. H. P. Pernice, C. Xiong, T. Baehr-Jones, M. Hochberg, and H. X. Tang, Nature 456, 480 (2008). [2] T. W. Ebbesen, C. Genet, and S. I. Bozhevolnyi, Phys. Today 61, 44 (2008). [3] R. F. Oulton, V. J. Sorger, T. Zentgraf, R.-M. Ma, C. Gladden, L. Dai, G. Bartal, and X. Zhang, Nature 461, 629 (2009). [4] S. Chu, Science 253, 861, 1991. [5] J. Homola, Anal. Bioanal. Chem. 377, 528 (2003). [6] I. H. El-Sayed, X. Huang, and M. A. El-Sayed Nano Lett. 5, 829 (2005). [7] I. Romero, J. Aizpurua, G. W. Bryant, and F. J. Garcia de Abajo, Opt. Express 14, 9988 (2006). [8] J. Zhang, Y. Fu, M. H. Chowdhury, and J. G. Lakowicz, Nano Let. 7, 2101 (2007). [9] H. Szmacinski, J. R. Lakowicz, J. M. Catchmark, K. Eid, J. P. Anderson, L. Middendorf, Appl. Spectrosc. 62, 733 (2008). [10] W. Li, Pedro H. C. Camargo, X. Lu, and Y. Xia, Nano Lett. 9, 485 (2009). [11] K. L. Wustholz, A.-I. Henry, J. M. McMahon, R. G. Freeman, N. Valley, M. E. Piotti, M. J. Natan, G. C. Schatz, and R. P. Van Duyne, J. Am. Chem. Soc. 132, 10903 (2010). [12] M. Rycenga, P. H. C. Camargo, W. Li, C. H. Moran, and Y. Xia, J. Phys. Chem. Lett. 1, 696 (2010). [13] W. A. Murray and W. L. Barnes, Adv. Mater. 19, 3771 (2007). 18 [14] K. L. Kelly, E. Coronado, L. L. Zhao, and G. C. Schatz, J. Phys. Chem. B 107, 668 (2003). [15] M. Sukharev, P. R. Sievert, T. Seideman, and J. B. Ketterson, J. Chem. Phys. 131, 034708 (2009). [16] S. A. Maier, M. L. Brongersma, P. G. Kik, S. Meltzer, A. A. G. Requicha, and H. A. Atwater, Adv. Mater. 13, 1501 (2001). [17] E. C. Le Ru and P. G. Etchegoin, Principles of Surface-Enhanced Raman Spectroscopy: And Related Plasmonic Effects (Elsevier, 2009). [18] M. Sukharev and T. Seideman, J. Phys. Chem. A 113, 7508 (2009). [19] Y. Fedutik, V. V. Temnov, O. Schops, U. Woggon, and M. V. Artemyev, Phys. Rev. Lett. 99, 136802 (2007). [20] X. Xu, B. Sun, P. R. Berman, D. G. Steel, A. S. Bracker, D. Gammon, and L. J. Sham, Science 317, 929 (2007). [21] V. K. Komarala, A. L. Bradley, Y. P. Rakovich, S. J. Byrne, Y. K. Gunko, and A. L. Rogach, Appl. Phys. Lett. 93, 123102 (2008). [22] A. A. Toropov, T. V. Shubina, V. N. Jmerik, S. V. Ivanov, Y. Ogawa, and F. Minami, Phys. Rev. Lett. 103, 037403 (2009). [23] D. E. Gomez, K. C. Vernon, P. Mulvaney, and T. J. Davis, Nano Lett. 10, 274 (2010). [24] X.-R. Su, W. Zhang, L. Zhou, X.-N. Peng, and Q.-Q. Wang, Opt. Express 7, 6516 (2010). [25] M. Achermann, J. Phys. Chem. Lett. 1, 2837 (2010). [26] M. L. Andersen, S. Stobbe, A. S. Sorensen, and P. Lodahl, Nat. Phys. 7, 215 (2010). [27] D. Ratchford, F. Shafiei, S. Kim, S. K. Gray, and X. Li, Nano Lett. 11, 1049 (2011). [28] E. Dulkeith, A. C. Morteani, T. Niedereichholz, T. A. Klar, J. Feldmann, S. A. Levi, F. C. J. M. van Veggel, D. N. Reinhoudt, M. Moller, and D. I. Gittins, Phys. Rev. Lett. 89, 203002 (2002). [29] C. Bonnand, J. Bellessa, and J. C. Plenet, Phys. Rev. B 73, 245330 (2006). [30] S. Pan, Z. Wang, and L. J. Rothberg , J. Phys. Chem. B 110, 17383 (2006). [31] T. Uwada, R. Toyota, H. Masuhara, and T. Asahi , J. Phys. Chem. C 111, 1549 (2007). [32] W. Ni, Z. Yang, H. Chen, L. Li, and J. Wang , J. Am. Chem. Soc. 130, 6692 (2008). [33] A. Salomon, C. Genet, and T. W. Ebbesen, Angew. Chem. 121, 8904 (2009). [34] N. I. Cade, T. Ritman-Meer, and D. Richards , Phys. Rev. B 79, 241404(R) (2009). [35] S. Hayashi, A. Maekawa, S. C. Kim, and M. Fujii, Phys. Rev. B 82, 035441 (2010). 19 [36] T. Sen, S. Jana, S. Koner, and A. Patra, J. Phys. Chem. C 114, 707 (2010). [37] C. Garcia-Segundo, H. Yan, and M. S. Zhan, Phys. Rev. A 75, 030902(R) (2007). [38] E. Vetsch, D. Reitz, G. Sague, R. Schmidt, S. T. Dawkins, and A. Rauschenbeutel, Phys. Rev. Lett. 104, 203603 (2010). [39] P. Xu, X. He, J. Wang, and M. Zhan, Opt. Lett. 35, 2164 (2010). [40] F. Le Kien and K. Hakuta, Phys. Rev. A 81, 063808 (2010). [41] Y. Sun and Y. Xia, Science 298, 2176 (2002). [42] R. W. Ziolkowski, J. M. Arnold, and D. M. Gogny, Phys. Rev. A 52, 3082 (1995). [43] G. Slavcheva, J. M. Arnold, I. Wallace, and R. W. Ziolkowski, Phys. Rev. A 66, 063418 (2002). [44] A. Fratalocchi, C. Conti, and G. Ruocco Phys. Rev. A 78, 013806 (2008). [45] K. Lopata and D. Neuhauser, J. Chem. Phys. 131, 014701 (2009). [46] S. K. Gray and T. Kupka, Phys. Rev. B 68, 045415 (2003). [47] H.-P. Breuer and F. Petruccione, The theory of open quantum systems (Oxford University Press, New York, 2002). [48] Note that as in many studies of this type, a 2-dimensional solution does not correspond to a two dimensional system but to the 3-dimensional system, of infinite extent in the z-direction, obtained by displacing the xy cross-section along the z axis. The densities defined above remain per-unit-volume quantities. [49] http://budker.berkeley.edu/ADM/index.html. [50] A. Taflove and S. C. Hagness, Computational electrodynamics: the finite-difference time- domain method, 3rd ed. (Artech House, Boston, 2005). [51] K. S. Yee, IEEE Trans. Antennas Propagat. 14, 302 (1966). [52] J. P. Berenger, Perfectly matched layer (PML) for computational electromagnetics in Synthesis lectures on computational electromagnetics ed. C. A. Balanis (Morgan and Claypool Publ., 2007). [53] J. A. Roden and S. D. Gedney, Microw. Opt. Techn. Let. 27, 334 (2000). [54] P. S. Pacheco, Parallel Programming with MPI (Morgan Kaufmann Publishing, 1993). [55] http://plasmon.poly.asu.edu. [56] E. Hairer, S. P. Norsett, and G. Wanner, Solving ordinary differential equations I: nonstiff problems (Springer, 2000). [57] In the actual calculations ρ22 and ρ33 exhibit small low frequency oscillations that we eliminate 20 by time-averaging. [58] J. Andreasen and H. Cao, J. Lightwave Technol. 27, 4530 (2009). [59] Y. Huang, Z. Zhang, and S. Ho, in Frontiers in Optics, OSA Technical Digest (CD) (Optical Society of America, 2008), paper FThP5. [60] J. Andreasen and H. Cao, Phys. Rev. A 82, 063835 (2010). [61] We have numerically verified that all results of our simulations scale linearly with the incident intensity. [62] K. Misawaa, H. Yaob, T. Hayashib, and T. Kobayashi, Chem. Phys. Lett. 183, 113 (1991). [63] S. Prasad and R. J. Glauber, Phys. Rev. A 82, 063805 (2010). [64] J. Gersten and A. Nitzan, J. Chem. Phys. 75, 1139 (1981). [65] J. D. Jackson, Classical Electrodynamics, Ch. 4 (John Wiley and Sons, New York 1962). [66] Y. Sugawara, T. A. Kelf, J. J. Baumberg, M. E Abdelsalam, and P. N. Bartlett, Phys. Rev. Lett. 97, 266808 (2006). [67] P. Vasa, R. Pomraenke, G. Cirmi, E. De Re, W. Wang, S. Schwieger, D. Leipold, E. Runge, G. Cerullo, and C. Lienau, ACS Nano 12, 7559 (2010). [68] D. D. Lekeufack, A. Brioude, A. W. Coleman, P. Miele, J. Bellessa, L. D. Zeng, and P. Stadelmann, Appl. Phys. Lett. 96, 253107 (2010). [69] E. Prodan and P. Nordlander, J. Chem. Phys. 120, 5444 (2004). [70] T. Gruner and D.-G. Welsch, Phys. Rev. A 53, 1818 (1996). [71] H. T. Dung, L. Knoll, and D.-G. Welsch, Phys. Rev. A 57, 3931 (1998). [72] S. Scheel, L. Knoll, and D.-G. Welsch, Phys. Rev. A 58, 700 (1998). [73] A. Tip, Phys. Rev. A 57, 4818 (1998). [74] A. Tip, Phys. Rev. A 69, 013804 (2004). [75] A. Tip, L. Knoll, S. Scheel and D.-G. Welsch, Phys. Rev. A 63, 043806 (2001). 21 FIG. 1: (Color online). Panel A shows the energy level diagram of a two-level two-dimensional atom with black arrows indicating optically induced transitions by the TE mode and red arrows representing spontaneous decay. Panel B depicts schematics of the simulations with the detection point shown as a red diamond (in the lower left corner), where the y-component of the Poynting vector is calculated. Rxyabsorbing boundaryabsorbing boundaryabsorbing boundaryabsorbing boundarySy110 nm110 nmωaspxpyplane wave excitation(A)(B) !22 FIG. 2: Scattering intensity as a function of the incoming frequency ωin calculated within SPM approach (solid line) and CW scheme (circles). Normalized absorption (see Eq.(12)) is shown as squares. Simulations are performed for the cluster with the following parameters: ωa = 3.1 eV, R = 25 nm, na = 7 × 1025 m−3, γ1 = 1012 s−1, γp = 1013 s−1, µsp = 25 Debye. 3.053.13.153.23.25win [eV]00.250.50.751signal [normalized]23 FIG. 3: (Color online). Linear optics of atomic clusters with ωa = 3.1 eV, γ1 = 1012 s−1. Panel A: scattering intensity as a function of the incident frequency for three atomic densities, na - black solid line: na = 2.5 × 1025 m−3 (ideal gas at atmospheric pressure and room temperature), red dashed line: na = 7× 1025 m−3, and blue dash-dotted line: na = 1026 m−3. Other parameters are: R = 25 nm, γp = 1013 s−1, µsp = 25 Debye. Panel B: same is in panel A for three cluster's radii, R - black (solid), red (dashed), and blue (dash-dotted) lines show scattering intensity results for R = 15 nm, 25 nm, and 35 nm, respectively. Other parameters are: na = 1026 m−3, γp = 1013 s−1, µsp = 25 Debye. Panel C: same as in panels A and B (now shown in logarithmic scale), but for four pure dephasing rates, γp - black (solid) line: γp = 2 × 1013 s−1, red (dashed) line: 1γp = 1013 s−1, blue (dash-dotted) line γp = 2× 1012 s−1, and green (dotted) line shows the data without pure dephasing γp = 0 s−1. Other parameters are: R = 25 nm, na = 1026 m−3, µsp = 25 Debye. Panel D: same as in panels A-C, but for three values of the matrix element of the dipole moment, µsp - black (solid), red (dashed), and blue (dash-dotted) lines correspond to atomic systems characterized by µsp = 10 Debye, µsp = 25 Debye, and µsp = 40 Debye, respectively. Other parameters are: R = 25 nm, na = 1026 m−3, γ1 = 1012 s−1, γp = 1013 s−1. 24 FIG. 4: (Color online). Scattering intensity (shown in double logarithmic scale) at the high fre- quency resonance, ωres, as a function of the atomic density, na, for three pure dephasing rates, γp - blue circles: γp = 2 × 1013 s−1, black rectangles: γp = 1013 s−1, and red triangles: γp = 2 × 1012 s−1. Other parameters are: ωa = 3.1 eV, γ1 = 1012 s−1, R = 25 nm, na = 1026 m−3, µsp = 25 Debye. The dashed straight lines represent fitting for each set of data demonstrating nearly ideal quadratic dependence on na at low densities. 102310241025102610(cid:239)710(cid:239)610(cid:239)510(cid:239)410(cid:239)310(cid:239)210(cid:239)1100na [m(cid:239)3]Sy((cid:116)res) / Sinc((cid:116)res) na1.9725 FIG. 5: (Color online). Spatial distributions of EM intensity (normalized to the incident intensity) in logarithmic scale at the low frequency resonance (left panel) and the high frequency resonance (right panel). Cluster's parameters are: ωa = 3.1 eV, γ1 = 1012 s−1, γp = 1013 s−1, R = 25 nm, na = 1026 m−3, µsp = 25 Debye. 1.000.800.600.400.200.002.231.791.340.890.450.00ωin = 3.09 eVωin = 3.18 eV26 FIG. 6: (Color online). Scattering intensity as a function of the incident frequency for the core- shell particle shown in the inset. Black solid line shows the data for the silver nanoparticle without atomic shell, red dashed line presents simulations for the hollow atomic shell, and blue dash-dotted line demonstrates results obtained for atomic shell with a silver core. Parameters of the simulations are: na = 2.5 × 1025 m−3, ωa = 3.61 eV, γ1 = 1012 s−1, γp = 1013 s−1, µsp = 25 Debye
1907.02255	1	1907	2019-07-04T07:25:02	Resonant Acoustic Wave Assisted Spin-Transfer-Torque Switching of Nanomagnets	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	We report the possibility of achieving an order of magnitude reduction in the energy dissipation needed to write bits in perpendicular magnetic tunnel junctions (p-MTJs) by simulating the magnetization dynamics under a combination of resonant surface acoustic waves (r-SAW) and spin-transfer-torque (STT). The magnetization dynamics were simulated using the Landau-Lifshitz-Gilbert equation under macrospin assumption with the inclusion of thermal noise. The resonant magnetization dynamics in the magnetostrictive nanomagnet build over few 10s of cycles of SAW application that drives the magnetization to precess in a cone with a deflection of ~45 degrees from the perpendicular direction. This reduces the STT current density required to switch the magnetization direction without increasing the STT application time or degrading the switching probability in the presence of room temperature thermal noise. This could lead to a pathway to achieve energy efficient switching of spin transfer torque random access memory (STTRAM) whose lateral dimensions can be scaled aggressively despite using materials with low magnetostriction by employing resonant excitation.	cond-mat.mes-hall	cond-mat	RESONANT ACOUSTIC WAVE ASSISTED SPIN-TRANSFER-TORQUE SWITCHING OF NANOMAGNETS Austin Roe1, Dhritiman Bhattacharya1, and Jayasimha Atulasimha1, 2 * 1Department of Mechanical and Nuclear Engineering, Virginia Commonwealth University, Richmond, VA 23284, USA 2Department of Electrical and Computer Engineering, Virginia Commonwealth University, Richmond, VA 23284, USA * Corresponding author: [email protected] We report the possibility of achieving an order of magnitude reduction in the energy dissipation needed to write bits in perpendicular magnetic tunnel junctions (p-MTJs) by simulating the magnetization dynamics under a combination of resonant surface acoustic waves (r-SAW) and spin-transfer-torque (STT). The magnetization dynamics were simulated using the Landau-Lifshitz-Gilbert equation under macrospin assumption with the inclusion of thermal noise. The resonant magnetization dynamics in the magnetostrictive nanomagnet build over few 10s of cycles of SAW application that drives the magnetization to precess in a cone with a deflection of ~45o from the perpendicular direction. This reduces the STT current density required to switch the magnetization direction without increasing the STT application time or degrading the switching probability in the presence of room temperature thermal noise. This could lead to a pathway to achieve energy efficient switching of spin transfer torque random access memory (STTRAM) whose lateral dimensions can be scaled aggressively despite using materials with low magnetostriction by employing resonant excitation. The quest to overcome the limitations of random access memory (RAM) to achieve energy efficient and higher speed computing architectures has motivated research to replace volatile CMOS memory devices. One alternative is the use of nanoscale magnetic tunnel junctions (MTJ) that implement non-volatile memory. The magnetization orientation of a MTJ's soft magnetic layer can be switched using different methods [1 -- 7]. One such method utilizes spin transfer torque (STT) which involves applying a current through a hard magnetic layer to polarize electrons that subsequently exert a torque on the magnetization of the soft layer to switch its direction. However, these STT based MTJ devices require ~100 fJ/bit [8] to switch the magnetization which is 1000 times more than the ~100 aJ energy required to switch CMOS devices [9]. This inefficiency could prevent the widespread adoption of pure current driven spin transfer torque (STT) switching [2,10,11]. Another newer that addresses some of the shortcomings of STT is spin orbit torque (SOT) induced switching [4,12,13]. However, it results in a 3-terminal memory device that could impede aggressive scaling. technology Strain generated via surface acoustic waves (SAW) can also be used to write a bit in an MTJ [14] by controlling the magnetization of its soft magnetostrictive layer. SAW can be created from an interdigitated transducer (IDT) fabricated on a piezoelectric substrate which typically produces Rayleigh (transverse) waves. Rotating the magnetization through the use of SAW is very energy efficient, however, only a ~90° rotation is possible unless dipole coupling [15] or sequential stress along multiple directions is used [16]. Mixed mode SAW and STT is a potential alternative to overcoming the large write energy requirement of STT and complexity of SAW devices. Previous efforts have explored this concept in nanomagnets with in-plane magnetization using low frequency SAW [17], which allows for the quasistatic rotation of the magnetization to an approximately known deflection. STT can be applied to achieve switching once this maximum deflection is reached. However, this approach requires a much higher magnitude SAW, especially in scaled nanomagnets. This is because the stress levels required to produce a large deflection of the magnetization increases with decreased volume of the nanomagnets as the uniaxial magnetic anisotropy Ku increases to ensure an energy barrier ~1 eV between the "0" and "1" states (KuΩ ≈ 1 eV, where Ω is the volume). Due to this, mixed mode switching has potential issues scaling below 100 nm diameter, even if moderately magnetostrictive materials such as FeGa [18] are used with low Gilbert damping [19]. High magnetostrictive materials such as Terfenol-D [20] will not necessarily achieve a larger magnetization deflection with low stress levels due to the bidirectional coupling between magnetization and strain [21]. This further motivates our resonance approach to overcome the limitations associated with quasistatic SAW excitation, allowing for competitive scalability to smaller lateral dimensions. perpendicular-to-plane magnetization. We study a hybrid resonant SAW and STT scheme to switch the magnetization of nanomagnets for both in-plane and 1 demonstrates how these devices can be realized. The SAW is applied over an entire array of nanomagnets and thus adds very little to the energy dissipation per bit switched. Once the nanomagnets have reached maximum deflection (Fig. 1(b) for in-plane, and (c) for out-of-plane), spin-transfer-torque can be applied to minimize the spin current required to switch as compared to a non-stressed state. Fig. Modelling of the magnetization dynamics was performed by solving the Landau-Lifshitz-Gilbert (LLG) equation [22] with inclusion of damping like spin transfer torque term [23]: (cid:4666)1(cid:3397)(cid:2009)(cid:2870)(cid:4667)(cid:3031)(cid:3014)(cid:4652)(cid:4652)(cid:3377)(cid:3031)(cid:3047)(cid:3404)(cid:2011)(cid:1839)(cid:4652)(cid:4652)(cid:3377)(cid:3400)(cid:1834)(cid:4652)(cid:4652)(cid:3377)(cid:3032)(cid:3033)(cid:3033)(cid:3398)(cid:3080)(cid:3082)(cid:3014)(cid:3294)(cid:3427)(cid:1839)(cid:4652)(cid:4652)(cid:3377)(cid:3400)(cid:3435)(cid:1839)(cid:4652)(cid:4652)(cid:3377)(cid:3400)(cid:1834)(cid:4652)(cid:4652)(cid:3377)(cid:3032)(cid:3033)(cid:3033)(cid:3439)(cid:3431)(cid:3397) (cid:3080)(cid:3082)(cid:3014)(cid:3294)(cid:2010)(cid:2013)(cid:3435)(cid:1839)(cid:3400)(cid:4666)(cid:1839)(cid:3043)(cid:3400)(cid:1839)(cid:4667)(cid:3439) (cid:2010)(cid:3404) (cid:4698) (cid:1328)(cid:3091)(cid:3116)(cid:3032)(cid:4698) (2) (1) (cid:3011)(cid:3039)(cid:3295)(cid:3283)(cid:3014)(cid:3294) where M is the is the magnetization, γ is the gyromagnetic ratio, α is the Gilbert damping coefficient, Ms is the saturation magnetization and Mp is the magnetization in the direction of the STT polarization, ℏ is the Planck constant, μ0 is the permeability of free space, e is the charge of an electron, J is the current density of the STT, and lth is the thickness of the nanomagnets. The effective field was calculated from the total energy of the system: (cid:1834)(cid:4652)(cid:4652)(cid:3377)(cid:3032)(cid:3033)(cid:3033)(cid:3404)(cid:3398) (cid:2869)(cid:3091)(cid:3116)(cid:3078)(cid:3031)(cid:3006)(cid:3031)(cid:3014)(cid:4652)(cid:4652)(cid:3377) (cid:1831)(cid:3404)(cid:1831)(cid:3046)(cid:3047)(cid:3045)(cid:3032)(cid:3046)(cid:3046) (cid:3028)(cid:3041)(cid:3036)(cid:3046)(cid:3042)(cid:3047)(cid:3045)(cid:3042)(cid:3043)(cid:3052)(cid:3397)(cid:1831)(cid:3046)(cid:3035)(cid:3028)(cid:3043)(cid:3032) (cid:3028)(cid:3041)(cid:3036)(cid:3046)(cid:3042)(cid:3047)(cid:3045)(cid:3042)(cid:3043)(cid:3052)(cid:3397)(cid:1831)(cid:3017)(cid:3014)(cid:3002) (4) (cid:1831)(cid:3046)(cid:3035)(cid:3028)(cid:3043)(cid:3032) (cid:3028)(cid:3041)(cid:3036)(cid:3046)(cid:3042)(cid:3047)(cid:3045)(cid:3042)(cid:3043)(cid:3052)(cid:3404)(cid:4672)(cid:3091)(cid:3116)(cid:2870)(cid:4673)(cid:2007)(cid:3427)(cid:1840)(cid:3031)_(cid:3051)(cid:3051)(cid:1839)(cid:3051)(cid:2870)(cid:3397)(cid:1840)(cid:3031)_(cid:3052)(cid:3052)(cid:1839)(cid:3052)(cid:2870)(cid:3397) (cid:1840)(cid:3031)_(cid:3053)(cid:3053)(cid:1839)(cid:3053)(cid:2870)(cid:3431) (cid:1831)(cid:3046)(cid:3047)(cid:3045)(cid:3032)(cid:3046)(cid:3046) (cid:3028)(cid:3041)(cid:3036)(cid:3046)(cid:3042)(cid:3047)(cid:3045)(cid:3042)(cid:3043)(cid:3052)(cid:3404)(cid:3398)(cid:2871)(cid:2870)(cid:4670)(cid:2019)(cid:3046)(cid:2026)(cid:2007)(cid:4671)sin(cid:2870)(cid:2016)(cid:3036)sin(cid:2870)(cid:2038)(cid:3036) (cid:1831)(cid:3017)(cid:3014)(cid:3002)(cid:3404)(cid:1837)(cid:3046)(cid:2868)cos(cid:2870)(cid:2016)(cid:3036) (3) (7) (5) (6) with Ω representing the volume of the nanomagnets, Nd_xx, Nd_yy, and Nd_zz the demagnetization factors in the respective directions, λs the saturation magnetostriction, σ the stress produced by the SAW on the nanomagnets, θi and φi the polar and azimuthal angles, respectively, of the magnetization, and Ks0 the surface anisotropy constant. We note the effective field due to stress is calculated from uniaxial stress only (cyclic tension and compression) due to the SAW. The stress in the in-plane direction orthogonal to SAW propogation, which sees the opposite stress (cyclic compression and tension) due to Poisson's effect, is neglected. This does not change the magnetization dynamics qualitatively, and only makes the stress amplitude we estimate conservative (a smaller stress will produce the same effective field if stress in the other direcition is accounted for). Table I lists the values of the material properties of Fe81Ga19 used in these simulations. It was chosen as it has moderate magnetostriction and low Gilbert damping [18,19]. Table I: FeGa material properties Parameters Fe0.81Ga0.19 Saturation Magnetization (Ms) 0.8x106 A/m Gilbert Damping (α) Gyromagnetic Ratio (γ) Magnetostriction (λs) 0.015 2.2x105 350 ppm The magnetization dynamics were simulated as follows. The resonant SAW was applied for several nanoseconds to build the maximum deflection of the magnetization. The magnitude of the SAW was chosen so that no switching from only the SAW excitation occurs. When the maximum deflection was reached, the minimum STT current density needed to maintain low switching error probability (for example, 99.9% switching in case of out-of-plane switching limited by the number of computations we could perform) was applied. The time of removal of the SAW (as long as this was after STT application was completed) made no difference to the final switching probability. IN-PLANE SWITCHING WITH SAW + STT: Simulation of elliptical nanomagnetic disks with in-plane magnetic anisotropy is shown in Fig 1(b) with dimensions of 40 x 30 nm along the easy and hard axis, respectively, and a thickness of 6 nm. SAW was applied to the nanomagnets and the resonance frequency, which creates the maximum deflection from the magnetic easy axis at the given SAW magnitude, was found. The SAW frequency was found to be double that of the frequency at which the magnetization precesses about the easy axis. When the magnet is compressed along its long (easy) axis, the magnetization rotates towards the magnetic hard axis on one side, and when a tensile stress is applied, the magnetization returns to the easy axis. This process isrepeated with magnetization moving towards the opposite hard axis in the next cycle, leading to a doubling of the frequency for the applied SAW. If resonance is achieved, the deflection of the magnetization from the easy axis will increase until a maximum deflection is reached over a few 10s of cycles. Without thermal noise (Fig 2(a)), the STT current density required (15x1011 A/m2 with a SAW of 23 MPa at 10.7 GHz) is lowered drastically compared to when no SAW is applied. To achieve switching in the same time without the application of SAW, a current density ~3 times larger was needed. This translates to nearly one order of magnitude lower energy (3 times current is 9 times write energy, assuming similar application time). However, in the in-plane case, this potential energy saving is severely negated by thermal noise effects as discussed next. to achieve 99% switching of THERMAL NOISE: The magnitude of the STT current density needed the magnetization from the positive to the negative y-axis in the presence of thermal noise and applied SAW of 10 MPa at 9.42 GHz was 10x1011 A/m2, shown in Fig. 2(c). Now when the SAW was removed the same current density magnitude sufficed to achieve 99% switching, indicating that the SAW was not helping lower the STT write current for the in-plane switching in the presence of thermal noise (Fig. 2 b). This can be explained as follows. Prior to STT application, a large difference can be seen between the magnetization deflection from the thermal noise without SAW to that with thermal noise with SAW. This indicates that the SAW indeed produced higher deflection even with the thermal noise. However, once STT of sufficient magnitude is applied, there is no difference in switching probability between the no-SAW and SAW case. In previous calculations without thermal noise, it was possible to determine apriori the specific timing to apply the STT so that the magnetization is at a point of maximum deflection. However, due to thermal fluctuations, while there is higher deflection with SAW, the inability to apply STT at the maximum deflection negated the benefit of any added deflection from the SAW and no difference was seen between the switching probability in the two cases, as shown in Fig. 2(b) and 2(c). OUT-OF-PLANE SWITCHING WITH SAW + STT: Simulation of the magnetization dynamics of cylindrical disks with a diameter of 50 nm and thickness 1.5 nm with perpendicular magnetic anisotropy is shown in Fig. 1 (c). Initially, the magnetization points in the out-of-plane direction (+z-axis), and a continuous SAW was applied across the magnets to find the resonant frequency, which is dependent on the anisotropy of the nanomagnets and magnitude of the SAW (as the oscillation is non-linear). This SAW applies compression and tension along the y-axis of the nanomagnet (and vice versa along the x-axis). This cyclical stressing on the magnet rotates the magnetization from pointing directly out-of-plane to begin to precess about the z- axis, as shown in Fig. 1(c). As a result of the resonant SAW, the magnetization precesses further from the z-axis and the cone of rotation becomes larger as the magnetization approaches the x-y plane of the nanomagnet. Once the maximum deflection of the magnetization from the out-of-plane direction was achieved, spin-transfer-torque is applied. As in the previous case, once the magnetization is switched and the STT current is withdrawn, the SAW can be concurrently removed or continued to run for a few cycles before withdrawal as it is not enough to switch the magnetization on its own. Previously with the in-plane calculations, the inability to selectively apply the STT when the deflection was at a maximum was an issue. However, the manner in which the magnetization precesses around the out-of-plane axis eliminates the need to time the application of the spin torque. This is because for a given polar deflection, θ of the magnetization, which is reached and stabilized after a few cycles, the azimuthal angle, Φ shown in Fig. 1 at which the magnetization is oriented when the STT is applied does not affect its efficacy. Fig. 3(a) shows the effect of SAW on deflection of the magnetization with and without thermal fluctuations. Without any SAW, thermal noise alone deflects the magnetization to a maximum of ~10° from the z-axis. On the other hand, when SAW of 100 MPa at a resonant frequency of 9.95 GHz is applied and the magnetization allowed to build to its maximum deflection, an average deflection of ~45° is seen in both the presence and absence of thermal noise. The thermal noise merely makes the deviation from the mean deflection of ~45° much higher than the case of precession without thermal noise. Fig. 3(b) shows the trajectories of magnetization switching in the presence of thermal noise for applied resonant SAW and STT. At a fixed current density and STT application time the switching probabilities (trajectories that switch) are much higher with SAW than without the SAW. OUT-OF-PLANE SWITCHING STT ONLY VS. SAW + STT ENERGY COMPARISON: At a current density of 1.9x1011 A/m2 applied for 0.7 ns along with a SAW magnitude of 100 MPa at 9.95 GHz, 99.9% of the simulated nanomagnets successfully switched from the positive to the negative z-axis. Without any SAW, at this same current density and STT application time, roughly 89% of the magnets switched. This error could be reduced to 99.9% in the no SAW case if the current density was doubled to 3.8x1011 A/m2, as shown in Fig. 3(c). Simulation of SAW at both 30 MPa at 11.36 GHz and 60 MPa at 10.79 GHz was also conducted and a current density of 3x1011 A/m2 and 2.2x1011 A/m2, respectively, was required to achieve 99.9% switching probability. Keeping the current density magnitude fixed at 2x1011 A/m2 and adjusting the STT application time yielded similar results. While previously, 99.9% of nanomagnets switched at 100 MPa and 0.7 ns, the case with no SAW needed double the amount of time (1.4 ns) to ensure the same switching probability. With required current density being just half that compared to the case without any SAW, the energy savings are four times that of the pure STT case. With materials such as Rare Earth substituted YIG [24] low magnetostriction but extremely small damping, at least an order of magnitude energy saving with similar or better error rates could be achieved. This could open the pathway to achieve desired scalability and be competitive with current CMOS memory implementations while having the added advantage of non-volatility. that have In summary, we have shown that application of resonant SAW and STT can be more energy efficient in switching nanomagnetic soft layer of p-MTJs than switching with only STT. This process can be further optimized to increase the energy savings and switching time. These theoretical results could stimulate experimental work, ultimately resulting in the development of more energy efficient STTRAM. A.R., D.B. and J.A are supported National Science Foundation SHF Small grant #CCF-1815033. References: M. T. Alam, M. J. Siddiq, G. H. Bernstein, M. Niemier, W. Porod, and X. S. Hu, IEEE Trans. Nanotechnol. 9, 348 (2010). D. C. Ralph and M. D. Stiles, J. Magn. Magn. Mater. 320, 1190 (2008). M. Yamanouchi, D. Chiba, F. Matsukura, and H. Ohno, Nature 428, 539 (2004). D. Bhowmik, L. You, and S. Salahuddin, Nat. Nanotechnol. 9, 59 (2014). S. Manipatruni, D. E. Nikonov, and I. A. Young, Appl. Phys. Express 7, 103001 (2014). N. Tiercelin, Y. Dusch, V. Preobrazhensky, and P. Pernod, J. Appl. Phys. 109, 07D726 (2011). [11] [12] [13] L. Berger, Phys. Rev. B 54, 9353 (1996). L. Liu, C.-F. Pai, Y. Li, H. W. Tseng, D. C. Ralph, and R. A. Buhrman, Science 336, 555 (2012). I. Mihai Miron, G. Gaudin, S. Auffret, B. Rodmacq, A. Schuhl, S. Pizzini, J. Vogel, and P. Gambardella, Nat. Mater. 9, 230 (2010). [14] N. D'Souza, A. Biswas, H. Ahmad, M. Salehi Fashami, M. Mamun Al-Rashid, V. Sampath, D. Bhattacharya, M. Ahsanul Abeed, J. Atulasimha, and S. Bandyopadhyay, Nanotechnology 29, 49 (2018). [15] V. Sampath, N. D'Souza, G. M. Atkinson, S. Bandyopadhyay, and J. Atulasimha, Appl. Phys. Lett. 109, 102403 (2016). [16] A. K. Biswas, H. Ahmad, J. Atulasimha, and S. Bandyopadhyay, Nano Lett. 17, 3478 (2017). [17] A. K. Biswas, S. Bandyopadhyay, and J. Atulasimha, Appl. Phys. Lett. 103, 232401 (2013). [18] A. Clark, M. Wun-Fogle, J. B. Restorff, and T. A. Lograsso, Mater. Trans. 43, 881 (2002). [19] D. B. Gopman, V. Sampath, H. Ahmad, S. Bandyopadhyay, and J. Atulasimha, IEEE Trans. Magn. 53, (2017). [20] [21] L. Sandlund, M. Fahlander, T. Cedell, A. E. Clark, J. B. Restorff, and M. Wun‐Fogle, J. Appl. Phys. 75, 5656 (1994). Z. Xiao, R. Lo Conte, C. Chen, C.-Y. Liang, A. Sepulveda, J. Bokor, G. P. Carman, and R. N. Candler, Sci Rep. 8, 5207 (2018). [22] T. L. Gilbert, IEEE Transactions on Magnetism, 40 (6) 3443 (2004). [23] [24] J. Z. Sun, Phys. Rev. B 62, 570, 2000. E. R. Rosenberg, L. Beran, C. O. Avci, C. Zeledon, B. Song, C. Gonzalez-Fuentes, J. Mendil, P. Gambardella, M. Veis, C. Garcia, G. S. D. Beach, and C. A. Ross, Phys. Rev. Mater. 2, 94405 (2018). [1] [2] [3] [4] [5] [6] [7] [8] K. Roy, S. Bandyopadhyay, and J. Atulasimha, Appl. Phys. Lett. 99, 063108 (2011). J. J. Nowak, R. P. Robertazzi, J. Z. Sun, G. Hu, J.-H. Park, J. Lee, A. J. Annunziata, G. P. Lauer, R. Kothandaraman, E. J. O'Sullivan, P. L. Trouilloud, Y. Kim, and D. C. Worledge, IEEE Magn. Lett. 7, 1 (2016). [9] [10] S. Datta, V. Q. Diep, and B. Behin-Aein, (2014). J. C. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996). (b) Figure 1: (a) MTJ array switched with resonant SAW and STT (b) Magnetization dynamics with resonant SAW + STT switching of in- plane magnetization (c) Magnetization dynamics with resonant SAW + STT switching of out-of-plane magnetization. (a) (c) Figure 2: In-plane magnetization dynamics simulations with (a) no thermal noise, (b) resonant STT and SAW, and (c) STT only and no SAW. NOTE: Angle shown is the azimuthal angle φ with ~90° degree being the initial orientation and both ~270° and ~ -90° representing a successful switch, while return to ~90° degree being an unsuccessful switching event. (a) (b) (c) Figure 3: Out-of-plane magnetization dynamics simulations with (a) comparison of resonant SAW with and without thermal noise to purely thermal noise and (b) switching of out-of-plane magnetization with resonant SAW and STT (angle shown is the polar angle, θ) (c) Switching probability vs. STT current density for three different SAW magnitudes as well as for no SAW appled.
1602.03136	5	1602	2016-06-10T23:37:46	Solvable multistate model of Landau-Zener transitions in cavity QED	[ "cond-mat.mes-hall", "math-ph", "math-ph", "quant-ph" ]	We consider the model of a single optical cavity mode interacting with two-level systems (spins) driven by a linearly time-dependent field. When this field passes through values at which spin energy level splittings become comparable to spin coupling to the optical mode, a cascade of Landau-Zener (LZ) transitions leads to co-flips of spins in exchange for photons of the cavity. We derive exact transition probabilities between different diabatic states induced by such a sweep of the field.	cond-mat.mes-hall	cond-mat	Solvable multistate model of Landau-Zener transitions in cavity QED Nikolai A. Sinitsyna and Fuxiang Lia,b a Theoretical Division, Los Alamos National Laboratory, Los Alamos, NM 87545, USA and b Center for Nonlinear Studies, Los Alamos National Laboratory, Los Alamos, NM 87545 USA (Dated: June 14, 2016) We consider the model of a single optical cavity mode interacting with two-level systems (spins) driven by a linearly time-dependent field. When this field passes through values at which spin energy level splittings become comparable to spin coupling to the optical mode, a cascade of Landau-Zener (LZ) transitions leads to co-flips of spins in exchange for photons of the cavity. We derive exact transition probabilities between different diabatic states induced by such a sweep of the field. I. INTRODUCTION Cavity quantum electrodynamics (cavity QED) is the study of light interaction with discrete quantum systems, such as spins of electrons and atoms, at conditions of significant cou- pling to a single photon. Traditional realizations of cavity QED in atomic physics [1] have recently been extended to numerous solid state applications, including superconductors [2], spins of defects and quantum dots [3] and fiber optics [4]. One of the most fundamental models in cavity QED is the Tavis-Cummings model [5] that describes coupling of Ns two-level systems, which we will call spins for simplicity, to a single optical cavity mode and an external field. The Hamil- tonian of this model is given by [6] Ns(cid:88) Ns(cid:88) Ns(cid:88) H = ωa†a− ∆ σi + i σi + g (a† σ− i + aσ+ i ), (1) i=1 i=1 i=1 where ω and a are, respectively, the frequency and the annihi- lation operator of the cavity mode, and σ± i are ith spin's rais- ing/lowering matrices. Parameter g describes spin couplings to the optical mode. This coupling is assumed to be identical for all spins, as in the original Tavis-Cummings model [5]; ∆ is the Zeeman-like splitting induced by the external field. We include the possibility of spins to experience additional local level splittings described by arbitrary parameters i. We also introduced projection operators σi ≡ (1i + σi z)/2, to spin "up" states. Here 1i is a unit matrix acting in the ith spin subspace, and σi z is the Pauli z-matrix of the ith spin. This model can be physically derived from a more general Dicke model [7] by applying the rotating wave approximation. The latter is well-justified in numerous available realizations of the cavity QED. The model (1) conserves the number of excitations, i.e. the Hamiltonian commutes with the operator of the number of bosons plus the number of spins up: Ne = a†a + σi. (2) i=1 Therefore, it is usually possible to diagonalize the matrix Hamiltonian numerically for small Ns and Ne. However, the Ns(cid:88) size of the phase space grows quickly with Ne and Ns, so fur- ther approximations are usually invoked, such as the degen- eracy assumption that all parameters i are identical. Various limits and extensions of the model (1) at constant parameters have been solved with algebraic Bethe ansatz [8]. Cavity QED is usually discussed in the context of achieving control of quantum states of two level systems and the photon mode. For example, practical goals can be to implement gates for quantum computation with spin qubits or to use spins as emitters of strongly nonclassical photon states. Such a control requires the ability to change parameters of the system in time. Therefore, time-dependent versions of the cavity QED models currently garner considerable scientific attention [9, 10]. Landau-Zener (LZ) transitions [11] induced by a linear sweep of parameters in cavity QED models have been sug- gested as a tool to produce a strongly non-classical light source and entangle qubits in an optical cavity [12, 13]. In the case of the model (1), this corresponds to the assumption that ∆ = βt, (3) with some constant sweeping rate β and t is time. Beyond cavity QED, similar models of linearly driven interacting spin systems have been recently discussed for applications in ultra- cold atomic gases [14, 15], LZ-interferometry [16], quantum dots [17], quantum control [18], and quantum coherence [19]. Considering Eq. (3) at t → ±∞, one can disregard the last sum in the Hamiltonian (1) so that its eigenstates become si- multaneously the eigenstates of the photon number operator, a†a, and spin polarization operators, σi z. We will call such states the diabatic states. Imagine then that at large nega- tive time t, the system is in the state without photons in the cavity mode and with all spins polarized in the diabatic state 0;↑1↑2 . . . ↑Ns(cid:105). As time changes, eventually, spin splittings become comparable to the coupling strength g so that some of the spins flip by emitting photons. Since photons are bosons, the increase of their number enhances the coupling of the pho- ton mode to all the spins, which facilitates new transitions and so on. Such nonlinear effects can be used in practice to induce strongly correlated spin-photon states. Importantly, at large positive time t all spin levels again split out of the resonance so that all transitions terminate. There is a characteristic time ∼ g/β, during which transi- tions between diabatic states are essential. We will assume sufficiently high quality optical cavity, for which this time is 6 1 0 2 n u J 0 1 ] l l a h - s e m . t a m - d n o c [ 5 v 6 3 1 3 0 . 2 0 6 1 : v i X r a much shorter than the time of photon loss, so we can disre- gard dissipation. One can define then the following scattering problem: Given Eqs. (1), (3) and that at t → −∞ the system is in one of the diabatic states of the Hamiltonian H, marked by an index n, what is the probability Pn(cid:48)n ≡ Pn→n(cid:48) to find the system in a diabatic state with an index n(cid:48) at t → +∞? We will call such a scattering problem exactly solvable if there is a procedure to write down any element Pn(cid:48)n of the transi- tion probability matrix P explicitly in terms of the well known special functions of model parameters. The time-dependent problem with the Hamiltonian (1) un- der assumption (3) looks very complex at first view. In order to find scattering probabilities between microstates one has to solve a system of many coupled differential equations with linearly changing time-dependent coefficients. Solutions of such equations show violant oscillations, which decay with long power-law tails in time. In addition, the number of cou- pled equations is growing exponentially with Ns. Therefore, generally, such systems remain challenging to study even nu- merically at Ns > 4. In this article we will argue, however, that the scattering problem for this model is exactly solvable, i.e., we will pro- vide an iterative procedure to obtain transition probabilities between any pair of states in any sector of the model in a fi- nite number of steps. We will also investigate some of the consequences of this solution. The plan of our article is as follows. In section 2, we will describe the connection of the model (1)-(3) to the multistate LZ problem and explore simplest situations that reduce to the previously solved models. In section 3, we will present so- lution of the model (4) in the eight-state sector, which corre- sponds to one of the simplest nontrivial cases of the Hamilto- nian (1) that do not reduce to the already known solvable mul- tistate LZ models. We discuss higher dimensional sectors in section 4. In section 5, we consider the degenerate case i = 0 for all i and derive the full transition probability matrix for the four-state sector, and investigate transition probabilities from the fully polarized initial conditions at arbitrary size of the phase space. In section 6, we will summarize our major find- ings and discuss perspectives to explore other strongly driven explicitly time-dependent interacting quantum problems. II. DRIVEN TAVIS-CUMMINGS MODEL AS A MULTISTATE LANDAU-ZENER PROBLEM (2), we can replace the operator(cid:80)Ns We will start our discussion with slightly rewriting the Hamiltonian in order to reduce the number of independent pa- rameters. Due to conservation of the number of excitations i=1 σi with Ne − a†a in any sector with a constant Ne. The term with Ne commutes then with the rest of the Hamiltonian and can be safely re- moved by a gauge transformation. Hence, we can rewrite the fist two terms in (1) as (ω +βt)a†a. If only the evolution from infinitely large negative to infinitely large positive time values is considered, the redefinition of time, t → t − ω/β, does not affect boundary conditions and, consequently, the scattering probabilities. On the other hand, it removes parameter ω from 2 √ Ns(cid:88) Ns(cid:88) the Hamiltonian. Moreover, rescaling time as t → t/ β in √ the Schrodinger equation, i ψ(t) = H(t)ψ(t), where ψ(t) is β → g and the state vector, and redefining couplings as g/ β → i, parameter β drops out of this equation. There- i/ fore, the Hamiltonian of the problem simplifies: √ H = ta†a + i σi + g (a† σ− i + aσ+ i ). (4) i=1 i=1 Let us first consider Ne = 1. In this case, the available phase space consists of the state with a single photon: Ns + 1(cid:105) ≡ 1;↓1,↓2, . . . ↓Ns(cid:105), (5) and Ns states with a single spin excitation: i(cid:105) ≡ 0;↓1,↓2, . . . ,↑i, . . . ↓Ns(cid:105). (6) In order to enumerate different diabatic states, here and in what follows we will assume the convention in which the state with the minimal possible number of photons has index 1; di- abatic states with the same number of photons are labeled in the decreasing order of their energy at t → −∞ (when disre- garding coupling terms in the Hamiltonian), and states with a larger number of photons have higher indexes than the states with a smaller number of photons. Let 1 > 2 > . . . > N . Following this convention, the Hamiltonian (4) in the basis (5)-(6) has the matrix form:  H =  . 0 g . . . g 1 0 . . . 0 0 2 ... ... ... ... 0 . . . 0 Ns g t g . . . ... g g (7) At this point, we note that the Hamiltonian (7) and the more general Hamiltonian (4) belong to the class of, so-called, mul- tistate LZ models that describe the evolution of a number N of states according to the Schodinger equation with parameters that change linearly with time [20]: i dψ dt = H(t)ψ, H(t) = A + Bt. (8) Here, ψ is the state vector in a space of N states; A and B are constant Hermitian N × N matrices. One can always choose the, so-called, diabatic basis in which the matrix B is diag- onal, and if any pair of its elements are degenerate then the corresponding off-diagonal element of the matrix A can be set to zero by a time-independent change of the basis, that is Bij = δijβi, Anm = 0 if βn = βm, n (cid:54)= m ∈ (1, . . . N ). (9) Constant parameters βi are called the slopes of diabatic levels, nonzero off-diagonal elements of the matrix A in the diabatic basis are called the coupling constants, and the diagonal ele- ments of the Hamiltonian Hii = βit + εi, LZ formula for intersection of two levels: p = e−2πg2 , q = 1 − p. 3 (10) Only semiclassical trajectories that turn or pass forward in time are allowed, so there is, maximum, a single trajectory that connects any given pair of states for the Demkov-Osherov model. The probability of the full trajectory is then the prod- uct of all encountered pairwise passing or turning probabili- ties. If there is no such a semiclassical trajectory that connects two given diabatic states, the corresponding transition proba- bility is zero. For example, if evolution starts with the state Ns + 1(cid:105) with a single photon, then the probability to remain in this diabatic state at t → +∞ is given by Ns(cid:89) PNs+1,Ns+1 = p = pNs. (11) Similarly, e.g, i=1 P21 = 0, P22 = P11 = p, P13 = q2p. (12) We remind that we use the convention in which P21 means the probability of the transition from level-1 to level-2. Consider now another case that reduces to another multi- state LZ-model, which solution is known. Assume that we have an arbitrary minimal number of bosons NB but only two spins. The diabatic basis consists then of only four states: 1(cid:105) = NB ↑1↑2(cid:105), 3(cid:105) = NB + 1;↓1↑2(cid:105), 2(cid:105) = NB + 1;↑1↓2(cid:105), 4(cid:105) = NB + 2;↓1↓2(cid:105). (13) √ Let us denote g1 = g Hamiltonian (4) in the diabatic basis then reads: NB + 1, and g2 = g √ NB + 2. The FIG. 1. Diabatic levels and couplings between them described by the Hamiltonian (4) at Ne = 1. where εi ≡ Aii, are called the diabatic energies. Some of the multistate LZ models have been already studied with the pur- pose to obtain transition probabilities between diabatic states [21–28]. For example, the model (7) is a special case of the exactly solvable Demkov-Osherov model [21]. It is convenient to represent any multistate LZ-model in the time-energy diagram that plots diabatic energies with intersec- tions of diabatic energy levels marked by corresponding pair- wise coupling constants, as we show in Fig. 1 for the model (7). According to this figure, the Demkov-Osherov model de- scribes the case of a single level intersecting a band of parallel diabatic levels. It turns out that, irrespectively of the coupling strength, transition probabilities in the Demkov-Osherov model can be obtained following a simple rule: One should obtain a semi- classical trajectory that connects the initial and final states in the graph in Fig. 1 and assume that probabilities to pass, p, or turn, q, at any level intersection are described by the standard  1 + 2 + NBt g1 g1 0 H = g1 1 + (NB + 1)t g1 0 0 g2 2 + (NB + 1)t g2 (NB + 2)t  . 0 g2 g2 (14) In Fig. 2, we show diabatic levels of this model. Although slopes of the levels are plotted there for the case NB = 0, we note that arbitrary NB would merely increase the slopes of all levels by equal amount, keeping relative slopes between lev- els the same, so geometry and the chronological order of level crossings would not change. The model in Fig. 2 is recognized as a four-state generalized bow-tie model, which exact solu- tion is known [22]. The generalized bow-tie model generally describes the case when several diabatic levels interact with two other parallel levels but do not interact with each other directly. Each of the levels from the first set couples equally to the parallel levels. Moreover, all diabatic levels from the first set must intersect at a single point - right in the middle between the parallel levels. It is easy to check that those con- ditions are satisfied by levels of the Hamiltonian (14). It has been noticed by Demkov and Ostrovsky that the semi- classical procedure that we described above for the Demkov- Osherov model can be equally applied to determine exact solution of the bow-tie model [22]: As in the case of the Demkov-Osherov model, there is a unique path in Fig. 2 that preserves causality and connects any given pair of diabatic states at t → ±∞. Let us denote p1 = e−2πg2(NB +1), p2 = e−2πg2(NB +2), qi = 1 − pi; i = 1, 2. Then the exact form of the and transition probability matrix for the model (14) can be readily   , 4 pointed that transition probabilities between states 1(cid:105) and 3(cid:105) of the model (18) are the same as transition probabilities be- tween states 1(cid:105) and 4(cid:105) in the nondegenerate version of the four-state model (14), while transition probabilities from or to the level 2(cid:105) of the model (18) are given by summation of transition probabilities, respectively, from and to both parallel levels with amplitudes b1, b2 in the nondegenerate version of the model (14). For example, P (3) 12 = P12 + P13, P (3) 23 = P24 + P34, where the index "(3)" marks transition probabilities in the model (17), and probabilities Pij are defined in (15). Hence, the full transition probability matrix for the model (17) reads: P (3) = p2 1 q1(p1 + p2) (1 − p1 − p2)2 q2(p1 + p2) q1(p1 + p2) q1q2 q2(p1 + p2) q1q2 p2 2 (19) where the element P (3) 22 was determined merely from the fact that, for unitary evolution, the transition probability matrix must be doubly stochastic, i.e. the sum of its elements along any column or any raw should be equal to one. Here, we make a new observation that 1 + P (3) 22 = P22 + P33 + P23 + P32, (20) where, on the right hand side, we have the sum of all transition probabilities between parallel levels in the four-state model (14), and the left hand side is the sum of the probability to stay at the central level in the model (17) plus 1. This unit, 1, can be interpreted as the probability, Pb−→b−, to stay in the antisymmetric state with amplitude b− if the evolution starts with b− = 1. We will use this observation to derive transition probabilities in the higher dimensional sectors of the degener- ate version of the model (4). III. EIGHT DIMENSIONAL SECTOR A. Parameters of the Model and later on: (cid:112) Let us first introduce the notation to be used in this section gn = g NB + n, pn = e−2πg2(NB +n), qn = 1 − pn, (21) where n is a positive integer, and NB is the minimal number of photons that can be in the given sector of the model, i.e. NB = Ne − Ns for Ne ≥ Ns. Formally, NB is an integer parameter of the model (4), however, our subsequent discus- sion will apply equally to continuation of this parameter to any positive real value. In this sense, our following discussion goes beyond the scope of the model (1). We also note that neg- ative integer values of NB are also allowed as far as Ne > 0. In such a case, only the sector of states coupled by positive real couplings (21) should be considered physical. We will focus only on sectors with Ne ≥ Ns, although treatment of FIG. 2. Diabatic levels and couplings between them described by the Hamiltonian (14) at NB = 0. written using Fig. 2 and semiclassical rules:  p2 1 P = q1p1 p2q1 q1q2 q2 p2q2 2 p1p2 p1q2 p2 2 q1p2 p1p2 q2 p1q1 1 q1q2 p1q2 p2q2  . (15) One more important observation made by Demkov and Ostrovsky in [22] is that the degenerate case of a general- ized bow-tie model, which in our case corresponds to 1 = 2 = 0, reduces to a model, which exact solution can be trivially obtained from the knowledge of the transition prob- ability matrix of the nondegenerate case, such as (15). Let a(t), b1(t), b2(t), c(t) be the amplitudes of states, respectively, 1(cid:105),2(cid:105),3(cid:105),4(cid:105) defined in (13). In the degenerate case, i.e., at 1 = 2 = 0, the Schrodinger equation for these amplitudes with the Hamiltonian (14) corresponds to a set of four coupled differential equations: i a = NBta + g1(b1 + b2), ib1 = (NB + 1)tb1 + g1a + g2c, ib2 = (NB + 1)tb2 + g1a + g2c, i c = (NB + 2)tc + g2(b1 + b2). √ Let us now introduce new variables b± = (b1±b2)/ 2. Then the amplitude b− completely decouples while the remaining amplitudes satisfy a new differential equation: (16) (17) (18)  = H (3)  a  , b+ c b+ c i d dt  a  NBt √ g1 where H (3) = g1 2 √ √ 2 (NB + 1)t  . √ 0 2 g2 0 g2 2 (NB + 2)t Equation (17) corresponds to a new three-state LZ model in which all diabatic levels intersect in one point at t = 0. This model is also exactly solvable, and it is known as a stan- dard bow-tie model [23]. Comparing exact solutions for stan- dard and generalized bow-tie models, Demkov and Ostrovsky cases with Ne < Ns is similar. We will also introduce two parameters: N ≡ 2Ns, (22) and a nonnegative integer parameter M (n) that is the num- ber of spins "down" in the state number n, e.g., M (1) = 0, M (2) = 1, M (N ) = Ns, e.t.c.. The meaning of N is the number of independent connected states in the sector with Ns spins-1/2 when Ne ≥ Ns. 5 B. Diabatic and Adiabatic Levels The nearest sector of the model (4) with more than two spins has Ns = 3, which, for Ne ≥ Ns, corresponds to di- mensionality of the model N = 8 with the Hamiltonian:  . (23)  H = 1 + 2 + 3 g1 g1 g1 g1 0 0 0 0 1 + 2 + t 0 0 g2 g2 0 0 g1 0 1 + 3 + t g1 0 0 0 g2 0 g2 0 2 + 3 + t 0 g2 g2 0 0 g2 g2 0 1 + 2t 0 g2 0 g2 0 0 0 g3 2 + 2t 0 g3 0 0 g2 g2 0 0 0 0 0 0 g3 g3 3 + 2t g3 3t g3 Here, for simplicity, we dropped the term NBt1 from the main diagonal of the matrix (23) because this term can be removed by a time-dependent change of phases of all diabatic state am- plitudes [6, 20]; therefore it does not influence transition prob- abilities and can be disregarded. Note that the definition of couplings (21) still depends on NB. Our major method to study models like (23) is based on the observation that, surprisingly, all known exactly solvable models of the type (8) with a finite number of interacting states [20–22, 24, 27, 28] have exact solutions for the scattering ma- trix that coincide with predictions of the independent cross- ing approximation, which we described for cases of Demkov- Osherov and bow-tie models in previous section. Moreover, all such solvable models have two properties [27, 28]: (i) the absence of the dynamic phase effect on transition probabilities in the semiclassical framework, and (ii) the existence of exact crossings of eigenenergy levels (i.e. instantaneous eigenvalues of the Hamiltonian as func- tions of time) that correspond to each intersection of diabatic levels (i.e. diagonal elements of the Hamiltonian in the dia- batic basis) without direct couplings in the diabatic level dia- gram. Here by the dynamic phase we mean the trivial phase φdyn = − [βk(t)t + εk(t)] dt, (24) (cid:90) ∞ −∞ where k(t) is the index of the level along a semiclassical tra- jectory. Its time-dependence follows from the possibility of switching diabatic level indexes at level crossing points. LZ-integrability conditions (i)-(ii) have already been used by one of us to search for new solvable multistate LZ sys- tems. There are already three different such models that have been found and solved in [27, 28] despite rigorous mathemat- ical proof for their integrability is still missing. The major observation of the present article is that invariant sectors of FIG. 3. Diabatic levels and couplings between them described by the Hamiltonian (23). For convenience, equal terms are added to all slopes of diabatic levels: H → H − 1.5t1. This is equivalent to a change of time-dependent phases of state amplitudes, which does not change transition probabilities. Blue arrows demonstrate the unique semiclassical path connecting the state 6(cid:105) at t → −∞ to the state 5(cid:105) at t → +∞. the model (4), such as the 8 × 8 model (23), in the nonde- generate case (e.g., 1 > 2 > 3), also satisfy conditions of integrability (i)-(ii); therefore, corresponding matrices of transition probabilities can be obtained exactly by applying the independent crossing "approximation". The structure of diabatic levels of the model (23) is shown in Fig. 3. By examining all possible semiclassical trajecto- ries, one can conclude that condition (i) is trivially satisfied because there is always only a single semiclassical path con- necting any pair of two states, i.e., there is no interference of different trajectories connecting the initial and final states. Therefore, the dynamic phase factorizes and cancels upon tak- ing the absolute value of the transition amplitude. In order to test condition (ii), we diagonalized the Hamilto- nian (23) and obtained its eigenvalues as functions of time, as 6 C. Solution of Eight-State Model Consider the unique allowed semiclassical path from the level-6 to the level-5 that we mark by blue arrows in Fig. 3. FIG. 4. (Color online) Eigenvalues of the Hamiltonian (23) shown as functions of time (eigenenergy levels). Parameters are: NB = 1, 1 = 2.4, 2 = 0, 3 = −1., g = 0.35. For convenience, equal terms are added to the main diagonal of the Hamiltonian: H → H − 2.5t1. shown in Fig. 4 for NB = 1 and g = 0.35. At such couplings, these functions are already quite distorted from the graph of diabatic levels in Fig. 3. However, one can see visually that Fig. 4 has ten exact crossing points, which is just the same as the number of diabatic level crossings in Fig. 3 that are not marked by direct pairwise couplings. Certainly, a Hamiltonian eigenvalue level crossing that is visually looking exact may hide an invisible mini-gap due to higher order processes. In order to check that this is not the case, we developed a numerical algorithm to estimate the min- imal distance between a pair of eigenenergy levels. Accord- ing to this algorithm, we increase time near a crossing point in small steps δt and trace changes in the difference of eigenen- ergies, En−En+1, with nearby indexes. We keep going until this difference stops decreasing. Then we make one step back and decrease the size of the step by two orders of magnitude and then repeat the process. After several iterations, we could estimate the minimal difference En − En+1 for visually ex- act crossings to be less than 10−14 at parameters chosen in Fig. 4. This leaves no doubt that the crossing points in Fig. 4 are truly exact. Therefore, condition (ii) is satisfied. We also observed the phenomenon that has not appeared in previous studies of integrable LZ-systems with conditions (i)- (ii) satisfied. At increasing values of coupling parameter g, some of the pairs of exact crossing points may approach each other and, eventually, annihilate with each other. For example, by increasing g for the model (23) further, only six out of ten exact crossing points survive. It seems, however, that this fact does not break integrability of the model. Consequently, condition (ii) should be reformulated by requiring only that at sufficiently small but finite values of couplings diabatic level crossings without direct transitions should give rise to exact crossings of eigenenergy levels. FIG. 5. (Color online) Comparison of numerical calculations of transition probabilities with semiclassical solution (25). All dis- crete points correspond to results of direct numerical solution of the Schrodinger equation with the Hamiltonian (23) starting at t = −1000 and changing time up to t = 1000 with a time step size dt = 0.0001. Solid lines are theoretical predictions of Eq. (25). In (a) and (b), independence of transition probabilities of energy level splittings is tested for 1 = 0.5, 2 = 0, 3 = −. Other parameters are explained in figures. In (c-f), transition probabilities are shown as functions of coupling constant g with different initial conditions at parameter values 1 = 0.5, 2 = 0, 3 = −1. For better visibil- ity, not all possible transition probabilities are shown at given initial conditions. Since LZ integrability conditions (i)-(ii) are satisfied for the full model (23), the exact value of the probability of the tran- sition from the level-6 to the level-5 is the product of all pair- wise LZ transition probabilities encountered during passing or turning at level intersections with nonzero couplings along the semiclassical path. In this case, for example, P56 = p2q2 3. Similarly, we can read probabilities of transitions between any pair of states: (a)(b)(f)(c)(d)(e) P = q1p2 1 p1p2 2 q2 1p2 q2 1p1 p2 p3 2q1 p2q1p1 1 q2 q2 q1p2 2p3 2p2 2 p2 p2q2 p1q1p2 2p1 2 p2 p2 p2q2 2p1 1q1 1 q2q2 q1q2p3 q1q2p3 q2p2p3 3 q2p3p2 q1p2q2 p1p2q2 q2 p2 p1q1q2 1q2 2q2 q1q2q3 p1q2q3 q2p2q3 q2q3p3 p2p1q2 q3 2 D. Numerical Check In order to test our predictions we solved the Schrodinger equation for evolution with the Hamiltonian (23) numerically. We used the algorithm that was described in the supplemental material for Ref. [29]. According to it, quantum evolution is simulated from a large negative time and proceeds to a large positive time with very small time steps of a size dt. At each time step, the state vector is updated with an evolution opera- tor, which is approximated by (cid:16)1 − i H(t)dt/2 (cid:17)(cid:16)1 + i H(t)dt/2 (cid:17)−1 U (t, dt) = , (26) where H(t) is the time-dependent matrix Hamiltonian (23), and 1 is the unit matrix. Operator U(t) is unitary, and up to the order of o(dt2), it coincides with the true evolution operator, H(τ )dτ , where T is the time-ordering operator. T e−i(cid:82) t+dt t Figure 5 shows excellent agreement between numerical results (discrete points) and theoretical predictions (solid curves) for the 8 × 8 Hamiltonian (23) at all considered pa- rameter values. One important theoretical prediction of the solution (25) is that all transition probabilities are indepen- dent of parameters i, as far as we preserve their relative order In Figs. 5(a) and (b), we test this 1 > 2 > . . . > Ns. prediction by varying these parameters. Importantly, when they are close to zero, i.e., become much smaller than the coupling constants gi, the independent crossing approxima- tion cannot be justified, however, numerical solution still does not show any sign of deviations from analytical predictions of Eq. (25). Figures 5(c-f) show transition probabilities as func- tions of the coupling g for different NB and different initial state vectors. Again, perfect agreement of numerical results with Eq. (25) at arbitrary values of parameters confirms our expectation that Eq. (25) provides, actually, the exact nonper- turbative solution of the multistate LZ problem (23). We also tested solution (25) at "unphysical" non-integer values of pa- rameter NB (Fig. 6(b)) and found no difference between the theory and numerical results. In this sense applicability of our solution extends beyond the model (4). The described algorithm is easy-to-realize and it is highly effective when the number of spins is small, Ns < 4. In such cases, it produces estimates of transition probabilities with ac- curacy of two significant digits for twenty different parameter values in several minutes. However, in addition to exponen- tially growing number of equations at higher values of Ns, slow saturation of the solution at large time t and strong oscil- lations in time require increasingly smaller sizes of time-steps q3 2 p2p3q2 q2q1p1 q2p2q1 p3q2q1 q3q2q1 q2 p2 p3q3q2 2q2 3q2 q2p1p2 p3p2q2 q3p2q2 q2q2 q2p2p1 p3q2p1 q3q2p1 1 p2 p2q2 p2 3q3 2p3 3 p2 p2q2 p3q3p2 2p3 2 q2 q2 q3p2 2p1 2p2 2 p3 p2 p2q3p3 2q3 3 q2 3p3 q2 3p2 p2 2p3 q3p2 3  . 7 (25) and a larger full time interval for simulations, making such simulations unreasonably long to persue. It is likely, however, that this numerical approach can be improved by employing interpolation methods that can estimate the solution saturation value by tracing decay of the oscillation envelope, thus reduc- ing the simulation time interval considerably. Finally, we note that when a single spin-photon coupling g in (1) was replaced in our studies by a set of different param- eters, γi (cid:54)= γj for i (cid:54)= j, that described couplings of different spins to the photon mode, we found that integrability condi- tion (ii) was not satisfied: Minigaps opened up at intersections of diabatic levels without direct couplings. In this case, our numerical studies showed substantial deviations (not shown) from analytical predictions, such as Eqs. (15) and (25). These tests provided additional evidence that the agreement of semi- classical analytical predictions and numerical results are not typical for the considered problem and parameter values, ex- cept when semiclassical predictions become for some reason exact. IV. HIGHER DIMENSIONAL SECTORS Explicit solution development for cases with ever higher number of spins is an increasingly tedious task. We checked numerically that our observations for Ns = 3 work equally well at Ns = 4. For example, Fig. 6 shows eigenenergy levels in the eleven-dimensional sector with Ne = 2. As expected, there are plenty of exact crossing points to match all cross- ings of diabatic levels without direct couplings. The case with Ns = 4 generally leads to the sector dimensionality N = 16 (at Ne ≥ Ns), which would not be illuminating to develop in this article fully. We will show in the following section that the degenerate limit, i = 0, leads to a much smaller subspace dimensionality that we will be able to test numerically up to Ns ∼ 20. Here, instead, we would like to examine already ob- tained results and summarize the common properties that we believe are valid for any higher dimensional non-degenerate case. Let us compare solution (25) for the case Ns = 3 with (15) for Ns = 2. We can also extend this set by the case Ns = 1 that corresponds just to a two-state LZ model with P11 = P22 = p1 and P12 = P21 = q1. The following properties appear to be common for all three cases: (a) There is an algorithm to obtain an explicit expression for all transition probabilities by drawing the time-energy di- agram for diabatic energy levels, such as in Figs. 1-3. There 8 For example, if we know a compact expression for transition probabilities from the state n = 1, i.e., with a minimal num- ber of initial bosons, to any other state, then we can use the symmetry (28) to find a similar expression for transition prob- abilities from the state with index N, i.e. with a maximal number of initial bosons, in the same invariant sector of the model (4) at Ne ≥ Ns . (e) While writing a general closed expression that describes all possible transition probabilities is an interesting combina- torial problem that we will leave open, it is possible to guess what are transition probabilities for the physically most im- portant case when the initial level has index n(cid:48) = 1, i.e., when the evolution starts with the state having a minimal number of photons in the cavity mode (all spins up). By comparing cases with Ns = 1, 2, 3, we introduce a conjecture that, for arbitrary Ns value, the probability of a transition from level n(cid:48) = 1 to level n has the form Pn1 = qk pir r , (29)  M (n)+1(cid:89) r=1 M (n)(cid:89) M (n)+1(cid:88) k=1 r=1 where ir = Ns − M (n), (30) and where ir are nonnegative integer numbers. Their zero val- ues are allowed. Equation (30) does not specify the values of ir completely but this is not needed if we are interested only in the probability to find a specific number of flipped spins, m, at the end of the evolution because the number of all possible combinations of ir for all levels n with M (n) = m is equal to the number Ns!/m!(Ns − m)! of all levels with polarization m. Hence, all possible values of ir contribute to the probabil- ity to find polarization m after the evolution. One can prove this by noticing that the number of possibilities to organize products of m + 1 probability parameters p1, . . . , pm+1 into all possible products made of Ns−m of them, including terms of higher powers of the same probability parameter, can be mapped to the problem of allocating m+1 bosons into N −m different discrete states, which results in Ns!/m!(Ns − m)! different possibilities. Let P1→m be the sum of Ns!/m!(Ns − m)! probabilities of transitions from level-1 to all states with M (n) = m. Then, after using the property (c), we find (cid:32) m(cid:89) (cid:33) Ns−m(cid:88) P1→m = qk δi1+...+im+1,Ns−m pir r . k=1 i1,...,im+1=0 (31) A special case is the probability of the transition from level-1 to level-N, i.e., from the state with all spins fully polarized "up" into the state with fully polarized "down" spins. In such a case m+1(cid:89) r=1 PN 1 = P1N = qk. (32) Ns(cid:89) k=1 FIG. 6. (Color online) Eigenenergy levels of the Hamiltonian (4) in the sector with Ns = 4 spins and Ne = 2 excitations, which corresponds to totally eleven interacting states. Parameters are: 1 = 2.5, 2 = 1, 3 = −1., 4 = −2.; g = 0.32. For convenience, equal terms are added to the main diagonal of the Hamiltonian: H → H − 3t1. is then a unique path connecting any two diabatic states at t → ±∞, such as the path from level-6 to level-5 in Fig. 3, which satisfies conditions of causality and the assumption that diabatic levels can be switched only at level crossings with a nonzero direct coupling between corresponding dia- batic states. After identifying such a path, the transition prob- ability of the full trajectory is given by the product of all pass- ing and turning LZ probabilities at encountered level intersec- tions along this path. (b) A direct consequence of (a) is that we can write down the general formula for all elements along the main diago- nal of the transition probability matrix. These elements corre- spond to the probabilities of events that the system remains at the initial diabatic state after the evolution. Since there is always a unique semiclassical path for any transition in the model (4), we can identify such a path in the given case with the path at which the system never turns to other dia- batic levels. If the initial level has the number m of spins "down", then there are m diabatic states with indexes i such that M (i) = m − 1 that are directly coupled to the initial state with coupling gm, and there are (Ns − m) states with M (i) = m + 1 coupled to the initial diabatic state with cou- pling gm+1. Hence: Pnn = (pM (n))M (n)(pM (n)+1)Ns−M (n), (27) where function M (n) is defined after Eq. (22). (c) At Ne ≥ Ns, all elements of the transition proba- bility matrix are products of exactly Ns factors pi or qi, i = 1, . . . Ns. Expressions, in terms of pi and qi factors, for transition probabilities from any given level to all other states are different from each other. (d) The transition probability matrix, at Ne ≥ Ns, has a symmetry, namely, let us consider elements of this matrix as functions of pi and qi: Pnn(cid:48) = Pnn(cid:48)(p1, . . . , pNs , q1, . . . qNs ), then Pnn(cid:48)(p1, . . . , pNs, q1, . . . , qNs) = PN−n+1,N−n(cid:48)+1(pNs , . . . , p1, qNs , . . . , q1). (28) V. SINGLE SPIN COUPLED TO PHOTON MODE 9 In this section, we explore the degenerate case: i = 0 for all i ∈ 1, . . . , Ns. In the Schrodinger equation for amplitudes, any multiplet with m spins "up" leads to Ns!/m!(Ns − m)! equations that collapse only to a single equation for the sym- metric combination of amplitudes, while equations for other orthogonal state amplitudes in this multiplet completely de- couple. After decoupling non-symmetric states, the remain- ing set of equations is equivalent to quantum mechanical evo- lution with the following Hamiltonian: (cid:16) a† S− + a S+(cid:17) Hd = ta†a + g , (33) where S± are spin raising/lowering operators for a spin size S = Ns/2. This time-dependent spin-boson model has at- tracted lots of attention recently due to applications in the the- ory of dynamic transitions through a quantum critical point and applications to experiments on molecular condensate cre- ation by a passage of an ultracold atomic gas through a Fesh- bach resonance. A number of approximate methods have been used to study transition probabilities in the model (33), includ- ing Keldysh technique and adiabatic approximation [30, 31]. It is also known that some of the simpler solvable multistate LZ models correspond to specific limits of the model (33) [32, 33]. Therefore, behavior of the model (33) is considered well understood, at least in the limit of a large spin (S (cid:29) 1). Exact solution can, however, shed light on some cases that are hard to explore with approximate techniques, such as moder- ate spin sizes and probabilities of rare events. Here, we will use the same notation, as in previous section, to mark different diabatic states of the model (33), namely, the index n ∈ (1, . . . , 2S +1) will mark all diabatic levels starting from the lowest number of initial bosons. Spin polarization in the model (33) is related to the index n as Sz = S − n + 1. Let us first work out the case with S = 3/2 in detail. The Hamiltonian (33) then describes the sector with four states that are obtained from the eight-state sector (23) for the nondegen- erate model in the limit 1 = 2 = 3 = 0. Let 1(cid:105), . . . ,8(cid:105) be the eight diabatic states in the nondegenerate case. In the de- generate limit, diabatic levels with indexes n = 2, 3, 4 become degenerate with each other, as well as levels with indexes n = 5, 6, 7. As in the case of the four-state bow-tie model, let us change the diabatic basis and introduce new states: u+(cid:105) ≡ 1√ 3 u1(cid:105) ≡ 1√ 2 u2(cid:105) ≡ 1√ 6 v+(cid:105) ≡ 1√ 3 v1(cid:105) ≡ 1√ 2 v2(cid:105) ≡ 1√ 6 (2(cid:105) + 3(cid:105) + 4(cid:105)) , (34) (2(cid:105) − 4(cid:105)) , (2(cid:105) − 23(cid:105) + 4(cid:105)) , (5(cid:105) + 6(cid:105) + 7(cid:105)) , (5(cid:105) − 7(cid:105)) , (5(cid:105) − 26(cid:105) + 7(cid:105)) . (35) FIG. 7. (Color online) Transition probabilities for the degener- ate case with total spin S = 3/2. Different sub-figures corre- spond to different initial conditions. Discrete points correspond to results of numerical simulations of the evolution with the Hamil- tonian (36) from t = −1000 to t = 1000 with a time step size dt = 0.0001. Solid lines are theoretical predictions of Eq. (39). In all cases, NB = 1. In the basis of states 1(cid:105), u+(cid:105), v+(cid:105), 8(cid:105), the effective Hamil- tonian of the degenerate limit of the model (23) reads  0 √ 3g1 0 0 H (4) d = √ 3g1 t 2g2 0 0 2g2 √ 2t 3g3 √ 0 0 3g3 3t  . (36) One can check by using the spin operator algebra and the def- inition of parameters (21) that the Hamiltonian (36), indeed, corresponds to the sector S = 3/2 of the Hamiltonian (33). Our steps to derive transition probabilities in the model (36) will be analogous to derivation of Eq. (17) for the case of the three state bow-tie model, which also corresponds to the sec- tor S = 1 of the model (33). Thus, transition probabilities to/from the levels with extremal slopes (Sz = ±3/2) from/to states u+(cid:105) or v+(cid:105) are given by the sum of transition proba- bilities to all corresponding diabatic states in the model (23) that become degenerate. For example, in order to find the transition probability from level-2 to level-1 of the model (36) one should take the sum of transition probabilities from levels- 2,3,4 to level-1 of the model (23): P (4) 12 = q1(p2 2 + p1p2 + p2 1). In the case of transitions between u+(cid:105) and v+(cid:105), we have to include information about probabilities to stay/leave to/from the decoupled states. The state u1(cid:105) is coupled only to the state v1(cid:105) with coupling g2, and u2(cid:105) is coupled to v2(cid:105) also with (a)(b)(c)(d)coupling g2. Corresponding transition probabilities between those pairs of states are described by the standard 2-state LZ transition probability: Pu1,u1 = Pv1,v1 = Pu2,u2 = Pv2,v2 = p2, Pv1,u1 = Pv2,u2 = Pu1,v1 = Pu2,v2 = q2. (37) (38) By analogy with derivation of the probability to stay in the second level in the degenerate three-state bow-tie model that we described at the end of section 2, we equate the sums of transition probabilities between sets of "becoming- degenerate" levels in the basis 1(cid:105), . . . ,8(cid:105) of the model (23) and the sums of corresponding transition probabilities in the basis (34)-(35): P (4) 22 + Pu1,u1 + Pu2,u2 = (cid:88) i,j=2,3,4 Pij, 10 where Pij are elements of the eight-state probability matrix (25). Similarly, P (4) 32 + Pv1,u1 + Pv2,u2 = Pji. i=2,3,4 j=5,6,7 We can now summarize all such results in a single probability matrix for the model (36), which we test numerically in Fig. 7: (cid:88) (cid:88)  p3 1 + p1p2 + p2 1 q1(p2 2) A q1q2(p1 + p2 + p3) C q1q2q3 q1(p2 1 + p1p2 + p2 2) q1q2(p1 + p2 + p3) q1q2q3 q3q2(p3 + p2 + p1) q3(p2 3 + p2p3 + p2 C B q3q2(p3 + p2 + p1) q3(p2 3 + p2p3 + p2 2) 2) p3 3  , (39) P (4) = where C = q2(p2 2 + 2p2(p3 + p1) + p1p3 + q2 1 + q2 2 + q2 3 − 2), A = 3p2 2p1 + p1q2 1 + p3q2 2 + 2p2(q2 1 + q2 2 − 1), B = 3p2 2p3 + p1q2 2 + p3q2 3 + 2p2(q2 2 + q2 3 − 1). Here, we would like to point to one useful symmetry of the transition probability matrix (39): ij = P (4) P (4) ji . (40) This property is, actually, valid for any higher dimensional sector of the model (33) because, in the matrix form, this model belongs to the class of multistate LZ-chain models, in which only couplings between states with nearest indexes are nonzero. Indeed, for spin S, the only independent nonzero matrix elements of the Hamiltonian (33) are = (n − 1)t, and (cid:17) (cid:112)S(S + 1) − (S − n + 1)(S − n). H (2S+1) (cid:16) (cid:17) (cid:16) nn d H (2S+1) d = gn n,n+1 The symmetry (40) was rigorously proved in [34] for any LZ- chain model. The worked out examples for S = 1 and S = 3/2 demon- strate that there is a simple algorithm to obtain the transition probability matrix for arbitrary S. Symmetric combinations of "becoming-degenerate" diabatic states and extremal slope states of the model (4) correspond to the diabatic states of the desired sector of the model (33). For the latter, transition probabilities become combinations of ones in the nondegen- erate model (4) with Ns = 2S and transition probabilities of the lower-S sectors of the degenerate model. While there is no doubt that it is possible to automate the process of deriving explicit expressions for the full transition probability matrices and fully solve a few higher-dimensional sectors, the number of terms involved is growing very quickly with S. Instead, in the rest of this section, we will focus on the physically most interesting case when the system starts with a fully polarized state Sz = ±S, i.e., n = 1 or n = 2S + 1. First, we note that Eq. (31) applies equally to the model (33), where we can identify parameter Ns with 2S and parameter m with an index n − 1 of the level of the Hamiltonian (33) to which the transition is considered. So, in terms of parameters of the model (33), we have (cid:32)n−1(cid:89) (cid:33) 2S+1−n(cid:88) P1→n = qk δi1+...+in,2S+1−n pir r . k=1 i1,...,in=0 (41) Moreover, using the symmetry (28), we can write a formula for the transition from the state number 2S + 1: n(cid:89) r=1 (cid:32) 2S(cid:89) (cid:33) × qk k=n δi(n−1)+...+i2S ,n−1 P2S+1→n = n−1(cid:88) i(n−1),...,i2S =0 (42) 2S(cid:89) r=n−1 pir r . In Figs. 8 and 9, we compare theoretical predictions of Eqs. (41)-(42) with our results of the direct numerical solu- 11 (Color online) Transition probabilities P1→n ≡ P (2S+1) FIG. 8. from the state n = 1 (Sz = S) to other states for evolution with the Hamiltonian (33) in the sector with 16 states (S = 15/2) at different coupling strengths. Theoretical predictions (dots on vertical lines) of Eq. (41) are compared to results of direct numerical simulations (empty boxes) Time evolution is from t = −2000 to t = 2000 with time step dt = 0.00005. In all cases, NB = 0. n1 FIG. 9. (Color online) Comparison of theoretical predictions (dots on vertical lines) of Eq. (42) and numerical simulations (empty boxes) for evolution with the Hamiltonian (33) in the sector with 18 states (S = 17/2) at different coupling strengths. Time evolution is from t = −2000 to t = 2000 with time step dt = 0.00005. In all cases, NB = 0. Evolution starts with the fully polarized spin state Sz = −S with 2S bosons, which corresponds to n = 2S + 1 = 18. tion of the Schrodinger equation for two different sectors of the model (33). Since the matrix Hamiltonian of the model (33) has nonzero elements only along the main and the next to the main diagonals, we can accelerate our simulations by ap- plying the leap-frog algorithm to invert matrices, which scales as N Log(N ) with the size N of the matrix. This allows us to check our theoretical predictions for considerably larger phase space sizes than in previous sections. Therefore, Figs. 8 and 9 serve also as a test of our conjectures made about higher dimensional sectors of the model (4). Figure 8 corresponds to the initially polarized state without bosons. Subfigures (a)-(d) show results at progressively larger values of the coupling parameter g. At moderate values of coupling, there is a phase (case (b-c)) at which practically all states have substantially nonzero probabilities to be populated at the end of the evolution. This is the result of the positive feedback induced by the emission of bosons, which leads to amplification of quantum fluctuations. Figure 9 shows transition probabilities in the opposite pro- cess when the spin is initially fully polarized "down" but there are 2S photons that it can absorb. In this case, very small val- ues of g are sufficient to produce substantial transition proba- bilities, as shown in Fig. 9(a). By increasing g, the final dis- tribution has a shape of a localized wavepacket that moves towards the state with zero number of bosons with increasing g (Figs. 9(b-c)). Eventually, at large g, the final distribution becomes sharply peaked near the state with Sz = S. VI. DISCUSSION We explored the cavity QED model of spins interacting with a single optical mode in a linearly time-dependent field. This model and its degenerate version (33) have already en- countered in the literature with diverse applications, from cav- ity QED to transitions through the Feshbach resonance in molecular condensates [30–32]. Certainly, existence of an ex- act solution can help to extend the discussion of these applica- tions. For example, the semiclassical limit of the model (33) corresponds to an explicitly time-dependent classical Hamil- tonian dynamics with connections to the Painlev´e (PII) equa- tion [31]. It would be insightful to explore whether the semi- classical limit of our solution can lead to the exact solution of such a classical system. Despite a number of fully solvable multistate LZ models has been known, all such models describe relatively simple situations, such as interactions of many uncoupled to each other levels with a single level [21, 23]. Many known solvable cases are, in fact, reducible in the sense that they can be de- coupled into a set of independent Demkov-Osherov, bow-tie, or two-state LZ models by applying simple well-characterized symmetry transformations [24]. Only recently, conditions of integrability, which we discussed in section 3B, have been used by one of us to uncover a few more relatively small-size solvable models [27, 28]. Perhaps, the most important achievement of our work is 12 the demonstration that systems with very complex many- body interacting Hamiltonians can be solvable at conditions of arbitrarily strong nonadiabatic driving of model parame- ters. Surprisingly, our final solution for the transition prob- ability matrix appears to be relatively simple; it is expressed only via elementary functions. This is in contrast to the time- independent case, whose solution generally can be written only implicitly via the roots of nonlinear algebraic equations. Moreover, analysis of the independent crossing approxima- tion, which we used to obtain transition probability matri- ces in this article, appears to be even simpler than in models [27, 28] that were previously derived using the same integra- bility conditions. Namely, in the case of the model (4), there is no interference between different semiclassical trajectories that connect different initial and final states in the independent crossing approximation. This strongly suggests that the class of fully integrable multistate LZ models can be substantially extended and it is important to continue developing the theory of quantum integrability of multistate LZ systems in order to obtain the means to deal with strongly interacting nonstation- ary problems in quantum mechanics. Several questions need to be resolved within this theory. Until now, there is no mathematically rigorous proof of the LZ integrability conditions introduced in [27, 28]. Their validity is supported only by considerable amount of numerical tests for models that have been solved by using such conditions and by a number of models that have been solved analytically by alternative methods. It is unclear whether there are counterex- amples or further restrictions on these conditions. They do not provide a direct path to derive new solvable models. Rather they serve as a test that a model should pass in order to be solvable by the semiclassical ansatz. The most important clue that LZ integrability conditions provide is that they request the existence of exact adiabatic energy crossing points that must encounter when one is vary- ing a single model parameter, which is also often the prop- erty of quantum integrable systems solvable by the algebraic Bethe ansatz. Recently, the reasons for the presence of such crossing points were questioned, and a new type of a dynamic symmetry based on the existence of nontrivial "commuting partner Hamiltonians" was suggested to explain it [35]. Both Demkov-Osherov and bow-tie models have been shown to possess such commuting partners [36]. This observation may be used to uncover new candidate LZ models that can be ex- actly solved. Finally, we would like to mention that the Stokes phenomenon has peculiar properties in multistate LZ systems [34, 37]. Its relation to the LZ-integrability is another peace of the puzzle. Acknowledgment. Authors thank Avadh Saxena for useful discussions. The work was carried out under the auspices of the National Nuclear Security Administration of the U.S. De- partment of Energy at Los Alamos National Laboratory under Contract No. DE-AC52-06NA25396. Authors also thanks the support from the LDRD program at LANL. [1] A. D. Cimmarusti, Z. Yan, B. D. Patterson, L. P. Corcos, L. S. Orozco, and S. Deffner, Phys. Rev. Lett. 114, 233602 (2015) [2] J. M. Fink, et al, Phys. Rev. Lett. 103, 083601 (2009); M. Feng, et al, Nature Comm. 6, 7111 (2015) Nikitin, S. Y. Umanskii, "Theory of Slow Atomic Collisions", Springer Series in Chemical Physics (Book 30), Springer (2011) [12] D. C. McKay, R. Naik, P. Reinhold, L. S. Bishop, D. I. Schuster, [3] D. Solenov, S. E. Economou, and T. L. Reinecke, Phys. Rev. B Phys. Rev. Lett. 114, 080501 (2015); 87, 035308 (2013) [4] D. Roy, Scientific Reports 3: 2337 (2013) [5] M. Tavis and F. Cummings, Phys. Rev., 170, 379 (1968) [6] Physically more justified Hamiltonian may also contain terms of the type ak(t)1k, where 1k is the unit operator in the space the kth spin, and ak(t) is an arbitrary time-dependent parameter. Such terms split relative energies of different spins, however, being commuting with each other and with the rest of the Hamiltonian (1), they can be removed from the Schrodinger equation by a trivial gauge transformation: ψ(t) → exp (cid:16)−i(cid:82) t −∞ dt(cid:48) (cid:80) k ak(t(cid:48))1k [7] R. H. Dicke, Phys. Rev. 93, 99 (1954) [8] N. M. Bogolyubov, J. Math. Sciences, 100, 2051 (2006); Phys. ψ(t). (cid:17) Rev. Lett. 107, 100401 (2011) [9] A. S. Parkins, P. Marte, P. Zoller, and H. J. Kimble, Phys. Rev. Lett. 71, 3095 (1993); V. M. Bastidas, C. Emary, B. Regler, T. Brandes, Phys. Rev. Lett. 108, 043003 (2012) [10] J. Chena, L.F. Wei, Physics Lett. A, 379, 2549 (2015); T.i Fujii, S. Matsuo, N. Hatakenaka, S. Kurihara, and A. Zeilinger, Phys. Rev. B 84, 174521 (2011); A. V. Dodonov, J. Phys. A: Math. Theor. 47, 285303 (2014) [11] H. Nakamura, "Nonadiabatic Transition", World Scientific Publishing Company, 2nd Edition (2012); M. S. Child, "Molec- ular Collision Theory", Dover Publications (2010); E. E. [13] K. Saito, M. Wubs, S. Kohler, P. Hanggi, Y. Kayanuma, Euro- phys. Lett. 76, 22 (2006); G. Sun, et al., Appl. Phys. Lett. 94, 102502 (2009); D. Zueco, P. Hanggi, S. Kohler, New J. Phys. 10, 115012 (2008); M. Wubs, S. Kohler, P. Hanggi, Physica E 40, 187 (2007); J. Larson, EPL 107, 30007 (2014) Phys. Rev. A 73, 013823 (2006) [14] Y-A. Chen, S. D. Huber, S. Trotzky, I. Bloch, and E. Altman, Nature Phys. 7, 61 (2011); S F. Caballero-Benitez and R. Pare- des, Phys. Rev. A 85, 023605 (2012); W. Tschischik, M. Haque, R. Moessner, Phys. Rev. A 86, 063633 (2012) [15] V. A. Yurovsky, A. Ben-Reuven, and P. S. Julienne, Phys. Rev. A 65, 043607 (2002); B. E. Dobrescu, and V. L. Pokrovsky, Phys. Letters A 350, 154 (2006); M. Schecter, and A. Kamenev, Phys. Rev. A 85, 043623 (2012) [16] M. N. Kiselev, K. Kikoin and M. B. Kenmoe, EPL 104, 57004 (2013); F. Forster et al Phys. Rev. Lett. 112, 116803 (2014); S. Ganeshan, E. Barnes, and S. Das Sarma, Phys. Rev. Lett. 111, 130405 (2013); H. Ribeiro, J. R. Petta, and G. Burkard, Phys. Rev. B 87, 235318 (2013) [17] G. Cao, et al., Nature Comm. 4, 1401 (2012); J. M. Nichol, et al., Nature Comm. 6 7682 (2015); C. Zhou, Phys. Rev. A 89, 022337 (2014); L. Gaudreau, Nature Phys. 8, 54 (2012) [18] A. del Campo, M. M. Rams, and W. H. Zurek, Phys. Rev. Lett. 109, 115703 (2012); J. Keeling, A. V. Shytov, and L. S. Levitov, Phys. Rev. Lett. 101, 196404 (2008) [19] K. Saito, M. Wubs, S. Kohler, Y. Kayanuma, and P. Hanggi, Phys. Rev. B 75, 214308 (2007); V. N. Ostrovsky and M. V. Volkov, Phys. Rev. B 73, 060405 (2006) [20] S. Brundobler and V. Elser, J. Phys. A 26, 1211 (1993) [21] Yu. N. Demkov and V. I. Osherov, Zh. Exp. Teor. Fiz. 53, 1589 (1967) [Sov. Phys. JETP 26, 916 (1968)]; A. A. Rangelov, J. Piilo, and N. V. Vitanov, Phys. Rev. A 72, 053404 (2005) [22] Y. N. Demkov and V. N. Ostrovsky, Phys. Rev. A 61, 032705 (2000) [23] Yu. N. Demkov and V. N. Ostrovsky, J. Phys. B 28, 403 (1995); V. N. Ostrovsky and H. Nakamura, J. Phys. A 30, 6939 (1997); Y. N. Demkov and V. N. Ostrovsky, J. Phys. B 34, 2419 (2001); C. E. Carroll and F. T. Hioe, J. Phys. A 19, 1151 (1986) [24] N. A. Sinitsyn, Phys. Rev. B 66, 205303 (2002) [25] N. A. Sinitsyn, J. Phys. A 37 (44), 10691 (2004); M. V. Volkov and V. N. Ostrovsky, J. Phys. B: At. Mol. Opt. Phys. 38, 907 (2005) [26] M. V. Volkov and V. N. Ostrovsky, J. Phys. B: At. Mol. Opt. 13 Phys. 37, 4069 (2004); B. E. Dobrescu and N. A. Sinitsyn, J. Phys. B: At. Mol. Opt. Phys. 39, 1253 (2006) [27] N. A. Sinitsyn, J. Phys. A: Math. Theor. 48 195305 (2015) [28] N. A. Sinitsyn, Phys. Rev. B 92, 205431 (2015) [29] N. A. Sinitsyn, Phys. Rev. Lett. 110, 150603 (2013) [30] A. Altland and V. Gurarie, Phys. Rev. Lett. 100, 063602 (2008); A. Altland, V. Gurarie, T. Kriecherbauer, and A. Polkovnikov, Phys. Rev. A 79, 042703 (2009) [31] A.P. Itin, P. Torma, Phys. Rev. A 79, 055602 (2009) [32] D. Sun and A. Abanov and V. L. Pokrovsky, EPL 83, 16003 (2008) [33] N. A. Sinitsyn, Phys. Rev. A 87, 032701 (2013); V. L. Pokrovsky and N. A. Sinitsyn, Phys. Rev. B 65, 153105 (2002) [34] N. A. Sinitsyn, Phys. Rev. A 90, 062509 (2014) [35] E. A. Yuzbashyan, B. S. Shastry, J. Stat. Phys. 150, 704 (2013) [36] A. Patra, E. A. Yuzbashyan, J. Phys. A: Math. Theor. 48, [37] C. Sun, and N. A. Sinitsyn, J. Phys. A: Math. Theor. 48, 505202 245303 (2015) (2015)
1501.06958	2	1501	2015-11-03T06:39:12	Multicomponent fractional quantum Hall states with subband and spin degrees of freedom	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	In wide GaAs quantum wells where two electric subbands are occupied we apply a parallel magnetic field or increase the electron density to cause a crossing of the two $N=0$ Landau levels of these subbands and with opposite spins. Near the crossing, the fractional quantum Hall states in the filling factor range $1<\nu<3$ exhibit a remarkable sequence of pseudospin polarization transitions resulting from the interplay between the spin and subband degrees of freedom. The field positions of the transitions yield a new and quantitative measure of the composite Fermions' discrete energy level separations. Surprisingly, the separations are smaller when the electrons have higher spin-polarization.	cond-mat.mes-hall	cond-mat	Multicomponent fractional quantum Hall states with subband and spin degrees of freedom Yang Liu, S. Hasdemir, J. Shabani, M. Shayegan, L.N. Pfeiffer, K.W. West, and K.W. Baldwin Department of Electrical Engineering, Princeton University, Princeton, New Jersey 08544 (Dated: June 10, 2021) In two-dimensional electron systems confined to wide GaAs quantum wells we observe a remark- able sequence of transitions of the fractional quantum Hall states in the filling factor range 1 < ν < 3 near the crossing of two N = 0 Landau levels from different subbands and with opposite spins. The transitions attest to the interplay between the spin and subband degrees of freedom and can be ex- plained as pseudospin polarization transitions where the pseudospins are the two crossing levels. The magnetic field positions of the transitions yield a new and quantitative measure of the transition en- ergies or, equivalently, the composite Fermions' discrete energy level separations. Surprisingly, these energies are much larger than those reported for other systems with SU(2) symmetry; moreover, they are larger when the electron system is less spin polarized. I. INTRODUCTION Fractional quantum Hall states (FQHSs) are the hall- marks of an interacting two-dimensional electron system (2DES) at large perpendicular magnetic field (B⊥) when a Landau level (LL) is partially occupied.1,2 Adding ex- tra electronic (pseudospin) degrees of freedom leads to additional sets of LLs which are separated by, e.g., the Zeeman energy (EZ ), subband separation (∆SAS), or val- ley splitting energy (EV ), in systems with spin, subband, or valley degree of freedom, respectively. When two such LLs cross at a particular LL filling factor (ν), the QHS at that filling typically weakens or disappears.3 In very high quality samples, however, if two N = 0 LLs cross, the 2DES can form two-component FQHSs where the components are the crossing LLs (the pseudospins), and exhibit pseudospin polarization transitions as one tunes the pseudospin energy splitting, or the separation be- tween the two LLs. Such transitions have been reported for 2DESs confined to AlAs quantum wells (QWs) where the energy separation between the two occupied valleys is tuned via the application of strain so that the two lowest (N = 0) LLs of the two valleys cross.4 -- 7 Multi-component FQHSs and transitions between their pseudospin config- urations have also been studied in other 2DESs where two or more LLs with different spin or valley indices are close in energy.8 -- 17 Here we present a study of 2DESs confined to wide GaAs QWs where EZ and ∆SAS are much smaller than the Coulomb energy at large B⊥. Thus the lowest four LLs -- the S0↑, S0↓, A0↑ and A0↓ levels -- are close in en- ergy so that, in principle, a full description of FQHSs would require the inclusion of all four LLs and using SU(4) symmetry (S and A refer to symmetric and anti- symmetric subbands, ↑ and ↓ refer to up- and down-spin; see Fig. 1(a)). Via applying either a parallel magnetic field component (B, see Fig. 1(b)), or changing the electron density in the QW,18 we reduce ∆SAS and in- crease EZ , so that the S0↓ and A0↑ levels cross when ∆SAS = EZ; see Fig. 1(a). Near the crossing, we ob- serve a remarkable pattern of appearing and disappearing FQHSs in the filling range 1 < ν < 3, revealing the for- mation of FQHSs with both spin and subband degrees of freedom simultaneously.19 We can qualitatively describe the number of observed transitions and their positions in a simplified two-component picture with SU(2) sym- metry, where the two pseudospins are the S0↓ and A0↑ levels. However, our quantitative analysis reveals several puzzling features. We find that at a fixed ν, the tran- sition energies are much larger than reported previously in systems with SU(2) symmetry where the two pseu- dospins are either spin or valley levels. Also, the tran- sition energies are about twice larger when the 2DES is less spin-polarized. We discuss possible origins of these unexpected features. II. METHOD Our sample, grown by molecular beam epitaxy, is a 65-nm-wide GaAs QW bounded on each side by undoped Al0.24Ga0.76As spacer layers and Si δ-doped layers. We use a 4 × 4 mm2 square piece with alloyed InSn contacts at four corners, and an evaporated Ti/Au front-gate and an In back-gate to change the 2DES density (n) while keeping the charge distribution symmetric [Fig. 1(b)]. We measure the transport coefficients in a dilution re- frigerator with base temperature T ≈ 30 mK, using low- frequency (< 30 Hz) lock-in technique and a rotatable sample platform to induce B. At B = 0, the Fourier transform of the Shubnikov-de Haas oscillations exhibits two peaks that correspond to the electron densities in the two subbands. The difference between these densi- ties yields ∆SAS which is in excellent agreement with the results of our B = 0 self-consistent calculations. To cal- culate ∆SAS at finite B and B⊥ [Fig. 1(a)], we employ a perturbative simulation introduced in Ref. [20], where we assume B only mixes LLs from different subbands but does not change the QW potential. ν = 2 ΔSAS (a) 10 K EZ 2 4 B (T) 6 B B┴ 0.4 (b) B θ T ≈ 65 mK (c) 8/3 2 7/3 ÷ 2 46 45 T ≈ 135 mK 0.3 3 ν = 4/3 5/3 θ 41 40 37 30 25 0.2 0.1 A0↓ A0↑ S0↓ S0↑ 0 0.4 0.3 0.2 ) Ω k ( x x R 0.1 0 0 6 0 7 3 3 4 5 B┴ (T) 4 B┴ (T) W = 65 nm 8/3 7/3 12/5 13/5 2 T ≈ 65 mK (e) ν = 2 0.6 θ 41.1 40.5 40 39.4 38.1 37.5 36.2 35 32.5 30 27.5 25 0 0.4 0.2 0 4 7/5 10/7 4/3 5/3 8/5 11/7 5 6 B┴ (T) 7 (d) ν = 2 θ 48.9 46.1 44.9 43.6 42.4 40.5 38.1 36.9 35 0 FIG. 1. (color online) (a) Relevant energy levels for n = 2.12 × 1011 cm−2 electrons confined to a 65-nm-wide GaAs QW near filling factor ν = 2, calculated as a function of parallel magnetic field (B). (b) Charge distribution (red curve) and potential (black curve) calculated via solving Poisson-Schrodinger equations self-consistently at B = 0. The experimental geometry and the tilt angle θ are also shown. (c)-(e) Waterfall plots of Rxx vs B⊥, measured at the indicated temperatures and different θ. Each trace is shifted vertically. The observed FQHS transitions are marked by the black arrows. III. EXPERIMENTAL RESULTS Figure 1 (c) shows the longitudinal magnetoresistance (Rxx) traces for 1 < ν < 3, measured at n = 2.12 × 1011 cm−2, T ≃ 135 mK, and different tilt angles (θ). As we increase θ, the ν = 2 Rxx plateau narrows near θ ≃ 37◦ and then widens again at larger θ. The re- duction of the Rxx plateau width signals a weaken- ing of the integer QHS, which is widely used to iden- tify LL crossings in high quality 2DESs and at low temperatures.3,21,22 This weakening, but not disappear- ing, is similar to what is seen at certain LL crossings at other integral ν when interaction preserves the energy gap through QHS ferromagnetism.3,23,24 In our samples, ∆SAS is larger than EZ at B = 0. As we increase B, ∆SAS decreases and EZ increases, so that the S0↓ and A0↑ levels cross when ∆SAS ≃ EZ .25 In Fig. 1(a), we show the calculated energies of the four N = 0 LLs rela- tive to the S0↑ level as a function of B at ν = 2.20 We use a three-fold enhanced EZ to match the experimental observation that the ν = 2 crossing occurs near θ ≃ 37◦ (B ≃ 3.5 T). This enhancement of EZ is not surprising and has also been reported in previous studies.3,17,21,22 More interestingly, near θ ≃ 37◦, the FQHSs on both sides of ν = 2 show a rich series of transitions, as marked with arrows in Figs. 1(c)-(e).26 The ν = 4/3 FQHS is strong at both small and large θ, but becomes weak at θ ≃ 30◦ [Figs. 1(c) and (e)]. The ν = 8/3 FQHS also experiences one transition, at θ ≃ 45◦ [Figs. 1(c) and (d)]. Meanwhile, the ν = 5/3 and 7/3 FQHSs become weak twice: at θ ≃ 30◦ and 40◦ for ν = 5/3, and at θ ≃ 37◦ and 45◦ for ν = 7/3. Data taken at lower T ≃ 65 mK, shown in Figs. 1(d) and (e), reveal a more remarkable pattern of higher-order FQHS transitions. On the left side of ν = 2 [Fig. 1(d)], the ν = 13/5 FQHS weakens twice at θ ≃ 40.5◦ and 46.1◦. The ν = 12/5 FQHS, on the other hand, becomes weak three times, at θ ≃ 36.9◦, 42.4◦ and 46.1◦. On the high-field side of ν = 2, as seen in Fig. 1(e), the ν = 7/5 FQHS weakens twice, at θ ≃ 27.5◦ and 35◦, and the ν = 8/5 FQHS thrice, at θ ≃ 27.5◦, 36.2◦ and 39.4◦. We also observe in Fig. 2(e) three transitions at ν = 10/7 and four transitions at ν = 11/7.18 We summarize in Fig. 2(a) the values of B and B⊥ for all the observed transitions. 1 4/3 10/7 7/5 6 7 ν = 5/3 Δ > 0 ΔSAS > EZ 8/5 11/7 5 B┴ (T) Energy ) T ( B -3 -2 2 1 -1 2 4 3 8/3 13/5 12/5 7/3 2 3 (b) ℏωCF Δ (4,0) (3,0) 4 (3,1) 5 (a) -4 -3 νCF = -2 3 2 Δ < 0 ΔSAS < EZ (2,1) (2,2) (1,1) (1,3) (0,2) (1,2) (0,1) (2,0) (1,0) Δ > 0 Δ < 0 (0,3) (0,4) ±4 ±3 ±2 νCF = ±1 B (color online) (a) Summary of B and B⊥ where FIG. 2. pseudospin transitions are seen in Fig. 1. The dashed line represenst zero pseudospin splitting ∆ = 0 (see text). ∆ = ∆SAS − EZ is positive in the green region, and becomes negative in the yellow region. Typical error bars are shown for the q/3 FQHS transitions. (b) Schematic depiction of the CF Λ-level energies in a simplified two-pseudospin picture. The dashed black and solid red lines represent the CF Λ-levels with up- and down-pseudospins (the electron S0↓ and A0↑ levels). Each configuration is labeled with its pseudospin Λ- level filling factors (νCF↑, νCF↓), where ↑ and ↓ stand for the up- and down-pseudospins. IV. DISCUSSION The fact that all these transtions occur near the crossing of the S0↓ and A0↑ levels when ∆SAS ≃ EZ suggests that these are FQHS pseudospin polarization transitions.19 Such transitions can be readily understood in the framework of composite Fermions (CFs), quasi- particles formed by attaching two flux quanta to each electron.2,27,28 The CFs form their own discrete en- ergy levels, the so-called Λ-levels which are separated by the CF cyclotron energy ¯hωCF, and the FQHS at ν = νCF/(2νCF + 1) is the integer QHS of CFs with inte- gral Λ-level filling factor νCF. Since the other two energy levels, S0↑ and A0↓, are reasonably far in energy and are always full or empty,19 we neglect them and interpret our data in a simple (SU(2)) picture where the two crossing levels, S0↓ and A0↑, are the two pseudospins.29 The pseu- dospin splitting energy ∆ = ∆SAS − EZ is then tuned from positive to negative via applying B [Figs. 1(a) and 2(b)]. As discussed below, this simplified two-pseudospin model qualitatively explains our data. 3 In this model, the FQHS at ν = 4/3 has only one oc- cupied Λ level (νCF = 1).29 It therefore has two possible configurations with the Λ-level filling factors of the up- and down-pseudospin CFs being (νCF↑, νCF↓) = (1, 0) or (0,1), and should exhibit one transition at ∆ = 0 [see Fig. 2(b)], consistent with Fig. 2(a) data. The FQHS at ν = 7/5, on the other hand, has two filled Λ-levels (νCF = 2), three pseudospin configurations [(νCF↓, νCF↓) = (2,0), (1,1) and (0,2)], and two transitions when ∆ = ±¯hωCF [Fig. 2(b)], also in agreement with Fig. 2(a) data. Using similar logic, we can explain the three transitions ob- served for the ν = 10/7 FQHS which has νCF = 3 and four configurations [Fig. 2(b)]. By invoking particle- hole symmetry, which links the FQHSs at ν to (4 − ν) in our system (e.g., 4/3 to 8/3), and also utilizing negative CF fillings, e.g., νCF = −2 for the ν = 5/3 state, we can explain all the transitions summarized in Fig. 2(a). In general, a FQHS with νCF has νCF + 1 pseudospin configurations, and νCF pseudospin polarization transi- tions which occur whenever ∆ equals a multiple integer of ¯hωCF [see Fig. 2(b)]. Next we proceed to a quantitative analysis of Fig. 2(a) data. Figure 2(b) implies that, when νCF is odd, one transition occurs exactly at ∆ = 0. In Fig. 2(a), we first fit the dashed line through these transition points. We then focus on a particular ν, e.g. 8/5, and calculate ∆SAS as a function of B at this filling, as shown in Fig. 3(a). Using the value of B at which the ∆ = 0 transi- tion for ν = 8/5 occurs (B = 3.96 T), we determine a value for g∗ (≃ −1.34) so that ∆SAS = EZ at B = 3.96 T. We then plot EZ (= g∗µBB) at ν = 8/5 as a function of B in Fig. 3(a), and determine ∆ for the other two 8/5 transitions. (See Fig. 3(b) for the four pseudospin configurations of the ν = 8/5 FQHS.) Since these transi- tions are expected to occur when ∆ = ±2¯hωCF [see Fig. 2(b)], we find ¯hωCF = 5.5 K for ∆ > 0 and ¯hωCF = 2.5 K for ∆ < 0. Using this procedure we can deduce ¯hωCF for FQHSs at ν = 8/5, 12/5, and 10/7 which have a ∆ = 0 transition on the dashed line in Fig. 3(a). For the FQHSs at ν = 13/5, 7/3, 5/3 and 7/5, we assume ∆ = 0 at the intersection of the dashed line and the vertical lines that mark these fillings in Fig. 3(a) and, following a similar procedure, find ¯hωCF from the transitions' B values. In Fig. 3(c) we plot as a function of 1/(2νCF + 1) all the deduced values of ¯hωCF, normalized to the Coulomb energy (VC = e2/4πǫlB, where lB = p¯h/eB⊥ is the mag- netic length and ǫ is the dielectric constant), for different FQHSs. Figure 3(c) data allow us to make a quanti- tative comparison of the deduced ¯hωCF/VC to previous experimental reports. The dashed lines in Fig. 3(c) rep- resent ¯hωCF/VC , measured near ν = 1/2 and 3/2, from FQHS spin transitions in QWs with same well-width (65 nm) but much lower density where the S0↑ and S0↓ lev- els are closer in energy and the A0↑ level is well above S0↓.13,18,19 Figure 3(c) plot clearly shows that our mea- sured ¯hωCF/VC is much larger than the dashed lines. It is possible that the transition energies we deduce are somewhat exaggerated because of the inaccuracy of the (a) (b) νCF = -3 (ν = 8/5) ΔSAS Δ EZ Δ ℏωCF ν = 8/5 3 4 B (T) 5 11/7 ν = 10/7 (c) Δ < 0 Δ > 0 ) K ( y g r e n E 10 0 2 ) % ( 0 / C V F C 2ℏ ω 5/3 7/3 4 -0.4 8/5 12/5 -0.2 1 < ν < 2 2 < ν < 3 0 1/(2νCF+1) 7/5 13/5 0.2 0.4 FIG. 3. (color online) (a) Calculated ∆SAS and EZ as a func- tion of B at ν = 8/5. (b) Λ-level diagram showing the four different pseudospin configurations for ν = 8/5 (νCF = −3) in our two-component picture. The pseudospin polarization transitions are expected when ∆ = 0, ±2¯hωCF [see Fig. 2(b)]. (c) The Λ-level separation in units of VC = e2/(4πǫlB) as a function of 1/(2νCF + 1), deduced from ∆ at which the FQHS transitions are observed in Fig. 2(a). Data are shown for both 1 < ν < 2 (black) and 2 < ν < 3 (red). The dashed lines rep- resent data for 2DESs with only spin degree of freedom13; results from transitions near ν = 1/2 and 3/2 are shown in the ∆ < 0 and ∆ > 0 regions, respectively. 4 tor of two larger than the dashed lines in Fig. 3(c).18 We therefore conclude that ¯hωCF in our 2DES with both spin and subband degrees of freedom is larger than in a 2DES with only spin degree of freedom. The reason for this is not entirely clear. Perhaps the nature of pseudospins plays a role in determining the polarization energies and ¯hωC F . It is also possible that in our system the close proximity of the other two LLs (S0↑ and A0↓; see Fig. 1(a)) is important. Both of these interesting possibilities deserve further theoretical examination. We highlight two additional noteworthy features of Fig. 3(c). First, the results for 1 < ν < 2 approximately match those for 2 < ν < 3, suggesting that ¯hωCF does not depend on which LL (S0↓ or A0↑) hosts the CFs. Second, in our 2DES we tune the crossing of two N = 0 LLs with opposite spins while a lower level (S0↑) is fully occupied. This allows us to observe FQHS pseudospin transitions at a given ν for both ∆ < 0 and ∆ > 0, when the net spin orientation of the CFs is aligned or anti-aligned with the spin (↑) of majority electrons, respectively. Data of Fig. 3(c) indicate that ¯hωCF/VC is about twice larger for the case ∆ > 0 compared to the ∆ < 0 case. This is an important observation as it sheds light on the mysteri- ous asymmetry between ¯hωCF/VC for CFs near ν = 3/2 and 1/2, deduced from CF spin-polarization transitions in GaAs 2DESs,13 or CF valley-polarization transitions in AlAs 2DESs.6 In those 2DESs the polarization transition energies of FQHSs near 3/2, where electrons are partially spin- (or valley-) polarized, are much larger than those near 1/2 where they are fully polarized.6,13 Our data, af- forded by the tunability of the subband/spin degrees of freedom, provide clear evidence that the spin polarization of the total electron system can affect the CF properties. ACKNOWLEDGMENTS perturbative calculations we use to determine the B- dependence of ∆SAS. To test this possibility, we made measurements on another QW with slightly narrower well width (55 nm).18 In this sample, we induce the crossing of the S0↓ and A0↑ levels via applying B and also by tuning n at B = 0. The ¯hωCF deduced from the tilt- ing data matches Fig. 3 data, but the n-tuning results are somewhat smaller,18 suggesting the inaccuracy of our perturbative calculations.30 However, we emphasize that even the n-tuning data which do not rely on calculations of ∆SAS, yield ¯hωC F /VC values that are about a fac- We acknowledge support through the NSF (DMR- 1305691), and DOE BES (DE-FG02-00-ER45841) for measurements, and the NSF (Grant MRSEC DMR- 1420541), the Gordon and Betty Moore Foundation (Grant GBMF4420), and Keck Foundation for sample fabrication and characterization. A portion of this work was performed at the NHMFL, which is supported by the NSF Cooperative Agreement No. DMR-1157490, the State of Florida, and the DOE. We thank S. Hannahs, G. E. Jones, T. P. Murphy, E. Palm, A. Suslov, and J. H. Park for technical assistance, and J. K. Jain for illu- minating discussions. 1 D. C. Tsui, H. L. Stormer, and A. C. Gossard, Phys. Rev. Lett. 48, 1559 (1982). 2 J. K. Jain, Composite Fermions (Cambridge University Press, Cambridge, UK, 2007). 3 Y. Liu, J. Shabani, Phys. Rev. B 84, 195303 (2011). and M. Shayegan, 4 N. C. Bishop, M. Padmanabhan, K. Vakili, Y. P. and M. Shayegan, Shkolnikov, E. P. De Poortere, Phys. Rev. Lett. 98, 266404 (2007). 5 M. Padmanabhan, T. Gokmen, Phys. Rev. B 80, 035423 (2009). 6 M. Padmanabhan, T. Gokmen, and M. Shayegan, and M. Shayegan, Phys. Rev. B 81, 113301 (2010). 7 M. Padmanabhan, T. Gokmen, Phys. Rev. Lett. 104, 016805 (2010). and M. Shayegan, S0↓ and A0↑ levels. 20 N. Kumada, K. 5 8 J. P. Eisenstein, H. L. Stormer, L. Pfeiffer, and K. W. West, Phys. Rev. Lett. 62, 1540 (1989). 9 R. G. Clark, S. R. Haynes, A. M. Suckling, J. R. Mal- and C. T. Foxon, lett, P. A. Wright, J. J. Harris, Phys. Rev. Lett. 62, 1536 (1989). 10 L. W. Engel, S. W. Hwang, T. Sajoto, D. C. Tsui, and M. Shayegan, Phys. Rev. B 45, 3418 (1992). 11 R. R. Du, A. S. Yeh, H. L. Stormer, D. C. Tsui, L. N. Pfeif- fer, and K. W. West, Phys. Rev. Lett. 75, 3926 (1995). 12 I. V. Kukushkin, K. v. Klitzing, and K. Eberl, Phys. Rev. Lett. 82, 3665 (1999). 13 Y. Liu, S. Hasdemir, A. W´ojs, J. K. Jain, L. N. Pfeif- and M. Shayegan, fer, K. W. West, K. W. Baldwin, Phys. Rev. B 90, 085301 (2014). 14 T. M. Kott, B. Hu, S. H. Brown, and B. E. Kane, Phys. Rev. B 89, 041107 (2014). 15 C. Dean, A. Young, P. Cadden-Zimansky, L. Wang, H. Ren, K. Watanabe, T. Taniguchi, P. Kim, J. Hone, and K. Shepard, Nature Physics 7, 693 (2011). 16 B. E. Feldman, A. J. Levin, B. Krauss, D. A. and A. Yacoby, Abanin, B. I. Halperin, J. H. Smet, Phys. Rev. Lett. 111, 076802 (2013). 17 J. Falson, D. Maryenko, B. Friess, D. Zhang, Y. Kozuka, and M. Kawasaki, J. H. Smet, A. Tsukazaki, Nat. Phys. 11, 347 (2015). 18 See Supplemental Material.. 19 As reported in Refs. [8-13], spin transitions of FQHSs around ν = 3/2 and 1/2 are seen in GaAs 2DESs as a function of density or tilt angle. In a 65-nm-wide QW sam- ple, we also see such spin transitions at low densities near n = 1.2×1011 cm−2 (see Ref. [13] for details). In this work, at a very high density (n = 2.12 × 1011 cm−2) and with ad- ditional (parallel) magnetic field when the S0↑ level should be fully occupied and no spin transitions are expected, we observe additional transitions which, in the simplest and most plausible explanation, are induced by the crossing of moda, K. Muraki, Y. Hirayama, Phys. Rev. B 77, 155324 (2008). Iwata, K. Tagashira, Y. Shi- and A. Sawada, 21 W. Desrat, F. Giazotto, V. Pellegrini, M. Governale, and L. Sorba, F. Beltram, F. Capotondi, G. Biasiol, Phys. Rev. B 71, 153314 (2005). 22 X. C. Zhang, I. Martin, and H. W. Jiang, Phys. Rev. B 74, 073301 (2006). 23 E. P. De Poortere, E. Tutuc, S. J. Papadakis, and M. Shayegan, Science 290, 1546 (2000). 24 K. Muraki, T. Saku, and Y. Hirayama, Phys. Rev. Lett. 87, 196801 (2001). 25 At very large B, ∆SAS essentially vanishes and the 2DES becomes bilayer, supporting FQHSs formed in each layer independently. In our experiments, B is not large enough to reach such regime. 26 Reentrant integer QHSs, seen also as Rxx minima, have been reported at ν > 2 in single-subband systems when an N ≥ 1 LL is partially filled; see, e.g. J. Eisenstein et al., Phys. Rev. Lett. 88, 076801 (2002). However, in our system, only the N = 0 LLs are occupied for ν < 4, and no such phases are expected. 27 J. K. Jain, Phys. Rev. Lett. 63, 199 (1989). 28 B. P. A. Lee, I. Halperin, and N. Read, Phys. Rev. B 47, 7312 (1993). 29 Because in our system the lowest (S0↑) level is always fully occupied in the range of interest here, a FQHS at ν is equivalent to the ν ′ = ν − 1 FQHS in a system such as 2D electrons in an AlAs QW where the crossing levels are the lowest two (valley) S0 LLs; see, e.g., Refs. [4 -- 7]. 30 We indeed hope that our data will stimulate more precise future calculations; see, e.g., Refs. [31] and [32]. 31 T. S. Lay, T. Jungwirth, L. Smrcka, and M. Shayegan, Phys. Rev. B 56, R7092 (1997). 32 W. Pan, T. Jungwirth, H. L. Stormer, D. C. Tsui, A. H. MacDonald, S. M. Girvin, L. Smrcka, L. N. Pfeiffer, K. W. Baldwin, and K. W. West, Phys. Rev. Lett. 85, 3257 (2000).
1802.01305	1	1802	2018-02-05T09:05:32	Micromagnetic view on ultrafast magnon generation by femtosecond spin current pulses	[ "cond-mat.mes-hall" ]	In this Article we discuss a micromagnetic modelling approach to describe the ultrafast spin-transfer torque excitation of coherent and incoherent magnons on the nanoscale. Implementing the action of a femtosecond spin current pulse entering an orthogonally magnetized thin ferromagnetic film, we reproduce recent experimental results and reveal the factors responsible for the unequal excitation efficiency of various spin waves. Our findings are in an excellent agreement with the results of an analytical description of spin-wave excitation based on classical kinetic equations. Furthermore, we suggest an experimental design allowing for the excitation of laterally propagating spin waves beyond the optical diffraction limit. Our findings demonstrate that the classical micromagnetic picture retains its predictive and interpretative power on femtosecond temporal and nanometer spatial scales.	cond-mat.mes-hall	cond-mat	Micromagnetic view on ultrafast magnon generation by femtosecond spin current pulses Henning Ulrichs∗ I. Physical Institute, Georg-August University of Gottingen, Friedrich-Hund-Platz 1, 37077 Gottingen, Germany 8 1 0 2 b e F 5 ] l l a h - s e m . t a m - d n o c [ 1 v 5 0 3 1 0 . 2 0 8 1 : v i X r a Physical Chemistry Department, Fritz Haber Institute of the Max Planck Society, Faradayweg 4-6, 14195 Berlin, Germany (Dated: February 6, 2018) Ilya Razdolski In this Article we discuss a micromagnetic modelling approach to describe the ultrafast spin- transfer torque excitation of coherent and incoherent magnons on the nanoscale. Implementing the action of a femtosecond spin current pulse entering an orthogonally magnetized thin ferromagnetic film, we reproduce recent experimental results and reveal the factors responsible for the unequal excitation efficiency of various spin waves. Our findings are in an excellent agreement with the results of an analytical description of spin-wave excitation based on classical kinetic equations. Furthermore, we suggest an experimental design allowing for the excitation of laterally propagating spin waves beyond the optical diffraction limit. Our findings demonstrate that the classical micromagnetic picture retains its predictive and interpretative power on femtosecond temporal and nanometer spatial scales. I. INTRODUCTION Stimulated by the seminal experiment by Beaurepaire et al.1 and the quest for ultrafast opto-magnetic record- ing, an immense amount of knowledge on the incoher- ent laser-induced spin dynamics (i.e. ultrafast demag- netization) in a large variety of materials has been ac- cumulated over the years2. Simultaneously, femtosec- ond optical excitation of coherent spin dynamics was discovered3,4, encompassing an extremely broad range of timescales which are governed by the intrinsic properties of magnetic systems. The temporal limitation for the excited spin modes is often pertinent to the pulse dura- tion of the light source, typically on the order of 10-100 fs. However, the spectrum of the accessible inhomoge- neous (with nonzero wavevector k) spin wave modes is governed by the spatial inhomogeneity scale of the exci- tation. In other words, on top of the temporal require- ments, non-uniform spin wave modes can only be gener- ated if their wavevectors k are contained in the spectrum of the spatially inhomogeneous stimulus. For the excita- tion with visible (VIS) or near-infrared (NIR) light, the optical penetration depth δ ≈ 10 − 15 nm serves as a good estimation for the excitation limit of perpendicular spin waves in metallic media5. Optical excitation of the in-plane propagating spin waves is even more restrictive, as the allowed k values are governed by the diffraction- limited beam spot size ((cid:38) 103 − 104 nm)6,7. Yet, in recent experiments, a strong interfacial local- ization of the spin transfer torque exerted by spin polar- ized currents enabled the excitation of spin waves with much larger wavevectors8,9. In particular, perpendicu- lar standing spin waves (PSSW) with f = 0.55 THz and k ∼ 1 nm−1 have been detected8. The wavelengths of these excitations approach the exchange length lex (a few nm in Fe10), where macroscopic spin models are likely to break down11. Enabling the expansion of ultrafast photo- magnonics12,13 onto the nanometer scale, these findings simultaneously question the applicability of conventional modeling of spin dynamics in these extreme conditions. For conventional magnonics, micromagnetic simulation is an indispensable tool for both prediction, and interpre- tation of static and dynamic magnetic properties14–20. In this article, we show that the micromagnetic mod- elling approach is also suitable for ultrafast processes on nanometer scales. In particular, we set up a micromag- netic model to first reproduce the recent experimental findings8 of ultrafast optical magnon generation. In this study, Fe/Au/Fe trilayers were optically pumped from one side, generating a spin current pulse which traverses the Au spacer and then interacts with the second Fe layer, resulting in the excitation of high-frequency spin dynamics in the THz domain. Here we develop a micro- magnetic model featuring ultrafast spin-transfer torque perturbation and verify that it can accurately reproduce the experimentally observed spin dynamics. We further identify important factors governing the excitation effi- ciency and energy transfer into the PSSW modes in thin ferromagnetic films. Complementing a recent theoreti- cal work21 on laser-generated superdiffusive spin trans- port in non-collinear spin valve systems and resulting macrospin dynamics in ferromagnets, our results open the door to understanding spin current-driven magnetism on the nanoscale. Later on, we include thermally ac- tivated, incoherent magnetic fluctuations in the model, and show that ultrafast spin currents can effectively cool or heat such thermal magnon ensembles, in agreement with experimental observations on slower timescales22. In the outlook, we outline the topological possibilities for the spin current-mediated generation of in-plane propa- gating, large-k spin waves beyond the optical diffraction limit. 2 mated with an analytic function (1) with t0 = 50 fs, τ1 = 10 fs, and τ2 = 150 fs, closely reproducing the re- sults of Ref.24. We use the software package mumax3 (Ref.25) to model the effect of a spin current pulse on the local magnetization inside the Fe film by augmenting the Landau-Lifshitz-Gilbert equation with the spin transfer torque (STT) term τSTT proposed by Slonczewski26: τSTT = γ Nz dµ0M0 jsM × M × p. (2) The reported in-plane excursion of the magnetization of my = My = 0.023 directly after the spin current pulse M0 arrival allows us to determine the respective current den- sity to be used in the simulations. For this purpose, we systematically varied the current density and analyzed the temporal evolution of the in-plane component my. The maximum excursion appears shortly after the spin current pulse maximum, which is in agreement with Ref.8 (see Figure 4(b) therein). Figure 2 shows that the max- imum depends linearly on the applied current density. The linear interpolation intersects with the horizontal dashed line defined by the experimental value of my at j0 = 5.9 · 1012A/m2. This result in an excellent agree- ment with the one obtained from the spin transfer density (7 µB/nm2) evaluated in Ref.8 (≈ 6 · 1012 A/m2), thus reinforcing our micromagnetic model. Figure 2. Determination of j0 from the dependence of the maximum of my on the current density j. The horizontal dashed line marks the my experimental value which corre- sponds to j0 = 5.9 · 1012 A/m2 (see the vertical dashed line). Further, this j0 value is realistic, as it corresponds to the ∼ 10% spin transport-induced demagnetization of a 10 nm-thick Fe film within 200 fs, in agreement with the latest results obtained within the superdiffusive transport model21. Moreover, it is close to the values reported in other works (1012 − 1013 A/m2, Ref.27, and 1013 A/m2, Ref.28). Similar numbers can be further obtained from the work of Choi et al.29 (∼ 1012 A/m2). Figure 1. Model details. (a) Sketch of the micromagnetic model. Besides incorporating the spin current, the model optionally includes thermal fluctuations. (b) Spatio-temporal dependence of the spin current entering the film, assuming λSTT = 2 nm. II. MODEL DETAILS Our model is visualized in Figure 1(a). It consists of a single ferromagnetic layer of Fe, with a thickness of d = 14.2 nm. We take as material parameters a saturation magnetization of µ0M0 = 2.1 T, an exchange constant of A = 19 pJ/m (corresponding to the exchange stiffness 2· meV in Fe from Ref.23), and a uniaxial D = 280 A anisotropy along x with strength Ku = 45956 J/m3, and assume a Gilbert damping factor of α = 0.008. For the demonstration of coherent magnon generation, we simulate a cube of size 1.16 × 1.16 × 14.2 nm3 subdi- vided into Nx × Ny × Nz = 2 × 2 × 24 cells. Periodic boundary conditions in x and y direction enlarge this cube into an infinitely extended film. For the demon- stration of incoherent magnon creation and annihilation, we simulate a larger cube of size 150 × 150 × 14.2 nm3 subdivided into Nx × Ny × Nz = 256 × 256 × 24 cells. Note that in this case, an additional magnetic field rep- resenting thermal fluctuations is switched on. The spin current js with polarization p enters the system at z = 0, and has the following empirical spatio-temporal form: js = 2e j0e−z/λSTT e−t/τ2 1 + e−(t−t0)/τ1 . (1) Here the penetration depth of the spin current is λSTT = 2 nm, as estimated in Ref.8, and the temporal profile of the spin current pulse (Fig. 1,b) is approxi- 3 Figure 3. Magnetization dynamics driven by the ultrafast spin-transfer torque. (a) Spatio-temporal plot of the dynamic part (cid:104)mdyn (cid:105) of laterally averaged dynamic magnetization components. The left panel of (b) shows a spatial Fourier transform of the (cid:104)mz(cid:105) data shown in the right panel of (a), and the right panel of (b) shows the spin-wave dispersion f (k) from Eq. (3). The dashed lines relate the numerical response to the analytic theory. i III. RESULTS A. Coherent magnon generation In the first numerical experiment, we prepared a spin current pulse with polarization p (cid:107) y ⊥ m. Then, ac- cording to Eq. (2) the spin torque will be τSTT ⊥ m. The subsequent spatially-resolved spin dynamics was com- puted for a total time of 1 ns after the spin current pulse peak. Figure 3(a) shows mdyn , which is the dynamic part of the laterally averaged magnetization component (cid:104)mi(cid:105)x,y(z, t) (i = x, y, z) in each layer of the Fe film for the first 8 ps. One can see how the spin current induces the formation of a localized wave packet which then ex- pands. Note that the quadratic dispersion of exchange- dominated spin waves30 i f (k) = γµ0 2π Han + 2A M0 k2 Han + 2A M0 k2 + M0 γ (3) is responsible for the quick spatial broadening of the spin- 2π ≈ 28 GHz/T is the gyromag- wave packet. Here, netic ratio, Han = 2Ku/M0 is the in-plane crystalline anisotropy field. The front of the wave packet travels with a characteristic speed of about 7 nm/ps correspond- ing to the group velocity of the magnons with the largest wavevectors contained in the excitation spectrum. After about 2 ps, the pulse has reached the surface of the Fe (cid:115)(cid:18) (cid:19) (cid:18) · (cid:19) film. The spin dynamics for times t > 2 ps is formed by a complex interference pattern which can be well ana- lyzed by a Fourier transformation in time, as shown in Figure 3(b). There, the local squared Fourier amplitude FFT{(cid:104)mz(cid:105)x,y}(z, f ) is depicted. In agreement with the A2 experiment8, this representation reveals that the pattern from Figure 3(a) can be understood as a superposition of standing spin waves. This is emphasized by the right panel in Figure 3(b), which shows the dispersion (3), plotted as a function of a dimensionless wave number κ = kd/π. Prominent spin dynamics can be found at integer κn = knd/π = 0, 1, 2, . . ., and the corresponding eigenmode frequencies fn = f (κn). Besides reproducing experimental findings, the micro- magnetic model allows for an identification of the param- eters that govern the mode-specific excitation efficiency. As such, we will now discuss the energy transfer into the different spin wave modes. For this purpose, we analyze in detail squared FFT amplitude spectra at the interface of the Fe film, z = 0. This choice is motivated by the first order boundary conditions for the spin wave modes ensuring the largest amplitude of the oscillations of the spin projections at the interfaces of the Fe film30. The integration of the A2 FFT(f ) spectrum in the vicinity of the peaks corresponding to the excitation of the different spin-wave modes yields the partial energies εn associated with each of the modes: 4 Figure 4. PSSW excitation efficiency. (a) Squared spatial β2 and temporal θ2 Fourier amplitudes of the spin-current pulse (1) as a function of the PSSW mode index for d = 14.2 nm and λSTT = 2 nm. The lines are shown as a guide to the eye. (b) Additional ξ factor as a function of the PSSW mode index for two different Fe film thicknesses d (14.2 nm, open squares, and 70 nm, full squares). (c) Correlation-like double logarithmic plot of the partial energy εn versus the product of the various factors which govern the energy transfer, β2 · θ2 · ξ. The different colors indicate various PSSW eigenmodes while within a single color, the data points are obtained for different values of λSTT. The solid line is a linear fit to the data. (cid:90) fn+∆f fn−∆f εn = A2 FFT(f ) df (4) This quantity is proportional to the energy transferred into this mode. In what follows, we shall focus on various factors which contribute to εn, aiming at understanding their importance for a comparative analysis of the excita- tion efficiency of the PSSW modes in thin ferromagnetic films. In the Supplementary Information we develop an an- alytic formalism based on Holstein-Primakoff transfor- mations, which is capable of deriving the contributing factors in detail. Importantly, the energy supplied by a spin current pulse js(z, t) is proportional to the product β2(kn) · θ2 (ωn) of the spatial and temporal Fourier powers of js(z, t), evaluated at k = kn, and ω = ωn. These two factors are shown in Fig. 4(a) for λSTT = 2 nm. Note that the temporal factor θ2 is only important when the oscillation period Tn = 1/fn approaches the duration of the spin current stimulus ∼ τ2. Thus, in our case for n < 5 and ∼ 250 fs spin current pulse duration, the spa- tial factor β2 plays a dominant role in determining the relative excitation efficiency of the PSSW modes. 1−Ek 2−Ek Further, we identify the material parameter-dependent susceptibility of the different spin-wave modes which, to- gether with their ellipticity Ek, gives rise to another fac- tor ξ = Γk (see Supplementary Information for de- ω2 k tails). Figure 4(b) shows the dependence of ξ on the mode number n for two different film thicknesses. For thin films exemplified here as d = 14.2 nm, the ξ factor peaks at n = 1 and further decays for higher modes. As this dependence is similar to the behaviour of βkn , the role of ξ for the relative excitation efficiency consists in emphasizing the mode with n = 1, consistent with the experimental data8. However, for thicker films, the max- imum of ξ is shifted towards higher modes (n = 5 for d = 70 nm). Note that because in those films both ω and k only slightly increase with n, for small n (when knλSTT (cid:28) 1, ωnτ2 (cid:28) 1) the other factors β2(kn) and 5 to enable direct comparison with the experimental peak- to-noise ratio rexp = Aexp . We arrive at the following condition for λSTT: 1 AN r15(λSTT) (cid:62) Aexp 1 AN , (6) In Figure 5 we show the dependence of the peak ratios on λSTT. The intersection of the red (experimental) and black (simulated) curves indicates the lower boundary for λSTT of about 1.56 nm. B. Incoherent magnon creation and annihilation As a second numerical experiment, we prepared a spin current pulse whose polarization p is collinear to x, and thus to the Fe magnetization m. Then, according to Equation (2) at temperature T = 0 the spin torque τSTT vanishes, and no coherent spin dynamics can be expected. At finite temperatures this expectation is, however, mis- leading due to the magnetic fluctuations present in the ferromagnet. Practically, these fluctuations manifest as a reduction of the average magnetization. Averaging in time gives the transverse components (cid:104)my,z(cid:105)t = 0, and the longitudinal component (cid:104)mx(cid:105)t < 1. The larger the temperature, the smaller is the longitudinal magne- tization component. Switching on a spin current pulse with p collinear to x acts only on the transverse compo- nents, which are momentarily nonzero. It is well known from conventional magnonics, that the resulting torque is damping- or antidamping-like, and that thermal fluc- tuation will therefore be suppressed or enhanced22. This sort of magnon cooling or heating should also manifest on ultrafast time-scales in either an increase, or a further decrease of the longitudinal magnetization component. In the following we will discuss simulation results ob- tained for λSTT = 2 nm at T = 100, 200, 300 K. In Figure 6(a) we plot the temporal form of the spin current pulse according to Eq. (1). In Fig. 6(b) we show the temporal evolution of the spatially averaged longitudinal magneti- zation component (cid:104)mx(cid:105)xyz for T = 100 K. When the spin current pulse penetrates into the film, the fluctuations in- crease (decrease), in case of p (cid:25)(cid:22) x (p (cid:107) x). Therefore, si- multaneously the longitudinal magnetization component further decreases (increases). Note that we only show sin- gle time series in Fig. 6(b). The deterministic nature of the thermal noise in the numerical simulation enables us to apply a normalization procedure to the case when no spins are injected into the ferromagnet. Normalizing and shifting yields the quantity (cid:104)∆mx(cid:105)xyz = which is shown in Fig. 6(c). This representation clearly shows that the maximum of the change in the magnetic moment is shifted by 360 fs with respect to the maxi- mum of the spin current, marked by the vertical dashed line. It can be shown that the front of the transient nor- malized (cid:104)∆mx(cid:105)xyz can be approximated with the time- (cid:104)mx(cid:105)xyz(js(cid:54)=0) (cid:104)mx(cid:105)xyz(js=0)−1, Figure 5. Determination of a lower boundary for the spin transfer torque characteristic depth λSTT. Simulated (black) and experimental (red) peak ratios as a function of λSTT. The red shaded region indicates the error bar around the data from the experiments. The vertical dashed line shows the estimated lower boundary λc. θ2 (ωn) are both almost constant. As such, up to much larger n, the dynamical response is dominated by ξ and thus can be enhanced at higher (n > 1) order spin wave modes. Supported by the analytic theory, we expect in summary a linear relation εn ∝ β2(kn) · θ2 (ωn) · ξ. (5) Indeed, Figure 4(c) shows that a linear scaling law holds over four decades. The data shown here were obtained by varying λSTT between 0.5 nm and 5 nm. The excel- lent agreement between the predictions of the analytic calculations and the results of numerical simulations em- phasizes that all significant factors are accounted for in Equation (5). Having outlined the factors contributing to the exci- tation efficiency of the PSSW modes in thin films, we can now build a bridge to the experiment. We note that in Ref.8 an upper boundary for λSTT has been identi- fied, based on the spin current pulse ability to excite the spin wave mode with n = 4. Here, the above men- tioned formalism enables the determination of a lower λSTT boundary. The critical conditions for that rely on the fact that the excitation efficiency for n = 5 mode was found insufficient for its unambiguous detection in the experimental data. Clearly, for smaller λSTT the n = 5 mode will be more strongly excited. As such, we can calculate the ratio r15 = s15 · A1/A5 for various λSTT, where An is the Fourier amplitude of the n-th PSSW mode at the interface. The correction factor s15 takes into account that in the experiment the MOKE in-depth sensitity function w(z) is responsible for the fact that different PSSW modes contribute unequally to the to- tal MOKE signal. We calculate w(z) using an optical transfer matrix method31, and determine s15, in order 6 integrated spin current pulse profile, indicating the accu- mulative nature of the effect. At later times, both signals decay approximately exponentially with a time constant of 1.79(1) ps. Since in our modelling (both numerical and analytical) this decay originates in Gilbert damping only, one can deduce that the spin wave modes with fre- quencies of a few THz dominate the dynamic response. We note that the change of the longitudinal moment is with only 10−3 rather small. As a matter of fact, it increases proportional to the temperature, as Fig. 6(c) shows. Recall that so far we were discussing spatial av- erages. By applying the normalization procedure to each layer of the simulated film, we have obtained the spa- tially resolved data shown in Fig. 6(d). There one can see that at the injection side the change increases by one order of magnitude, compared to the averaged dynam- ics. Note that Fig. 6(d) looks qualitatively similar to the left panel in Fig. 3(a). However, in contrast to the data shown there, here the spin injection does not cause spatio-temporal coherence of the spin dynamics. Instead, the data plotted in Fig. 6(c) should be interpreted as a spatio-temporal modulation of the density of thermal magnons in the film. IV. OUTLOOK: OPTICAL SPIN-WAVE EXCITATION BEYOND THE DIFFRACTION LIMIT So far we have been discussing the spin wave (magnon) eigenmodes which are quantized in the direction perpen- dicular to the film. When considering optical excitation of laterally propagating magnons, one has to acknowl- edge that this process is usually limited by diffraction. As already mentioned in the introduction, this implies that optical excitation of magnons with wavelengths be- low the light wavelength is rather inefficient. We will now show that spin currents offer a unique possibility to over- come this limitation. Consider a multilayer as in Ref.8, but with the top Fe layer patterned into stripes, as de- picted in Figure 7(a). There, the spin current generation, and subsequently the magnon generation in the bottom Fe layer is governed by the geometry of the stripes instead of the laser spot size. To model this situation, we again employed the spatio-temporal shape of the spin current pulse given by Eq. (1). In addition, a lateral mask de- fined by six stripes (width w = 10 nm and spatial period s = 20 nm) was imposed to model the structure shown in Fig. 7(a). As such, the lateral cross-section of the laser- generated spin current pulse reproduces the mask stripe pattern. We used 8192 × 1 × 10 cells of size 0.53 nm3, which were enlarged again into a film by applying peri- odic boundary conditions in x and y-direction. In Fig. 7(b) we show a map of the transient magne- tization component (cid:104)mz(cid:105)a,yz(x, t), averaged across y, z, and across a width of a = 2 nm around equidistantly probed locations at a distance x towards the center of the stripes. The data clearly show the lateral emission Suppression and enhancement of Figure 6. incoherent magnon dynamics by spin currents. (a) Temporal shape of the spin-current pulse js(t) of the two opposite polarities. (b) Simulated time dependence of the spatially averaged longitu- dinal magnetization component (cid:104)mx(cid:105) at T = 100 K. (c) Varia- tions of spatially averaged longitudinal magnetization compo- nent (cid:104)mx(cid:105), normalized to that obtained in the case of no spin current injection. Multiple lines show the results calculated for three various temperatures, as indicated by dark arrows. The dashed line in (a) through (c) indicates the peak time of the spin current pulse. (d) Layer-resolved, only laterally av- eraged, normalized change of the longitudinal magnetization component (cid:104)mx(cid:105) obtained at the temperature of 200 K. 7 of a spin wave packet, moving with a velocity of about 2500 ± 200 m/s. A Fourier transformation in time and space reveals the spectral properties of the spin dynam- ics, as shown in Fig. 7(c). Here, one can see that the dis- persion f (k) of dipole-exchange spin-waves in Fe30 (see the dashed line in Fig. 7(c)) falls on top of the regions of strong response in the spectrum. Reflecting the peri- odicity of the stripe pattern, the strongest response can be seen around f = 75 GHz, and kx = 2π/s. The group ∂k (2π/s) ≈ 2600 m/s velocity at this wave number vg = ∂ω is in a good agreement with the propagation speed of the spin-wave packet in Fig. 7(b). V. CONCLUSIONS In summary, we have shown that the action of ultra- fast spin currents penetrating into a magnetic thin film can be modelled with great qualitative and quantitative agreement with the experimental data by including a Slonczewski-like spin-torque term in the micromagnetic equation of motion. Depending on the polarization of the spin current, the resulting torque can either create or an- nihilate coherent or incoherent magnons. In particular, we reproduced the recent experimental demonstration of coherent magnon generation by spin currents, and ob- tained further insights into the spatial scales involved in this process. We have identified the factors contribut- ing to the relative excitation efficiency and shown that the linear proportionality law holds over four orders of magnitude. Lastly, employing numerical simulations, we complemented the experimentally estimated constraints on the characteristic spin transfer torque depth λSTT in Fe. Further, our analysis of the spin current excitation of incoherent magnons indicates that the simulated ul- trafast heating and cooling should be detectable by magneto-optical methods. We note here that the heat- ing effects which are not explicitly accounted for in our model might introduce additional complications. Both thermal and spin current-driven signals will be overlaid in time, but can in principle be distinguished by their symmetry properties. Considering thinner ferromagnetic films and larger densities of the injected spin current, as in Ref.32, our modelling suggests that the hot electrons- driven spin-transfer torque can be a relevant and viable Ansatz to understand the observed spin dynamics. We nevertheless acknowledge that under more extreme con- ditions a transient change of the magnetization itself, and increased thermal fluctuations cannot be neglected in or- der to obtain a complete picture. Then, more elaborate modelling techniques such as those described in Ref.33 are needed. In the outlook section we discussed a way to over- come the diffraction limit when exciting propagating in- plane spin waves. The basic idea relies on the fact in Fe/Au/Fe trilayer spin valve trilayers8,24, nanostructur- ing of the top, laser-excited layer enables spatial tailor- Figure 7. Excitation of the in-plane propagating spin waves. (a) Conceptual device design. Femtosecond laser pulse ex- cites the periodically striped top Fe layer, resulting in gen- eration of a spatially-modulated spin current pulse. The lat- ter reaches the underlying orthogonally magnetized Fe layer. The resulted spin transfer torque excites propagating spin waves. Magnifying the active area, the inset shows the spin orientations in the Fe layers, and the spin current flow in the non-magnetic spacer. The stripe period is denoted with s, while L = 500 nm is the maximum probing length. (b) Spatio-temporal plot of the averaged out-of-plane magneti- zation component (cid:104)mz(cid:105)a,yz(x, t). (c) 2-dimensional Fourier transform of the data shown in (b), indicating efficient ex- citation of spin waves with wavenumbers governed by the stripes periodicity s, i.e. at k = 2π/s. The white dashed line shows the dispersion of dipole-exchange spin waves according to Ref.30. ing of the spin current profile. In the simplest case, spin currents can only be excited where the top Fe layer ex- ists, thus introducing an in-plane inhomogeneity into the STT stimulus. We have shown that geometrical pat- terning enables the excitation of propagating spin waves with wavelengths considerably smaller than the optical diffraction limit of typically used VIS to NIR-VIS laser sources. Note that the proposed device design lifts the restriction pertinent to the use of epitaxial Fe films for setting up their magnetizations directions. Indeed, the latter can be achieved exploiting the shape anisotropy of the stripes, and an external magnetic field can be used to ensure an orthogonal magnetization in the bot- tom Fe layer, instead of relying on magneto-crystalline anisotropy. As such, spin current-driven excitation of high-frequency spin waves in amorphous ferromagnets (e.g. low-loss CoFeB) or even insulating materials (such as yttrium iron garnet attracting increased attention re- cently) remains an intriguing perspective. To conclude, we emphasize that this work shows that experimental observations from conventional magnonics and recent ultrafast experiments can be explained on equal theoretical footings. We are convinced that our findings open a fruitful perspective for the application of the predictive and interpretative power of micromagnetic simulation in experimental ultrafast magnetism. H.U. acknowledges financial support by the Deutsche Forschungsgemeinschaft within project A06 of the SFB 1073 'Atomic scale control of energy conversion'. The au- thors thank A. Melnikov and C. Seick for valuable com- ments and M. Wolf for continuous support. VI. SUPPLEMENT A. Details of the micromagnetic model In Eq. (1) we neglect the actual propagation of the spin current pulse inside the Fe film with the Fermi velocity ∼ 1 nm/fs, due to the fact that this speed is much larger than the phase and group velocities of the involved magnons. Furthermore, following theo- retical considerations34,35, a field-like torque term was not taken into account in Eq. (2) due to the condition d (cid:29) λSTT. Further experimental support can be ob- tained from Fig. 4(b) in Ref.8, where the P-MOKE sig- nal proportional to mz responds to the STT stimulus with a significant delay. On the contrary, the L-MOKE response immediately follows the spin current-driven ac- cumulation of magnetic moment in Fe, corroborating the dominant role of the damping-like torque term in the STT-induced spin dynamics. If a field-like torque term would be active, the P-MOKE signal mz should respond to the spin current excitation directly. Note that the focus of this work is on accurate mod- elling of exchange-dominated high-frequency spin dy- namics. Given the complicated anisotropy of the spin- wave dispersion30 in the dipolar regime, an accurate re- 8 production of the Ferromagnetic Resonance (FMR) mode and all degenerate in-plane propagating modes is beyond the scope of this work. The reference experiment8 sug- gests that the excitation efficiency of this mode is smaller than that for the first-order PSSW (with n = 1). In our simulations, the experimentally observed ratio of these efficiencies can be achieved by including an additional small external field of Hext = 5 Oe, pointing along y. Variations of the strength of this transverse field affect the FMR mode only while the properties of the PSSW modes remain unchanged. Experimentally, the inciden- tal presence of a transverse magnetic field cannot be ex- cluded either. However, due to uncertainties in its mag- nitude, we keep Hext = 0 , and accept that the excitation efficiency of the FMR mode can be potentially overesti- mated in our model. B. A classical Hamiltonian view 1. General approach The theoretical concepts for the following description were first outlined by H. Suhl36 and V.S. L'vov37. This theory can be regarded as analytic micromagnetic mod- elling. Note that it has been successfully employed to describe spin-current driven magnetization dynamics in conventional magnonic studies38–41. Consider a thin fer- romagnetic film of thickness d = 14.2 nm, supporting magnons with amplitudes bk, frequencies ωk, and relax- ation rates Γk. According to Suhl, the magnetization dy- namics, as described by the Landau-Lifshitz (LL) equa- tion M = −γµ0M × Heff , (7) can be analyzed into plane spin-waves, and the dynamics of these modes can be described by simple kinetic equa- tions. The first step is to linearize the LL-equation (7), and then apply the first Holstein-Primakoff transforma- tion (HPT) to calculate m+ = akeikr = my + i mz, (8) (cid:88) k (cid:88) where mi = Mi/M0. The second HPT takes into ac- count ellipticity of the precession in a tangentially mag- netized film. It finally maps ak to the amplitudes bk. In total, the HPTs diagonalize the Hamiltonian H, which generates the LL equation (7). Without dissipation and interactions, H then simply reads: H = ωkbkb∗ k. (9) k Note that for the dispersion ωk = ω(k) we take the approximation by Eq. (3). The canonical equation of motion for the spin-wave amplitudes is then 9 whereas mz(t, z) = 0. This result implies that one can obtain the temporal shape of the spin current pulse θ(t) by taking the time-derivative of my(t) probed in the corresponding MOKE geometry (e.g. by L-MOKE in Fig. 4(b) from Ref.8). Furthermore, here it is seen why we call this process coherent. In a stroboscopic pump-probe experiment, one always induces deterministic growth of a transverse magnetization component with the same phase. For longer time scales, one needs to consider all terms in Eq. (16). Note that a constant spin current θ(t) = θ0 leads to a new equilibrium orientation of the magneti- zation, whereas a time-limited spin current pulse θ(t) pumps energy into the different spin wave modes. Here we assume that the perturbation is small so that the ori- entation of the magnetization remains unchanged. Ap- plying a time-domain Fourier transformation to Eq. (16) and its complex conjugate we get: iωbk + [iωk + Γk] bk − θ(ω)βk = 0, k + [−iωk + Γk] b∗ k = 0. k − θ∗(ω)β∗ iωb∗ (19) Thus, the power spectrum of a given mode k reads: pk(ω) = bk b∗ k = θ(ω)θ∗(ω)βkβ∗ k − ω2 + Γ2 ω2 k k + 2iωΓk . (20) To obtain the absorbed partial energy εk introduced in Eq. (4), we integrate pk(ω) over the frequency range: (cid:20)(cid:90) ∞ (cid:21) εk ∼ Im pkdω 0 ≈ θ(ωk)θ∗(ωk)βkβ∗ ≈ θ(ωk)θ∗(ωk)βkβ∗ k k tan−1 Γk ωk ωk Γk ω2 k . (21) Here, we can already identify the factors θ2(ω) = θ(ωk)θ∗(ωk), and β2(k) = βkβ∗ k from Eq. (5). Confirm- ing intuitive expectations, both spectral Fourier powers of the spatial and temporal factors directly impact the partial energy uptake of a particular spin current-driven spin wave mode. Furthermore, we now explicitly de- rive the additional factor ξ which appears crucial for the comparisons of the excitation efficiency between different modes. Recall that in the numerical simulation, we actu- ally analyze the dynamics of the out-of-plane component mz. When passing back from the amplitudes bk to the magnetization, one has to take into account the elliptic- ity Ek of the precession. We thus have to acknowledge that (see e.g. chapter 1 in Ref.42) bk + iωkbk = 0. (10) Adding a Gilbert-like dissipation term to the Eq. (7), one gets: bk + [iωk + Γk] bk = 0, (11) where the relaxation rate is given by Γk = αωH ∂ω ∂ωH , (12) with ωH = γµ0H + γµ0 2A M0 k2. 2. Coherent magnon generation with p ⊥ M For the case of our spatially inhomogeneous spin cur- rent, similar to Ref.38 we first introduce the quantity β(t, r) = γ 2eµ0MS j0e−z/λSTT e−t/τ2 1 + e−(t−t0)/τ1 . (13) It is then convenient to consider a Fourier representation of β, and separate out the time-dependence: (cid:88) k (cid:88) (cid:90) t 0 (cid:90) t 0 β(t, r) = θ(t) βkeikr. (14) If p ⊥ M, linearizing the STT term given by Eq. (2) results for the first HPT in: m+ = my + i mz − θ(t) βkeikr. (15) The rate equation for a particular mode amplitude bk then reads: k bk + [iωk + Γk] bk − θ(t)βk = 0. (16) On short time-scales t (cid:28) 1 , the second term in Eq. (16) can be neglected. Direct integration yields, if the initial dynamic amplitude is small: , 2π ωk Γk bk(t) = βk θ(t(cid:48))dt(cid:48). (17) Summation over the all k modes and Fourier transforma- tion back into real space gives my(t, z) = β(z) θ(t(cid:48))dt(cid:48), (18) mk,z2 mk,y2 = 1 − Ek, (22) where (cid:32) 1 + Ek = (cid:33)−1 . (23) γµ0H + γµ0 γµ0M0 2A M0 k2 When analyzing the partial energies of the modes found in the spectrum of mz, the mode specific correction yields for the third factor in Eq. (5): ξ = 1 − Ek 2 − Ek . Γk ω2 k (24) Incoherent magnon generation with p (cid:107) M 3. We will now consider spin injection into a thermally If we assume a spatially occupied magnon ensemble. homogeneous spin current, whose polarization p (cid:107) M, the linearized contribution of the STT term to the LL- equation (7) results in: m+ = my + i mz + β(t)m+, (25) with the temporal dependence of β given by Eq. (1). The rate equation for bk reads38: bk + [iωk + Γk + β(t)] bk = Fk, (26) , and F represents a thermal noise where β(t) = γ js(t) dµ0M0 source. Note that in Eq. (27) higher order terms in bk, which result from the second HPT, are not taken into account38. Note that the product β(t)bk in Eq. (27) explains why the excitation process is incoherent: The random phase of a thermally driven magnon gets im- printed on the spin current term. Therefore in a pump- probe experiment, the temporal evolution of the indi- vidual mode's phases will differ from shot to shot in a random fashion. should again consider β = θ(t)(cid:80) If the spin current is spatially inhomogeneous, one k βkeikr. Then, in prin- ciple, different βk induce a mixing between the magnon modes bk. The rate equation then reads: bk + [iωk + Γk] bk + δ(k(cid:48) + k(cid:48)(cid:48) − k)βk(cid:48)bk(cid:48)(cid:48) = Fk. (27) (cid:88) k(cid:48),k(cid:48)(cid:48) For the experimental situation of a laterally homoge- neous, but vertically inhomogeneous spin current, the mixing couples the modes with different kz but equal kx and ky. To simulate the dynamics of a thermal magnon ensemble, we consider a volume of v = dL2, and quan- tize the wavenumbers according to kx = nπ L , ky = mπ L , 10 d , with L = 10d. Modes up to m, n = ±256, kz = oπ and o = 24 are taken into account. For the implemen- tation of Fk we have chosen a Gaussian random number generator, which obeys (cid:104)FkF ∗ k(cid:48)(cid:105) = 2δ(k − k(cid:48))Γk kBT ωk . (28) In the absence of a spin current, this provides in the temporal average an equilibrium magnon number density of beq k 2 = (cid:104)bkb∗ k(cid:105) = kBT ωk . (29) of magnons (cid:80) To probe the dynamics, we determine the total number k. Since each magnon carries 2µB, k bkb∗ this reduces the magnetization to (cid:115) (cid:115) (cid:88) k (cid:88) k mx = 1 − 2µB vMs bkb∗ k (30) Similar to that in the analysis of the numerical simula- tions output, we then normalize the results by the equi- librium magnetization meq x meq x = 1 − 2µB vMs and shift by 1 to obtain ∆mx = − 1. mx meq x kBT ωk , (31) (32) Note that for comparing this analytic model with the micromagnetic model, we neglected the cross coupling terms in Eq. (27) since those cancel out in the statistical average. Figure 8 shows the results obtained for positive and negative spin current pulses of equal magnitude, at three different temperatures. The data look quite similar to those shown in Figure 6(c). For instance, for positive spin injection, after a fast initial increase which satu- rates 350 fs after the maximum of the injection density, the magnon numbers decrease approximately exponen- tially with a decay time of 1.48(1) ps. This value is quite close to the one obtained in the numerical simulations (1.79(1) ps). Small differences in decay time and in the overall magnitude can be attributed to the fact that in the rate equations (27) interactions between the magnons are not taken into account. In contrast, in the micromag- netic simulation, the underlying fully nonlinear Landau- Lifshitz equation captures these processes. Also, slightly different compositions of the magnon ensembles in the analytic and numerical description can give rise to differ- ent decay rates. 11 Figure 8. Comparison of the spin dynamics according to the classical Hamiltonian model and to the numerical micromag- netic modelling. The red and blue lines are obtained for the opposite polarities of the spin current pulse. (a) Relative vari- ations of averaged magnon number (cid:104)∆mx(cid:105) from Eq. (32) as a function of time, obtained from the Hamiltonian model given by Eq. (27). Computation for different temperatures as in- dicated. (b) Corresponding micromagnetic simulation data, also shown in Figure 6(c). The dashed line indicates the peak time of the injected spin current pulse. ∗ [email protected] 1 E. Beaurepaire, J.-C. Merle, A. Daunois, and J.-Y. Bigot, Phys. Rev. Lett. 76, 4250 (1996). 2 A. Kirilyuk, A. V. Kimel, and T. Rasing, Rev. Mod. Phys. 82, 2731 (2010). netic Materials 320, 1238 (2008). 12 M. Djordjevic and M. Munzenberg, Phys. Rev. B 75, 012404 (2007). 13 B. Lenk, H. Ulrichs, F. Garbs, Physics Reports 507, 107 (2011). and M. Munzenberg, 3 J. M. Kikkawa and D. D. Awschalom, Phys. Rev. Lett. 80, 14 R. Hertel, W. Wulfhekel, and J. Kirschner, Phys. Rev. 4313 (1998). Lett. 93, 257202 (2004). 4 J.-Y. Bigot, M. Vomir, and E. Beaurepaire, Nat. Phys. 5, 15 V. V. Kruglyak, S. O. Demokritov, and D. Grundler, Jour- 515 (2009). 5 M. van Kampen, C. Jozsa, J. T. Kohlhepp, P. LeClair, L. Lagae, W. J. M. de Jonge, and B. Koopmans, Phys. Rev. Lett. 88, 227201 (2002). 6 T. Satoh, Y. Terui, R. Moriya, B. A. Ivanov, K. Ando, E. Saitoh, T. Shimura, and K. Kuroda, Nat. Photon. 6, 662 (2012). 7 Y. Au, M. Dvornik, T. Davison, E. Ahmad, P. S. Keat- ley, A. Vansteenkiste, B. Van Waeyenberge, and V. V. Kruglyak, Phys. Rev. Lett. 110, 097201 (2013). 8 I. Razdolski, A. Alekhin, N. Ilin, J. P. Meyburg, V. Rod- datis, D. Diesing, U. Bovensiepen, and A. Melnikov, Na- ture Communications 8, 15007 (2017). 9 M. L. M. Lalieu, P. L. J. Helgers, and B. Koopmans, Phys. Rev. B 96, 014417 (2017). 10 E. H. Frei, S. Shtrikman, and D. Treves, Phys. Rev. 106, 446 (1957). 11 D. Berkov and J. Miltat, Journal of Magnetism and Mag- nal of Physics D: Applied Physics 43, 264001 (2010). 16 S.-K. Kim, Journal of Physics D: Applied Physics 43, 264004 (2010). 17 J. Iwasaki, M. Mochizuki, and N. Nagaosa, Nature Nan- otechnology 8, 742 (2013). 18 M. Dvornik, Y. Au, and V. V. Kruglyak, "Micromagnetic simulations in magnonics," in Magnonics: From Funda- mentals to Applications, edited by S. O. Demokritov and A. N. Slavin (Springer Berlin Heidelberg, Berlin, Heidel- berg, 2013) pp. 101–115. 19 H. Ulrichs, V. E. Demidov, and S. O. Demokri- tov, Applied Physics Letters 104, 042407 (2014), https://doi.org/10.1063/1.4863660. 20 C. Banerjee, P. Gruszecki, J. W. Klos, O. Hellwig, M. Krawczyk, and A. Barman, Phys. Rev. B 96, 024421 (2017). 21 P. Bal´az, M. Zonda, K. Carva, P. Maldonado, and P. M. Oppeneer, ArXiv e-prints (2017), arXiv:1710.02083 [cond- 12 mat.mes-hall]. 22 V. E. Demidov, S. Urazhdin, E. R. J. Edwards, M. D. Stiles, R. D. McMichael, and S. O. Demokritov, Phys. Rev. Lett. 107, 107204 (2011). 23 H. A. Mook and R. M. Nicklow, Phys. Rev. B 7, 336 (1973). 24 A. Alekhin, I. Razdolski, N. Ilin, J. P. Meyburg, D. Diesing, V. Roddatis, I. Rungger, M. Stamenova, S. Sanvito, U. Bovensiepen, and A. Melnikov, Phys. Rev. Lett. 119, 017202 (2017). 25 A. Vansteenkiste, J. Leliaert, M. Dvornik, M. Helsen, F. Garcia-Sanchez, and B. V. Waeyenberge, AIP Advances 4, 107133 (2014), http://dx.doi.org/10.1063/1.4899186. 26 J. Slonczewski, Journal of Magnetism and Magnetic Ma- terials 159, L1 (1996). 27 A. Melnikov, I. Razdolski, T. O. Wehling, E. T. Papaioan- nou, V. Roddatis, P. Fumagalli, O. Aktsipetrov, A. I. Lichtenstein, and U. Bovensiepen, Phys. Rev. Lett. 107, 076601 (2011). 28 M. Battiato and K. Held, Phys. Rev. Lett. 116, 196601 (2016). 29 G.-M. Choi and D. G. Cahill, Phys. Rev. B 90, 214432 (2014). 30 B. A. Kalinikos and A. N. Slavin, Journal of Physics C: Solid State Physics 19, 7013 (1986). 31 J. Zak, E. Moog, C. Liu, and S. Bader, Journal of Mag- netism and Magnetic Materials 89, 107 (1990). 32 E. Turgut, C. La-o vorakiat, J. M. Shaw, P. Grychtol, H. T. Nembach, D. Rudolf, R. Adam, M. Aeschlimann, C. M. Schneider, T. J. Silva, M. M. Murnane, H. C. Kapteyn, and S. Mathias, Phys. Rev. Lett. 110, 197201 (2013). 33 R. F. L. Evans, D. Hinzke, U. Atxitia, U. Nowak, R. W. Chantrell, and O. Chubykalo-Fesenko, Phys. Rev. B 85, 014433 (2012). 34 M. A. Zimmler, B. Ozyilmaz, W. Chen, A. D. Kent, J. Z. Sun, M. J. Rooks, and R. H. Koch, Phys. Rev. B 70, 184438 (2004). 35 C. Abert, H. Sepehri-Amin, F. Bruckner, C. Vogler, M. Hayashi, and D. Suess, Phys. Rev. Applied 7, 054007 (2017). 36 H. Suhl, Journal of Physics and Chemistry of Solids 1, 209 (1957). 37 V. S. L'vov, Wave Turbulence Under Parametric Excita- tion (Springer-Verlag Berlin Heidelberg, 1994). 38 S. M. Rezende, F. M. de Aguiar, and A. Azevedo, Phys. Rev. Lett. 94, 037202 (2005). 39 A. N. Slavin and P. Kabos, IEEE Transactions on Mag- netics 41, 1264 (2005). 40 A. Slavin and V. Tiberkevich, Phys. Rev. Lett. 95, 237201 (2005). 41 V. E. Demidov, S. Urazhdin, H. Ulrichs, V. Tiberkevich, A. Slavin, D. Baither, G. Schmitz, and S. O. Demokritov, Nature Materials 11, 1028 (2012). 42 A. Gurevich and G. Melkov, Magnetization Oscillations and Waves (Taylor & Francis, 1996).
1808.08861	2	1808	2018-08-28T05:13:38	Transverse instability of dark solitons in spin-orbit coupled polariton condensates	[ "cond-mat.mes-hall", "nlin.PS", "physics.optics" ]	We consider dark solitons and their stability in spin-orbit coupled polariton condensates. The system supports spinor solitons of two types: conventional (symmetric) dark solitons and asymmetric half-dark solitons. They demonstrate essentially different behavior upon variation of the strength of spin-orbit coupling. One-dimensional spin-orbit coupled dark solitons are usually unstable, while half-dark solitons can be stable. Two-dimensional dark solitons at early stages of the development of transverse instabilities turn into asymmetric snaking patterns and later into sets of vortex-antivortex solitons with notably different shapes. Depending on the sign of spin-orbit coupling two distinct instability scenarios are possible for such solitons, in which vortices in one component correspond to vortices or antivortices in other component. The decay of two-dimensional half-dark solitons results in the formation of half-vortex chains.	cond-mat.mes-hall	cond-mat	a Transverse instability of dark solitons in spin-orbit coupled polariton condensates Dmitry A. Zezyulin1 and Yaroslav V. Kartasov2,3 1ITMO University, St. Petersburg 197101, Russia 2ICFO-Institut de Ciencies Fotoniques, The Barcelona Institute of Science and Technology, 08860 Castelldefels (Barcelona), Spain and 3Institute of Spectroscopy, Russian Academy of Sciences, Troitsk, Moscow, 108840, Russia We consider dark solitons and their stability in spin-orbit coupled polariton condensates. The system supports spinor solitons of two types: conventional (symmetric) dark solitons and asymmetric half-dark solitons. They demonstrate essentially different behavior upon variation of the strength of spin-orbit coupling. One-dimensional spin-orbit coupled dark solitons are usually unstable, while half-dark solitons can be stable. Two-dimensional dark solitons at early stages of the development of transverse instabilities turn into asymmetric snaking patterns and later into sets of vortex-antivortex solitons with notably different shapes. Depending on the sign of spin-orbit coupling two distinct instability scenarios are possible for such solitons, in which vortices in one component correspond to vortices or antivortices in other component. The decay of two-dimensional half-dark solitons results in the formation of half-vortex chains. Introduction. Dark solitons are the fundamental soli- ton solutions of the nonlinear Schrodinger equation, which exist when the product of coefficients determining dispersion and nonlinearity strength is negative [1]. One- dimensional (1D) dark soliton represents a density dip residing on the modulationally stable background. Dark solitons were observed in diverse systems, including opti- cal fibers [2, 3] and Bose-Einstein condensates [4]. Being stable entities in the 1D case, dark solitons in 2D usually suffer from transverse "snake" instabilities [5], leading to their decay into sets of vortex-antivortex solitons ob- served experimentally [6, 7], see also reviews [8, 9]. In the case, when dark solitons are generated in supersonic flows past localized obstacles, their snake instability turns into convective one [10]. Such solitons elongate faster than they are destroyed by the instability, that allowed their observation in polariton condensates [11]. Polariton condensates forming in planar microcavities and exhibiting strong nonlinear effects due to repulsive exciton-exciton interactions, represent an ideal platform for the exploration of physics of dark and vortex solitons [12]. Polariton condensates support formation of oblique dark [11] and bright [13] solitons, nonequilibrium dark solitons supported by resonant [14, 15] and nonresonant [16 -- 19] pump, spontaneous formation [20] and nucleation [21 -- 23] of vortices in the flow past obstacles, etc. Polar- ization phenomena arising from spin degree of freedom substantially enrich evolution of condensates [24] leading to new types of spinor solitons, such as half-dark [25, 26] and half-vortex solitons [27]. Inclusion of polarization effects requires taking into account TE-TM splitting of polariton energy levels, that can be interpreted as an ef- fective spin-orbit coupling (SOC) [28, 29]. This coupling dramatically changes properties and the very structure of polariton solitons [30]. The effect of SOC was discussed upon generation of oblique half-dark solitons [25], but its influence on sta- bility of dark polariton solitons remains practically unex- plored. Thus, in [17] a reduced to usual linear coupling form of SOC was considered. The only work [19] deal- ing with decay of two-dimensional dark polariton solitons did not take into account polarization effects. The im- pact of SOC on stability of dark solitons was considered in Bose-Einstein condensates [31, 32] and optical waveg- uides [33], where this coupling has completely different form and physical origin than coupling in polaritonic sys- tems. The aim of this Letter is to show that even small SOC drastically affects the development of the transverse in- stability (TI) of dark polariton solitons, leading to com- pletely different dynamical patterns from those for scalar dark solitons. They are composed from vortex-antivortex pairs, whose disposition depends on the sign of SOC. Due to SOC, vortices and antivortices in emerging pairs ac- quire different shapes. Two different regimes are encoun- tered, when for positive/negative coupling strength the appearance of vortex in one spin component is followed by the formation of antivortex/vortex in other compo- nent. We also found that half-dark solitons split into half-dark vortex pairs. The model. We consider spin-orbit coupled polariton condensate in a 2D microcavity. We are interested in quasi-conservative dynamics governed by the normalized equations [30]: i∂tψ± = (cid:2)− 1 2 ∇2 + ψ±2 + σψ∓2(cid:3) ψ± + β(∂x ∓ i∂y)2ψ∓, In the experiment, the unavoidable losses are expected to affect the polariton dynamics only quantitatively (see e.g. [34]), e.g. one may expect gradual shrinkage of instabil- ity band due to decrease of peak amplitude. According to the recent experiments [35], the polariton lifetime can be increased to hundreds of picoseconds in high-quality microcavity samples that substantially exceeds time in- tervals considered below. ψ(x, y, t) = (ψ+, ψ−)T is the spinor wavefunction in circular polarization basis; ∂t,x,y are the partial derivatives; ∇2 = ∂ 2 y is the Laplacian; x +∂ 2 2 FIG. 2. Maximal instability increment as a function of mod- ulation frequency ky and SOC strength β for dark (a) and halfdark (b) solitons. cus on symmetric modes. The solution (1) immediately shows that the increase of β from negative to positive values leads to soliton steepening, while in the limit β → 1/2 (which however may not be practically reach- able) the soliton becomes infinitely narrow for a fixed µ, see Fig. 1(c). Besides conventional dark solitons, the system also supports asymmetric half-dark solitons with significantly different shapes of the two components. The latter can be found numerically and their transformation with β is illustrated in Figs. 1(a,b). For small β values half-dark solitons feature a prominent density dip only in the u+ component, with u− component being nearly uniform. The increase of SOC strength to larger positive or nega- tive values leads to steepening of the u+ component and development of a dip surrounded by two maxima in the u− component. Remarkably, for half-dark solitons inver- sion of the sign of β does not change the shape of solution. This is in clear contrast to usual dark solitons that have different widths for positive and negative β values, see Fig. 1(c). This suggests that stability properties of half- dark solitons depend on β only, while for dark solitons the sign of β is important too. Transverse instability. First, we address the spec- tral instability of dark solitons with respect to small- amplitude two-dimensional perturbations. We consider perturbed solutions in the form ψ± = e−iµt[u±(x) + ±(x)e−iky y+λ∗t], where p±(x) and q±(x) p±(x)eiky y+λt+q∗ are small perturbations localized in x, ky is the spa- tial frequency of perturbation in the y direction, and λ = λre + iλim is a complex eigenvalue, whose real part λre determines the instability growth rate. Lineariza- tion around stationary solution u± leads to the eigen- value problem for λ, which can be solved numerically for different values of SOC strength β and frequency ky. It is well known [5] for β = σ = 0 that usual dark soli- tons are unstable for the modulation frequencies within the interval 0 < ky2 < 1. This situation changes signif- icantly when SOC strength becomes nonzero, and weak FIG. 1. Modification of the shape of u+ (a) and u− (b) com- ponents of halfdark soliton and of u± components (c) of dark soliton upon variation of SOC strength β. Here and in figures below µ = 1 and σ = −0.05. time t and spatial coordinates x, y are measured in units which ensure = 1 and m∗ = 1, where m∗ is the ef- fective polariton mass. The model accounts for repulsive interactions of polaritons with the same spin, while small σ < 0 characterizes the attraction of polaritons with op- posite spins (we set σ = −0.05, that is the experimentally relevant value); β is the SOC strength proportional to the difference of effective masses of TE and TM polaritons. Soliton solutions. We look for stationary quasi-one- dimensional solutions that depend only on one spatial coordinate x and are uniform in y. Using the substi- tution ψ± = e−iµtu±(x), where µ is the chemical po- tential, and solving the resulting system for u±(x), we have encountered solutions of two types. Solution of the first type is obtained using simple reduction in the form u−(x) = ±u+(x), i.e., it represents either symmetric (up- per sign) or antisymmetric (lower sign) state, where u+ is the conventional tanh-shaped dark soliton: u+(x) = r µ 1 + σ tanh(cid:18)r µ 1 ∓ 2β x(cid:19) , (1) where sign in the denominator should be chosen accord- ing to type of solution. The properties (including stabil- ity) of symmetric and antisymmetric modes are similar (the properties of one family are obtained from those of other family by the inversion of β sign), so we fo- 3 FIG. 3. Decay of the dark soliton at β = 0. ψ+ and ψ− components are shown in the left and right columns, respec- tively. For t = 27 we also show phases arg ψ±. All distri- butions are shown within (x, y) ∈ [−15, 15] × [−33.4, 33.4] windows. See detailed dynamics in Visualization 1. FIG. 5. Decay of dark soliton at β = 0.1. ψ+ and ψ− are shown in left and right columns, respectively, within (x, y) ∈ [−15, 15] × [−33.4, 33.4] windows. For t = 27 both amplitudes and phases are shown. See Visualization 3. solitons with representative chemical potential µ = 1 on the plane of parameters (ky, β). Notice that instability band is asymmetric in β as it was supposed above. It no- tably expands with increase of β. Fig. 2(a) shows that even weak attraction of cross-polarized polaritons leads to the 1D instability of dark solitons at ky = 0. More- over, the increase of SOC strength β to large (positive or negative) values clearly enhances one-dimensional in- stability. For β > 0.1 the instability associated with ky = 0 is the strongest one, i.e. the corresponding in- crement is maximal among all ky. Thus for large β, if instability is seeded by noisy perturbations, the snaking may not occur and dynamics will be ruled mostly by the 1D instability. Therefore, proceeding to investigation of dynamical development of TI, we limit the considera- tion to relatively small values of β. In Figs. 3, 4 and 5 we illustrate the development of TI seeded by small ini- tial perturbations with the same frequency ky ≈ 0.5 for three representative values of β. In the absence of SOC (β = 0, Fig. 3), TI results in the development of sym- metric snaking pattern (identical in both components) from input dark soliton stripe, which then breaks into alternating vortices and antivortices, i.e., topological de- fects with winding numbers equal to +1 and −1. For FIG. 4. Decay of the dark soliton at β = −0.1. ψ+ and ψ− are shown in left and right columns, respectively, within (x, y) ∈ [−15, 15] × [−33.4, 33.4] windows. For t = 50 both amplitudes and phases are shown. See Visualization 2. cross-attraction σ is taken into account. In Fig. 2(a) we plot the maximal instability increment computed for dark nonzero β (Fig. 4 and 5) the emerging snaking patterns are strongly asymmetric and different in two components, they break into segments of unequal lengths, such that longer segments rotate in the (x, y) plane and direction of rotation is opposite in the ψ+ and ψ− components. Notice the appearance of elements with three close vor- tices that alternate with antivortices in Fig. 4 that was never observed in scalar systems. Strong asymmetry in shapes of vortices and antivortices in the chain due to SOC is a distinctive feature of this system. At later stages of the instability development, one observes the formation of vortex-antivortex pairs. At the same time, the comparison of lower panels of Figs. 4 and 5 reveals qualitative difference of phase patterns for opposite signs of SOC. Indeed, in Fig. 4 the position of each vortex in ψ+ component approximately coincides with a vortex in ψ− component, and each antivortex in ψ+ corresponds to an antivortex in ψ−. In Fig. 5 one observes completely different picture where each vortex in ψ+ component cor- responds to an antivortex in ψ−, and, vice versa, each antivortex in ψ+ corresponds to a vortex in ψ−. This phenomenon does not exist in scalar systems. Now we turn to TI of asymmetric half-dark solitons. Their instability increments are plotted in Fig. 2(b) on the (ky, β) plane. The dependence is symmetric in β. In contrast to dark solitons [Fig. 2(a)], half-dark ones are most unstable with respect to perturbations with nonzero modulation frequency ky (except for large val- ues of β close to ±1/2). Moreover, for small values of SOC strength β < 0.2, 1D half-dark solitons are com- pletely stable, i.e., λre = 0 at ky = 0. Half-dark soli- tons demonstrate very unusual TI, dynamics depicted in Fig. 6, where one again observes the asymmetric snaking pattern, which however is well-pronounced only in the ψ+ component, while ψ− component features only rela- tively weak density modulations, even though the SOC strength in this case was relatively large (β = 0.4). Re- spectively, at later stages of TI development one observes the excitation of the chains of half-vortices (i.e. vortices appear only in ψ+ component), where nearly circularly symmetric states alternate with extremely asymmetric ones. To conclude, SOC significantly affects stability of 1D and 2D dimensional dark and half-dark solitons in po- lariton condensates. SOC leads to unusual asymmetric snaking pattern at early stages of TI development, and at later stages solutions break into vortex-antivortex pairs such that the resulting phase distribution can be distinc- tively different for opposite signs of SOC. FUNDING INFORMATION Government of Russian Federation (Grant 08-08). Ministry of Education and Science of Russian Federation (Megagrant No. 14.Y26.31.0015). Severo Ochoa program 4 FIG. 6. Decay of the half-dark soliton with β = 0.4. ψ+ and ψ− correspond to left and right columns, respectively. For t = 35 both amplitudes and phases are shown. All panels are shown within (x, y) ∈ [−10, 10] × [−33.4, 33.4] windows. See Visualization 4. (SEV-2015-0522) of the Government of Spain, Fundacio Cellex, Fundaci Mir-Puig. [1] V. E. Zakharov and A. B. Shabat, "Exact theory of two-dimensional self-focusing and one-dimensional self- modulation of wave in nonlinear media," Sov. Phys. J. Exp. Theor. Phys. 34, 62 (1972). [2] P. Emplit, J. P. Hamaide, R. Reynard, C. Froehly, and A. Barthelemy, "Picosecond steps and dark pulses through nonlinear single-mode fibers," Opt. Commun. 62, 374 (1987). [3] D. Krokel, N. J. Halas, G. Giuliani, and D. Grischkowsky, Dark-pulse propagation in optical fibers", Phys. Rev. Lett. 60, 29 (1988). [4] S. Burger, K. Bongs, S. Dettmer,W. Ertmer, K. Seng- stock, A. Sanpera, G. V. Shlyapnikov, and M. Lewen- stein, "Dark solitons in Bose-Einstein condensates," Phys. Rev. Lett. 83, 5198 (1999). [5] E. A. Kuznetsov, S. K. Turitsyn, "Instability and collapse of solitons in media with a defocusing nonlinearity," Sov. Phys. J. Exp. Theor. Phys. 67, 1583 (1988). [6] A. V. Mamaev, M. Saffman, and A. A. Zozulya, "Prop- agation of dark stripe beams in nonlinear media: Snake instability and creation of optical vortices," Phys. Rev. Lett. 76, 2262 (1996). [7] Z. Dutton, M. Budde, C. Slowe, and L. V. Hau, "Ob- servation of quantum shock waves created with ultra- compressed slow light pulses in a Bose-Einstein conden- sate," Science 293, 663 (2001). [8] Y. S. Kivshar and B. Luther-Davies, "Dark optical soli- tons: physics and applications," Phys. Rep. 298, 81 (1998). [9] Y. S. Kivshar and D. E. Pelinovsky, "Self-focusing and transverse instabilities of solitary waves," Phys. Rep. 331, 117 (2000). [10] A. M. Kamchatnov and L. P. Pitaevskii, "Stabilization of solitons generated by a supersonic flow of Bose-Einstein condensate past an obstacle," Phys. Rev. Lett. 100, 160402 (2008). [11] A. Amo, S. Pigeon, D. Sunvitto, V. G. Sala, R. Hivet, I. Carusotto, F. Pisanello, G. Lemenager, R. Houdre, E. Giacobino, C. Ciuti, and A. Bramati, "Polariton super- fluids reveal quantum hydrodynamic solitons," Science 332, 1167 (2011). [12] I. Carusotto and C. Ciuti, "Quantum fluids of light," Rev. Mod. Phys. 85, 299 (2013). [13] M. Sich, D. N. Krizhanovskii, M. S. Skolnick, A. V. Gor- bach, R. Hartley, D. V. Skryabin, E. A. Cerda-Mendez, K. Biermann, R. Hey, and P. V. Santos, "Observation of bright solitons in a semiconductor microcavity," Nat. Photon. 6, 50 (2012). [14] A. V. Yulin, O. A. Egorov, F. Lederer, and D. V. Skryabin, "Dark polariton solitons in semiconductor mi- crocavities," Phys. Rev. A 78, 061801(R) (2008). [15] V. Goblot, H. S. Nguyen, I. Carusotto, E. Galopin, A. Lematre, I. Sagnes, A. Amo, and J. Bloch, "Phase- controlled bistability of a dark soliton train in a polariton fluid," Phys. Rev. Lett. 117, 217401 (2016). [16] Y. Xue and M. Matuszewski, "Creation and abrupt decay of a quasistationary dark soliton in a polariton conden- sate," Phys. Rev. Lett. 112, 216401 (2014). [17] F. Pinsker and H. Flayac, "On-demand dark soliton train manipulation in a spinor polariton condensate," Phys. Rev. Lett. 112, 140405 (2014). [18] T.-W. Chen, S.-D. Jheng, W.-F. Hsieh, S.-C. Cheng, "Dark solitons in a spinor microcavity-polariton conden- sate," Superlattices Microstruct. 98, 96 (2016). [19] L. A. Smirnov, D. A. Smirnova, E. A. Ostrovskaya, and Yu. S. Kivshar, "Dynamics and stability of dark soli- tons in exciton-polariton condensates," Phys. Rev. B 89, 235310 (2014). [20] K. G. Lagoudakis, M. Wouters, M. Richard, A. Baas, I. Carusotto, R. Andr, Le Si Dang, and B. Deveaud Pldran, "Quantized vortices in an excitonpolariton condensate," Nat. Phys. 4, 706 (2008) [21] G. Nardin, G. Grosso, Y. Leger, B. Pietka, F. Morier- Genoud, D. Deveaud-Pledran, "Hydrodynamic nucle- ation of quantized vortex pairs in a polariton quantum fluid," Nat. Phys. 7, 635 (2011). [22] S. Pigeon, I. Carusotto, and C. Ciuti, "Hydrodynamic nucleation of vortices and solitons in a resonantly excited polariton superfluid," Phys. Rev. B 83, 144513 (2011). [23] G. Grosso, G. Nardin, F. Morier-Genoud, Y. Leger, and B. Deveaud-Pledran, "Soliton instabilities and vortex 5 street formation in a polariton quantum fluid," Phys. Rev. Lett. 107, 245301 (2011). [24] A. Werner, O. A. Egorov, and F. Lederer, "Spin dynam- ics of dark polariton solitons," Phys. Rev. B 85, 115315 (2012). [25] H. Flayac, D. D. Solnyshkov, and G. Malpuech, "Oblique half-solitons and their generation in exciton-polariton condensates," Phys. Rev. B 83, 193305 (2011). [26] R. Hivet, H. Flayac, D. D. Solnyshkov, D. Tanese, T. Boulier, D. Andreoli, E. Giacobino, J. Bloch, A. Bra- mati, G. Malpuech, and A. Amo, "Half-solitons in a po- lariton quantum fluid behave like magnetic monopoles," Nat. Phys. 8, 724 (2012). [27] K. G. Lagoudakis, T. Ostatnick, A. V. Kavokin, Y. G. Rubo, R. Andr, B. Deveaud-Pldran, "Observation of half-quantum vortices in an exciton-polariton conden- sate," Science 326, 974 (2009). [28] V. G. Sala, D. D. Solnyshkov, I. Carusotto, T. Jacqmin, A. Lemaitre, H. Teras, A. Nalitov, M. Abbarchi, E. Ga- lopin, I. Sagnes, J. Bloch, G. Malpuech, and A. Amo, "Spin-orbit coupling for photons and polaritons in mi- crostructures," Phys. Rev. X 5, 011034 (2015). [29] S. Dufferwiel, F. Li, E. Cancellieri, L. Giriunas, A. A. P. Trichet, D. M. Whittaker, P. M. Walker, F. Fras, E. Clarke, J. M. Smith, M. S. Skolnick, D. N. Krizhanovskii, "Spin textures of exciton-polaritons in a tunable micro- cavity with large TE-TM splitting," Phys. Rev. Lett. 115, 246401 (2015). [30] H. Flayac, I. A. Shelykh, D. D. Solnyshkov, and G. Malpuech, "Topological stability of the half-vortices in spinor exciton-polariton condensates," Phys. Rev. B 81, 045318 (2010). [31] V. Achilleos, J. Stockhofe, P. G. Kevrekidis, D. J. Frantzeskakis, and P. Schmelcher, "Matter-wave dark solitons and their excitation spectra in spin-orbit cou- pled Bose-Einstein condensates," Europhys. Lett. 103, 20002 (2013). [32] Y. V. Kartashov and A. M. Kamchatnov, "Wave patterns generated by a flow of a two-component Bose-Einstein condensate with spin-orbit interaction past a localized obstacle," Europhys. Lett. 107, 10008 (2014). [33] Y. V. Kartashov, V. V. Konotop, and B. A. Malomed, "Dark solitons in dual-core waveguides with dispersive coupling," Opt. Lett. 40, 4126 (2015). [34] M. Walker, L. Tinkler, D. V. Skryabin, A. Yulin, B. Royall, I. Farrer, D. A. Ritchie, M. S. Skolnik, D. N. Krizhanovskii, "Ultra-low-power hybrid light-matter soli- tons." Nat. Commun. 6, 8317 (2015). [35] Y. Sun, P. Wen, Y. Yoon, G. Liu, M. Steger, L. N. Pfeif- fer, K. West, D. W. Snoke, K. A. Nelson, "Bose-Einstein Condensation of Long-Lifetime Polaritons in Thermal Equilibrium." Phys. Rev. Lett. 118, 016602 (2017).
1712.09062	1	1712	2017-12-25T12:34:32	Fundamental intrinsic lifetimes in semiconductor self-assembled quantum dots	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci", "quant-ph" ]	The self-assembled quantum dots (QDs) provide an ideal platform for realization of quantum information technology because it provides on demand single photons, entangled photon pairs from biexciton cascade pro- cess, single spin qubits, and so on. The fine structure splitting (FSS) of exciton is a fundamental property of QDs for thees applications. From the symmetry point of view, since the two bright exciton states belong to two different representations for QDs with C2v symmetry, they should not only have different energies, but also have different lifetimes, which is termed exciton lifetime asymmetry. In contrast to extensively studied FSS, the investigation of the exciton lifetime asymmetry is still missed in literature. In this work, we carried out the first investigation of the exciton lifetime asymmetry in self-assembled QDs and presented a theory to deduce lifetime asymmetry indirectly from measurable qualities of QDs. We further revealed that intrinsic lifetimes and their asymmetry are fundamental quantities of QDs, which determine the bound of the extrinsic lifetime asymmetries, polarization angles, FSSs, and their evolution under uniaxial external forces. Our findings provide an important basis to deeply understanding properties of QDs.	cond-mat.mes-hall	cond-mat	Fundamental intrinsic lifetimes in semiconductor self-assembled quantum dots Wen Xiong,1 Jun-Wei Luo,2, 3, 4, ∗ Xiulai Xu,5 Ming Gong,6, 3, † Shu-Shen Li,2, 3, 7 and Guang-Can Guo6, 3 1Department of Physics, Chongqing University, Chongqing 401331, China 2State Key Laboratory of Superlattices and Microstructures, Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, China 3Synergetic Innovation Center of Quantum Information and Quantum Physics, University of Science and Technology of China, Hefei, Anhui 230026, China 4College of Materials Science and Opto-Electronic Technology, University of Chinese Academy of Sciences, Beijing 100049, China 5Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, Chinese Academy of Sciences, Beijing 100190, P. R. China 6Key Laboratory of Quantum Information, University of Science and Technology of China, CAS, Hefei, 230026, Peoples Republic of China 7College of Materials Science and Opto-Electronic Technology, University of Chinese Academy of Sciences, Beijing .100049, China (Dated: December 27, 2017) The self-assembled quantum dots (QDs) provide an ideal platform for realization of quantum information technology because it provides on demand single photons, entangled photon pairs from biexciton cascade pro- cess, single spin qubits, and so on. The fine structure splitting (FSS) of exciton is a fundamental property of QDs for thees applications. From the symmetry point of view, since the two bright exciton states belong to two different representations for QDs with C2v symmetry, they should not only have different energies, but also have different lifetimes, which is termed exciton lifetime asymmetry. In contrast to extensively studied FSS, the investigation of the exciton lifetime asymmetry is still missed in literature. In this work, we carried out the first investigation of the exciton lifetime asymmetry in self-assembled QDs and presented a theory to deduce lifetime asymmetry indirectly from measurable qualities of QDs. We further revealed that intrinsic lifetimes and their asymmetry are fundamental quantities of QDs, which determine the bound of the extrinsic lifetime asymmetries, polarization angles, FSSs, and their evolution under uniaxial external forces. Our findings provide an important basis to deeply understanding properties of QDs. PACS numbers: 78.67.Hc, 42.50.-p, 73.21.La The self-assembled quantum dots (QDs) provide a promis- ing platform for realizing on-demand entangled photon pairs from the biexciton-exciton-vacuum cascade process[1], which are essential for practical quantum communication[2–7]. However the major obstacle in realizing this goal comes from the non-degeneracy of the two intermediate bright exciton states (see Fig. 1a), in which their energy difference, called fine structure splitting (FSS), is much larger than the homoge- neous broadening of the emission lines (Γ ∼ 1 µeV[8–10]), thus the "which-way" information is erased and only classi- cally corrected photons instead of maximally entangled pho- ton pairs can be created from this cascade process. In the past decade, strenuous efforts have been devoted to eliminate this splitting by applying various experimental techniques, includ- ing thermal annealing[11–15], electric field[16–23], magnetic field[24–27] and external stress[28–38] etc.. However, none of them is efficient. In recent years, the entangled photon pairs were demonstrated in a way that first picking out the QDs with small FSS from a QDs ensemble after post annealing and then eliminating the FSS using magnetic fields (see more details in the first experiment by Stevenson et al. in 2006[27]). It is worth to note that only a tiny fraction of QDs in experimen- tal grown samples can be used to achieve entangled photon pairs. Moreover, these devices can only work at low temper- ature since the emitted exciton energies are very close to the emission lines from the wetting layer[39]. The mechanism underlying the difficulty of eliminating the FSS comes fun- damentally from the low symmetry of self-assembled QDs. The higher symmetry of the bulk materials are impossible to be restored in QDs by the above mentioned techniques[11– 13, 16–18, 25–28]. Gong et al. based on a proposed minimal two-band model [40] uncovered that the FSS is impossible to be tuned to zero for a general QD with a single external force. However, the FSS can be eliminated by two independent ex- ternal forces[41]. Employing three external forces, it is even possible to construct a wavelength tunnable entangled pho- ton emitter[42–44], opening the door for interfacing between QDs and other solid state systems and even between dissim- ilar QDs. Recently, this proposal has indeed been realized in experiments [45–49], in which the wavelength of exciton can be tuned in the range of few meV. From the symmetry point of view, since the two bright ex- citon states belong to two different representations for QDs with C2v symmetry[50], they should not only have different energies, but also have different lifetimes. The lifetime dif- ference between two bright exciton states are termed as ex- citon lifetime asymmetry (see Fig. 1a). In contrast to ex- tensively studied FSS, the investigation of this anisotropy ef- fect is still missed in literatures. In this work, we carried out the first investigation of the exciton lifetime asymmetry in self-assembled QDs and presented a theory to deduce life- time asymmetry indirectly from measurable qualities of QDs. We further revealed that the lifetime asymmetries are funda- mental quantities of QDs and unraveled some exact relations (a) Biexciton (c) T3xx T4xx FSS Exciton T3x T4x Vacuum Irreducible Representation (b) CB VB Exciton Standard Basis Functions FIG. 1. (Color online) Direct consequence of low symmetry in QDs. (a) The two bright states (solid lines) of exciton have different ener- gies and lifetimes when they belong to different irreducible represen- tations. (b) The basic irreducible representations of conduction band (CB) and valance band (VB) and exciton, and their basis for QDs with C2v symmetry. (c) In the presence of weak C1 potential, the direct coupling between the two bright states lead to wave function mixing, thus the emission will deviate from [110] and [1¯10] direc- tions. between FSS, polarization angle and lifetime asymmetry for exciton and biexciton in QDs. These exact relations are ver- ified by performing atomistic simulations of self-assembled QDs using the empirical pseudopotential method[51, 52]. Atomistic Simulation Method. We employ the empirical pseudopotential method[51, 52] to simulate the electronic and optical properties of self-assembled QDs. We model the In- GaAs/GaAs QDs by embedding it a much larger GaAs super- cell with periodic boundary condition and minimize the total strain energy using the valence force field model[53, 54]. The single particle wave functions are determined by, 1 2∇2 + Vps(r))ψi = Eiψi, (− (1) where Vps = Vsoc +Piα Viα(r − Riα) is the empirical pseu- dopotential and i is the site index, Riα is the position of the atom type α, and Vsoc the spin-orbit coupling term. The posi- tion of each atom in the supercell is obtained by minimizing the total strain energy. The exciton and biexciton energies are then calculated by employing the configuration interac- tion method taking into account the Coulomb interaction and exchange-correlation interaction [51, 52, 55]. Theoretical Modelling. We carry out the symmetry anal- ysis of QDs as given in Ref. 40 by decomposing the QD Hamiltonian H into two parts: H = H2v + V1, where H2v is a predominant term having a C2v symmetry and includes the kinetic energy, Coulomb interactions and all the potentials with C2v symmetry. The second term V1 represents the re- maining potentials lowering the symmetry of QDs from C2v 2 to C1. V1 can be treated as a perturbation to H2v consider- ing its weak effect on energy levels of QDs. Two bright states of the H2v Hamiltonian are denoted as 3i = Γ2 − iΓ4i and 4i = Γ2 + iΓ4i, respectively, where Γii (i = 1, 2, 3, 4) are the irreducible representations of the C2v point group (see Fig. 1b and the symmetry table in 50). We refer quantities of the H2v Hamiltonian to intrinsic. Since these two bright exciton states belong to two different irreducible representations, they must have different energies with an energy separation of FSS, and different lifetimes (τ3η and τ4η, respectively) with a time difference termed as intrinsic lifetime asymmetry δτη. Hence, τ3η = τη + δτη/2 and τ4η = τη − δτη/2; here, τη is averaged lifetime (hereafter η = x for exciton and η = xx for biexci- ton). Since the V1 potential is inevitable and uncontrollable in experimental grown QDs, the intrinsic lifetimes and their asymmetry can not be measured directly in experiments. To simplify the effective model for total QD Hamiltonian H, we take advantage of two additional features. Firstly, the time-reversal symmetry for exciton, T 2 = +1 (excitons have integer spin), ensures that the wave functions of two bright exciton states can be real simultaneously[56]. Secondly, the spin selection rule forbids the mixture between dark exciton states (m = ±2) and bright exciton states (m = ±1) even in the presence of V1 term. The dark exciton states can only be probed by coupling to bright states through applying external (in-plane) magnetic fields[25–27]. We thus can safely neglect the dark states and construct the effective model based only on two bright exciton states of the H2v as following, H = E0 + δσz + κσx (2) where E0 is the mean energy of two bright exciton states, σx and σz are Pauli matrices acting on two bright states with eigenvalues of m = ±1, and δ = h3H3i − h4H4i and κ = h3H4i. The bright states of the total Hamiltonian H can be constructed as a linear combination of 3i and 4i, ψ4(cid:19) = u(θ)(cid:18)3i 4i(cid:19) , (cid:18)ψ3 − sin(θ) cos(θ)(cid:19) , (3) u(θ) = (cid:18) cos(θ) sin(θ) where θ is the polarization angle as schematic shown in Fig. 1c and u(θ) is a so(2) rotation matrix. We learn that tan(θ) = (δ + ∆/2)κ−1, where ∆ = 2√δ2 + κ2 the magnitude of ex- citon FSS. The values of κ and δ can be determined from ex- perimental measurements of FSS and polarization angle. In an ensemble of QDs, the V1 potential is usually a random po- tential, thus κ and δ can be treated as two independent random numbers[57], for which reason the QDs with similar structural profile may have fairly different optical properties. If two bright exciton states have measurable extrinsic life- times T3η and T4η, which are usually deduced experimentally by fitting the time-resolved photoluminescence spectrum to a single exponential decaying function. The difference in life- times gives rise to the extrinsic lifetime asymmetry δTη, δTη = T3η − T4η, T3η + T4η = 2Tη, (4) where Tη is the mean extrinsic lifetime measured from off- resonant excitation. In this measurement, we have assumed TABLE I. Summarized parameters for different QDs from atomistic simulation. We consider the lens (L), elongated (E) and pyramind (Py) QDs with different sizes (diameter sizes along [110] and [1¯10] directions and height along z direction). The exciton energy, FSS and polarization angle without external force are shown in columns fourth to sixth. The last eight columns present the extrinsic and intrinsic lifetime asymmetries. 3 (d[110], d[1¯10], h) EX(eV) ∆(µeV) θ Tx(ns) # QD(P/L) 1.56 0 25.0, 25.0, 3.0 InAs(L) 1 1.56 24.0, 24.0, 3.0 0 2 InAs(L) 1.84 0 25.0, 25.0, 4.0 InAs(L) 3 1.56 0 25.0, 20.0, 2.5 InAs(L) 4 1.59 0 25.0, 22.7, 3.0 InAs(L) 5 14.65 24.0, 24.0, 4.0 1.21 In0.6Ga0.4As(L) 6 174.52 1.29 25.0, 25.0, 3.0 7 In0.6Ga0.4As(L) 157.68 1.23 25.0, 25.0, 4.0 In0.6Ga0.4As(L) 8 122.35 1.23 25.0, 22.7, 3.0 9 In0.6Ga0.4As(E) 109.16 1.13 10 In0.6Ga0.4As(E) 25.0, 30.0, 4.5 11 In0.6Ga0.4As(Py) 25.0, 25.0, 3.0 163.30 1.43 164.01 1.30 12 In0.6Ga0.4As(Py) 25.0, 25.0, 4.0 14.7 16.3 8.2 6.0 6.1 1.29 1.35 3.37 7.48 4.23 7.18 6.44 1.01 1.02 0.98 1.02 0.99 1.24 1.25 1.23 1.24 1.21 1.28 1.25 δTx(ps) τx(ns) 1.56 134.00 1.56 142.77 1.84 172.55 1.56 385.34 1.59 244.85 1.21 -1.48 1.29 15.52 1.23 6.87 -46.45 1.23 -147.54 1.13 -3.0 1.43 1.30 -8.64 δτx(ps) Txx(ns) 134.00 142.77 172.55 385.34 244.85 -1.70 15.81 9.65 108.85 188.57 -3.59 -10.19 1.67 1.67 1.96 1.68 1.71 1.33 1.40 1.33 1.35 1.26 1.53 1.41 δTxx(ps) τxx(ns) 1.67 144.45 1.67 153.65 1.96 184.84 1.68 418.26 1.71 265.04 1.33 -2.52 1.40 18.1 1.33 8.15 -52.97 1.35 -166.49 1.26 -2.91 1.53 1.41 -11.62 δτxx(ps) 144.45 153.65 184.84 418.26 265.04 -2.89 18.45 11.45 124.14 212.80 -3.49 -13.70 that the spin information in the electron-hole pairs are to- tally lost during the relaxation from the wetting layer to the ground state of exciton, thus the two bright states are equally populated. However, those extrinsic lifetime asymmetries are not well defined for following reasons. Firstly, the time- resolved photoluminescence spectrum is in fact composed by two exponential decay functions with two slightly different lifetimes[58–60], and thus T3η and T4η can only be obtained in the sense of best fitting. Secondly, the V1 potential, which induces inter-state mixing (see Eq. 3), is not the major origin of lifetime asymmetries [61, 62]. A more accurate descrip- tion of lifetime asymmetries should be defined by the model of H2v, instead of the total Hamiltonian H. Because lifetime asymmetries are in general much smaller than the mean lifetimes, we are ready to obtain 1 T3η = cos2(θ) τ3η + sin2(θ) τ4η , 1 T4η = sin2(θ) τ3η + cos2(θ) τ4η . (5) To the leading term of δτη, T3η = τη + cos(2θ) 2 δτη, T4η = τη − cos(2θ) 2 δτη. (6) The above results are identical to fitting the photolumines- cence spectrum to a single exponential decaying function by minimizing the following functional, − t τ4η − e− t Tη 2dt, (7) 0 − t cne τ3η + (1 − cn)e Fn = Z ∞ where n = 3, 4, with c3 = cos2(θ) for 3i and c4 = sin2(θ) for 4i. Assuming Tη = τη + xδτη (where x ≪ 1), we obtain Fη = (−1 + 2cn + 2x)2δτ 2/16τ + O(δτ /τ )3. It is straightforward to obtain its solution, which is identical to ones given in Eq. 6. We therefore see that definitions given in Eq. 5 and Eq. 7 are equivalent in the small asymmetry limit. The unusual effects of low symmetry perturbation potential. (i) Lifetime sum rule: The weak C1 potential will not alter the averaged exciton lifetime, which is determined as, T3η + T4η = τ3η + τ4η, τη = Tη. (8) The equality relation τη = Tη indicates that the mean lifetime is independent of V1 potential, which is manifested in the in- vestigated QDs ensembles as shown in Fig. 3. (ii) Lifetime asymmetry relation: The extrinsic lifetime asymmetry is de- termined as δTη = T3η − T4η = cos(2θ)δτη ≤ δτη. (9) We see that, to the leading term, the extrinsic lifetime asym- metries depends only on the intrinsic lifetime asymmetries δτη and the polarization angle θ, and is independent of the mean lifetimes τη and Tη. Interestingly, while the V1 potential can enhance the FSS and polarization angle, it will unexpectedly suppress the magnitude of δTη, which is upper bounded by δτη. Moreover, when δτη = 0 (in QDs with high symme- tries) or θ = ± π 4 (polarized along Γ2 and Γ4 directions), the change of low symmetry potential will never induce a finite extrinsic lifetime asymmetry. To verify above predictions we carry out atomistic simula- tions for single QDs as well as QDs ensembles. The calculated results for various types of single pure InAs/GaAs QDs and alloyed In(Ga)As/GaAs QDs are summarized in Table I. For pure InAs/GaAs QDs, we find that, as expected since V1 = 0, the polarization angle θ = 0 (or π/2), and δτη = δTη, fol- lowing exactly Eq. 9. The magnitude of both intrinsic and extrinsic lifetime asymmetries δTη and δτη are in range of 0.1 - 0.4 ns, and much smaller than the mean lifetimes Tη and τη, which are around 1.5-2.0 ns. From Table I we see that the lifetime sum rule given in Eq . 8 holds for all calcu- lated QDs, including alloyed ones. Although the polarization angles θ in alloyed QDs deviate significantly away from the [110] and [1¯10] directions caused by wave functions mixing, δTη ≤ δτη holds in all QDs. Therefore, we demonstrate that the low symmetry perturbation potential can remarkably suppress the lifetime asymmetries along with the derived re- lation given in Eq. 9. We further consider the alloyed QDs ensembles, in which the FSSs, polarization angles, and mean lifetimes Tη fluctu- ate from dot to dot in a wide range. We arbitrarily choose (a) (b) (c) { s d n a B d e t i c x E (d) { s d n a B d e t i c x E FIG. 2. (Color online) Effect of external stress on QDs. (a) and (b) show the role of stress along [110] and [100] direction, respectively. The black and blue boxes represent the QDs before and after being stressed. (c) and (d) show the coupling between the bright states and the excited states coupled due to stress along different directions. In these two cases the strain Hamiltonian may have different symme- tries, thus the coupling between the bright states and excited states and the direction coupling among the two bright states are totally different. an alloyed InGaAs/GaAs QD from Table I (No. 9) and then generate 50 different replica with randomly placed In and Ga atoms for a specific composition of 60%, which mimics an experimentally grown QDs ensemble. The calculated results for this ensemble are shown in Fig. 3. As expected, the FSSs, polarization angles θ, mean lifetimes Tη as well as lifetime asymmetries δTη fluctuate from dot to dot in a wide range within an ensemble. Because all modeled dots within the en- semble have the same shape, size and alloy composition, they share always the same H2v Hamiltonian, which means their intrinsic properties should be similar. Indeed, the deduced in- trinsic lifetime asymmetries δτη are rather insensitive to alloy atom fluctuations in an ensemble. The observed fluctuations in extrinsic quantities of FSSs, θ, Tη, and δTη are attributed to alloy fluctuation induced change of the V1 potential. We find that δTη ≤ δτη for all dots (see Fig. 3d-e), and the ra- tios between extrinsic and intrinsic lifetime asymmetries fall on a cosine curve as shown in Fig. 3f, in consistent with the prediction given in Eq. 9. The alloy-induced V1 perturbation potential remarkably reduces extrinsic lifetime asymmetries δTη to around zero in compared with their intrinsic lifetime asymmetry around 0.1 ns. The optical anisotropy relying on the intrinsic lifetimes. The signals collected along φ direction after off-resonance ex- citation can be written as Iη(φ) ∝ 1 + cos(2θ) cos(2(φ − θ)) δτη 2τη , (10) (a) l θ e g n A . l o P (c) ) s n ( x x T , x T 4 (b) ) V e µ ( S S F 10 5 10 20 30 40 50 0 0 10 20 30 40 50 (d) ) s n ( x T δ , x τ δ 0.3 0.2 0.1 0 -0.1 -0.2 δτ x δTx Tx Txx 10 20 30 40 50 0 10 20 30 40 50 4 3 2 1 0 0 1.5 1.4 1.3 1.2 1.1 0 (e) ) s n ( x x T δ , x x τ δ 0.3 0.2 0.1 0 -0.1 -0.2 0 δτ xx δTxx 40 10 Number of QDs 20 30 (f) 1 α τ δ / α T δ 0.5 0 -0.5 -1 1 50 x xx 2 3 Pol. Angle θ 4 FIG. 3. (Color online) Effect of random fluctuation the lifetime asymmetries in QDs ensemble. (a) and (b) show the calculated po- larization angles and FSSs in QDs ensemble. (c) The extrinsic mean lifetimes for exciton and biexciton (by assuming Eq. 8). (d) - (e) The intrinsic and extrinsic lifetime asymmetries in QDs ensemble for exciton and biexcition, where the intrinsic lifetime asymmetries are determined using the wave functions of 3i and 4i for QDs with H2v symmetry. The horizontal blue lines represent the mean value of δτη, which are 0.10 ns (∼ 9%τx) and 0.11 ns (∼ 10%τxx), respec- tively. (f) The ratio between δTη and δτη will collapse to a cosine function (blue solid line) according to Eq. 9. where φ is the angle relative to [110] direction. Here, we also assume that both bright states are equally occupied from off resonance excitation like we developing Eq. 8. We gain the maximum degree of linear polarization ρη,max when φ = θ or θ + π/2, ρη,max = Imax − Imin Imax + Imin cos(2θ)δτη 2τη = (cid:12)(cid:12)(cid:12) . (11) (cid:12)(cid:12)(cid:12) ≤ (cid:12)(cid:12)(cid:12) δτη 2τη(cid:12)(cid:12)(cid:12) We learn that the degree of polarization is determined fully by polarizaton angle θ, lifetime asymmetry δτη, and mean life- time τη, considering that the transition between two bright ex- citon states are strictly forbidden by selection rule. The degree of polarization is upper bounded by δτη/τη/2. Regarding δτη ≪ τη, the degree of polarization is restrict to small mag- nitudes despite the large varying from dot to dot. Note that the polarization discussed here should different from the lin- ear polarization defined in some of experiments, in which the excitation and measurement are performed along two orthog- onal direction, thus the degree of polarizations are generally in the order of 80% - 90% out of the available experimental data[63–66] . In the latter case polarization is complicated (a) l θ e g n A . l o P 3.5 3 2.5 -200 (c) 25 ) V e µ ( S S F 20 15 10 5 0 -200 (b) 3 2.5 2 1.5 Pure QD Alloyed QD -100 0 100 F // [100] (MPa) -100 0 100 F // [100] (MPa) 200 -200 (d) 40 30 20 10 200 0 -200 -100 0 100 F // [110] (MPa) -100 0 100 F // [110] (MPa) (a) 2.5 ) s n ( x 4 , x 3 τ 2 1.5 200 1 -200 (c) 800 600 ) s p ( x 400 τ δ 200 0 -200 200 τ 3x τ 4x τ 3x τ 4x -100 0 100 F // [100] (MPa) -100 0 100 F // [100] (MPa) (b) 2.5 2 1.5 200 1 -200 (d) 600 400 200 0 -200 -400 200 -200 τ 3x τ 4x τ 3x τ 4x -100 0 100 F // [110] (MPa) Alloyed QD Pure QD -100 0 100 F // [110] (MPa) 5 200 200 FIG. 4. (Color online) Polarization angle and FSS in pure and al- loyed QDs under stress. The left (a, c) and right (b, d) columns show the results for stress along [100] direction and [110] direction, respectively. The open symbols represent the data from atomistic simulation while the solid line is the fitting using the effective two- band model with the following parameters in unit of µeV for δ, κ and µeV/MPa for α, β: (i) For pure QDs, δ=2.73, κ=-1.69, α[100]=0, β[100]=0.05 and α[110]=-0.12, β[110]=0. (ii) For alloyed QDs, δ=4.1, κ=-0.02, α[100]=0, β[100]=0.05 and α[110]=-0.19, β[110]=0. ) α τ 2 / α τ δ ( / ρ . l o P f o . g e D e v i t l a e R 1 0.5 0 -0.5 0 ρ ' mα ρ× ' α ρ+α ' 0.2 0.4 0.6 Pol. Angle (θ/π) 0.8 1 FIG. 5. (Color online) Relative relations between different degree of polarizations as a function polarization angle for an alloyed quantum dots. since carrier scattering, spin flipping and lifetime asymmetry all contribute to its non-unity. For the measurement along [110] and [1¯10] directions (let φ = 0), ρη,× = I[110] − I[1¯10] I[110] + I[1¯10] = cos2(2θ)δτη 2τη , (12) which is smaller than ρη,max by a factor of cos(2θ). We further prove straightforwardly that the degree of linear polarization along the x and y directions is ρη,+ = sin(2θ)ρη,max. The relations between these degree of polarizations are presented in Fig. 5. Although the small extrinsic lifetime asymmetry FIG. 6. (Color online) Linear and quadratic relations for in- trinsic lifetime asymmetries. The solid squares show the pure QDs, while the solid circles show the alloyed QDs along [100] (left column) and [110] (right column) direction obtained from atomistic simulation. The solid lines are fitted using quadratic function of F , and in the solid lines we have used the param- eters (in unit of ns/MPa for ξx i ) as follows (i) For pure QDs along [100] direction: γx 3=- 4=1.83×10−6; For pure QDs along 2.29×10−6, γx 4=-4.20×10−4, [110] direction: γx 4=-2.30×10−8. ξx (ii) For alloyed QDs along [100] direction: 4=3.82×10−6; 3=3.45×10−4, ξx γx For alloyed QDs along [110] direction: γx 3=- 2.18×10−7, γx 4=-2.58×10−7. i and ns/MPa2 for γx 3=-2.31×10−8, γx 3=-5.53×10−6, γx 4=-3.51×10−5, ξx 3=7.00×10−4, ξx 3=6.73×10−6, ξx 4=-4.95×10−4, ξx 4=7.53×10−7, ξx 3=3.52×10−4, ξx maybe hard to measure directly by fitting to two exponential decay functions, the relative polarization angle θ, the mean lifetime τη (see Eq. 8) as well as the degree of polarization ρη,i can be measured precisely, thus making the intrinsic life- time asymmetries δτη rather feasible to get in experiments. Once we obtain intrinsic lifetime asymmetry δτη, the extrin- sic lifetime asymmetry δTη can be instead deduced using Eq. 9. The optical polarization relying on the intrinsic lifetimes Since exciton and biexciton states possess the same polariza- tion angle θ, we obtain ρx,i ρxx,i = δτx δτxx · τxx τx , i = {max,×, +}, (13) which is also independent of the measured φ. The deduced relations of the degree of polarizations with intrinsic lifetimes τη and lifetime asymmetry δτη of the H2v term along various directions can be examined with the data of atomistic calcula- tion shown in Fig. 3. For this QDs ensemble, δτη ∼ 0.1 ns, τη ∼ 1.25 − 1.35 ns, thus ρη,max 6 δτη/2τη ∼ 4%, which is in good agreement with atomistic simulation predicted up- per bound of the degree of polarization of ∼ 4% as shown in Fig. 3a. We may also calculate the ratio between the degree of polarizations for exciton and biexciton, which is indepen- dent of directions of optical measurements. We find this ratio being ρx,i/ρxx,i ∼ 1.08. We expect this ratio becomes larger for QDs with large shape asymmetry (such as elongation). In principle, the evolution of linear polarization as a function of angle φ (Eq. 10) can be obtained by carefully calibrating the light path in experiments. The Quadratic and Strong Nonlinear responses of the ex- In the presence of weak trinsic lifetimes to external stress. external force F , the effective Hamiltonian becomes[40–43], H = E0 + (αF + δ)σz + (κ + βF )σx, (14) where, the extra perturbative term δV = VsF is responsible for the applied external force. The symmetry of δV depends strongly on the directions and the ways the forces been applied (see Fig. 2). Here, we only consider the case with single ex- ternal force. The additivity of δV ensures that multiple forces can be treated in the same way as the single force. Notice that two unperturbed (intrinsic) bright exciton states 3i and 4i are also functions of F due to the H2v Hamiltonian depends on F . For the stress applied along the [110] direction, the strained Hamiltonian still keeps the C2v symmetry (see Fig. 2a, c), thus the stress induced coupling between the two bright states is forbidden. This is different from the case along the [100] direction, where the stress induces not only coupling of bright exciton states to highly excited states arising from other bands (termed as inter-band coupling), but also coupling between two bright exciton states (termed as intra-band coupling) (see Fig. 2b, d). The linear coupling dominates the inter-band cou- pling due to the large energy separation between the ground s band and the excited p and d bands, while the nonlinear effect may become significant in intra-band coupling due to much smaller energy difference between these two bright ex- citon states. This difference has remarkable consequences to the lifetime asymmetries. According to perturbation theory, we rewrite the intrinsic lifetimes as, 1 τ η i (F ) = 1 τ η i + γη i F + ξη i F 2, (15) for i = 3, 4 and η = {x, xx}. Here, we have introduced γη i and ξη to characterize the linear and quadratic dependence i on external force, respectively. When the contributions of the second and third terms are small in compared with 1/τ η i , we obtain, τ η i (F ) = τ η i − (τ η i )2[γη i F + ξη i F 2 − (γη i )2τ η i F 2], (16) 3 +τ η 4 )/2 and δτη = (τ η which is again a quadratic function of F . It it expected that τη = (τ η 4 )/2 are also linear and/or quadratic functions of F . With these defined intrinsic lifetime asymmetries, the extrinsic lifetimes and their asymmetries can be obtained via, 3 −τ η δTη(F ) = cos(2θ(F ))δτη(F ), Tη(F ) = τη(F ). (17) (a) ) s n ( x 4 , x 3 T 2 1.8 1.6 1.4 1.2 1 -200 (c) 300 250 200 150 ) s p ( x T δ 100 50 0 -200 T3x T4x T3x T4x (b) 2.4 2 1.6 1.2 T3x T3x T4x T4x -100 0 100 F // [100] (MPa) -100 0 100 F // [100] (MPa) 200 -200 (d) 800 600 400 200 0 -200 -400 200 -200 -100 0 100 F // [110] (MPa) Alloyed QD Pure QD -100 0 100 F // [110] (MPa) 6 200 200 FIG. 7. (Color online) Strong nonlinearity in extrinsic lifetimes and extrinsic lifetime asymmetries. The solid squares, and solid angles show the results for alloyed and pure QD, respectively. The solid line is computed using parameters from Fig. 6, and using Eq. 17. The striking consequence is that even under a weak force the extrinsic lifetime asymmetry δTη is not necessary to be a sim- ple quadratic function of F due to the presence of the strong nonlinear cosine term. We can derive more exact relations via a combination of current results and results in previous literatures[40, 42, 43]. We next attempt to study the evolution of optical proper- ties of QDs under external forces, which enables us to verify the last and the most intriguing prediction (Eq. 17) in this work. The properties of QDs fluctuate strongly from dot to dot, therefore the investigation in single QDs could provide more convincing evidences for our predictions. We consider the pure InAs/GaAs QD (No. 3 in Table I) and alloyed In- GaAs/GaAs QD (No. 12) under uniaxial stress along [110] and [100] directions, respectively. Fig. 4 shows calculated FSSs and polarization angles as functions of stress for these two QDs employing the atomistic method, accompanying fit- ted results using the two-level model. We demonstrate that we can tune the FSS to zero upon applying a stress along the [110] direction, which does not lower the QD symmetry (see Fig. 2), for the pure QDs, but it is impossible for alloyed QDs where the achievable minimum lower bound of FSS is sig- nificant larger than the spontaneous broadening of the spectra Γ[8–10]. This large minimum FSS implies that we can not eliminate the FSS of alloyed QDs by using one single exter- nal force. Fig. 6 shows calculated intrinsic lifetimes and their asym- metries as functions of applied stress for both pure InAs/GaAs and alloyed InGaAs/GaAs QDs. For stress applied along the [110] direction, we find that τ3x and τ4x are linear func- tions of stress F , as shown in Fig. 6b-d. To further ver- ify their linear feature, we fit these data to quadratic func- tions. We gain tiny coefficients of the quadratic terms for both pure InAs/GaAs and alloyed InGaAs/GaAs QDs (ξx 3,4 ∼ 10−7 − 10−8 ns/MPa2, see fitted data in Fig. 6), which illus- trate both intrinsic lifetimes and their asymmetries are linear against [110] stress. In striking contrast to [110] stress ap- plied along the [100] direction gives rise to different response of the lifetime of τ3x and τ4x because it causes direct coupling between two bright exciton states (see mechanism in Fig. 2). From Fig. 6 we see that both τ3x and τ4x exhibit quadratic rela- tions against the applied [100] stress F . As expected, the cor- responding lifetime asymmetry δτx also displays a quadratic function of F . The similar features can also be found for tran- sition from biexciton to exciton states. The extrinsic lifetime asymmetries have more complicated response behaviours to applied stress and will exhibit strong nonlinearity even under weak force due to direct coupling between the two bright states induced by the C1 symmetry potential V1. Fig. 7 shows atomistic calculated results. In the pure InAs/GaAs QD the V1 potential is absent, and thus T3x and T4x should possess perfect linear functions of F for stress applied along the [110] direction and quadratic func- tions for stress applied along the [100] direction. The re- sults shown in Fig. 7 well support the theoretical predic- tion. These stress-responses are identical to τη, as shown in Fig. 6b,d, in the sense that θ = 0. However, for al- loyed InGaAs/GaAs QD, strong nonlinearity behaviours of both mean extrinsic lifetimes and extrinsic lifetime asymme- tries are expected for stress applied along [100] and [110] di- rections. Since Ti(F ) = τi(F ) (see Eq. 17), we can observe the strong nonlinearity in both the extrinsic lifetimes Tiη and the extrinsic lifetime asymmetries δTη. Moreover, while the extrinsic lifetime asymmetries are generally very small due to wave function mixing effect, we find that the extrinsic lifetime asymmetries in Fig. 7 c-d, and the intrinsic lifetime asymme- tries in Fig. 6c-d, can be pronouncedly enhanced by exter- nal stress from tens of ps to 0.2 - 0.7 ns, that is, by at least one order of magnitude. This enhanced asymmetries may lead to direct measurement of lifetime asymmetries with only time-resolved photoluminescnce spectrum, in which the time- resolved spectrum should be fitted using two exponential de- cay functions. The similar strong nonlinearity effect has also been observed for transition from biexciton to exciton states in a reason discussed above. These results can be well de- scribed by the data in Fig. 6 and are in fully accordance with our theoretical predictions. Summary. We introduced lifetime asymmetry, which is a new concept, into self-assembled QDs. We revealed that in- trinsic lifetimes are fundamental quantities of QDs, which determine the bound of the extrinsic lifetime asymmetries, polarization angles, FSSs, and their evolution under uniax- ial external forces. These predictions can be direct measured or extracted from experiments using the state-of-the-art tech- niques, such as the measured linear polarization as well as the optical properties of QDs under external forces. We verified these predictions using atomistic simulations. We found that the intrinsic lifetime asymmetries can be of the order of few 7 hundred picoseconds in pure InAs/GaAs QDs, but the extrin- sic lifetime asymmetries can be much smaller in alloyed In- GaAs/GaAs QDs. However, the lifetime asymmetries are sus- ceptible to external forces and its directions, thus can be more conclusively verified by investigating their behaviours under external forces. These exact relations represent a complete description of the optical properties of QDs. Our findings pro- vide an important basis to deeply understanding properties of QDs. Acknowledgement. M.G. is supported by the National Youth Thousand Talents Program (No. KJ2030000001), the USTC start-up funding (No. KY2030000053), the na- tional natural science foundation (NSFC) under grant No. GG2470000101). X. X. is supported by National Basic Re- search Program of China under No. 2014CB921003, the NSFC under Nos. 11721404, 91436101 and 61675228; the Strategic Priority Research Program of the Chinese Academy of Sciences under No. XDB07030200 and XDPB0803, and the CAS Interdisciplinary Innovation Team. J. L. is sup- ported by NSFC under Nos. 61121491, 11474273, 11104264 and U1530401, and National Young 1000 Talents Plan. X. W. is supported by NSFC under No. 61404015 and funda- mental research funds for the central universities (2015CD- JXY300001). ∗ Email: [email protected] † Email: [email protected] [1] Benson, O.; Santori, C.; Pelton, M.; Yamamoto, Y. Phys. Rev. Lett. 2000, 84, 2513. [2] Pan, J.-W.; Bouwmeester, D.; Weinfurter, H.; Zeilinger, A. Phys. Rev. Lett. 1998. 80, 3891. [3] Gisin, N.; Ribordy, G.; Tittel, W.; Zbinden, H. Rev. Mod. Phys. 2002, 74, 145. [4] Duan, L.-M.; Lukin, M. D.; Cirac, J. I.; Zoller, P. Nature 2001, 414, 413. [5] Xu, J.-S.; Li, C.-F.; Gong, M.; Zou, X.-B.; Shi, C.-H.; Chen, G.; Guo, G.-C. Phys. Rev. Lett. 2010, 104, 100502. [6] Huang, Y.-F.; Liu, B.-H.; Peng, L.; Li, Y.-H.; Li, L.; Li, C.-F.; Guo, G.-C. Nat. Commun. 2011, 2, 546. [7] Yin, J.; Cao, Y.; Li, Y.-H.; Liao, S.-K.; Zhang, L.; Ren, J.-G.; Cai, W.-Q.; Liu W.-Y. et al. Science 2017, 356, 1140. [8] Gammon, D.; Snow, E. S.; Shanabrook, B. V.; Katzer, D. S.; Park, D. Phys. Rev. Lett. 1996, 76, 3005. [9] Bayer, M.; Ortner, G.; Stern, O.; Kuther, A.; Gorbunov, A. A.; Forchel, A.; Hawrylak, P.; Fafard, S.; Hinzer, K.; Reinecke, T. L.; Walck, S. N.; Reithmaier, J. P.; Klopf, F.; Schafer, F. Phys. Rev. B 2002, 65, 195315. [10] Seguin, R.; Schliwa, A.; Rodt, S.; Potschke, K.; Pohl, U. W.; Bimberg, D. Phys. Rev. Lett. 2005, 95, 257402. [11] Langbein, W.; Borri, P.; Woggon, U.; Stavarache, V.; Reuter, D.; Wieck, A. D. Phys. Rev. B 2004, 69, 161301(R). [12] Tartakovskii, A. I.; Makhonin, M. N.; Sellers, I. R.; Cahill, J.; Andreev, A. D.; Whittaker, D. M.; Wells, J-P. R.; Fox, A. M.; Mowbray, D. J.; Skolnick, M. S.; Groom, K. M.; Steer, M. J.; Liu, H. Y.; Hopkinson, M. Phys. Rev. B 2004, 70, 193303. [13] Ellis, D. J. P.; Stevenson, R. M.; Young, R. J.; Shields, A. J.; Atkinson, P.; Ritchie, D. A. Appl. Phys. Lett. 2007, 90, 011907. 8 [14] Seguin, R.; Schliwa, A.; Germann, T. D.; Rodt, S.; Potschke, K.; Strittmatter, A.; Pohl, U. W.; Bimberg, D.; Winkelnkemper, M.; Hammerschmidt, T.; Kratzer, P. Appl. Phys. Lett. 2006, 89, 263109. [15] Tartakovskii, A. I.; Makhonin, M. N.; Sellers, I. R.; Cahill, J.; Andreev, A. D.; Whittaker, D. M.; Wells, J-P. R.; Fox, A. M.; Mowbray, D. J.; Skolnick, M. S.; Groom, K. M.; Steer, M. J.; Liu, H. Y.; Hopkinson, M. Phys. Rev. B 2004, 70, 193303. [16] Gerardot, B. D.; Seidl, S.; Dalgarno, P. A.; Warburton, R. J.; Granados, D.; Garcia, J. M.; Kowalik, K.; Krebs, O.; Karrai, K.; Badolato, A.; Petroff P. M. Appl. Phys. Lett. 2007, 90, 041101. [17] Kowalik, K.; Krebs, O.; Lemaıtre, A.; Laurent, S.; Senellart, P.; Rastelli, A.; Schmidt, O. G. Phys. Rev. B 2014, 89, 115309. [36] Trotta, R.; Wildmann, J. S.; Zallo, E.; Schmidt, O. G.; Rastelli, A. Nano Lett. 2014, 14, 3439. [37] Trotta, R.; Mart´ın-S´anchez, J.; Wildmann, J. S.; Piredda, G.; Reindl, M.; Schimpf, C.; Zallo, E.; Stroj, S.; Edlinger, J.; Rastelli, A. Nat. Commun. 2015, 7, 10375. [38] Hofer, B.; Zhang, J.; Wildmann, J.; Zallo, E.; Trotta, R.; Rastelli, A.; Schmidt, O. G. Appl. Phys. Lett. 2017, 110, 151102. [39] Hafenbrak, R.; Ulrich, S. M.; Michler, P.; Wang, L.; Rastelli, A.; Schmidt, O. G. New J. Phys. 2007, 9, 315. [40] Gong, M.; Zhang, W.; Guo, G.-C.; He, L. Phys. Rev. Lett. 2011, Voisin, P.; Gaj, J. A. Appl. Phys. Lett. 2005, 85, 041907. 106, 227401. [18] Vogel, M. M.; Ulrich, S. M.; Hafenbrak, R.; Michler, P.; Wang, L.; Rastelli, A.; Schmidt, O. G. Appl. Phys. Lett. 2007, 91, 051904. [41] Sapienza, L.; Malein, R. N. E.; Kuklewicz, C. E.; Kremer, P. E.; Srinivasan, K.; Griffiths, A.; Clarke, E.; Gong, M.; Warburton, R. J.; Gerardot, B. D. Phys. Rev. B 2013, 88, 155330. [19] Ghali, M.; Ohtani, K.; Ohno, Y.; Ohno, H. Nat. Commun. 2012, [42] Wang, J.; Gong, M.; Guo G.-C.; He, L. Appl. Phys. Lett. 2012, 3, 661. 101, 063114. [20] Bennett, A. J.; Pooley, M. A.; Stevenson, R. M.; Ward, M. B.; Patel, R. B.; Boyer de la Giroday, A.; Skold, A.; Farrer, I.; Nicoll, C. A.; Ritchie, D. A.; Shields, A. J. Nature Phys. 2010, 6, 947. [21] Tang, J., Cao, S., Cao, Y., Sun, Y., Geng, W, Williams, D. A., Jin, K., Xu, X., Appl. Phys. Lett. 105, 041109 (2014). [22] Mar, J. D. and Baumberg, J. J. and Xu, X. L. and Irvine, A. C. [43] Wang, J.; Gong, M.; Guo G.-C.; He, L. Phys. Rev. Lett. 2015, 115, 067401. [44] Trotta, R.; Mart´ın-S´anchez, J.; Daruka, I.; Ortix, C.; Rastelli, A. Phys. Rev. Lett. 2015, 114, 150502. [45] Zhang, J.; Wildmann, J. S.; Ding, F.; Trotta, R.; Huo, Y.; Zallo, E.; Huber, D.; Rastelli, A.; Schmidt, O. G. Nat. Commun. 2015, 6, 10067. and Williams, D. A., Phys. Rev. B 93, 045316 (2016). [46] Chen, Y.; Zhang, J.; Zopf, M.; Jung, K.; Zhang, Y.; Keil, R.; [23] Mar, J. D. and Xu, X. L. and Sandhu, J. S. and Irvine, A. C. and Hopkinson, M. and Williams, D. A., Applied Physics Letters, 97, 221108 (2010). Ding, F.; Schmidt, O. G. Nat. Commun. 2016, 7, 10387. [47] Keil, R.; Zopf, M.; Chen, Y.; Hoefer, B.; Zhang, J.; Ding, F.; Schmidt, O. G. Nat. Commun. 2016, 8, 15501. [24] Mrowi´nski, P.; Musiał, A.; Mary´nski, A.; Syperek, M.; Misiewicz, J.; Somers, A.; Reithmaier, J. P.; Hofling, S.; Se¸k, G. Appl. Phys. Lett. 2015, 106, 053114. [48] Zhang, J.; Ding, F.; Zallo, E.; Trotta, R.; Hofer, B.; Han, L.; Kumar, S.; Huo, Y.; Rastelli, A.; Schmidt, O. G. Nano Lett. 2013, 13, 5808-5813. [25] Stevenson, R. M.; Young, R. J.; See, P.; Gevaux, D. G.; Cooper, K.; Atkinson, P.; Farrer, I.; Ritchie, D. A.; Shields, A. J. Phys. Rev. B 2006, 73, 033306. [49] Huang, H.; Trotta, R.; Huo, Y.; Lettner, T.; Wildmann, J. S.; Mart´ın-S´anchez, J.; Huber, D.; Reindl, M.; Zhang, J.; Zallo, E.; Schmidt, O. G.; Rastelli, A. ACS Photonics 2017, 4, 868. [26] Puls, J.; Rabe, M.; Wunsche, H.-J.; Henneberger, F. Phys. Rev. B 1999, 60, 16303. [27] Stevenson, R. M.; Young, R. J.; Atkinson, P.; Cooper, K.; Ritchie, D. A.; Shields, A. J. Nature 2006, 439, 179. [28] Kuklewicz C. E.; Malein, R. N. E.; Petroff, P. M.; Gerardot, B. [50] Koster,G. F.; Dimmock, J. O.; Wheeler, R. G.; Statz, H. Prop- erties of the Thirty-Two Point Groups (MIT Press, Cambridge, MA, 1963). [51] Wang, L. W.; Zunger, A. Phys. Rev. B 1999, 59, 15806. [52] Williamson, A. J.; Wang, L. W.; Zunger, A. Phys. Rev. B 2000, D. Nano Lett. 2012, 12, 3761. 62, 12963. [29] Ding, F.; Singh, R.; Plumhof, J. D.; Zander, T.; Kr´apek, V.; Chen, Y. H.; Benyoucef, M.; Zwiller, V.; Dorr, K.; Bester, G.; Rastelli, A.; Schmidt, O. G. Phys. Rev. Lett. 2010, 104, 067405. [30] Plumhof, J. D.; Kr´apek, V.; Ding, F.; Jons, K. D.; Hafenbrak, R.; Klenovsk´y, P.; Herklotz, A.; Dorr, K.; Michler, P.; Rastelli, A.; Schmidt, O. G. Phys. Rev. B 2011, 83, 121302(R). [31] Rastelli, A.; Ding, F.; Plumhof, J. D.; Kumar, S.; Trotta, R.; Deneke, Ch.; Malachias, A.; Atkinson, P.; Zallo, E.; Zander, T.; Herklotz, A.; Singh, R.; Kr´apek, V.; Schroter, J. R.; Kiravittaya, S.; Benyoucef, M.; Hafenbrak, R.; Jons, K. D.; Thurmer, D. J.; Grimm, D.; Bester, G.; Dorr, K.; Michler, P.; Schmidt, O. G. Phys. Status Solidi B 2012, 249, 687. [32] Trotta, R.; Atkinson, P.; Plumhof, J. D.; Zallo, E.; Rezaev, R. O.; Kumar, S.; Baunack, S.; Schroter, J. R.; Rastelli, A.; Schmidt, O. G. Adv. Mater. 2012, 24, 2668. [33] Trotta, R.; Zallo, E.; Ortix, C.; Atkinson, P.; Plumhof, J. D.; van den Brink, J.; Rastelli, A.; Schmidt, O. G. Phys. Rev. Lett. 2012, 109, 147401. [53] Keating, P. N. Phys. Rev. 1966, 145, 637. [54] Martin, R. M. Phys. Rev. B 1970, 1, 4005. [55] Franceschetti, A.; Fu, H.; Wang, L. W.; Zunger, A. Phys. Rev. B 1999, 60, 1819. [56] Sakurai, J. J.; Modern Quantum Mechanics, San Fu Tuan (Ben- jamin/Cummings, Menlo Park, CA, 1985), p. 276. [57] Gong, M.; Hofer, B.; Zallo, E.; Trotta, R.; Luo, J.-W.; Schmidt, O. G.; Zhang, C. Phys. Rev. B 2014, 89, 205312. [58] Mukai, K.; Ohtsuka, N.; Shoji, H.; Sugawara, M. Phys. Rev. B 1996, 54, R5243. [59] Inokuma, T.; Arai, T.; Ishikawa, M. Phys. Rev. B 1990, 42, 11093. [60] van Sark, W. G. J. H. M.; Frederix, P. L. T. M.; Bol, A. A.; Gerritsen, H. C.; Meijerink, A. ChemPhysChem 2002, 3, 871. [61] Singh, R.; Bester, G. Phys. Rev. Lett. 2009, 103, 063601. [62] Dupertuis, M. A.; Karlsson, K. F.; Oberli, D. Y.; Pelucchi, E.; Rudra, A.; Holtz, P. O.; Kapon, E. Phys. Rev. Lett. 2011, 107, 127403. [34] Plumhof, J. D.; Trotta, R.; Kr´apek, V.; Zallo, E.; Atkinson, P.; Kumar, S.; Rastelli, A.; Schmidt, O. G. Phys. Rev. B 2013, 87, 075311. [63] Kulakovskii, V. D.; Bacher, G.; Weigand, R.; Kummell, T.; Forchel, A.; Borovitskaya, E.; Leonardi, K.; Hommel, D. Phys. Rev. Lett. 1999, 82, 1780. [35] Kumar, S.; Zallo, E.; Liao, Y. H.; Lin, P. Y.; Trotta, R.; Atkin- son, P.; Plumhof, J. D.; Ding, F.; Gerardot, B. D.; Cheng, S. J.; [64] Ulrich, S. M.; Strauf, S.; Michler, P.; Bacher. G.; Forchel, A. Appl. Phys. Lett. 2003, 83, 1848. [65] Astakhov, G. V.; Kiessling, T.; Platonov, A. V.; Slobodskyy, T.; Mahapatra, S.; Ossau, W.; Schmidt, G.; Bruuner, K.; Molenkamp, L. W. Phys. Rev. Lett. 2006, 96, 027402. [66] Dzhioev, R. I.; Zakharchenya, B. P.; Ivchenko, E. L.; Korenev, V. L.; Kusraev, Y. G.; Ledentsov, N. N.; Ustinov, V. M.; Zhukov, A. E.; Tsatsulnikov, A. F. Phys. Solid State 1998, 40, 790. 9
1504.02328	1	1504	2015-04-09T14:31:42	Observation of quantum transport features in graphene devices fabricated utilizing a nano-manipulating probe technique	[ "cond-mat.mes-hall" ]	A novel method for fast fabrication of mesoscopic multilayered graphene electronic devices utilizing nanoprobes to exfoliate graphite flakes is developed. The magnetoresistance of these devices exhibit pronounced Shubnikov-de Haas oscillations at magnetic fields above 4 T and at temperatures below 30 K. From the analysis of the SdH oscillations we show that multilayer graphene devices have a carrier density and effective mass (m*= 0.042me - 0.083me) comparable to those of bilayer and trilayer graphene. The quantum lifetime in this multilayered graphene is in the range 22 to 90 fs corresponding to a disorder-broadening of 5 to 15 meV.	cond-mat.mes-hall	cond-mat	Observation of quantum transport features in graphene devices fabricated utilizing a nano-manipulating probe technique Christopher Coleman, Davie Mtsuko, Chris Botha† and Somnath Bhattacharyyaa) Nano-Scale Transport Physics Laboratory, School of Physics, and DST/NRF Centre of Excellence in Strong Materials, University of the Witwatersrand, Johannesburg, Private Bag 3, WITS 2050, South Africa † Present address: Orbilectron Consulting (Pty) Ltd., Port Elizabeth, South Africa A novel method for fast fabrication of mesoscopic multilayered graphene electronic devices utilizing nanoprobes to exfoliate graphite flakes is developed. The magnetoresistance of these devices exhibit pronounced Shubnikov-de Haas oscillations at magnetic fields above 4 T and at temperatures below 30 K. From the analysis of the SdH oscillations we show that multilayer graphene devices have a carrier density and effective mass (m= 0.042me - 0.083me) comparable to those of bilayer and trilayer graphene. The quantum lifetime in this multilayered graphene is in the range 22 to 90 fs corresponding to a disorder-broadening of 5 to 15 meV. ___________________________________________________________________________ a) Author to whom correspondence should be addressed. Electronic mail: [email protected] 1 Introduction: Graphene, the two-dimensional allotrope of carbon has since its discovery in 2004 attracted an immense amount of scientific attention [1-3]. This is due to the unique and fascinating properties of this material such as high current densities of orders of magnitude larger than in copper and also the existence of Dirac fermions at the k-points [4,5]. Since this discovery, graphene has been hailed as a super material with the possibility of replacing silicon for electronic purposes and a potential material in the field of micro and nano-electronics. There are however many challenges in creating functional devices with graphene, these range from the isolation of single or few layered graphene sheets to the transfer and manipulation of the material for device fabrication. Conventional methods used in the fabrication of these devices include delicate and complex processes such as electron beam lithography coupled with reactive plasma etching as well as techniques using scanning probe microscopy specifically AFM-lithography [1,6,7]. In addition to being relatively costly these processes have proved to be time consuming with an extremely low output of functioning devices, leading to a search for more expeditious and effective routes for the device fabrication. In this work we develop a novel method of fast and contamination-free fabrication of few layer graphene devices. This involves mechanically peeling off thin sheets of multilayered graphene from bulk exfoliated graphite stock, placement of the multilayers onto prefabricated Hall bar and waveguide contact pads, in order to create devices for Hall effect and magnetoresistance (MR) measurements as well as high frequency measurements. By comparison with results obtained in single layer, bilayer and trilayer graphene a strong understanding of the origin of the evolution of transport characteristics in the transition from single layer to graphite can be established. Magnetic field-dependent studies on single layer graphene have shown Shubnikov de Haas oscillations which get damped as the temperature increases or the bias current is increased as well as carrier lifetimes up to 53 fs and effective masses 0.033me [8]. 2 The influence of interlayer effects on SdH oscillations and transport parameters in multilayer graphene devices is of great interest. Temperature damped SdH measurements on bilayer graphene in one study have found hole and electron effective masses of 0.034me and 0.046me, respectively, with the suppressed m values (with respect to single particle values) being attributed to a renormalization of band structure induced by electron-electron interaction [9]. In other studies on trilayer graphene effective mass of 0.054me were obtained showing that effective mass increases with increasing number of layers [10]. It is of interest to see if this trend continues as the number of layers is further increased. In works on multilayered graphene systems it was also established that the electric field effect is prominent and that the frequency of the SdH oscillations varies with a change in the gate voltage of the devices [11- 13]. Quantum Hall effect studies in ABC stacked trilayer graphene devices, have revealed a new type of quasiparticle with l= 3 chiral degree and a cubic dispersion predicted by theoretical works [14]. Despite the complexity of multilayer graphene systems such a richness in transport physics observed in bilayer and trilayer is a strong drive of our studies of multilayer systems. Experimental details: The multilayered graphene stock samples were obtained by the conventional scotch tape method developed by Konstantin Novoselov and Andre Geim whereby highly ordered pyrolytic graphite is continually mechanically cleaved using an adhesive tape until few layers remain [1]. These layers are visible under an optical microscope even at low magnification with different thicknesses or number of layers rather, appearing as different colored flakes due to interference effects. This method is usually continued until one arrives at single layer graphene. As can be imagined this method is relatively time consuming. Also other complex e-beam lithography procedures (including mapping, resist coating, write field alignment, additional metallization and lift-off steps) need to be carried out in order to produce devices 3 for the electrical characterization. In our approach, the multilayered graphene samples on the SiO2 substrates are annealed at a temperature of 1000 oC in argon gas environment, then imaged using a JEOL J1007F SEM. Using nano-manipulators (Kleindek-Germany), with tips having a radius of curvature of approximately 100 nm, thin strips of layered graphene were peeled off of the larger flakes. It was found that these strips adhered relatively well to the manipulating probe and once separated from the larger flakes were easily transported by moving the probe to any desired location within the imaging range. It was found, however, that placing the graphene strips down onto the desired location was difficult due to substrate charging, this was overcome by grounding the substrate in the SEM. In this case multilayered graphene strips were placed and positioned onto prefabricated contact pads designed in six-probe Hall bar configuration. It was observed that using this method introduces ripples and folds in the material, this is unwanted as it would affect the transport properties of the device. Attempts were made at using the nano-probe tips to iron out these ripples, this technique showed some success with larger wider ripples but was unable to smooth out sharp thin folds in the multilayered sheet. The multilayered strips where then welded into place with platinum tabs using a combination of SEM beam and gas injection system (OmniProbe USA). Once the multilayered strips were secured, the devices were annealed in argon to remove any contaminants and then subjected to Raman analysis. Although the described process has yielded only multilayered sheets of graphene thus far, the technique is far less time consuming and provides a mechanism to suspend sheets directly across contact electrodes. Utilizing this method also eliminates much of the risk of contaminating and functionalizing the graphene samples with chemicals commonly used in device fabrication. Fig. 2 shows a Raman spectrum for the multilayered graphene device created using the above method. The prominent features of the multilayered sample used in this study 4 include the 2D peak at 2700 cm-1 which has a lower intensity than the G peak at 1580 cm-1. This is unlike single layer graphene which has a 2D peak with much greater intensity than that of the G peak. The shape of the 2D band is routinely used to determine the number of layers in few layered samples. This is possible due to the unique peak shapes which can be fitted with multiple Lorentzians specific to the number of layers up to 4 layers, beyond this number the 2D peak is too similar to that of graphite to determine number of layers. The sample used in this study shows a 2D peak that can be fitted with two Lorentzian curves, the one at higher frequency being more prominent, as is expected for graphite. This implies that the sample is composed of more than 4 layers as the unique spectra are only observed up to this point. Also noted is the small D peak at 1350 cm-1. This Raman peak is an indication of disorder in the material. The introduction of disorder is mostly due to folding and rippling of the sheet layers which was found to be unavoidable. Results and Discussion: In bench-top measurements the resistance was found to be in the Ω-range. This is expected as having a multilayered system allows for multiple conduction channels, increasing the conductivity compared to single or very few layered graphene devices where resistance is typically in the kΩ range. The devices were then cooled to room temperature and characterized over a large temperature range. The Hall bar device was analyzed using a cryogenic cryostat system with the capabilities of achieving measurements in fields up to 12 T. The resistance-temperature dependence at various fields, perpendicular to the sample surface, is shown in Fig. 3(a). It is noted that initially the device exhibits a decrease in resistance as temperature is raised, but only up until 25K, at this point the resistance increases with temperature. This change indicates a crossover of conduction mechanism from insulating to metallic behavior. Also noted are resistance fluctuations which occur at high fields above 4T and temperatures below 35 K. To investigate this oscillatory behavior the 5 MR measurement of the device was undertaken (See Fig. 3(b)). Here it was found that at low temperatures below 25 K, initially the device shows a negative MR in the low field regime (from 0 T to 4 T), above 4 T however the MR starts to oscillate with increasing amplitudes the magnetic field is ramped up. These oscillations are noted to be damped as the temperature is increased. As the magnetic field is increased the Fermi sphere splits into Landau tubes. The change is temperature kB∆T shifts the Landau levels with respect to the Fermi level producing resonances when the Fermi level coincides with the Landau levels. As temperature phonon- induced disorder broadening smears the oscillations. This oscillatory behavior persists until 35 K, above this temperature no oscillations were observed. This is consistent with the interpretation of the R-T data as well as previous studies that have shown that SdH oscillations are observable in mesoscopic graphite and thin layered graphene [1,8]. From this MR data, the resistance as a function of inverse magnetic field was plotted at temperatures 5 K, 10 K, 15 K and 20 K. By determining the SdH frequency it is able to calculate carrier density for the device by the equation: 𝑣 = 𝑛𝑒ℎ 𝑒𝐵 . Here, ne is the carrier density, B the applied magnetic field with h and e being Planck's constant and the fundamental charge. The carrier density was found to decrease as temperature increased and also to increase slightly with increasing magnetic field. There were two exceptions to this general trend; the MR measurements made at 2K and at 5K. At these low temperatures it was expected that SdH oscillations would be most prominent, that however was found not to be the case. At these two lowest temperatures oscillatory behavior did start at 4T applied field but the strength of the oscillations are miniscule as compared to that of the oscillations at higher temperatures. From the carrier density we can evaluate the effective mass 𝑚∗ = ( ℏ 𝑣𝐹 )√𝜋𝑛𝑒 . For carrier density in the range 5  1012cm-2 to 2  1013 we find the mean effective mass of in our multilayer graphene device to be in the range 0.042me to 0.083me. This is comparable with 6 the effective mass measured in bilayer layer and trilayer graphene and reported in literature [9] and the theoretical values of effective mass of AB stacked bilayer graphene (𝑚𝐴𝐵 ≈ 𝛾1 2𝑣2 ⁄ ≈ 0.035𝑚𝑒) and ABA stacked trilayer graphene (𝑚𝐴𝐵𝐴 = √2𝑚𝐴𝐵 ≈ 0.05𝑚𝑒) [14]. Our measurements confirm that the increase in number of layers leads to higher effective mass compared to effective mass of monolayer graphene. Apart from the carrier density shown in Fig. 3(c), we can extract two extra transport parameters from the SdH oscillation data analysis namely the carrier lifetime or the quantum scattering time and the phase coherence length. To extract the quantum lifetime we make use of the general form of MR of a two-dimensional electron system (2DES) in the SdH oscillation regime which is given by [8]: 𝑅𝑋𝑋 = 𝑅0 [1 + 𝜆 ∑ ∞ 𝑆=1 𝐷(𝑠𝑋)𝑒𝑥𝑝 (− 𝑠𝜋 𝑤𝑐𝜏 ) × cos (𝑠 ℏ𝑆𝐹 𝑒𝐵 − 𝑠𝜋 + 𝑠𝜙0)] , from which to the first order we can write the amplitude of oscillations as:𝐴𝑒𝑥 = 𝜆𝐷(𝑋)𝑒𝑥𝑝 (− 𝜋 𝑤𝑐𝜏 ), where 𝑤𝑐 = 𝑒𝐵/𝑚∗ is the cyclotron frequency and 𝐷(𝑋) = 2𝜋𝑘𝐵𝑇/ℏ𝑤𝑐 𝑠𝑖𝑛ℎ(2𝜋𝑘𝐵𝑇/ℏ𝑤𝑐) is the damping factor. Hence by plotting the amplitudes (Aex/D(X)) versus 1/Bex of the oscillation at extrema as shown in Fig. 1, we find the carrier lifetime in the range 22 fs to 90 fs for effective mass 0.065 me. This is comparable to a lifetime of 54 fs observed in single layer graphene [8]. In Fig. 2 we plot the lifetime as a function of temperature. The 𝜏 range corresponds to a disorder-broadening values ( 𝛤 = ℏ/2𝜏) in the range 4.6 to 15 meV. This is comparable to disorder broadening (5.5 to 8.0 meV) observed in bilayer graphene [9]. The phase coherence length is estimated using formula ΔB = φ0/Lφ 2= to be 66 nm at 25 K at 6 T and 75 nm at 6 T at 2 K. [15]. The inverse period of the SdH oscillation can also be used to estimate the cross sectional area of the Fermi surface pocket perpendicular to the magnetic field using the formula ∆(1/𝐵) = 2𝜋𝑒 ℏ𝐴⁄ [16,17]. For a 10 T field at 25 K we find the extremal Fermi surface area to be ~91017 m-2. However, the exact morphology of the Fermi surface pockets is not known. 7 We next examine some peculiar features of the SdH oscillations. It is seen that as the temperature is increased the amplitude of oscillations in the high field regime slightly reduces. This damping of oscillation is due to phonon scattering which reduces the phase coherence length. It is also observed that the extrema shift location on the 1/B axis as the temperature is varied. This can be explained by small shifts in Landau level positions due to inelastic phonon scattering. We note the manifestation of beat frequencies in the MR with the field arranged perpendicular to the sample. The formation of these beat frequencies is generally understood to indicate coherent transport between layers whereby electrons are transferred directly between layers ensuring that intralayer momentum is conserved. As a consequence interference between wavefunctions on adjacent layers is possible [18]. It is also possible for this mode of interlayer transport to be weakly incoherent as the tunneling events between layers are uncorrelated due to the electrons being scattered multiple times within the layers before tunneling between the layers. This beat behavior is related to the warping of the Fermi surface and the interference of cyclotron orbits of different radii, the so called short and long orbits [18]. Furthermore it has been experimentally demonstrated that transport between few layer graphene is strongly related to the stacking order of the layers and breakdown of coherent transport can occur due to twisting or misalignment whereby the stacking order is changed from the Bernal to rhombohedral configuration [19]. Taking all these aspects into consideration it is quite natural to link the modulated SdH oscillations to these interlayer interference effects. The irregularity in the oscillations may also arise from the modulation of the SdHO by two other kinds of oscillations known as magneto-intersubband oscillations (MISO) and phonon induced resistance oscillations whose origin can be traced to scattering- assisted coupling of carrier states in different Landau levels [20]. 8 Conclusion: In summary, we have demonstrated a quick and efficient way of producing multilayer to a few layer graphene devices. The proposed method is faster than conventional procedures and eliminates much of the risk for contamination but has the disadvantage of being an extremely delicate process. This method does not allow for the peeling off of single layer graphene owing to the large tip radius of the manipulating probe. It has however proven to be effective at creating graphene devices in the mesoscopic regime such that it can be regarded to hold much potential for fast device fabrication for research purposes in the field of mesoscopic layered systems. The low temperature measurement exhibit quantum effects in the form of SdH oscillations. From the analysis of the SdH oscillations it is seen that this multilayer graphene device has carrier density, effective mass, quantum lifetime and phase coherence length comparable to those of bilayer and trilayer graphene devices. Acknowledgements: SB acknowledges the CSIR-NLC for establishing the laser ablation set up which produced the some of the graphene samples and NRF (SA) NNEP as well as nanotechnology flagship grant. 9 References [1] K.S. Novoselov, A.K. Geim, S.V. Morozov, D. Jiang, Y. Zhang, S.V. Dubonos, I.V. Grigorieva, A.A. Firsov, Science 306, 666 (2004). [2] K.S. Novoselov, D. Jiang, F. Schedin, T. J. Booth, V.V. Khotkevich, S.V. Morozov, and A.K. Geim, Proc. Natl. Acad. Sci. USA 102,10451 (2005). [3] A.K. Geim, and K.S. Novoselov, Nature Materials 6, 183 (2007). [4] S.V. Morozov, K.S. Novoselov, M.I. Katsnelson, F. Schedin, D.C. Ellias, J.A. Jaszczak, A.K. Geim, Phys. Rev. Lett. 100, 016602 (2008). [5] K.I. Bolotin, K.J. Sikes, Z. Jiang, M. Klima G. Fudenberg, J. Hone,P. Kim, H.L. Stormer Solid State Commun.146, 351(2008). [6] M. Wendel, S. Kuhn, H. Lorentz, J.P. Kotthaus, and M. Holland, Appl. Phys. Lett. 65, 1775 (1994). [7] B. Vasic, M. Kratzer, A. Matković, A. Nevosad, U. Ralević, D. Jovanović, C. Ganser, C. Teichert and R. Gajić, Nanotechnology 24, 015303 (2003). [8] Z. Tan, C. Tan, L. Ma, G.T. Liu , and C.L. Yang, Phys. Rev. B 84, 115429 (2011). [9] K. Zou, X. Hong, and J. Zhu,Phys. Rev. B 84, 085408 (2011). [10] M. F. Craciun, S. Russo, M. Yamamoto, J.B. Oostinga, A.F. Morpurgo, S. Tarucha. Nature 4, 383 (2009). [11] E. V. Castro, K.S. Novoselov, S.V. Morozov, N.M.R. Peres, J.M.P Lopes dos Santos, J. Nilsson, F. Guinea, A.K. Geim, A.H. Castro Neto, Phys. Rev. Lett. 99, 216802 (2007). [12] K. S. Novoselov, E. McCann, S. V. Morozov, V. I. Fal’ko, M. I. Katsnelson, U. Zeitler, D. Jiang, F. Schedin, and A. K. Geim, Nat. Phys.2, 177 (2006). [13] T. Taychatanapat, K. Watanabe, T. Taniguchi, P. Jarillo-Herrero, Nat.Phys.7, 621 (2011). [14] L. Zhang, Y. Zhang, J. Camacho, M. Khodas, and I. Zaliznyak. Nat. Phys. 7, 953 (2011). 10 [15] N.E. Staley, C.P. Puls, and Y. Liu, Phys. Rev. B 77, 155429 (2008). [16] F. Y. Yang, K. Liu, K. Hong, D. H. Reich, P. C. Searson, C. L. Chien, Y. Leprince- Wang, K. Y. Zhang, and K. Han, Phys Rev B 61, 6631 (2000). [17] J. Kim, D. Kim, T. Chang, and W. Lee, Appl. Phys. Lett. 105, 123107 (2014). [18] P. Moses and R. H. McKenzie, Phys. Rev. B 60, 7998 (1999). [19]Y. Kim et al., Phys. Rev. Lett. 110, 096602 (2013). [20] O. E. Raichev, Phys. Rev. B 81, 195301 (2010). 11 Figure Captions: Fig. 1. SEM images of the main steps followed in the proposed method in creating multilayered devices. (a) Thin strips of layered graphene are peeled off of larger flakes. (b) The strips are transported to desired location using the probe tip. (c) Gas injection is used to make platinum tabs to keep strips in place. (d) The final device ready for a four or six probe measurement. Fig. 2. Raman spectrum of multilayered graphene. Inset: Decomposition of 2D peak. Fig. 3. (a) Resistance temperature dependence of four-probe device at various magnetic fields (0 T, 4 T, 8 T, and 12 T). (b) The resistance versus positive applied field for temperatures 15 K and 20 K showing oscillation peaks. (c) Calculated carrier density as a function of applied magnetic field at different temperatures. (d) A plot of the resistance versus applied positive field for temperatures 2K, 15 K and 25 K. Fig. 4. A comparison of the longitudinal resistance as a function of inverse applied magnetic field. Although oscillations do not occur at the same field there is a correspondence in the period at different temperatures (a) 25 K, (b) 20 K, (c) 15 K, and (d) 10 K. Fig. 5. (a) A plot of oscillation extrema amplitude (Aex) versus inverse of magnetic field. (b) The variation of quantum scattering time with temperature. 12
1710.04786	2	1710	2017-12-02T03:15:09	Tunable Intrinsic Plasmons due to Band Inversion in Topological Materials	[ "cond-mat.mes-hall" ]	The band inversion has led to rich physical effects in both topological insulators and topological semimetals. It has been found that the inverted band structure with the Mexican-hat dispersion could enhance the interband correlation leading to a strong intrinsic plasmon excitation. Its frequency ranges from several $\mathrm{meV}$ to tens of $\mathrm{meV}$ and can be effectively tuned by the external fields. The electron-hole asymmetric term splits the peak of the plasmon excitation into double peaks. The fate and properties of this plasmon excitation can also act as a probe to characterize the topological phases even in the lightly doped systems. We numerically demonstrate the impact of the band inversion on plasmon excitations in magnetically doped thin films of three-dimensional strong topological insulators, V- or Cr-doped (Bi, Sb)$_2$Te$_3$, which support the quantum anomalous Hall states. Our work thus sheds some new light on the potential applications of topological materials in plasmonics.	cond-mat.mes-hall	cond-mat	Tunable Intrinsic Plasmons due to Band Inversion in Topological Materials Furu Zhang,1 Jianhui Zhou,2, 3, ∗ Di Xiao,2 and Yugui Yao1, † 1Beijing Key Laboratory of Nanophotonics and Ultrafine Optoelectronic Systems, School of Physics, Beijing Institute of Technology, Beijing 100081, China 2Department of Physics, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, USA 3Department of Physics, The University of Hong Kong, Pokfulam Road, Hong Kong, China The band inversion has led to rich physical effects in both topological insulators and topological semimetals. It has been found that the inverted band structure with the Mexican-hat dispersion could enhance the interband correlation leading to a strong intrinsic plasmon excitation. Its fre- quency ranges from several meV to tens of meV and can be effectively tuned by the external fields. The electron-hole asymmetric term splits the peak of the plasmon excitation into double peaks. The fate and properties of this plasmon excitation can also act as a probe to characterize the topological phases even in the lightly doped systems. We numerically demonstrate the impact of the band inver- sion on plasmon excitations in magnetically doped thin films of three-dimensional strong topological insulators, V- or Cr-doped (Bi,Sb)2Te3, which support the quantum anomalous Hall states. Our work thus sheds some new light on the potential applications of topological materials in plasmonics. Introduction.–Plasmons are ubiquitous collective den- sity oscillations of an electron liquid and can occur in metals, doped semiconductors and semimetals [1, 2]. The frequency of plasmon excitation is usually proportional to the density of states (DOS) of carriers at the Fermi level in the long wavelength limit [3]. In systems with vanishing DOS at the Fermi level, however, it had been demonstrated that strong correlation, higher order term in the effective mass, and the chiral anomaly could give rise to exotic plasmon excitations at zero temperature in the context of two dimensional (2D) Anderson insula- tors [4], graphene [5], HgTe quantum well [6], and three- dimensional (3D) Weyl semimetals [7], respectively. In this paper, we reveal a new mechanism based on the band inversion with the Mexican-hat dispersion that can be used to excite and manipulate intrinsic plasmon excita- tions in topological materials even with vanishing DOS. The band inversion has played a critical role in the transport properties of both topological insulators and topological semimetals, which have attracted intensive interest in condensed matter physics [8–11]. However, there has been little study of the effect of band inversion on the optical properties of topological materials [12]. Recently, the experimental observation of the quantum anomalous Hall (QAH) effect [13, 14] has been reported in magnetically doped ultrathin films of topological in- sulators V- or Cr-doped (Bi, Sb)2Te3 [15–17], in which its band topology can be tuned by external fields. On the other hand, some recent experimental efforts have been made on observing the plasmon excitations in the bulk or surface of both the magnetic and nonmagnetic 3D topological insulators [18–26]. The frequency of these plasmon excitations ranges from terahertz to far infrared and facilitates a wide variety of applications of topologi- cal insulators, such as in information and communication, chemical and biological sensing, and medical sciences [27– 29]. Therefore, these QAH topological materials provide us with a promising platform to controllably create and manipulate collective excitations via the band inversion mechanism, a new control knob for the field of plasmon- ics [1]. In this paper, we show that the inverted band struc- ture with the Mexican-hat dispersion could enhance the interband correlation, leading to a strong intrinsic plas- mon excitation. We theoretically examine the relevant physics in both the intrinsic and lightly doped thin films of topological insulators. The frequency of the resulting intrinsic plasmon can be effectively tuned by the external fields. The peak of plasmon excitation is split into dou- ble peaks by the electron-hole asymmetry. Consequently, the plasmon excitation provides us with an effective tool to identify the topological phases. Model of the QAH systems.–To demonstrate our the- ory, we begin with the following Hamiltonian that de- scribes the band inversion in the magnetically doped thin films of topological insulators V- or Cr-doped (Bi, Sb)2Te3 [30] H = H0 + m 2 τ0 ⊗ σz, (1) where m is the exchange field originating from the mag- netic dopants, effectively acting as a Zeeman field. H0 is given by [31, 32] H0 = −Dk2 +(cid:18) h+ (k) V V h− (k) (cid:19) , (2) with h± (k) = (cid:18) ±( ∆ 2 − Bk2) −ivF k+ ivF k− 2 − Bk2) (cid:19) , ∓( ∆ where h± (k) describes the 2D Dirac fermions with a k- dependent mass. Here k = (kx, ky) is the 2D wave vector, y, vF is the effective ve- locity, the D term breaks the electron-hole symmetry, ∆ is the hybridization of the top and bottom surface states k± = kx ± iky and k = qk2 x + k2 of the thin film, and V measures the structural inversion asymmetry between the top and bottom surfaces. The Pauli matrices τ0 and σz act on the pseudospin space re- lated to the top and bottom surfaces and the real spin degree of freedom, respectively. In this paper we primar- ily consider the insulating phase that requires D < B. The actual value of the parameters describes realistic ma- terials [14, 15]. Specifically, vF = 3.0 eV· A, V = 0.03 eV and B = −30 eV · A2. Let us first discuss the topological properties of the Hamiltonian in Eq. (1). As the exchanged field m increases, the system will undergo a topological phase transition from a trivial phase to a nontrivial one or vice versa. The critical value of the exchanged filed is m0 = −√∆2 + 4V 2. It is useful to introduce a dimen- sionless parameter ξ = m/m0 to mark the topological phase transition. When ξ < 1, the system is in a topo- logically trivial phase, and when ξ > 1, it is in a non- trivial phase with chiral edge states [15]. The band gap between the two inner bands closes at the critical point with ξ = 1. Fig. 1(a) clearly exhibits that in the triv- ial phase, the band edge have an obvious Mexican-hat like inverted band structure. The joint density of states near the band edge of this inverted band exhibits a 1D behavior and possesses a Van Hove singularity at a finite wave vector, in which the corresponding DOS diverges, as shown in Fig. 1(c). On the other hand, in the nontriv- ial phase [see Fig. 1(b)], all the four bands are parabolic, i.e., uninverted [30]. The DOS near these band edges does not possess a strong singularity [Fig. 1(d)]. In ad- dition, Figs. 1(e)-1(f) show a great enhancement of the overlap of wave functions due to the band inversion. Band inversion mechanism for plasmons.–We start from the general dielectric function at arbitrary wave vec- tor q and frequency ω within the random phase approx- imation [3] ε(q, ω) = 1 − V (q)Π(q, ω), (3) where V (q) = 2πe 2/κq is the Fourier transform of the 2D Coulomb interaction, κ is the effective background dielec- tric constant. The noninteracting polarization function Π(q, ω) has the form [33] Π(q, ω) = Xλ,λ′ Z d2k (2π)2 f (Eλk) − f (Eλ′k′ ) ω + Eλk − Eλ′k′ + iη Fλλ′ (k, k′), (4) in which the overlap of eigenstates Fλλ′ (k, k′) is given by 2 , (5) Fλλ′ (cid:0)k, k′(cid:1) = (cid:12)(cid:12)(cid:10)k, λλ′, k′(cid:11)(cid:12)(cid:12) where (cid:12)(cid:12)λ′, k′(cid:11) is the periodic part of the Bloch wave function with k′ = k + q. The band indices λ and λ′ run over all relevant energy bands. η is related to the electron lifetime due to the Landau damping. Eλk is the energy dispersion of the effective Hamiltonian, and f (x) = [exp{β (x − µ)} + 1]−1 is the Fermi distribution (a) (b) 2 -0.02 -0.01 0.00 k / Å-1 0.01 0.02 -0.02 -0.01 0.01 0.02 0.00 k / Å-1 (c) Van Hove singularity (d) 0.00 0.01 0.02 0.03 0.00 0.01 0.02 0.03 / eV (e) / eV (f) 0.4 0.3 0.2 V e / ) k ( E S O D 0.030 0.015 0.000 -0.015 -0.030 0.004 0.003 0.002 0.001 0.000 0.014 1 - Å / y k 0.007 0.000 -0.007 -0.014 -0.014 -0.007 0.000 kx / Å-1 0.007 0.014 -0.014 -0.007 0.007 0.014 0.000 kx / Å-1 0.1 0.0 Figure 1. (color online) The band structures of the two in- ner bands of the thin film of topological insulators described by Eq. (1) in the trivial (a) and nontrivial (b) phases, re- spectively. (c) and (d) show the corresponding DOS. (e) and (f) are the intensity plots of the overlap of wave functions F2,3(k, k′). Parameters: ∆ = −0.01 eV, D = 10.0 eV · A2, η = 10−5 eV and q = 2 × 10−3 A−1. function with β = 1/kBT . For simplicity, we assume zero temperature T = 0 K throughout this paper. It is well known that the plasmons can be revealed as sharp peaks in the energy-loss function, defined as the imaginary part of the inverse dielectric function [3] Loss (q, ω) = Im(cid:20)− 1 ε (q, ω)(cid:21) , (6) which can be probed in various spectroscopy experiments [34], such as the electron energy-loss spectroscopy and angle-resolved photoemission spectroscopy. The polarization function in Eq. (4) consists of both the interband and the intraband parts. When the Fermi level lies in the band gap, at zero temperature, the in- traband transition is forbidden. Thus, the interband correlation will determine the plasmon excitations. In conventional semiconductors or insulators, the interband correlation decays rapidly because of small wave function overlap in Eq. (5). This is why the plasmon excitations usually occur in doped systems, in which the intraband correlation becomes nonzero and dominates over the in- terband correlation. Here we notice that the band inversion could enhance 3 3.0 2.4 1.8 1.2 0.6 0.0 V e / V e / 0.05 0.04 0.03 0.02 0.01 0.00 0.05 0.04 0.03 0.02 0.01 0.00 (a) 0.003 0.006 (c) (b) 0.009 0.012 0.015 0.003 0.006 5 4 3 2 1 0.009 0.012 0.015 (d) 0.012 0.015 0 0.000 0.005 0.003 0.006 0.009 q / Å-1 0.015 0.020 0.010 / eV Figure 3. (color online) The evolution of the plasmon dis- persion in the intrinsic QAH systems of thin films of V- or Cr-doped (Bi, Sb)2Te3 for ξ= 0.5 (a), 0.8 (b), 1.0 (c), respec- tively. The color bar denotes intensities of Loss (q, ω). (d) Plots of the energy-loss function at q = 2 × 10−3 A−1 for dif- ferent values of ξ. The effective dielectric constant is κ = 5. Other parameters are the same as those in Fig. 1. phase transition occurs (ξ=1), the band gap closes, and the dispersion curve becomes shortest. Our numerical calculations also show that, after the topological phase transition (ξ>1), the plasmon becomes almost invisible due to the weak interband correlation. Thus, the band inversion strongly modifies the properties of this inter- band plasmon dispersion. It is one of the main results in this work. The zeros in the real part of the dynamical dielectric function, Re[ε(q, ω)], can also be used to determine the plasmon excitations. Fig. 4(a) depicts the numerical re- sults of Re [ε(q, ω)] (green line), Im [ε(q, ω)] (orange line) and the energy loss function for ξ = 0. It is clear that there exist two peaks (labeled by P1 and P2) in plasmon dispersion. The plasmon P2 has a higher energy and a weak damping rate, which is characterized by the imagi- nary part of the dielectric function. Compared with P2, the plasmon P1 has a lower energy and a stronger damp- ing rate. We also note that a smaller dielectric constant κ can weaken the damping rate and thus enhance plas- mon P1. We plot the dispersions for both plasmons in Fig. 4(b) and find that the energy split of this double peak can reach to tens of meV. Physically, plasmon P2 (P1) stems from the sum (difference) of processes (II) and (III) in Fig. 2. Thus, the double-peak fine struc- ture in the plasmon dispersions is a direct consequence of the electron-hole asymmetry in Eq. (2), which breaks the degeneracy between two interband processes (II) and Figure 2. Schematically decomposing the interband transition between a pair of inverted bands into four distinct processes in the uninverted band picture. Processes (I) and (IV) cor- respond to the interband transitions, whereas processes (II) and (III) corresponding to the effective intraband transitions shall greatly enhance the interband coherence and hence are critical to the strong intrinsic plasmon excitation. the interband correlation and give rise to a strong in- trinsic plasmon excitation. To make the physics more transparent, we divide the interband transition between the two inverted bands into four different processes, as shown in Fig. 2. It is clear that the strong overlap of interband transition in the inverted band picture above corresponds to the effective intraband transition in the uninverted band picture, labeled by the processes (II) and (III) in Fig. 2. These two processes are responsi- ble for the band-inversion enhanced interband transition due to the large wave function overlap [Figs. 1(e)-1(f)]. On the other hand, the joint DOS near the highly de- generate band minima of this inverted conduction band exhibits an 1D behavior and possesses a Van Hove sin- gularity, in which the corresponding DOS diverges, as shown in Fig. 1(c). In addition, there exists a weak dis- continuous point at the conduction band maximum in Fig. 1(c). The band-inversion induced Van Hove sin- gularity has also played a crucial role in unconventional superconductivity [35], and ferromagnetism [36]. In the following, we show that the large interband correlation together with the Van Hove singularity is essential to give rise to strong intrinsic plasmon excitations [37]. Plasmon in intrinsic QAH systems.–To determine the plasmon dispersion in intrinsic QAH systems of thin films of V- or Cr-doped (Bi, Sb)2Te3, we numerically calculate the energy loss function Loss (q, ω) with several values of ξ and present the corresponding results in Figs. 3(a)- 3(d). When ξ is small, the system has a clear band in- version structure [see Fig. 1(a)]. Meanwhile, there is a strong plasmon excitation in the low frequency region (ω < 0.05 eV), as shown in Fig. 3(a). The plasmon dispersion scales approximately linearly with the wave vector q. The energy of this plasmon ranges from sev- eral meV to tens of meV and can be effectively tuned by external fields and other parameters. As ξ increases, the height of the peak of the Mexican hat decreases. Ac- cordingly, the plasmons becomes weakened and their dis- persion curves become short. The plasmon frequency also has an obvious red-shift due to the shrink of the gap between the two inner bands. When the topological (a) P1 40 30 20 10 0 ] Re[ ] Im[ Im[-1/ ] P2 ] / 1 - [ m I , ] [ m I , ] [ e R 0.012 (a) 0.010 V e / 0.008 (b) =0.8 0.012 0.010 0.008 0.006 0.006 0.0006 0.0009 0.0012 q / Å-1 0.0015 0.0018 0.0012 0.0015 0.0021 0.0018 q / Å-1 0.0024 4 4.0 3.2 2.4 1.6 0.8 0.0 0.02 0.03 0.04 / eV 0.05 0.06 (b) P1 P2 0.06 0.05 0.04 0.03 0.02 V e / 0.01 0.000 0.002 0.004 0.008 0.010 0.012 0.006 q / Å Figure 4. (color online) The double-peak structure in plasmon dispersion induced by the electron-hole asymmetric term in the thin films of V- or Cr-doped (Bi, Sb)2Te3 (a). The real part, the imaginary part of the dielectric function, and the energy-loss function are labeled by the green, yellow and blue lines, respectively. (b) shows the energy dispersion of double plasmons (P1 and P2) determined by the peaks of the energy loss function. The parameters are ∆ = −0.02 eV, D = 28 eV · A2, κ = 1, ξ = 0, and q = 8 × 10−3 A−1. (III) in Fig. 2. It is useful to point out that for the in- trinsic gapped graphene [39, 40], silicene and other buck- led honeycomb lattices [42, 43], the polarization function within the random phase approximation satisfies the re- lation Re[Π(q, ω)] ≤ 0 or Re[ε(q, ω)] ≥ 1 [39]. As a re- sult, when the Fermi level lies in the band gap and has a vanishing DOS, these systems hardly develop a plasmon excitation at zero temperature. Several remarks are in order here. First, this band in- version enhancement mechanism for interband plasmons is also applicable to the quantum spin Hall effect in HgTe quantum wells [8, 9], and in InAs/GaSb quantum wells [44]. Unlike the effective model in Eq. (1) we study here, the inverted bands in these systems are usually in the topologically nontrivial phase. Secondly, in Ref. [6], the dynamical polarization function and weak plasmons in HgTe quantum wells were calculated. But the mechanism to enhance and manipulate in-gap plasmon mode was not Figure 5. (color online) The energy dispersion of the plas- mons for a lightly doped QAH system of a thin film of V- or Cr-doped (Bi, Sb)2Te3 with µ = 0.004 eV and κ = 5 for ξ = 0.5 (a) and 0.8 (b), respectively. The blue dashed lines determined by minima of the single particle excitation energy, guide to the eye, denote the edge of electron-hole continuum, whereas the color bar is for intensities of Loss (q, ω). Other parameters are identical to those in Fig. 1. discussed. In addition, our present work only reveals the effect of the band inversion on the longitudinal bulk plas- mon excitations. The case for transverse plasmon excita- tions will be reported elsewhere. Note that the plasmon excitations studied here essentially differ from the edge magnetoplasmon with no magnetic fields observed in the topological nontrivial phase [25]. Plasmon in lightly doped QAH systems.–In reality, the system might be metallic due to defects, self-doping, and charge transfer from substrates, gating, etc. To gain more insights into the intrinsic plasmon excitation, we briefly discuss the impact of finite doping [45]. For the lightly doped case, the interband correlation is still strong enough to rise the novel interband plasmon, whereas a usual plasmon comes from the intraband transition. On the other hand, for the heavily doped case, the interband correlation will be significantly suppressed such that the interband plasmon excitation disappears. However, the usual intraband plasmon excitation continues to exist. Figs. 5(a) and 5(b) depict the energy dispersion of plas- mons for µ = 4.0 meV measured from absolute zero. The novel interband plasmon starts from a relatively large wave vector in Fig. 5(a) and becomes weak in Fig. 5(b) and finally disappears near the topological phase transi- tion, similar to the intrinsic case in Fig. 3. Therefore, the novel interband plasmon still exists and could sig- nify the topological phase transition in the lightly doped QAH systems. As shown in Fig. 5(a), the usual intra- band plasmon starts from a very small wave vector and shows the energy dispersion ω (q) ∝ √q for small q, simi- lar to the conventional 2D electron gases [2] and 2D Dirac fermions [34, 39–43]. After q excesses 1.2× 10−3 A−1, its dispersion greatly deviates from √q. Numerical results show that the intraband plasmon survives and remains noticeable even when ξ > 1. Thus, this intraband plas- mon is insensitive to the topology of energy bands. In addition, the frequency of the intraband plasmon shows a continuous redshift as increasing ξ. In summary, it has been shown that the band inver- sion with the Mexican-hat dispersion indeed leads to a strong intrinsic plasmon excitation with highly tunable frequencies in topological materials. The electron-hole asymmetric term could induce a double-peak fine struc- ture in the plasmon dispersion. We have further pointed out that this intrinsic plasmon excitation can be used to characterize the topological phase transition via bulk measurements. The above physics has been numerically examined in both the intrinsic and lightly doped thin films of magnetic topological insulators V- or Cr-doped (Bi, Sb)2Te3. Therefore, our work paves the way for the potential applications of topological materials in ter- ahertz and infrared plasmonics. A series of recent experimental advances have been made in the detection of plasmons in 3D topological insu- lators, in particular, in their magnetic thin films. We ex- pect that at the current status of experimental progress, the predicted anomalous plasmons and the double-peak structure will soon be tested experimentally. We are grateful to Hao-Ran Chang, Wen-Yu Shan and Zhi-Ming Yu for enlightening discussions. F.Z. and Y.Y. acknowledge fundings from the MOST Project of China (Nos. 2014CB920903 and 2016YFA0300603), the National Nature Science Foundation of China (Grant Nos. 11574029 and 11734003). J.Z. and D.X. were sup- ported by AFOSR Grant No. FA9550-14-1-0277. J.Z. was also supported by the Research Grant Council, Uni- versity Grants Committee, Hong Kong under Grant No. 17301116 and C6026-16W. ∗ [email protected] † [email protected] [1] S. Maier, Plasmonics: Fundamentals and Applications (Springer, New York, 2007). [2] G. Giuliani and G. Vignale, Quantum theory of the electron liquid (Cambridge University Press, Cambridge, UK, 2005). [11] N. P. Armitage, E. J. Mele, and A. Vishwanath, 5 arXiv:1705.01111 [cond-mat.str-el]. A. Tan and Z. [12] L. M. Rappe, Phys. Rev. Lett. 116, 237402 (2016). [13] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). [14] R. Yu, W. Zhang, H.-J. Zhang, S.-C. Zhang, X. Dai, and Z. Fang, Science 329, 61 (2010). [15] C.-Z. Chang, J. Zhang, X. Feng, J. Shen, Z. Zhang, M. Guo, K. Li, Y. Ou, P. Wei, L.-L. Wang, et al., Science 340, 167 (2013). [16] A. J. Bestwick, E. J. Fox, X. Kou, L. Pan, and D. Goldhaber-Gordon, L. Wang, K. Phys. Rev. Lett. 114, 187201 (2015). [17] C.-Z. Chang, W. Zhao, D. Y. Kim, H. Zhang, B. A. Assaf, D. Heiman, S.-C. Zhang, C. Liu, M. H. Chan, and J. S. Moodera, Nat. Mater. 14, 473 (2015). [18] S. Raghu, S. B. Chung, X.-L. Qi, and S.-C. Zhang, Phys. Rev. Lett. 104, 116401 (2010). [19] R. E. V. Profumo, R. Asgari, M. Polini, and A. H. Mac- Donald, Phys. Rev. B 85, 085443 (2012). [20] P. Di Pietro, M. Ortolani, O. Limaj, A. Di Gaspare, V. Giliberti, F. Giorgianni, M. Brahlek, N. Bansal, N. Koirala, S. Oh, P. Calvani, and S. Lupi, Nat Nano 8, 556 (2013). [21] J.-Y. Ou, J.-K. So, G. Adamo, A. Sulaev, L. Wang, and N. I. Zheludev, Nat. Commun. 5 (2014). [22] A. Politano, V. M. Silkin, I. A. Nechaev, M. S. Vitiello, L. Viti, Z. S. Aliev, M. B. Babanly, G. Chiarello, P. M. Echenique, and E. V. Chulkov, Phys. Rev. Lett. 115, 216802 (2015). [23] A. Kogar, S. Vig, A. Thaler, M. H. Wong, Y. Xiao, D. Reig-i Plessis, G. Y. Cho, T. Valla, Z. Pan, J. Schneeloch, R. Zhong, G. D. Gu, T. L. Hughes, G. J. MacDougall, T.-C. Chiang, and P. Abbamonte, Phys. Rev. Lett. 115, 257402 (2015). S. Babakiray, [24] Y. D. Glinka, T. A. John- Lederman, son, M. B. Holcomb, Nat. Commun. 7, 13054 (2016). and D. [25] A. Mahoney, J. Colless, L. Peeters, S. Pauka, E. Fox, X. Kou, L. Pan, K. Wang, D. Goldhaber-Gordon, and D. Reilly, arXiv:1703.03122 [cond-mat.mes-hall]. [26] X. Jia, S. Zhang, R. Sankar, F.-C. Chou, W. Wang, and K. Kempa, E. W. Plummer, J. Zhang, X. Zhu, J. Guo, Phys. Rev. Lett. 119, 136805 (2017). [27] B. Ferguson and X.-C. Zhang, Nat Mater 1, 26 (2002). [28] M. Tonouchi, Nat Photon 1, 97 (2007). [29] R. Soref, Nat Photon 4, 495 (2010). [30] H.-Z. Zhao, S.-Q. and Lu, A. Shen, [3] D. Pines, Elementary Excitations in Solids: Lectures on Protons, Electrons, and Plasmons, Phys. Rev. Lett. 111, 146802 (2013). Vol. 5 (Westview Press, Boulder, 1999). and M. Shahbazyan V. [4] T. E. Raikh, S.-C. Zhang, Nat. Phys. 5, 438 (2009). [31] H. Zhang, C.-X. Liu, X.-L. Qi, X. Dai, Z. Fang, and Phys. Rev. B 53, 7299 (1996). [32] W.-Y. Shan, H.-Z. Lu, and S.-Q. Shen, [5] S. Gangadharaiah, A. M. Farid, and E. G. Mishchenko, New Journal of Physics 12, 043048 (2010). Phys. Rev. Lett. 100, 166802 (2008). [6] S. Juergens, P. Michetti, and B. Trauzettel, Phys. Rev. Lett. 112, 076804 (2014). [7] J. Zhou, H.-R. Chang, and D. Xiao, [33] See Supplemental Material for the derivation of the non- interacting polarization function in Eq. (4) in the main text, which includes Ref. [2]. [34] A. Grigorenko, M. Polini, and K. Novoselov, Nat. Pho- Phys. Rev. B 91, 035114 (2015). ton. 6, 749 (2012). [8] M. Z. Hasan and C. L. Kane, [35] G. Goldstein, C. Aron, and C. Chamon, Rev. Mod. Phys. 82, 3045 (2010). Phys. Rev. B 92, 020504 (2015). [9] X.-L. Qi and S.-C. Zhang, [36] T. Cao, Z. Li, and S. G. Louie, Rev. Mod. Phys. 83, 1057 (2011). [10] A. Bansil, H. Lin, and T. Das, Rev. Mod. Phys. 88, 021004 (2016). Phys. Rev. Lett. 114, 236602 (2015). [37] In some literatures, a sign change of mass in the Dirac fermions with a constant mass is also called the band inversion [38]. However, it only exchanges the parity of wave functions but does not lead to any enhancement of the interband correlation or the Van Hove singularity. Thus, this case is not the key focus of the present paper. (Springer-Verlag, Insulators [38] S.-Q. Shen, Topological Berlin, 2012). [39] X.-F. Wang and T. Chakraborty, Phys. Rev. B 75, 033408 (2007). [40] P. K. Pyatkovskiy, J. Phys.: Condens. Matter 21, 025506 (2009). [41] A. Scholz, T. Schliemann, Stauber, and J. Phys. Rev. B 86, 195424 (2012). [42] H.-R. Chang, J. Zhou, H. Zhang, and Y. Yao, 6 Phys. Rev. B 89, 201411 (2014). [43] C. J. Tabert and E. J. Nicol, Phys. Rev. B 89, 195410 (2014). [44] C. Liu, T. L. Hughes, X.-L. Qi, K. Wang, and S.-C. Zhang, Phys. Rev. Lett. 100, 236601 (2008). [45] We utilize the rigid-band approximation, in which the energy band structure is assumed not be changed by a slightly doping except a shift of the Fermi level. In this work, we focus on the n-doped case, namely, the Fermi level crosses the conduction bands. The p-doped case can be calculated in a similar way. SUPPLEMENTARY MATERIAL FOR "TUNABLE INTRINSIC PLASMONS DUE TO BAND INVERSION IN TOPOLOGICAL MATERIALS" This supplemental material contains a detailed derivation of Eq. (4) in the main text. We start with the polarization function within the one-loop approximation at a finite temperature T Π (q, iωm) = kBT +∞ Xn=−∞ 1 L2 Xk tr(cid:2)G (k, iΩn) G(cid:0)k′, iΩn + iωm(cid:1)(cid:3) , (7) where ωm = 2mπkBT and Ωn = (2n + 1) πkBT are Matsubara frequencies, L2 is the area of a 2D system, and k′ = k + q. Using the Matsubara Green's function in the eigenstate basis, G (k, iΩn) = Xλ λkihλk iΩn − Eλk + µ with µ being the chemical potential and H (k)λki = Eλk λki, one gets iΩn − Eλk + µ · L2 Xk Xs,p Xn=−∞Xλ,λ′ Π (q, iωm) = kBT λki hλk hsp +∞ 1 , (cid:12)(cid:12)λ′k′(cid:11)(cid:10)λ′k′(cid:12)(cid:12) iΩn + iωm − Eλ′k′ + µ spi . (8) (9) After making use of the orthogonal relation hλkspi = δλsδkp and summing over s and p, one has iΩn + iωm − Eλ′k′ + µ (cid:12)(cid:12)(cid:10)λkλ′k′(cid:11)(cid:12)(cid:12) iΩn − Eλk + µ · Xn=−∞Xλ,λ′ Π (q, iωm) = kBT L2 Xk +∞ 1 1 1 2 By use of the summation formula [2], kBT +∞ Xn=−∞ 1 iΩn − (x1 − µ) · 1 iΩn + iωm − (x2 − µ) = f (x1) − f (x2) iωm + x1 − x2 , . (10) (11) (12) performing the summation over n leads to Π (q, iωm) = Xλ,λ′ 1 L2 Xk f (Eλk) − f (Eλ′k′) iωm + Eλk − Eλ′k′ (cid:12)(cid:12)(cid:10)λkλ′k′(cid:11)(cid:12)(cid:12) 2 . Carrying out the analytic continuation from Matsubara frequencies iωm → ω + iη, one obtains the retarded nonin- teracting polarization function Replacing (cid:0)1/L2(cid:1)Pk with R d2k 1 L2 Xk Π (q, ω) = Xλ,λ′ (2π)2 , one immediately reaches f (Eλk) − f (Eλ′k′ ) ω + Eλk − Eλ′k′ + iη (cid:12)(cid:12)(cid:10)λkλ′k′(cid:11)(cid:12)(cid:12) 2 . Π (q, ω) = Xλ,λ′Z d2k (2π)2 [f (Eλk) − f (Eλ′k′)] ω + Eλk − Eλ′k′ + iη (cid:12)(cid:12)(cid:10)λkλ′k′(cid:11)(cid:12)(cid:12) 2 , which is the noninteracting polarization function in Eq. (4) in the main text. (13) (14)
1206.0099	1	1206	2012-06-01T07:04:07	Metallic phase of the quantum Hall effect in four-dimensional space	[ "cond-mat.mes-hall" ]	We study the phase diagram of the quantum Hall effect in four-dimensional (4D) space. Unlike in 2D, in 4D there exists a metallic as well as an insulating phase, depending on the disorder strength. The critical exponent $\nu\approx 1.2$ of the diverging localization length at the quantum Hall insulator-to-metal transition differs from the semiclassical value $\nu=1$ of 4D Anderson transitions in the presence of time-reversal symmetry. Our numerical analysis is based on a mapping of the 4D Hamiltonian onto a 1D dynamical system, providing a route towards the experimental realization of the 4D quantum Hall effect.	cond-mat.mes-hall	cond-mat	Metallic phase of the quantum Hall effect in four-dimensional space J. M. Edge,1 J. Tworzyd lo,2 and C. W. J. Beenakker1 1Instituut-Lorentz, Universiteit Leiden, P.O. Box 9506, 2300 RA Leiden, The Netherlands 2Institute of Theoretical Physics, Faculty of Physics, University of Warsaw, Hoza 69, 00 -- 681 Warsaw, Poland (Dated: June 2012) We study the phase diagram of the quantum Hall effect in four-dimensional (4D) space. Unlike in 2D, in 4D there exists a metallic as well as an insulating phase, depending on the disorder strength. The critical exponent ν ≈ 1.2 of the diverging localization length at the quantum Hall insulator-to-metal transition differs from the semiclassical value ν = 1 of 4D Anderson transitions in the presence of time-reversal symmetry. Our numerical analysis is based on a mapping of the 4D Hamiltonian onto a 1D dynamical system, providing a route towards the experimental realization of the 4D quantum Hall effect. PACS numbers: 64.70.Tg, 72.15.Rn, 73.43.Nq To understand our three-dimensional world it is help- ful and rewarding to ask how physics would differ in other dimensions. The interest in two-dimensional and four-dimensional worlds goes back to the 19th century [1 -- 3], inspired by the development of non-Euclidean ge- ometry. In modern times, nanotechnology has brought us the 2D "Flatland" in semiconductor heterostructures and in graphene [4], while ultracold atomic lattices offer a way to simulate 4D dynamics by encoding the extra dimension in an internal degree of freedom of the atoms [5]. These studies of physics in other dimensions are par- ticularly interesting if they reveal effects that lack a 3D counterpart. The integer quantum Hall effect is a cele- brated example, first discovered in 2D [6], then general- ized to 4D and 8D [7, 8], and now known to exist in any even-dimensional space [9 -- 11]. In each case an integer- valued topological invariant (a Chern number C [10 -- 12]) counts the number of gapless surface modes, observable as the quantized Hall conductance in 2D. Disorder lo- calizes single-particle excitations in the bulk but not on the surface, producing an insulating quantum Hall fluid bounded by a conducting surface. Topological phase transitions (at which C switches from one integer to another) have been extensively stud- ied in the 2D quantum Hall effect [13], but not for the higher dimensional generalizations. As we will show here, in 4D the quantum Hall insulators with a different topo- logical invariant are separated in phase space by a metal- lic region, so that a topological phase transition is com- posed of a pair of metal-insulator transitions. In 2D an intermediate metallic phase requires the presence of time- reversal symmetry (as in the quantum spin-Hall effect [14]), but in 4D there is no such requirement. We find that the 4D quantum Hall insulator-to-metal transition has a different critical behavior than the 4D Anderson transitions in time-reversally-symmetric systems [15, 16]. To determine the phase diagram of the 4D quantum Hall effect we proceed as follows: We start from a Dirac Hamiltonian [7, 10] of electrons with 4D momentum p coupled to SU(2) spin and valley degrees of freedom σ, τ . This model has time-reversal symmetry, so it is in the universality class of the quantum spin-Hall effect (class AII in the classification of Anderson transitions [18]). In order to bring it in the universality class A of the quan- tum Hall effect, we add time-reversal-symmetry breaking disorder. Following the lines set out in Ref. [19] for the 2D quantum Hall effect, we then map the 4D Hamiltonian problem onto a 1D dynamical system. Here we use this dimensional reduction to arrive at an efficient numerical simulation, but in principle this method could be used to simulate the 4D quantum Hall effect in ultracold atomic lattices [20]. The Dirac Hamiltonian has the form [10, 21] 4 Xi=0 4 v Xi=1 H(p) = 2 arctanv viΓi, (1) cos pi, sin p1, sin p2, sin p3, sin p4! , (2) v = K µ + Γ = (τx ⊗ σ0, τz ⊗ σ0, τy ⊗ σx, τy ⊗ σy, τy ⊗ σz). The vectors v, Γ have components vi, Γi (i = 0, 1, 2, 3, 4). The Pauli matrices σα, τα (α = x, y, z), with unit matri- ces σ0, τ0, act on the spin and valley degrees of freedom. The two parameters K, µ are real numbers. (3) The Hamiltonian (1) is a 4 × 4 matrix dependent on the momentum vector p = (p1, p2, p3, p4), in the Brillouin zone pi < π of a 4D square lattice (lattice constant and set equal to unity). It satisfies the time-reversal symmetry relation H(p) = T H(−p)T −1, T = (iσy ⊗ τx)K, (4) with K the operator of complex conjugation. The spec- trum consists of a pair of twofold degenerate bands E±(p) = ±2 arctanv, (5) with a band gap centered at the Fermi level E = 0. The topological invariant (the second Chern number) depends on the effective mass parameter µ, according to [10, 22] 10 3 sign (µ) − sign (µ) 0 if 0 < µ < 2, if 2 < µ < 4, if µ > 4. C =  Changes in C are signaled by a closing and reopening of the band gap at time-reversally invariant momenta [pi = −pi (mod 2π)], where v(p) vanishes. We now include a spin-dependent potential V (x), with position operator x = (x1, x2, x3, x4) = i∂/∂p, to break both translational invariance and time-reversal symme- try. To arrive at a dynamical system the potential is added stroboscopically in time, (6) 1 Metal K / 1 0.1 0.01 C = 0 Insulator C = +3 C = −1 0 1 2 3 4 µ 5 6 7 8 2 103 101 10−1 10−3 10−5 10−7 H(t) = V (x) + H(p) ∞ Xn=−∞ δ(t − n). (7) The time evolution of a wave function over one period is given by the Floquet operator F = e−iH(p)e−iV (x). (8) (We measure position in units of the lattice constant and set the stroboscopic period equal to unity.) Periodically time-dependent models of this type are called quantum kicked rotators [23]. A first advantage of the stroboscopic description is that a simple separable potential, V (x) = V1(x1) − 4 Xi=2 ωixi, V1(x1) = x2 1 τz ⊗ σ0, (9) can be used to simulate d-dimensional disorder -- pro- vided the frequencies ωi/2π are incommensurate irra- tional numbers [24, 25]. Note that the τz matrix anti- commutes with T from Eq. (4), so this potential breaks time-reversal symmetry. A second advantage is that the d-dimensional dynam- ics with a single stroboscopic period can be mapped ex- actly onto a 1-dimensional dynamics with d stroboscopic periods [26]. For that purpose we choose an initial wave function of the form Ψ(p, t = 0) = ψ(p1, t = 0) 4 Yi=2 δ(pi − αi). (10) During one period the momentum pi (i = 2, 3, 4) is in- cremented to pi + ωi (mod 2π), so we may replace the 4D dynamics by a 1D dynamics with a time-dependent Floquet operator F (t) = e−iH(p1,ω2t+α2,ω3t+α3,ω4t+α4)e−iV1(x1). We introduce a Bloch number q and calculate the time dependence of the wave function ψ(p1, t) = eiqp1 χ(p1, t). The state χ(p1, t) is a 2π-periodic function of p1, so it is a superposition of a discrete set of eigenstates eimp1 of x1. (11) FIG. 1: Phase diagram of the 4D quantum Hall effect (broken time-reversal symmetry, class A). The colour scale represents D = ∆2/√t at t = 105. The phase boundary separating metal and insulator appears approximately at D ≈ 0.08 (white). A more precise determination of the phase boundary from the scaling analysis is indicated by the red curves (with circles marking the two transitions analyzed in Table I). Because of the ±µ symmetry only positive values of µ are shown. We truncate the set to M states, m ∈ {1, 2, . . . M}, with periodic boundary conditions at the end points. The cal- culation of ψ at integer t then amounts to subsequent multiplications of the Floquet operator, which can be done efficiently using the Fast Fourier Transform algo- rithm [19]. Ranges of M up to 103 and t up to 6· 107 are included in the simulations. The metallic and insulating regions in the phase dia- gram are obtained by initializing χ(x1, t) at x1 = m0 and calculating the mean square displacement at time t, ∆2(t) =Xm (m − m0)2χ(x1 = m, t)2. (12) The overbar indicates an average over initial conditions and over the Bloch number q. In the large-t limit one has ∆(t) → constant in the insulating phase (localization) and ∆2(t) ∝ t in the metallic phase (regular diffusion). On the critical line separating these two phases ∆2(t) ∝ √t (anomalous diffusion). The resulting phase diagram as a function of effective mass µ and disorder strength 1/K is shown in Fig. 1. The clean limit of the model is reached for K → ∞ [because then V (x) is negligibly small relative to H(p)]. In this limit the system switches between topologically distinct insulating phases at µ = 0, 2, 4, see Eq. (6). Disorder introduces a metallic phase in between the in- sulating phases, so that now a change in C happens via two metal-insulator transitions. At fixed µ the metallic phase exists for a finite range of disorder strengths, van- ishing both for weak and strong disorder. The insulat- ing phase that appears at the largest disorder strengths (1/K & 10) is disconnected from the insulating phases at smaller disorder, but presumably it is topologically trivial (C = 0). 10 1 Metal K / 1 0.1 0.01 C = 0 Insulator C = +3 C = −1 0 1 2 3 4 µ 5 6 7 8 3 103 101 10−1 10−3 10−5 10−7 FIG. 3: Phase diagram of the 4D quantum spin-Hall effect (preserved time-reversal symmetry, class AII). The colour scale is as in Fig. 1. effect, by replacing the τz matrix in Eq. (9) with the unit matrix τ0. The potential V (x) then commutes with T and preserves time-reversal symmetry. The phase dia- gram is qualitatively the same, see Fig. 3, at least for not too strong disorder. The insulating phase for 1/K & 10 is now connected to the C = 0 phase at weaker disorder, so we definitely know that this is a trivial insulator. Table I (last two rows) shows that a significant effect of time-reversal symmetry appears in the critical exponent ν ≈ 1.0 for class AII -- well below what we found for class A. Within numerical accuracy, the critical exponent which we find for the 4D quantum spin-Hall effect agrees with the semiclassical formula [15, 16, 30] νsc = 1 2 + 1 d − 2 (14) for d-dimensional Anderson transitions with time- reversal symmetry. In conclusion, we have demonstrated that the phase di- agram of the quantum Hall effect depends sensitively on the dimensionality. While in 2D only insulating phases exist, in 4D a metallic phase appears as well, for an in- termediate range of not too strong and not too weak dis- order. Topological phase transitions in the 4D quantum Hall effect occur via consecutive insulator-to-metal-to- insulator transitions, rather than via a single insulator- to-insulator transition as in 2D. This altogether different phenomenology is made possible by the fact that bro- ken time-reversal symmetry prevents the formation of a metal in 2D but not in higher dimensions. One direction for further theoretical research is to- wards a semiclassical scaling theory of 4D quantum Hall transitions. In 2D there is no hope for such a theory, but in 4D the situation is more promising in view of the close agreement between numerics and semiclassics that we have found for the critical exponent in the quantum spin-Hall effect. From the experimental point of view, the mapping of a scalar 3D Hamiltonian onto a 1D dynami- cal system has been realized in cold atoms with a pulsed FIG. 2: Finite-time scaling of the mean-square displacement at the quantum Hall insulator-to-metal transition (left circle in Fig. 1). The data points result from the numerical simula- tion, the curves are fits to the scaling law (13). The crossing identifies the critical point µc, separating the metallic phase (µ > µc) from the insulating phase (µ < µc). symmetry topological critical class A A AII AII invariant C exponent ν 1.18 ± 0.05 1.21 ± 0.03 0.99 ± 0.02 1.02 ± 0.04 −1 0 −1 0 TABLE I: Critical exponent ν for the metal-insulator transi- tions indicated by circles in Figs. 1 and 3. The value C of the topological invariant at the insulating side of the transition is indicated, as well as the symmetry class: class A (quantum Hall effect) or class AII (quantum spin-Hall effect). To characterize the metal-insulator transition we have performed a finite-time scaling analysis, along the lines of Refs. [19, 27]. One-parameter-scaling in 4D requires that the mean-square-displacement obeys the scaling law ∆2(t) √t = D(ξ−4t), ξ ∝ µ − µc−ν. (13) All microscopic parameters enter only through the local- ization length ξ, which has a power law divergence at the critical point µc. We determine the critical exponent ν by expanding D in a power series near µc, including also finite-time corrections to scaling [28]. A typical scaling plot is shown in Fig. 2 [29]. Results are listed in Table I (first two rows). As ex- pected from general considerations [14], there is no de- pendence of ν on the value of the topological invariant C at the insulating side of the transition. Within numeri- cal accuracy, ν ≈ 1.2 for the transition from a metal to either a trivial or a nontrivial quantum Hall insulator. The calculations described so far are for a system with broken time-reversal symmetry, in class A of the quan- tum Hall effect. To investigate the role played by this symmetry we have repeated the calculations in the time- reversally-symmetric class AII of the quantum spin-Hall optical lattice, driven by three incommensurate frequen- cies [20, 27]. Our 4D-to-1D mapping adds the complexity of an additional driving frequency, but more importantly would require control over an internal degree of freedom of the atoms to produce the required 4 × 4 matrix struc- ture. We thank A. R. Akhmerov, J. P. Dahlhaus, and E. P. L. van Nieuwenburg for useful discussions. This re- search was supported by the Dutch Science Foundation NWO/FOM, by an ERC Advanced Investigator Grant, and by the EU network NanoCTM. 4 [1] G. F. Rodwell, Nature 8, 8 (1873). [2] E. A. Abbott, Flatland (1884, reprinted by Dover Publi- cations, New York, 1992). [3] C. H. Hinton, Speculations on the Fourth Dimension (1884, reprinted by Dover Publications, New York, 1980). [4] A. K. Geim and A. H. MacDonald, Physics Today 60 (8), 35 (2007). [5] O. Boada, A. Celi, J. I. Latorre, and M. Lewenstein, Phys. Rev. Lett. 108, 133001 (2012). [6] K. von Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). [7] S.-C. Zhang and J. Hu, Science 294, 823 (2001). [8] B. A. Bernevig, J. Hu, N. Toumbas, and S.-C. Zhang, Phys. Rev. Lett. 91, 236803 (2003). [9] G. Meng, J. Phys. A 36, 9415 (2003). [27] G. Lemari´e, J. Chab´e, P. Szriftgiser, J. C. Garreau, B. Gr´emaud, and D. Delande, Phys. Rev. A 80, 043626 (2009). [28] K. Slevin and T. Ohtsuki, Phys. Rev. B 80, 041304 (2009). [29] For details of the scaling analysis, see the Supplemental Information (Appendix). [30] Eq. (14) was derived in Ref. [16] for a time-reversal sym- metry operator that squares to +1 (orthogonal symme- try), while we have T 2 = −1 (symplectic symmetry). Apparently the critical exponent is not sensitive to this difference. Appendix A: Finite-time scaling analysis of the [10] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B metal-insulator transitions 78, 195424 (2008). [11] S. Ryu, A. Schnyder, A. Furusaki, and A. Ludwig, New J. Phys. 12, 065010 (2010). [12] D. J. Thouless, M. Kohmoto, P. Nightingale, and M. den Nijs, Phys. Rev. Lett. 49, 405 (1982). [13] B. Huckestein, Rev. Mod. Phys. 67, 357 (1995). [14] S. Ryu, C. Mudry, H. Obuse, and A. Furusaki, New J. Phys. 12, 065005 (2010). [15] A. M. Garc´ıa-Garc´ıa and E. Cuevas, Phys. Rev. B 75, 174203 (2007). [16] A. M. Garc´ıa-Garc´ıa, Phys. Rev. Lett. 100, 076404 (2008). [17] D. L. Shepelyansky, arXiv:1102.4450. [18] F. Evers and A. D. Mirlin, Rev. Mod. Phys. 80, 1355 (2008). [19] J. P. Dahlhaus, J. M. Edge, J. Tworzyd lo, and C. W. J. Beenakker, Phys. Rev. B 84, 115133 (2011). [20] J. Chab´e, G. Lemari´e, B. Gr´emaud, D. Delande, P. Szrift- giser, and J. C. Garreau, Phys. Rev. Lett. 101, 255702 (2008). [21] The factor (2/v) arctan v in the Hamiltonian (1) can be omitted without changing the topological invariant. It is included following Ref. [19], to flatten the band structure and reduce finite-size effects. [22] M. F. L. Golterman, K. Jansen, and D. B. Kaplan, Phys. Lett. B 301, 219 (1993). [23] F. M. Izrailev, Phys. Rep. 196, 299 (1990). [24] F. Borgonovi and D. L. Shepelyansky, J. Physique I 6, 287 (1996); Physica D 109, 24 (1997). [25] We performed our simulations for two sets of irrational incommensurate frequencies: {ω2/2π, ω3/2π, ω4/2π} = {1/λ, 1/λ2, 1/√2} and {ln 3, ln 4, ln 5}, with λ3 = 1 + λ. Both sets gave equivalent results, within the numerical accuracy. Data given in the text is for the first set of frequencies. [26] G. Casati, I. Guarneri, and D. L. Shepelyansky, Phys. Rev. Lett. 62, 345 (1989). Here we give the details of the scaling analysis pre- sented in the main text, following the established proce- dure [28]. Close to the transition point µc, the anoma- lous diffusion coefficient D = ∆2/√t obeys the single- parameter scaling law (13), which we can write in the form lnD = F (t1/4ν u), u = µ − µc + O(µ − µc)2. a Finite-time term t−yG(t1/4ν u) to lnD, with y > 0 the leading irrel- evant exponent. We expand the functions F and G in a power series in scaling (A1) corrections to add t1/4νu, and expand u in a power series in µ − µc, lnD = lnDc + N1 Xk=1 c(1) k (t1/4ν u)k + N3 Xk=1 u = µ − µc + N2 Xk=2 c(2) k (µ − µc)k. c(3) k t−y(t1/4ν u)k−1, (A2) (A3) We fit the unknown parameters to the numerical data for ∆2(µ, t), averaged over 500 − 1000 initial conditions, varying µ around µc for six values of t in the range 104 − 6 · 107. The number of fit parameters N1, N2, N3 was increased until we reached a value for chi-squared per degree of freedom (χ2/ndf) close to unity. We have performed this scaling analysis for two sets of incommensurate frequencies, in order to make sure that our results for the critical exponents do not depend on the precise choice of driving frequencies. The two sets are 5 • Set Ω1: ω2 = 2π/λ, ω3 = 2π/λ2, ω4 = 2π/√2, where λ is the real solution to the equation x3 − x − 1 = 0 [24]. • Set Ω2: ω2 = 2π ln 3, ω3 = 2π ln 4, ω4 = 2π ln 5. Results for both frequency sets are given in Tables II and III, and summarized for frequency set Ω1 in Table I. class C freq. set µc N1 N2 N3 A −1 A −1 A 0 A 0 3.51 1 3.51 2 4.30 1 4.30 1 Ω1 Ω2 Ω1 Ω2 2 1 1 2 ν χ2/ndf 0 1.18 ± 0.05 0 1.21 ± 0.06 1 1.21 ± 0.03 1 1.20 ± 0.03 0.97 1.01 0.89 1.05 TABLE II: Parameters for the scaling analysis of the metal- insulator transition in the 4D quantum Hall effect (symmetry class A). class C freq. set µc N1 N2 N3 AII −1 AII −1 0 AII AII 0 3.65 1 3.62 1 4.22 1 4.24 1 Ω1 Ω2 Ω1 Ω2 1 1 1 2 ν χ2/ndf 1 0.99 ± 0.02 1 0.98 ± 0.04 1 1.02 ± 0.04 2 1.07 ± 0.05 1.16 0.78 0.88 0.91 TABLE III: Parameters for the scaling analysis of the metal- insulator transition in the 4D quantum spin-Hall effect (sym- metry class AII).
1105.4551	2	1105	2011-10-23T07:31:21	Angle Dependence of Landau Level Spectrum in Twisted Bilayer Graphene	[ "cond-mat.mes-hall" ]	In the context of the low energy effective theory, the exact Landau level spectrum of quasiparticles in twisted bilayer graphene with small twist angle is analytically obtained by spheroidal eigenvalues. We analyze the dependence of the Landau levels on the twist angle to find the points, where the two-fold degeneracy for twist angles is lifted in the nonzero modes and below/above which massive/massless fermion pictures become valid. In the perpendicular magnetic field of 10\,T, the degeneracy is removed at $\theta_{{\rm deg}}\sim 3^\circ$ %angles around 3 degrees for a few low levels, specifically, $\theta_{\rm deg}\simeq 2.56^\circ$ for the first pair of nonzero levels and $\theta_{\rm deg}\simeq 3.50^\circ$ for the next pair. Massive quasiparticle appears at $\theta<\theta_{{\rm c}}\simeq 1.17^\circ$ in 10\,T, %angles less than 1.17 degrees. which match perfectly with the recent experimental results. Since our analysis is applicable to the cases of arbitrary constant magnetic fields, we make predictions for the same experiment performed in arbitrary constant magnetic fields, e.g., for B=40\,T we get $\theta_{\rm c}\simeq 2.34^\circ$ and the sequence of angles $\theta_{\rm deg} = 5.11, 7.01, 8.42,...$ for the pairs of nonzero energy levels. The symmetry restoration mechanism behind the massive/massless transition is conjectured to be a tunneling (instanton) in momentum space.	cond-mat.mes-hall	cond-mat	Angle Dependence of Landau Level Spectrum in Twisted Bilayer Graphene Min-Young Choi, Young-Hwan Hyun, and Yoonbai Kim Department of Physics, BK21 Physics Research Division, and Institute of Basic Science Sungkyunkwan University, Suwon 440-746, Korea∗ In the context of the low energy effective theory, the exact Landau level spectrum of quasiparticles in twisted bilayer graphene with small twist angle is analytically obtained by spheroidal eigenvalues. We analyze the dependence of the Landau levels on the twist angle to find the points, where the two-fold degeneracy for twist angles is lifted in the nonzero modes and below/above which massive/massless fermion pictures become valid. In the perpendicular magnetic field of 10 T, the degeneracy is removed at θdeg ∼ 3◦ for a few low levels, specifically, θdeg ≃ 2.56◦ for the first pair of nonzero levels and θdeg ≃ 3.50◦ for the next pair. Massive quasiparticle appears at θ < θc ≃ 1.17◦ in 10 T, which match perfectly with the recent experimental results. Since our analysis is applicable to the cases of arbitrary constant magnetic fields, we make predictions for the same experiment performed in arbitrary constant magnetic fields, e.g., for B = 40 T we get θc ≃ 2.34◦ and the sequence of angles θdeg = 5.11, 7.01, 8.42, · ·· for the pairs of nonzero energy levels. The symmetry restoration mechanism behind the massive/massless transition is conjectured to be a tunneling (instanton) in momentum space. PACS numbers: 81.05.ue 02.30.Gp 71.70.Di 73.22.Pr I. INTRODUCTION One of the most prominent hallmarks of the existence of massless Dirac fermions in monolayer graphene was the experimental discovery of an unusual quantum Hall effect [1]. In other words, the observation of the Landau level (LL) pattern of the massless Dirac fermions, in the pres- ence of a perpendicular magnetic field B, En ∼ ±√Bn (n = 0, 1, 2,··· ), confirmed the characteristic band struc- ture of graphene, long after its theoretical prediction [2]. The quasiparticles in bilayer graphene (BLG) of Bernal stacking are massive and described by the following Hamiltonian. For a given valley, say K, in the low energy continuum limit, it is H 0 K = − 2 m (cid:18) 0 ∂2 ¯∂2 0 (cid:19) , (1) where m is the effective mass of the quasiparticle in BLG and we have introduced complex coordinates z ≡ (x + iy)/√2 on the graphene plane. The derivative operators are defined as ∂ ≡ ∂/∂z and ¯∂ ≡ ∂/∂ ¯z, respectively. The LL spectrum of BLG is thus different from that of monolayer graphene and is given by eigenvalues of the following Hamiltonian in a magnetic field B: K = −ω(cid:18) 0 a†2 0 (cid:19) , H B a2 (2) that is, by En = ±ωpn(n − 1) [3], where ω = eB/m is the cyclotron frequency. The lowering/raising operators a/a† satisfy the usual harmonic oscillator algebra and are given by suitable combinations of covariant derivatives representing the magnetic field B. ∗Electronic address: mychoi22, yhhyun, [email protected] When graphene layers stack together, the interlayer couplings significantly change the nature of quasiparti- cles, like the BLG case considered above. Surprisingly, massless Dirac fermions survive the stacking in multi- layer structure grown on SiC [4]. The main reason for this stunning effect is thought to be the decoupling of twisted layers [5],[6]-[14]. Persistence and properties of massless Dirac fermions at small twist angles are, how- ever, under some controversy. According to ab initio calculations [15], the decoupling occurs at any values of twist angle and massless Dirac fermions are essentially those of monolayer graphene. On the other hand, the tight-binding analysis [5] indicates a strong role played by the interlayer coupling, which considerably affects the nature of quasiparticles. These results are based on the band structure calculation in the absence of magnetic field. In a recent experiment [16], the issue of angle de- pendence of the LL's in the presence of magnetic field is addressed by combining scanning tunneling microscopy and LL scanning tunneling spectroscopy. They measured some critical angles at which two-fold degeneracy due to the presence of two massless Dirac fermions is lifted and where the massless Dirac fermion picture breaks down. Specifically, the degeneracy can be seen at angles above roughly θdeg ∼ 3◦ for a few low levels in the magnetic field of 10 T, and the critical value of twist angle be- low/above which massive/massless LL spectrum is shown is θc ≃ 1.16◦ for 10 T. An effective Hamiltonian to obtain the corresponding LL's is suggested recently, but only the zero-modes are analytically investigated [17]. In order to study the angle dependence we should have controls over the nonzero-modes. In this paper, we exactly solve the LL spectrum of the Hamiltonian proposed in [17] to give the concrete values of angles, which show very precise agreements with the measured values for the magnetic field 10 T [16]. Since the LL spectrum is obtained in analytic form, we can predict θc and θdeg for every nonzero pair of the LL's in the presence of arbitrary constant magnetic field (under a plausible assumption on the physical continuity of the LL spectrum as functions of twist angle about which we dis- cuss in Appendix A). Furthermore, our analytic result for LL's (18) smoothly interpolates between the spec- tra known before for massive/massless quasiparticles in BLG, and allows one to get systematic power series cor- rections for both of two sides of the spectrum. A natural question one can ask here is: "What is the symmetry restoration mechanism behind the transition between the massive/massless spectra?" Based on our exact results, we anticipate that the non-perturbative symmetry restoration mechanism is a tunneling (instan- ton) in the momentum (reciprocal) space [18]. The remaining parts of this paper are organized as fol- lows. In Sec. II, we briefly review the construction of the low-energy effective Hamiltonian for charged quasi- particles in twisted BLG [17] and solve the associated LL problem analytically by invoking a differential-equation representation of the eigenproblem. In Sec. III, we re- veal the distinction between two asymptotic regions in LL spectrum, obtained in Sec. II, considered as func- tions of twist angle. Finally, in Sec. IV, we wrap this paper up with a short summary and some discussions. An appendix is devoted to discuss the change in the LL spectrum driven by the Fermi speed renormalization as twist angle increases and to advocate the validity of the analysis made in Sec. III. II. EXACT LANDAU LEVELS IN TWISTED BILAYER GRAPHENE Twisted bilayer graphene is a structure specified by a rotational mismatch given by an angle θ with respect to the perfect Bernal (AB) stacked bilayer graphene. This twisted pattern is not difficult to find but can be seen on the surface of graphite, for an example. In low en- ergy limit, it shows a drastically different electronic band structure from that of the Bernal-stacked BLG. Its low energy quasiparticles are two massless Dirac fermions rather than one massive fermion, per each valley (K/−K) [5]. For a small θ, the apices of the associated Dirac cones are separated by ∆K = K − Kθ ≃ 4πθ/3√3 acc in reciprocal space, where acc ≃ 1.42 A is the distance be- tween two adjacent carbon atoms in the hexagonal lat- tice. A commensurate rotation with a periodic Moir´e pattern occurs at the angles θi: cos θi = 3i2 + 3i + 1 2 3i2 + 3i + 1 , i = 0, 1, 2,··· , (3) and the superlattice structure is specified by basis vectors t1 = ia1 +(i+1)a2 and t2 = −(i+1)a1 +(2i+1)a2, where a1, a2 are the Bravais lattice basis vectors in the single layer hexagonal lattice [5]. The lattice constant of the superlattice is given by t1 = t2 = acc √9i2 + 9i + 3 . 2 The reciprocal lattice is spanned by G1 = G2 = 4π 9i2 + 9i + 3 4π 9i2 + 9i + 3 [(3i + 1)a1 + a2] , [−(3i + 2)a1 + (3i + 1)a2] . (4) The first Brillouin zone for twisted BLG is depicted in Fig. 1(a) and the corresponding low energy band struc- ture in the K-valley is shown in Fig. 1(b). Electronic properties of twisted BLG and related systems are cur- rently under intensive debates [6]-[14]. FIG. 1: (Color online) (a) First Brillouin zone for twisted BLG. The first Brillouin zone of the upper layer (dashed hexagon) is rotated by an angle θ with respect to that of the lower layer (full hexagon). (b) Low energy band structure near the K-valley of twisted BLG. The dispersion relation for 2 k + ∆K 2 , (∆Kx + i∆Ky), and m(θ) 2 , then we quasiparticles is given by E(k, ¯k) = ± 2 where k ≡ 1√2 is θ-dependent effective mass. If we set κ ≡ k − ∆K get the massless behavior E ≈ ±√2vFκ near κ = 0, where vF ≡ ∆K √2 m(θ) (kx + iky), ∆K ≡ 1√2 is the renormalized Fermi speed. m(θ)k − ∆K In a recent paper [17], an effective Hamiltonian for low energy dynamics of twisted bilayer graphene is proposed. Neglecting commensuration effects between two layers, the Hamiltonian describing twisted BLG around the pair (K, Kθ) reads Htw(k) = (cid:18)HD(k + ∆K P H† ⊥ 2 ) P H⊥ HD(k − ∆K 2 )(cid:19) , (5) where HD is the Dirac Hamiltonian for monolayer graphene and H⊥ is the hopping matrix between the two layers. According to the analysis of the Moir´e pattern in twisted BLG [5], for small twist angle θ, there are three different types of H⊥: H⊥ ∈ (cid:26)−t⊥(cid:18)1 1 1 1(cid:19) ,−t⊥(cid:18)e∓i 2π e±i 2π 3 (cid:19)(cid:27) , e∓i 2π (6) 1 3 3 where t⊥ is a coupling parameter which generally de- pends on θ. The θ-dependence of t⊥ is, however, very mild and the Slater-Koster calculation performed in Ref. [5] indicates that t⊥ ≃ 0.4 γ1 is nearly a constant, where γ1 ≃ 0.3 eV is the nearest interlayer coupling in the untwisted Bernal-stacked BLG. The first one in (6) 3 corresponds to the Fourier component with the crystal momentum G = 0 and the second and the third one with G = −G1,−(G1 + G2). The summation in the in- terlayer coupling term in (5) runs over these three types of the coupling matrices and the other components are suppressed. Due to this interlayer coupling, there is a slight difference of 1 meV order between the energies as- sociated with each Dirac points but this second-order ef- fect in perturbation is negligible. Assuming a simplified interlayer coupling under the condition t⊥ ≫ vF∆K (vF is the Fermi speed in the monolayer graphene) [24], which satisfy [a, a] = [a†, a†] = 0 and [a, a†] = 1. β is a complex parameter proportional to ∆K as β ≡ r 4eB ∆K = r 8eB (∆Kx + i∆Ky) . (12) We can make β real-valued simply by rotating the coor- dinates, β = r 8eB ∆K . (13) X H⊥ → −3 × 5 t⊥ 2 (cid:18)0 1 0 0(cid:19) , (7) After the rotation, for ψ = (ψ1 ψ2)T , the eigenvalue prob- lem for the Hamiltonian (10), HBψ = E ψ, reduces to the following one-component problem: one can obtain an effective two-band Hamiltonian which resembles (1) H eff tw = − 2 m(θ) (cid:18) 0 ¯∂2 − (∆K/2)2 ∂2 − (∆K/2)2 0 (cid:19) , (8) (cid:0)a†2 − β2(cid:1) (a2 − β2)ψ1 = λψ1 , (14) where λ = [E/ω(θ)]2. The remaining component ψ2 can be expressed in terms of ψ1, λ, and the lowering operator a (except the case λ = 0). Using the anti-holomorphic representation [19], where m(θ) = 15 t⊥/4 v2 F is θ-dependent effective mass due to the θ-dependent t⊥. In (7) the multiplication by 3 mimics the summation over three possible crystal momentum G, and another factor 5/2 is introduced to match the spectrum at θ = 0 (Bernal-stacked BLG). For θ = 0, the period of the superlattice is infinite and the summation over G is overcounting since G1 = G2 = 0. Thus the multiplication by 3 in (7) should be disregarded in this case and the interlayer coupling becomes that of the Bernal-stacked BLG: 5 t⊥ 2 (cid:18)0 1 0 0(cid:19) ≃ (cid:18)0 −γ1 H⊥(θ = 0) = − 0 (cid:19) . (9) 0 Then the effective Hamiltonian (8) goes to (1). Since ∆K is proportional to twist angle θ, the two-band ap- proximation made here is restricted to be applicable for very small angles only. Nevertheless, we believe that the LL spectrum given by (18) below is smoothly connected to the spectrum for larger angles, and the analysis made in Sec. III is trustable. The multiplicative factor in (7) also plays a crucial role in this context. We relegate the discussion of this issue to the appendix of this paper. In the presence of a perpendicular magnetic field, B, the Hamiltonian is written in the form HB = −ω(θ)(cid:18) 0 a2 − β2 a†2 − ¯β2 (cid:19) , (10) where ω(θ) ≡ eB/m(θ). The lowering/raising operators a/a† are given in terms of the covariant derivatives D ≡ ∂ + ie (Ax− iAy) is a complex vector potential. With the gauge choice A = − i A(z, ¯z) / ¯D ≡ ¯∂ + ie ¯A(z, ¯z), and A ≡ 1√2 2 B ¯z, we specifically have a† 7→ x , a 7→ d dx , (15) the eigenvalue problem (14) is expressed as a second order ordinary differential equation, (x2 − β2)(u′′ − β2u) − λu = 0 . (16) Here, the variable x is not the coordinate in the graphene plane and therefore the function u(x) representing eigen- states are not the wave function in coordinate space. By get the spheroidal equation of p = β2 and q = 1 [(x2 − 1)v′]′ +(cid:20)−p2(x2 − 1) − λ − rescaling x 7→ βx and setting u(x) = √x2 − 1 v(x), we x2 − 1(cid:21) v = 0 (17) which is a particular case b = s = 0 of the confluent Heun's equation: [(x2 − 1)v′]′ + [−p2(x2 − 1) + 2pbx− λ− x2−1 The eigenvalues of the confluent Heun's equations are λ(a) q,s,n−1(p, b), and then the corresponding LL's are given by the spheroidal eigenvalues ]v = 0 [20]. q2+s2+2qsx q2 E2 n = (ω(θ))2λ(a) 1,0,n−1(β2, 0) , (18) where the superscript (a) stands for 'angular'. The LL spectrum for a fixed value of β (twist angle) is depicted in Fig. 2. This analytic result reproduces the numerical calculation performed in Ref. [17], except the scale of B- dependence due to the multiplicative factor in (7). III. ANALYSIS OF SPECTRUM IN TWO ASYMPTOTIC REGIONS a = r eB ¯D, a† = −r eB D , (11) Let us consider the region of the parameter β in which the band structure is well-described by two massless Small β expansion [21] for the eigenvalues (18) gives, 4 E2 n (ω(θ))2 ≈ n(n − 1) − 2n(n − 1) (2n + 1)(2n − 3) β4 + 2n(n − 1)[n(n − 1)(4n(n − 1) − 39) + 63] (2n + 3)(2n + 1)3(2n − 3)3(2n − 5) β8 (19) + O(β12) . FIG. 2: (Color online) Landau levels of twisted bilayer graphene, given by the spheroidal eigenvalues. The character- istic energy scale of the van-Hove singularity has been chosen as ωβ2 = 2∆K2/8m = 0.1 eV. This scale corresponds to the twist angle of θ ≃ 3.27◦. Dirac fermions whose Fermi speed is renormalized to be smaller than that of the monolayer graphene. In this region, the LL's must show a two-fold degeneracy that reflects the existence of two copies of fermions. Fig. 2 reveals its tail, and Fig. 3 clearly indicates that degen- eracy. The lifting of this degeneracy signals breakdown of the description of electronic bands by two copies of massless fermions and it is caused by the contribution from a saddle point in the band located between the two Dirac cones. Twist-induced van-Hove singularity eventu- ally dominates the spectrum. Untwisted (β = 0) LL spectrum is given by the leading term in (19). The common factor n(n − 1) appears in all the higher order terms in (19) and the modes labeled by n = 0 and n = 1 remain to be zero-energy modes even if the corrections in higher powers of β are consid- ered. Therefore they are protected. Notice that we are neglecting any other potential lifts of degeneracy due to Zeeman effect, interactions, etc. This protection is due to the topological structure of the band in twisted BLG [17]. In twisted BLG, the two Dirac cones are not re- lated by time-reversal symmetry and they are described by the same Berry phase, so that they are of different topological structure from the Dirac cones in monolayer graphene. For sufficiently large β, the asymptotic expansion of the eigenvalues [21] gives E2 n (ω(θ))2 ≈ 4h n 2i2 2i β2 − 2h n −h n 2i3 1 β2 +··· , (20) where the square bracket denotes the integer part of n/2. The presence of this integer-valued function implies that there are asymptotic degeneracies between the each even level and the next odd one. The difference between the 2ℓ-th and (2ℓ + 1)-th LL's decreases exponentially and is vanishing at infinite β: 2ℓ+1 − E2 E2 (ω(θ))2 ≈ 2ℓ 24ℓ+3β4ℓ+2 (ℓ − 1)! ℓ! e−2β2 + ··· . (21) The exponential factor is independent of ℓ but the power of the monomial function is proportional to ℓ so that, in physics experiments, the degeneracy between the ad- jacent two levels of lower ℓ (or n) is lifted for smaller β (smaller θ), which is consistent with Fig. 3. From now on, we will set the external magnetic field B = 10 T, perpen- dicular to the plane of BLG. Since the value of β is large enough up to a very small value of θ (β ∼ 1 for θ ∼ 1.17◦), we can use the asymptotic formula (21) in order to esti- mate the point on which the two-fold degeneracy is lifted, under the assumption that the θ-dependence of ω(θ) (or that of t⊥) is mild enough. The values βdeg for ℓ = 1 where the n = 2 and n = 3 levels become non-degenerate and for ℓ = 2 (n = 4, 5) are, from (21), FIG. 3: (Color online) Landau level energies divided by θ- dependent energy scale ω(θ), for a fixed perpendicular mag- netic field B = 10 T, plotted against twist angle. If the θ- dependence of ω(θ) is mild enough, the qualitative behavior of the LL's is the same as E/ω(θ). and thus 24ℓ+3β4ℓ+2 deg (ℓ − 1)! ℓ! e−2β2 deg ≃ 1 , β(ℓ=1) deg ≃ 2.18 and β(ℓ=2) deg ≃ 2.99 , (22) (23) respectively. They correspond to the twisted angles θdeg ≃ 2.56◦ and ≃ 3.50◦ (see the Fig. 3), in agreement with the measured value, about 3◦, from Ref. [16]. As the twist angle θ decreases, β crosses the transi- tion point βc = 1 between the region where the small β expansion (19) can be trusted and the region where the large β expansion (20) is trustworthy. We will call the region β > βc massless region and β < βc massive region, respectively, because of an obvious reason from the behaviors of the LL's in each domain, (19) and (20). The critical point βc = 1 can roughly be considered as the point at which the van-Hove singularity eventually starts dominating the spectrum and the description of the band by two massless Dirac fermions breaks down. This critical value βc = 1 corresponds to the twist angle θc ≃ 1.17◦ as mentioned already. The measured value θ(measured) ≃ 1.16◦ [16] is very close to our theoretical c value, even though there is no exact criterion of fixing the value due to the continuity of the spectrum. Currently obtainable maximum value of a static mag- netic field is about 40 T. The critical value βc = 1 and the value of βdeg for each ℓ in (22) are independent of B and thus the only effect on the values of various an- gles for the 40 T magnetic field is a multiplication by the factor 2 = p40/10 from (13) to the angles for 10 T. Therefore θc ≃ 2.34◦ and the angles above which two-fold degeneracy can be seen are as tabulated below, in a high magnetic field B = 40 T. ℓ 1 ··· θdeg(◦) 5.11 7.01 8.42 ··· 2 3 This simple dependence of the specific angles θdeg and θc on the magnetic field B, that is, θdeg,c(B) = r B B0 θdeg,c(B0) , (24) where B0 means a reference value of magnetic field, say, B0 = 10 T allows us to draw the Fig. 4 to read the angles off for various values of B. IV. SUMMARY AND DISCUSSIONS To summarize, we solved the eigenvalue problem for LL's in twisted BLG with small twist angles and ana- lyzed the angle dependence of the spectrum in a fixed magnetic field. The corresponding eigenvalues (18) are given by the so-called spheroidal eigenvalues modulated by an unknown function ω(θ) of twist angle θ, but the modulation does not change the spectrum seriously. As the results, we got the angle θc below/above which the spectrum behaves as LL's for massive/massless fermions and also the angles θdeg below which the two-fold degen- eracies due to the Dirac-point splitting are lifted. The specific values for {θc, θdeg} for B < 50 T can be read off from the Fig. 4. They can be measured in principle by a 5 FIG. 4: (Color online) The dependence on B of the various specific angles. judicious experiment which combines the scanning tun- neling microscopy and the LL scanning tunneling spec- troscopy. The differences between each adjacent energy levels (21) show non-perturbative behavior in the expansion parameter, 1/β2. Indeed, the exponential factor in (21) gives us a hint about the symmetry restoration mecha- nism behind the transition described above. It is a typical signal of tunneling mechanism (instanton effect), in this case not in coordinate space but in momentum space [18]. The situation is analogous to the well-known double-well potential problem in quantum mechanics. The two Dirac points correspond to classically degenerate ground states, and the van-Hove energy scale plays the role of the height of potential barrier between them. By the tunneling, the Dirac electrons are delocalized in momentum space, thus become localized in coordinate space. The deformation of the band structure of BLG trig- gered by twisting is the Dirac-point splitting, as depicted in Fig. 1(b). The Dirac-point splitting also happens in untwisted Bernal-stacked BLG by the effect of external magnetic field which is parallel to the graphene plane [22, 23]. Therefore, in a tilted magnetic field with re- spect to the graphene plane, the LL spectrum of un- twisted BLG is expected to be the same as (18), if the parameter β ∝ Bk/√B⊥ represents the effect of the par- allel component Bk to the spectrum [23]. In principle, the magnetic field component parallel to the plane, which is easy to get by tilting the graphene sample in an external magnetic field, can split the Dirac point and make the LL spectrum doubly-degenerate. It is intriguing to in- vestigate the combined effect of twisting and inclination in a magnetic field, since in reality graphene samples al- ways deviate from the perfect plane by various physical reasons. Acknowledgments The authors benefited from useful discussions with C. Sochichiu. This work was supported by the Korea Re- search Foundation Grant funded by the Korean Gov- ernment with grant number KRF-2008-313-C00170 and 2011-0011660 (Y.K.). Appendix A: Renormalized Fermi speed and angle dependence The authors of Ref. [17] managed to get the two-band effective Hamiltonian (8) by assuming a simplified form of the interlayer coupling (7) under the condition that ξ ≡ vF∆K/t⊥ ≪ 1, as we reviewed in Sec. II. The reduction from the full Hamiltonian (5) to the effective one (8) is reminiscent of that of the Bernal-sracked BLG case. In fact, in addition to the simplified form assumed by the authors of Ref. [17], we need a numerical multi- plication factor in (7) in order to connect smoothly the spectrum and the Fermi speed to those for larger angles (ξ ≫ 1) obtained in Ref. [5] as we shall see below. The condition of small ξ, ξ ≪ 1, used to get the two-band effective Hamiltonian (8), can be understood and refined as follows. The minimum energy of the up- per band of the full Hamiltonian, approximately Emin ≃ 15 t⊥/2, should be much larger than the van-Hove en- ergy of the lower band EvH = 2∆K2/8m(θ) = F∆K2/30 t⊥, so that they cannot feel the existence 2v2 of each other (Fig. 5 (a)). This refined condition that t⊥ ≫ 1 vF∆K (ξ ≪ 15) yields the restriction on the range of angles 15 45√3 acc 4πvF ≃ 9.80◦ , θ ≪ t⊥ (A1) but this does not tell us sharply how small the twist angle θ should be. For example, a blind application of our formula to θ = 8◦ gives a divergent result for a small but finite B, say, B = 0.1 T. Then, how can we trust the analysis presented in Sec. III? Some of the predicted values of θc,deg might be in the region where the spectrum is not quite close to that given by (18). Let us recall that for sufficiently large angles, the spec- trum given by (18) shows the characteristic LL behavior of massless particles. The combination of twist angle θ ∼ 5◦ and magnetic field B = 10 T for instance corre- sponds to β ∼ 4.26, in the "massless region". Moreover, θ ∼ 5◦ and βc = 1 require B ∼ 182 T, a very large field strength. This simple observation means that the LL spectrum of twisted BLG (18) with θ ∼ 5◦ cannot be pushed to "massive region", unless a very strong mag- netic field is applied. Let us explain this point in more detail. As the twist angle increases the van-Hove energy, playing the role of a barrier between the two Dirac points, also increases as can be seen in Fig. 5 (b). Therefore the "tunneling" between the two Dirac points is suppressed 6 FIG. 5: (Color online) Schematic plots for the (upper half) band structure of the full Hamiltonian (5) along the ∆K- direction. (a) The case in which the minimum energy of the upper band is larger than the van-Hove energy. The two-band approximation can be applied without any difficulty. (b) The opposite case. The condition to derive the effective two-band Hamiltonian is not satisfied any more. and the Z2 symmetry is broken. Unless a very strong magnetic field is applied, the broken symmetry cannot be restored and the spectrum is described by two mass- less fermions degenerate in energy, for θ ∼ 5◦. This fact is clearly shown in Fig. 6, plotted for θ = 5◦. FIG. 6: (Color online) The LL spectrum (18) for θ = 5◦. Each adjacent levels are degenerate and in the "massless" region if the applied magnetic field is not very high. Indeed, the masslessness of quasiparticles near one of the two Dirac points was shown in Ref. [5] by applying perturbation theory under the opposite condition, that ξ ≫ 1. Their result indicates the Fermi speed renormal- ization vF/vF = 1 − 9/ξ2 , (A2) where vF is the renormalized Fermi speed. This result also seems strange since it tells us that the renormalized Fermi speed vanishes at ξ = 3, corresponding to θ ≃ 2◦, and becomes negative below that angle though it should be a positive quantity by definition. Therefore the spec- trum obtained in Ref. [5] seems valid at most in the region ξ ≫ 3. At any rate, the LL spectrum according to the Fermi speed renormalization shown in Ref. [5] is given by Eℓ = ±vF(cid:18)1 − 9 2v2 t2 ⊥ F∆K2(cid:19)√2eBℓ , (ℓ = 0, 1, 2,··· ) , (A3) that is, the LL for massless fermions modulated by the renormalized Fermi speed. Note, however, that the validity regions for the spectra (18) and (A3) can have some overlap for 3 ≪ ξ ≪ 15. Actually, as we remarked above, both the spectra at an intermediate angle such as θ = 5◦ are of massless charac- ter. Except the modulation function ω(θ) depending on the renormalized Fermi speed, they are the same unless the applied magnetic field is very very large. The renor- malized Fermi speed according to (8) and (20) (see also the caption of Fig. 1) is, to the first order in ξ/15, linear in ξ: vF vF = 2vF∆K 15 t⊥ = 2 15 ξ . (A4) All these circumstances make it plausible that the spec- tra (18) and (A3) are smoothly connected in the inter- mediate range of ξ (or θ) and, here comes the punchline, the curves representing the renormalized Fermi speed for these two cases are almost tangent to each other, at the angle θ ≃ 3.37◦ (ξ ≃ 5.20). The linear function f (ξ) which is exactly tangent to the curve vF/vF = 1− 9/ξ2 is f (ξ) = 2ξ/9√3. See the inset in Fig. 7. If the renormal- izaion of the Fermi speed is given by the curve depicted here, the LL's smoothly connecting (18) and (A3) should behave qualitatively as Fig. 7. The Fermi speed renor- maliztion affects the spectrum for angles θ & 3.37◦ to change the shape of its tail. Recall that the values of θc,deg predicted in Sec. III are independent of the (renormalized) Fermi speed since their predictions are based only on the spheroidal eigen- values themselves, i.e., En/ω. Therefore, if the discus- sion made in this appendix based on physical continuity of the spectrum is correct, the predicted values can re- main trustable. In other words, if one can find an exact formula for renormalized Fermi speed as a function of θ which interpolates the two asymptotic forms (A2) and (A4) (equivalently the modulation function ω(θ)), the exact LL spectrum is given by the spheroidal eigenval- ues modulated by it, (18), at any values of θ. Thus it is 7 extremely interesting to find such a exact interpolating function for the renormalized Fermi speed. A final remark is in order. Since the validity of the asymptotic form (A2) is uncertain in the intermediate region, the slope in the other asymptotic form (A4) is also uncertain accordingly. Still the inclusion of the factor 3 in (7) seems crucial, because otherwise the slope in (A4) must be modified into a too large number to have a chance of smooth interpolation. The other numerical FIG. 7: (Color online) The qualitative behavior of the LL's (B = 10 T) as functions of θ, modulated by the renormalized Fermi speed. Inset: expected behavior of the renormalized Fermi speed, interpolating (A2) and (A4) (the solid lines). factor 5/2 in (7) is the actual source of uncertainty, and it can be replaced by a number in some range -- roughly from 2 to 2.5. If we adopted the simplified interlayer coupling 0 0(cid:19) X H⊥ → −3 × 2t⊥(cid:18)0 1 (A5) instead of (7), the slope of (A4) becomes 1/6 which is a reasonable number to make the interpolation. Actually, the interlayer coupling matrix H⊥ = −2t⊥(cid:18)0 1 0 0(cid:19) ap- proximates each of the coupling terms in (6) much more closely, as one can see by performing the diagonalization of (5) after turning off the block diagonal Dirac Hamilto- nians. The coupling constant t⊥ should be close to γ1/2 in this case. [1] K. S. Novoselov, A. K. Geim, S. V. Morosov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos and A. A. Firsov, Nature 438, 197 (2005); Y. Zhang, Y.-W. Tan, H. L. Stormer and P. Kim, Nature 438, 201 (2005). [2] P. R. Wallace, Phys. Rev. 71 622 (1947). [3] E. McCann and V. I. Fal'ko, Phys. Rev. Lett. 96, 086805 (2006). [4] M. L. Sadowski, Phys. Rev. Lett. 97, 266405 (2006); M. Sprinkle et al., Phys. Rev. Lett. 103, 226803 (2009); D. L. Miller et al., Science 324, 924 (2009). 8 [5] J. M. B. Lopes dos Santos, N. M. R. Peres and A. H. arXiv:1103.3172[cond-mat]. Castro Neto, Phys. Rev. Lett. 99, 256802 (2007). [6] J. Haas et al., Phys. Rev. Lett. 100, 125504 (2008). [7] Guohong Li et al., Nature Physics 6, 109 (2010). [8] S. Shallcross, S. Sharma, E. Kandelaki and O. A. Pankra- tov, Phys. Rev. B 81, 165105 (2010). [9] R. Bistritzer and A. H. MacDonald, arXiv:1009.4203. [10] V. M. Apalkov and T. Chakraborty, arXiv:1102.5660. [11] G. Trambly de Laissardi`ere, D. Mayou and L. Magaud, Nano Lett. 10, 804 (2010). [12] E. J. Mele, Phys. Rev. B 81, 161405 (2010). [13] E. Su´arez Morell et al., Phys. Rev. B 82, 121407 (2010). [14] S. Shallcross et al., Phys. Rev. B 83, 153402 (2011). [15] S. Shallcross, S. Sharma and O. A. Pankratov, Phys. Rev. [18] R. L. Jaffe, Am. J. Phys. 78, 620 (2010). [19] L. D. Faddeev and A. A. Slavnov, Gauge Fields: An In- troduction to Quantum Theory, 2nd ed. (Westview Press, 1993). [20] A. Ronveaux, Heun's Differential Equations (Oxford Uni- versity Press, Oxford Oxfordshire, 1995). [21] I. Komarov, L. Ponomarev, and S. Slavyanov, Sfer- (Nauka, oidal'nye i kulonovskie sferoidal'nye funkcii 1979), in Russian; E. W. Weisstein, Spheroidal Wave Function, MathWorld-A Wolfram Web Resource, URL http://mathworld.wolfram.com/SpheroidalWaveFunction.html. [22] S. S. Pershoguba and V. M. Yakovenko, Phys. Rev. B 82, 205408 (2010). Lett. 101, 056803 (2008). [23] Y. Hyun, Y. Kim, C. Sochichiu, and M. Choi, preprint, [16] A. Luican, Guohong Li, A. Reina, J. Kong, R. R. Nair, K. S. Novoselov, A. K. Geim and E. Y. Andrei, Phys. Rev. Lett. 106, 126802 (2011). [17] R. de Gail, M. O. Goerbig, F. Guinea, G. Mon- tambaux and A. H. Catro Neto, "Topologically Pro- tected Zero Modes in Twisted Bilayer Graphene", arXiv:1008.0488. [24] This subtle assumption may be justified by the fact that the Hamiltonian remains in the same topological univer- sality class [17].
1805.09566	1	1805	2018-05-24T09:24:34	Beating pattern in radiation-induced oscillatory magnetoresistance in 2DES: coupling of plasmon-like and acoustic phonon modes	[ "cond-mat.mes-hall" ]	We present a microscopic theory on the observation of a beating pattern in the radiation-induced magnetoresistance oscillations at very low magnetic field. We con- sider that such a beating pattern develops as a result of the coupling between two oscillatory components: the first is a system of electron Landau states being harmon- ically driven by radiation. The second is a lattice oscillation, i.e., an acoustic phonon mode. We analyze the dependence of the beating pattern on temperature, radiation frequency and power. We conclude that the beating pattern is an evidence of the radiation-driven nature of the irradiated Landau states that makes them behave as a collective plasma oscillation at the radiation frequency. Thus, the frequency of such plasmons could be tuned from microwave to terahertz in the same nanodevice with an apparent technological application.	cond-mat.mes-hall	cond-mat	a Beating pattern in radiation-induced oscillatory magnetoresistance in 2DES: coupling of plasmon-like and acoustic phonon modes. J. Inarrea1,2 1Escuela Polit´ecnica Superior,Universidad Carlos III,Leganes,Madrid,28911,Spain 2Unidad Asociada al Instituto de Ciencia de Materiales, CSIC, Cantoblanco,Madrid,28049,Spain. (Dated: September 3, 2018) We present a microscopic theory on the observation of a beating pattern in the radiation-induced magnetoresistance oscillations at very low magnetic field. We con- sider that such a beating pattern develops as a result of the coupling between two oscillatory components: the first is a system of electron Landau states being harmon- ically driven by radiation. The second is a lattice oscillation, i.e., an acoustic phonon mode. We analyze the dependence of the beating pattern on temperature, radiation frequency and power. We conclude that the beating pattern is an evidence of the radiation-driven nature of the irradiated Landau states that makes them behave as a collective plasma oscillation at the radiation frequency. Thus, the frequency of such plasmons could be tuned from microwave to terahertz in the same nanodevice with an apparent technological application. PACS numbers: Beating patterns show up when there are two oscillatory contributions coexisting and coupled in the same physical system. For instance, beating patterns can be observed in magnetoresistance (Rxx) of two-dimensional electron systems (2DES) when there are two populated conduction electron subbands involved in the transport. This situation gives rise to the well-known magneto-intersubband scattering oscillations (MISO)1,2. They can also be obtained in Rxx of 2DES with strong Rashba spin-orbit coupling3,4. For both cases the two oscillatory subsystems are the two sets of broadened Landau levels of slightly different energies. On the other hand, two important physical effects in magnetotransport of 2DES 2 ) ( X X R 4 3 2 1 4 3 2 1 (a) 0.01 (b) f=105 GHz T=0.5 K 0.02 0.03 B(T) 0.04 0.05 6 8 10 12 14 16 18 20 22 24 w/wC FIG. 1: Calculated Rxx for a radiation frequency of 105 GHz. In panel a) Rxx vs B and in b) Rxx vs w/wc. Apart from the usual RIRO we observe a beat at low B with a node around B = 0.02T in panel a). In panel b) we observe a change in phase difference of π when crossing the node. were discovered more than a decade ago by Mani et al.,5: the radiation-induced magnetore- sistance oscillations (RIRO) and the even more striking of zero resistance states (ZRS)5,6. To date, there is not a clear consensus on the physical origin of such remarkable effects. After a huge number of experiments7 -- 23 and proposed theories24? -- 47 to explain them, we have to admit that they are still under debate. Soon after the discovery of RIRO and ZRS, another surprising experimental result re- garding RIRO was published49. It consisted in an unexpected beating pattern at very low B superimposed to RIRO. As with RIRO this beating pattern was radiation-induced. This subtle effect was overlooked by the scientific community and very little attention was paid. Nevertheless, there has recently been shown new experimental evidence presenting a similar beating pattern profile on RIRO at very low B50 too. As explained above, this radiation- induced beating patter indicates the presence of two comparable oscillatory contributions. However, possible physical scenarios giving rise to beating patterns, such as two populated electron subbands or Rashba spin-orbit coupling, cannot easily explain the obtained exper- imental results49,50. In this letter we develop a microscopic theory to explain the beating pattern in RIRO 3 ) ( X X R 9 8 7 6 5 4 3 2 1 30 GHz 40 GHz 58 GHz 75 GHz 90 GHz 105 GHz 0.01 0.02 0.03 B(T) 0.04 0.05 FIG. 2: Dependence of the beating pattern on the radiation frequency ranging from 30 GHz to 105 GHz. The node position does not change irrespective of the radiation frequency. based, on the one hand, on the physical effect of plasmon-phonon coupling in polar semiconductors51 -- 53. According to it, in polar semiconductors like GaAs, collective oscil- lations of electrical charges (plasmons) and lattice ions oscillations (phonons) can couple via Coulomb interaction. As a result, the initially individual (plasmon and phonon) modes re- adjust their frequencies to give rise to new hybrid plasmon-phonon coupled modes. On the other hand, our microscopic theory is based on the previous model for RIRO, the radiation- driven electron orbit model. This model, in turn, is based on the exact solution of the electronic wave function in the presence of a static magnetic field and radiation. In this model the electrons orbits or Landau States (LS) move back and forth harmonically driven by radiation, (driven-LS), at the radiation frequency (w). Thus, the guiding centers of the LS perform harmonic and classical trajectories making the system of driven-LS behave like a collective oscillation of electric charge, i.e. a plasmon-like mode. Now, this plasmon-like mode, with acoustic frequency, can couple with a collective lattice ions oscillation of simi- lar amplitude and frequency, an acoustic phonon mode. Thus, we can observe the rise of a beating pattern, for instance, in Rxx. Therefore, the observation of beats in the RIRO profile would be a clear evidence of the spatial swinging nature of the irradiated LS. This provides a source of excitation of acoustic plasmon-like modes in 2DES with a frequency ranging from the microwave (MW) to the terahertz (THz) part of the spectrum. 4 (a) T=1.0 K 0.9 K 0.8 K 0.7 K 0.6 K 3.0 0.5 K 2.5 0.01 (b) ) ( X X R 3.0 2.5 f=40 GHz 0.02 0.03 B(T) 0.04 T=1.0 K 0.9 K 0.8 K 0.7 K 0.6 K 0.5 K 2 3 4 5 6 7 w/wC 8 9 10 FIG. 3: Dependence of the beating pattern on temperature. In the upper panel we exhibit irra- diated Rxx vs B and in the lower panel the same vs w/wc. In the lower panel the node position moves to lower w/wc for decreasing T . In the upper panel we observe the opposite trend, the node moves to higher B for decreasing T . As we said above, the radiation-driven electron orbits model32,33 was developed to explain the striking effects of RIRO and ZRS. One of the main conclusion of this theory is that under radiation the LS oscillate, with their guiding centers, at the radiation frequency according to same without radiation and A = X(t) = X0 +A sin wt, where X(t) is the time dependent LS guiding center position, X0 is the −w2)2+γ 4 where, in turn, E0 is the radiation electric field and wc the cyclotron frequency. γ is a phenomenologically introduced damping factor m∗√(w2 eEo c for the electron scattering with the lattice ions. Following the physics of plasmon-phonon coupling we consider that the system of driven-LS, behaving like a "plasmon-like" mode, can couple with an acoustic phonon mode of similar frequency and amplitude. We derive classically and similarly to a system of coupled harmonic oscillators, the new frequencies of the hybrid modes54 and the guiding center position of the driven-LS mode. Firstly, and after some algebra, these frequencies are given by55 -- 57: 2w2 ± = (w2 + w2 ac) ±p(w2 − w2 ac) + 16λ2wacw (1) where w is the frequency of both, radiation and the driven-LS mode. wac is the frequency of the acoustic phonon mode and λ the plasmon-phonon coupling constant. If w is only slightly 5 different from wac, i.e., w ≃ wac, then it is straightforward to finally obtain that w± ≃ w±λ, where λ ≪ w. Secondly, the position of the guiding center of the hybrid driven-LS mode is now57 X(t) = X0 + A sin w+t + B sin w−t. With similar algebra as before, we introduce the damping that the hybrid driven-LS mode undergoes due to scattering with the lattice ions. The obtained expression for w± and X(t) are now: 2w2 ± = (w2 1 + w2 2) ±q(w2 1 − w2 2) + 16λ2w2w1 X(t) = X0 + e− γ 2 t[A sin w+t + B sin w−t] (2) (3) where w2 1 = w2 − γ2 4 ≃ w2 and w2 2 = w2 ac − γ2 4 ≃ w2 ac, considering that γ2 ≪ w2, w2 ac. With the new obtained expression for X(t) (Eq. 3) and according to the radiation-driven electron orbit model, we obtain for the average distance advanced by the electron in a scattering event, ∆X(t) = ∆X0 − e− γ 2 τ A[sin w+τ + sin w−τ ] = ∆X0 − 2Ae− γ 2 τ sin wτ cos λτ (4) where we have considered that the amplitudes of both modes are similar, i.e., A ≃ B and that w± ≃ w ± λ. The time, τ , according to the radiation driven electron orbit model58,59, is the "flight time", the time it takes the electron to jump due to scattering from one orbit to another and its value is given by τ = 2π wc . Following the same RIRO model and using the obtained ∆X(t), we end up with an expression for Rxx 58,59: Rxx ∝ ∆X0 − 2Ae−π γ wc sin(cid:18)2π w wc(cid:19) cos(cid:18)2π λ wc(cid:19) (5) where we want to stand out the essential part that explains the appearance of the beating pattern in RIRO. In Fig. 1 we exhibit calculated Rxx for a radiation frequency of 105 GHz. In Fig. 1a, Rxx vs B and in Fig. 1b, Rxx vs w/wc. Apart from the usual RIRO we observe a beat at very low B with a node around B = 0.02T in the upper panel. In the lower panel the vertical lines for integer values of the abscissa indicate a phase change of π in the Rxx oscillations when crossing the node. Eq. 5 readily explains the rise of a beat when w+ is just slightly different from w− or in other words, when λ ≪ w; the change of phase in π is due to the modulation of the slower function, i.e., the cosine function. In Fig. 2 we present the dependence of the 6 f=40 GHz 4.0 (a) P=2mW 3.5 3.0 1.5mW 1.3mW 1.1mW 0.9mW 2.5 0.7mW ) ( X X R 4.0 3.5 3.0 2.5 2.0 0.01 (b) 0.02 B(T) 0.03 0.04 P=2mW 1.5mW 1.3mW 1.1mW 0.9mW 0.7mW 2 3 4 5 6 7 w/wC 8 9 10 11 FIG. 4: Same as in Fig. 3 but in function of radiation power. beating pattern on the frequency, ranging from 30 GHz to 105 GHz. We observe a constant B-position for the node irrespective of the frequency. We find again the explanation in Eq. 5. We can tell that the node position depends on the cosine function where w does not show up. Thus, the node position is immune to w. However any variation of the coupling constant λ that shows up in the cosine will clearly affect the node position and even the number of beats that can be observed. In Fig. 3 we present the calculated results of the dependence of the beating pattern on temperature (T ) for a radiation frequency of 40 GHz and T from 0.5K to 1.0K. In Fig. 3a, we exhibit Rxx vs B and in Fig. 3b Rxx vs w/wc. For both panels the curves are shifted for clarity. Interestingly enough, as exhibited in the lower panel, the node position is not constant and moves to lower w/wc for decreasing T , and in the same way the beat gets more intense. In the upper panel we observe the opposite trend, the node moves to higher B for decreasing T but the intensity increase keeps the same as in the lower one. The displacement of the node and the intensity variation indicate that a changing temperature affects the driven-LS-phonon coupling and, in turn, λ. In Fig. 4 we study the beating pattern in function of the radiation power (P ). P runs from 0.7 mW to 2 mW. Similarly as in Fig.3, in panel a) we exhibit irradiated Rxx vs B and in panel b) the same vs w/wc. As with T , we observe a similar node displacement and beat intensity variation for decreasing P , proving that λ depends on P too. Thus, both quantities, T and P , affect the beat in 7 f=105GHz (a) P=1mW T=0.5K P=0.1mW T=0.1K 0.01 0.02 0.03 B(T) 0.04 0.05 0.06 (b) P=1mW T=0.5K 2.8 2.6 2.4 ) ( X X R 2.8 P=0.1mW T=0.1K 2.6 4 6 8 10 12 14 16 18 20 22 24 26 28 30 w/wC FIG. 5: Dependence of the beating pattern on radiation power and temperaature for a radiation frequency of 105 GHz. In the upper panel we exhibit irradiated Rxx vs B and the lower one Rxx vs w/wc. As expected, we observe an extra beat and node when going from P = 1mW and T = 0.5K to P = 0.1mW and T = 0.1K. the same way: a decrease of any of them makes the coupling stronger and λ bigger. And, on the other hand, an increase gives rise to a progressive destruction of the beat. The physical explanation can be readily obtained from our model. A higher T triggers a more intense scattering between the electrons in their orbits and the lattice ions. In the case of an increasing P , the amplitude of the driven-LS oscillations gets also bigger, and in turn, the probability for the electrons in their orbits to be scattered is also higher too. Thus, for both increasing quantities we obtain a similar damping effect on the driven-LS-acoustic phonon coupling that gets progressively destroyed. To study the dependence of λ, i.e., of the beating pattern on T and P , we have developed a phenomenological equation consisting. in adding a first order (linear) correction to λ in the variation of T and P . Thus, for T , λ = λ0 − λ1(T − T0) where λ0 and λ1 are constants and T0 = 0.5K in agreement with experimental values50. And a similar equation for P , λ = λ0 − λ1(P − P0) and P0 = 0.7mW . The calculated results are in qualitatively good agreement with experiments50. In Fig. 5 we exhibit the calculated results when simultaneously changing T and P . Ac- 8 cording to our model, for instance in the case of decreasing, this will lead to a strengthening of the beating pattern effect. Thus, we would expect the rise of more nodes and beats. This is presented in Fig. 5 for a radiation frequency of 105 GHz; in the upper panel Rxx vs B and the lower one Rxx vs w/wc. As expected, we observe the appearance of an extra beat and node when going from P = 1mW and T = 0.5K to P = 0.1mW and T = 0.1K. We observe the jump of π when crossing a node. Thus, in this figure we observe a total phase change of 2π when crossing two nodes in a row. Again this is explained by the effect of modulation of the cosine function on RIRO. This work is supported by the MINECO (Spain) under grant MAT2017-86717-P and ITN Grant 234970 (EU). Grupo de Matematicas aplicadas a la Materia Condensada, (UC3M), Unidad Asociada al CSIC. 1 P.T. Coleridge, Semicond. Sci. Technol., 5, 961-966, (1990). 2 T.H. Sander, S.N. Holmes, J.J. Harris, D.K. Maude and J.C. Portal, Phys. Rev. B. 58, 13856, (1998). 3 J. Luo, H. Munekata, F.F. Fang and P.J. Stiles, Phys. Rev. B. 38, 10142, (1988). 4 J.P. Heida, B.J. van Wees, J.J. Kuipers T.M. Klapwijk and G. Borghs, Phys. Rev. B. 57, 11911, (1998). 5 R.G. Mani, J.H. Smet, K. von Klitzing, V. Narayanamurti, W.B. Johnson, V. Umansky, Nature 420 646 (2002). 6 M.A. Zudov, R.R. Lu, N. Pfeiffer, K.W. West, Phys. Rev. Lett. 90 046807 (2003). 7 A. N. Ramanayaka, R. G. Mani, J. Inarrea, and W. Wegscheider, Phys. Rev. B, 85, 205315, (2012) 8 R. G. Mani, V. Narayanamurti, K. von Klitzing, J. H. Smet, W. B. Johnson, and V. Umansky, Phys. Rev. B, 69, 161306(R), (2004) 9 R. G. Mani, J. H. Smet, K. von Klitzing, V. Narayanamurti, W. B. Johnson, and V. Umansky , Phys. Rev. Lett. 92, 146801 (2004). 10 R. G. Mani, J. H. Smet, K. von Klitzing, V. Narayanamurti, W. B. Johnson, and V. Umansky , Phys. Rev. B69, 193304 (2004). 11 J. H. Smet, B. Gorshunov, C. Jiang, L. Pfeiffer, K. West, V. Umansky, M. Dressel, R. Meisels, 9 F. Kuchar, and K. von Klitzing , Phys. Rev. Lett. 95, 118604 (2005). 12 Tianyu Ye, Han-Chun Liu, Zhuo Wang, W. Wegscheider and Ramesh G. Mani, Sci. Rep. 5, 14880, (2015) 13 R. G. Mani, and A. Kriisa, Sci. Rep. 3, 3478, (2013) 14 R. G. Mani, Appl. Phys. Lett. 85, 4962, (2004) 15 R. G. Mani, W.B. Johnson, V.Umansky, V. Narayanamurti and K. Ploog, Phys. Rev. B79, 205320 (2009). 16 S. Wiedmann, G.M. Gusev, O.E. Raichev, A.K. Bakarov, and J.C. Portal, Phys. Rev. Lett., 105, 026804, (2010) 17 S. Wiedmann, G.M. Gusev, O.E. Raichev, A.K. Bakarov, and J.C. Portal, Phys. Rev. B, 81, 085311, (2010) 18 R. G. Mani, C. Gerl, S.Schmult, W. Wegscheider and V. Umansky, Phys. Rev. B81, 125320, (2010) 19 R.G. Mani, A.N. Ramanayaka and W. Wegscheider, Phys. Rev. B., 84, 085308, (2011) 20 Jesus Inarrea, R.G. Mani and W. Wegscheider, Phys. Rev. ,82 205321 (2010) 21 Tianyu Ye, Jesus Inarrea, W. Wegscheider and R.G. Mani Phys. Rev. B. 94, 035305, 2016 22 T. Herrmann, I. A. Dmitriev, D. A. Kozlov, M. Schneider, B. Jentzsch, Z. D. Kvon, P. Olbrich, V. V. Bel'kov, A. Bayer, D. Schuh, D. Bougeard, T. Kuczmik, M. Oltscher, D. Weiss, and S. D. Ganichev, Phys. Rev. B 94, 081301(R) (2016) 23 T. Herrmann, Z. D. Kvon, I. A. Dmitriev, D. A. Kozlov, B. Jentzsch, M. Schneider, L. Schell, V. V. Bel'kov, A. Bayer, D. Schuh, D. Bougeard, T. Kuczmik, M. Oltscher, D. Weiss, and S. D. Ganichev, Phys. Rev. B 96, 115449 (2017) 24 V.I. Ryzhii, R.A. Suris, B.S. Shchamkhalova , Sov. Phys. Semicond. 20, 1299, (1986) 25 A.C. Durst, S. Sachdev, N. Read, S.M. Girvin, Phys. Rev. Lett.91 086803 (2003) 26 X.L. Lei, S.Y. Liu, Phys. Rev. Lett.91, 226805 (2003) 27 V. Ryzhii and R. Suris, J.Phys.:Condens. Matter 15 6855 (2003). 28 V. Ryzhii, Phys. Rev. B 68 193402 (2003). 29 V. Ryzhii and V. Vyurkov, Phys. Rev. B 68 165406 (2003) 30 P.H. Rivera and P.A. Schulz, Phys. Rev. B 70 075314 (2004) 31 M. G. Vavilov, I. A. Dmitriev, I. L. Aleiner, A. D. Mirlin, and D. G. Polyakov, Phys. Rev. B, 70, 161306 (2004) 10 32 J. Inarrea and G. Platero, Phys. Rev. Lett. 94 016806, (2005) 33 J. Inarrea and G. Platero, Phys. Rev. B 72 193414 (2005) 34 J. Inarrea and G. Platero, Appl. Phys. Lett., 89, 052109, (2006) 35 J. Inarrea and G. Platero, Phys. Rev. B, 76, 073311, (2007) 36 J. Inarrea, Appl. Phys. Lett. 90, 172118, (2007) 37 J. Inarrea and G. Platero, Appl. Phys Lett. 93, 062104, (2008) 38 J. Inarrea and G. Platero, Phys. Rev. B, 78, 193310,(2008) 39 J. Inarrea, Appl. Phys Lett. 92, 192113,(2008) 40 Jesus Inarrea and Gloria Platero, Appl. Phys. Lett. 95, 162106, (2009) 41 J. Inarrea, Appl. Phys Lett. 90, 262101,(2007) 42 J. Inarrea, G. Platero and A. H. MacDonald, Physica Stat. Solid. A, 203,1148, (2006) 43 J. Inarrea, Appl. Phys Lett. 100, 242103,(2012) 44 Maria Busl, Gloria Platero, Antti-Pekka Jauho Phys. Rev. B, 85, 155449, (2012) 45 R Aguado and G Platero Phys. Rev. B 55, 12860, (1997) 46 J Inarrea and G Platero Europhys. Lett. 34 , 43, (1996) 47 J. Inarrea and G. Platero, Appl. Phys. Lett., 93, 062104, (2008) 48 Y.M. Beltukov and M.I. Dyakonov, Phys. Rev. Lett. 116, 176801, (2016) 49 R.G. Mani, J.H. Smet, K. von Klitizing, V. Narayanamurti, W.B. Johnson and V. Umansky, Phys. Rev. B 69 193304 (2004) 50 Q. Shi, M. Zudov, K. Baldwin, L. Pfeiffer and K. West, 33rd International Conference on Physics of Semiconductors (ICPS 2016), 31 July - 5 August 2016, Beijing, China. 51 B.B. Varga, Phys. Rev. 137 A1896 (1965). 52 Kunie Ishioka, Kristina Brixius, Ulrich Hofer, Avinash Rustagi, Evan M. Thatcher, Cristopher J. Stanton and Hrvoje Petek, Phys. Rev. B 92 205203 (2015). 53 Muneaki Hase and Masahiro Kitajima, J. Phys.: Condens. Matter 22 073201 (2010). 54 Lukas Novotny, Am. J. Phys. 78 1199 (2010). 55 C.G. Olson and D.W. Lynch, Phys. Rev. 177 1231 (1969) 56 J.D. Lee and Muneaki Hase, Phys. Rev. Lett. 101 235501 (2008) 57 P. Torma and W.L. Barnes, Rep. Prog. Phys. 78, 013901, (2015) 58 Jesus Inarrea, Euro. Phys Lett. 113 57004 (2016) 59 Jesus Inarrea, Scientific Reports, 7, 13573 (2017)
1606.07402	2	1606	2016-08-03T17:00:13	Dispersing artifacts in FT-STS: a comparison of set point effects across acquisition modes	[ "cond-mat.mes-hall" ]	Fourier transform scanning tunneling spectroscopy (FT-STS), or quasiparticle interference (QPI), has become an influential tool for the study of a wide range of important materials in condensed matter physics. However, FT-STS in complex materials is often challenging to interpret, requiring significant theoretical input in many cases, making it crucial to understand potential artifacts of the measurement. Here, we compare the most common modes of acquiring FT-STS data and show through both experiment and simulations that artifact features can arise that depend on how the tip height is stabilized throughout the course of the measurement. The most dramatic effect occurs when a series of dI/dV maps at different energies are acquired with simultaneous constant current feedback; here a feature that disperses in energy appears that is not observed in other measurement modes. Such artifact features are similar to those arising from real physical processes in the sample and are susceptible to misinterpretation.	cond-mat.mes-hall	cond-mat	Dispersing artifacts in FT-STS: a comparison of set point effects across acquisition modes A J Macdonald1,2, Y-S Tremblay-Johnston2,3, S Grothe1,2, S Chi1,2, P Dosanjh1,2, S Johnston4 and S A Burke1,2,3 1Department of Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada V6T 1Z1 2Stewart Blusson Quantum Matter Institute, University of British Columbia, Vancouver, BC, Canada, V6T 1Z4 3Department of Chemistry, University of British Columbia, Vancouver, BC, Canada V6T 1Z1 4Department of Physics and Astronomy, University of Tennessee, Knoxville, Tennessee 37996-1200, USA E-mail: [email protected] July 2016 Abstract. Fourier transform scanning tunnelling spectroscopy (FT-STS), or quasiparticle interference (QPI), has become an influential tool for the study of a wide range of important materials in condensed matter physics. However, FT- STS in complex materials is often challenging to interpret, requiring significant theoretical input in many cases, making it crucial to understand potential artifacts of the measurement. Here, we compare the most common modes of acquiring FT-STS data and show through both experiment and simulations that artifact features can arise that depend on how the tip height is stabilized throughout the course of the measurement. The most dramatic effect occurs when a series of dI/dV maps at different energies are acquired with simultaneous constant current feedback; here a feature that disperses in energy appears that is not observed in other measurement modes. Such artifact features are similar to those arising from real physical processes in the sample and are susceptible to misinterpretation. PACS numbers: 68.37.Ef, 74.55.+v, 68.37.Ef 6 1 0 2 g u A 3 ] l l a h - s e m . t a m - d n o c [ 2 v 2 0 4 7 0 . 6 0 6 1 : v i X r a Dispersing artifacts in FT-STS 1. Introduction for relating structural Scanning tunnelling microscopy (STM) is a powerful tool features to real-space electronic structure on the atomic scale. However, for materials with delocalized electronic states, the momentum dependent band structure provides an important representation for understanding bulk and surface electronic properties. This reciprocal space picture can be probed by STM via scattering of the electrons in the material by defects (e.g. impurities, step edges, and adatoms), which establish an interference pattern with scattering wavevectors that connect pieces of the underlying band structure of the material. By visualizing the Fourier transform of these interference patterns, information about the momentum dependence of the electronic structure becomes accessible, including the ability to map the electronic dispersion of both occupied and unoccupied bands, and locally correlate this with surface structure. This technique, known as Fourier-transform scanning tunnelling spectroscopy (FT-STS), or quasiparticle interference (QPI), has emerged as an important probe of a wide range of complex materials including superconductors [1, 2, 3, 4, 5, 6], topological insulators [7, 8, 9, 10, 11, 12, 13], and graphene [14, 15, 16]. With the remarkable stability of today's instrumentation, large data sets can be acquired yielding resolution in energy and momentum space rivaling that of state- of-the-art angle resolved photoemission [17] allowing new insight into physical processes such as electron- boson coupling [17, 18]. These advances in FT- STS, combined with the real-space resolution of STM, provide a unique view into the connection between physical and electronic structure, for example where there are different surface terminations with different electronic structure [12, 19, 20]. However, there is a catch; typically one must have some a priori knowledge of the underlying band structure to assign meaning to the features observed in FT-STS measurements because this relies on scattering processes from one part of the band structure to another. Simple 1-band systems can be analyzed directly, and this was first done for the surface states of noble metals [21, 22, 23, 24, 25] which have more recently been used to advance FT-STS as a probe for many-body effects [17, 26]. However, more complicated systems require complementary information about the band structure, and the patterns observed can become 1 extremely complex when scattering between multiple bands [3, 19, 27], and the possibility of selection rules for the allowed scattering processes [5, 7, 16, 28], come into play. These numerous features that may or may not disperse in energy can prove difficult to disentangle, usually requiring extensive theoretical support. In these cases it is all the more important that the data itself is well understood. Several different methods of acquiring FT-STS data have emerged, which can also influence the observed features. Therefore a careful comparison and analysis of these modes of acquisition and their potential artifacts is essential. Here, we return to a simple, well-known system: the Ag(111) surface state. We show through both experiment and theory that the way in which the tip height is maintained throughout the measurement -- "set point effects" -- can generate potentially misleading artifacts. 2. Methods The Ag(111) surface supports a well characterized two- dimensional Shockley surface state [29, 30, 31, 32, 33, 34, 35, 36, 17]. We chose this noble metal surface state because it serves as a well-studied, theoretically understood material with a single band, that is ideal for comparison of different FT-STS acquisition modes. Within the energy range we probed, the band dispersion can be well described by a free electron model of the form (k) = ¯h2k2(cid:107) 2m∗ − µ, (1) where (k) is the band dispersion, ¯h is the reduced Planck constant, m∗ is the effective electron mass, which for the Ag(111) surface state is approximately equal to 0.4 times the free electron mass [17, 30, 31, 34, 37, 38], and (0) = −µ = −65 meV is the chemical potential. All measurements were made at 4.2 K in ultrahigh vacuum with a pressure < 1 × 10−10 mbar using a commercial Createc STM. An electrochemically etched tungsten tip was used for all measurements, and was further prepared in situ by sputtering and annealing to remove the oxide layer. Initial contact with the Ag crystal likely results in an Ag terminated tip. The Ag(111) crystal was cleaned by 3 cycles of sputtering under 2.0 × 10−5 mbar Ar atmosphere and annealing at 500◦C to produce large, clean terraces with a low density of CO adsorbates that act as scattering Dispersing artifacts in FT-STS centers (see Figure 1a). Spatial calibration was performed once prior to all measurements by obtaining atomic resolution of the Ag(111) surface to ensure accurate real and reciprocal space measurements. Two main methods of collecting spatially resolved spectroscopic data were used. I-V spectroscopic grids were measured with varying size and spatial resolution (giving different pixel densities) and a thermally limited energy resolution of 1.5 meV. Each I-V spectrum consisted of 512 data points, which were Gaussian smoothed over 3 adjacent points in energy and averaged over 8-12 repeated measurements at each spatial location. The dI/dV was calculated numerically from these processed I-V spectra, giving dI/dV(x,y,V). Typical grid measurements took between 50 and 80 hours to complete. The dI/dV maps were taken using a lock-in amplifier with a bias modulation frequency of 1.017 kHz and amplitude of 5 mV. As these yield only one energy, a series of maps was acquired to investigate dispersion. The dI/dV map measurements were acquired with a simultaneous constant current topography, except for constant height measurements where the feedback was disengaged. Throughout the rest of the paper we will refer to these methods as "grids","maps", or "constant height maps" if no feedback was used. Once spectroscopic data is acquired, the FT-STS is obtained from the dI/dV image for each energy. Figure 1b) shows the real-space dI/dV image extracted from a grid measurement at the Fermi energy (EF ) for the region corresponding to the topography shown Surface state scattering from CO in Figure 1a). adsorbates and step edges is clearly visible. Since the CO adsorbates and step edges exhibit different scattering potentials, in order to attain a clean FT- STS, a real space correction is applied replacing the step edges (region between dashed lines in Figure 1b) with the average intensity of the dI/dV image. Figure 1c) then shows the Fourier transform of this dI/dV map into scattering or q-space. The bright ring is indicative of intraband scattering across the surface state band of Ag(111) and has a radius qF = 2kF at EF . Since the intraband surface state scattering is isotropic we can perform an angular average to improve the signal to noise of the scattering intensity, S(qr), where qr denotes the radial component of q. Additional intensity at the top and bottom of this ring is due to the non-isotropic scattering off of the step edges remaining after the real space filtering. To further reduce the influence of this different scatterer we take a restricted angular average only over the left and right quadrants between the dashed lines to produce Figure 1d): the azimuthally averaged line cut of the Fourier transformed data. The peak in Figure 1d) corresponds to the Fermi wave vector qF = 0.168± 2 0.002 A−1. By performing the process highlighted in Figure 1a)-d) at multiple energies it is possible to measure the scattering intensity as a function of energy and qr, S(qr,E), recovering the dispersion (qr) (see Figure 2). These steps were also used in Grothe et al. [17], and similar processing was performed on all of the data presented here as needed. Figure 1. Typical FT-STS data showing how the analysis proceeds from real space to q-space from a spectroscopic grid (Vs = −40 meV, It = 540 pA, 282 nm x 282 nm). (a) Real space topography showing multiple terraces separated by step edges. (b) Conductance map (dI/dV) at EF (V = 0); the white dashed lines indicate where step edges have been removed before taking the Fourier transform. (c) Absolute value of the Fourier transform of the dI/dV showing a dominant ring resulting from intra-band scattering of the Ag(111) surface state. Here the dashed lines show the restricted angular average in q-space. (d) The angular average of the absolute Fourier transform signal. The primary feature at qF = 2kF corresponds to the scattering vector of the surface state at EF . The smaller feature at lower qr is a set point driven effect. The two common methods described above for obtaining the dI/dV(x,y,V) needed to construct the FT-STS scattering dispersions differ in how the tip height is maintained. For grid measurements, where a full STS is acquired at each pixel, the tip height is usually stabilized at each point using consistent tunnelling parameters throughout to account for drift over the long timescale of the measurement. For dI/dV maps acquired with a lock-in amplifier at multiple energies, a constant current feedback is usually used to maintain the tip-sample distance and generate a simultaneous topographic image with each energy mapped. 3. Results Figure 2 shows measured by FT-STS in each of the scattering intensity S(qr,E) the two main (a) 0 nm2.3 nm4.3 pS20.4 pS(b) (c)qx (Å−1)qy (Å−1)θ−0.200.2−0.200.200.10.20.300.30.6qr=2kFqr (Å−1)S(qr,E=0)(d)(a) 0 nm2.3 nm4.3 pS20.4 pS(b) (c)qx (Å−1)qy (Å−1)θ−0.200.2−0.200.200.10.20.300.30.6qr=2kFqr (Å−1)S(qr,E=0)(d)100 nm100 nmDispersing artifacts in FT-STS 3 acquisition modes. Both measurements show the expected parabolic dispersion of the surface state band, but differ in other features. As determined previously in Grothe et al. [17], the intensity below the onset of the band is a product of the nature of the scatterer and varies somewhat in intensity between measurements regardless of acquisition mode depending on whether the dominant scatter is CO, other impurities, or if there is significant intensity from step edge scattering that remains after filtering. We therefore do not focus on this feature here. Instead we examine the dispersive QPI features related to the dispersing band structure of the system. Most notably, the grid measurement (Figure 2a) shows a faint, broad vertical feature slightly above qF = 2kF in addition to the expected intraband scattering. Whereas for the measurement made by acquiring dI/dV maps at different energies, an additional faint, and also somewhat broad feature appears that instead disperses, crossing qF = 2kF at EF . The dI/dV maps (Figure 2b) also show a strongly varying background intensity as a function of energy, strongest near EF , that is not seen in the grid measurement. The constant current maps have lower energy resolution compared to grid measurements over the same time scale, due to a difference in the speed of data acquisition between the two measurement techniques. We can now clearly see that acquiring the FT-STS dispersion using different methods yields qualitatively different results. Figure 2. Energy dispersion in q-space resulting from two different measurement techniques, spectroscopic grids and Though the surface state constant current lock-in maps. intensity differs dispersion is clear in both plots the overall significantly. Horizontal and vertical lines indicate EF and qF respectively and the dashed parabola comes from fitting the effective electronic mass to a free electron model. (a) Spectroscopic grid from a 60 x 60 nm2 area with set point bias of Vs = 100 mV and It = 100 pA. (b) Constant current maps taken with a lock-in amplifier over a 60 x 60 nm2 area and a current It = 100 pA. Lock-in parameters: f = 1.017 kHz and Vmod = 5 mV. To probe this in more detail, grid measurements were acquired using different bias voltages to stabilize the tip height at each pixel. We had previously noticed that stabilization biases, Vs, corresponding to energies well below the onset of the surface state band, µ, showed no prominent features other than the parabolic dispersion [17], thus minimizing the set point effect. Figure 3 shows each notable case: eVs < (0), EF > eVs > (0), and eVs > EF . As previously observed, there is no prominent feature for a stabilization bias below the onset of the band. For a stabilization bias between the onset of the band and EF , we see a faint, broad feature below 2kF , and for stabilization biases above EF , we see a faint, broad feature above 2kF . Therefore, for grid measurements, this additional feature depends on the bias used when the tip height is stabilized, with the feature crossing over 2kF at the Fermi energy, much like the additional feature observed for dI/dV maps. Figure 3. Comparison of three spectroscopic grids with three different set point conditions. (a) Grid with Vs = −100 mV, It = 100 pA, real space size 239 x 239 nm2 with 380 x 380 pixels (b) Grid with Vs = −40 mV, It = 540 pA, real space size 280 x 280 nm2 with 400 x 400 pixels (c) Grid with Vs = 100 mV, It = 100 pA, real space size 240 x 240 nm2 and 350 x 350 pixels. To make a more direct comparison between the different acquisition modes, Fourier transformed dI/dV maps were generated at the same energy, E = 50 meV, by four different acquisition methods (see Figure 4): a grid with Vs corresponding to the same energy as the energy examined, eVs = E = 50 meV (Figure 4b), a grid with opposite polarity Vs from the energy examined, eVs = −E = −50 meV (Figure 4c), a constant-current dI/dV map acquired with a lock-in amplifier where Vs always corresponds to the energy examined, eVs = E = 50 mV (Figure 4d), and a constant-height dI/dV map acquired with a lock-in amplifier where the tip height is stabilized only at the first pixel of the image (Figure 4e). The sharp peak seen in all four measurements at qr = 0.22 A−1 corresponds to the intraband scattering of the surface state at 50 meV. The grid with Vs = 50 mV is nearly identical to the constant current dI/dV map; this is expected if the additional feature is a set-point effect since the feedback is stabilized at each point with the same parameters for both measurements. In addition (a)qr=2kFqr (Å−1)Energy (meV)00.10.20.3−80−4004080qr (Å−1) MinMax(b)qr=2kF00.10.20.3(a)qr=2kFqr (Å−1)Energy (meV)eVs < ε(0)00.10.20.3−40−2002040(b)qr (Å−1)EF > eVs > ε(0)00.10.20.3qr (Å−1)eVs > EF (c)MinMax00.10.20.3Dispersing artifacts in FT-STS 4 to the ring corresponding to the expected scattering across the surface state band at qr = 0.22 A−1, there is a second relatively sharp feature between q = 2kF and the surface state band, indicated by the blue arrow in Figure 4a). However, for the grid acquired with Vs = −50 mV, while the intraband scattering at qr = 0.22 A−1 remains the same, there is now a much broader feature centered below 2kF (indicated by the red arrow), and the feature seen with Vs = +50 mV (blue arrow) is no longer observed. A constant height dI/dV measurement was also acquired, where the tip height is stabilized only in one position at the beginning of the measurement. Although the data has a larger low-frequency background, no clear secondary features are observed. Each of the secondary features appear above the background level of the other line cuts indicating that each are in fact additional features tied only to the measurement mode and parameters. The lack of any secondary features in the constant height measurement, along with the dependence of the additional features on the bias used to stabilize the tip height at each position for grids and maps, point to an influence of the spatially varying tip-height on the dI/dV measurement. When the tip is stabilized at each pixel, a constant current condition is met by the feedback circuit. That constant current condition depends on the integrated density of states, convolved with the transmission function of the tunnel junction. Since the density of states varies with both position and energy, the constant current topography will contain spatial modulations due to the electronic structure that depend on the bias applied, modulating the physical tip-sample separation. As the dI/dV signal also contains the transmission function, which depends on tip-sample separation, it is perhaps not surprising that extraneous features are observed in FT-STS that depend on the energy used to stabilize the tip height. This dependence of the dI/dV on the tip height now explains the differences between grid and constant current dI/dV map measurements: for a grid only one stabilization bias is used for all energies probed, so a non-dispersing feature either above (positive Vs) or below (negative Vs) 2kF is observed, but for constant current dI/dV maps, the stabilization bias follows the energy being probed generating a secondary feature that disperses, crossing 2kF at EF (see Figure 2b). To understand this effect, we now examine the influence of the spatially varying tip height on the dI/dV signal. Whenever the feedback circuit is engaged, the STM tip is stabilized to a particular height above the surface, zs, determined by a user specified bias, Vs, and tunnelling current set point, Is. In the zero temperature, one-dimensional limit this tunnelling current, Is = I(x, y, zs, Vs), can be Figure 4. The effect of stabilization bias on the observed QPI pattern. (a) Line cuts in q-space taking an angular average and comparing between two grids with different stabilization bias', a constant current map, and a constant height map. (b-e) FT- STS at E = 50 meV from (b) a spectroscopic grid with Vs = 50 mV, Is = 100 pA, (c) a spectroscopic grid with Vs = −50 mV, Is = 100 pA, (d) constant current map at Vs = 50 mV, Is = 100 pA, and (e) constant height map at Vs = 50 mV, and initial current Is = 100 pA. For the constant height data a restricted azimuthal average similar to that described in the methods for grid measurements was used to reduce the influence of a step edge running across the top of the image and some additional artifacts introduced by applying a line-by-line subtraction to account for z-drift. approximated by [39] (cid:90) eVs Is = ρ(x, y, E)ρt(E − eVs)T (zs, Vs, E)dE, (2) 0 is the where ρ is the sample density of states, ρt tip density of states, and T is the tunnelling barrier transmission coefficient. Following previous work [39, 40] we assume a trapezoidal tunnelling barrier and Dispersing artifacts in FT-STS 5 estimate T (zs, Vs, E) using the WKB approximation which gives (cid:32) (cid:114) √ 2m ¯h 2 − zs (cid:33) φ + − E eVs 2 , (3) T (zs, Vs, E) = exp bias, to obtain the dI/dV. Taking the full derivative of I(x, y, z, V ) gives two terms with a zs dependence [39] dI(x, y, zs) dV ∝ eρ(eV )ρt(0)T (zs) − √ √ 2me φ 2¯h zsI(zs, V ).(7) where φ is the effective height of the tunnelling barrier and m is the free electron mass. We take φ = 4.65 eV, the average of the work function of the W tip (φ = 4.55 eV) and the Ag sample (φ = 4.74 eV) [31]. In the low bias approximation we follow Koslowski et al. [41] and let T (zs, Vs, E) ≈ T (zs, Vs). Inserting this into the equation for Is (cid:90) eVs (cid:0)−zs 0 Is = Is = e ρ(x, y, E)ρt(E − eVs)T (zs, Vs)dE √ (cid:1)(cid:90) eVs 2 ρ(x, y, E)ρt(E − eVs)dE. 2mφ ¯h 0 (cid:32) Rearranging to solve for zs √ zs = − ¯h 2 2mφ ln (cid:82) eVs 0 Is ρ(x, y, E)ρt(E − eVs)dE (4) (5) (cid:33) .(6) This equation holds the key to understanding how the set point parameters influence the FT-STS results differently for spectroscopic grids, constant current maps, and constant height maps. In grid acquisition zs = z(x, y, Vs, Is) and is set by the feedback at each pixel based on the values of Vs and Is. This makes the tip sensitive to lateral variations in the LDOS but since Vs remains the same at every pixel this at most introduces a single spatial frequency corresponding to a non-dispersing feature in q-space. This feature appears at approximately the average of all scattering q values between 0 and eVs as it is related to the integrated LDOS. This is not the case for constant current maps where the stabilization bias is tied to the map energy E for each map. This means that zs = z(x, y, Vs = E/e, Is) where Vs varies, changing the spatial features convolved into the dI/dV measurement. Since zs contains periodic spatial modulations at approximately the average of all q values between 0 and Vs = E/e, this leads to the presence of dispersing features in the FT-STS pattern. Constant height maps, on the other hand, have no sensitivity to lateral variations of the LDOS. In a constant height map the tip height is set at one position, xs, ys, at the start of the map and then the feedback is disengaged, excluding the possibility of spatially dependent, feedback induced artifacts ie. zs = z(xs, ys, Vs = E/e, Is). To properly simulate the contribution to the FT- STS from set point effects we analyze the derivative of the simulated I(x, y, z, V ) with respect to the applied Figure 5. Comparison of T-matrix simulations applying the analytical expressions derived here with the experiments for different stabilization bias conditions. Scaled colour intensity is the same across each row. The leftmost column with a)- c) simulates a grid measurement with Vs = −100 meV and compares it to an experimental dispersion in d) with the same Vs. The middle column e)-g) simulates a grid measurement with Vs = 100 meV with the experimental equivalent shown in h). The right column i)-k) shows constant current acquisition mode where the set point energy matches the scan energy as in constant current maps in l). The first row shows the LDOS in q-space as determined from T-matrix simulations. The second row shows the calculated T (zs) of this LDOS with b) Vs = −100 meV bias, f) Vs = 100 meV bias, and j) Vs varying with the energy probed. The third row shows the product of the LDOS with T (zs), which qualitatively agrees with the experimental data shown in the fourth row. Note: Energy and q resolution varies between the measurements. Comparing the magnitude of these terms using the work function and free electron mass the first term dominates under the experimental conditions considered here, consistent with Li et al. [31] Using these analytic results, the effect on the FT- STS pattern of the different measurement modes and their set point artifacts can be simulated, expanding (a)GridVs = −0.1 eV−50050GridVs = 0.1 eV(e)MapVs = E(i)ρ (qr,E)(b)Energy (meV)−50050(f)(j)T(zs)(c)−50050(g)(k)dI/dV(q)(d)00.150.3−50050(h)qr (Å−1)00.150.3(l)Exp00.150.3Dispersing artifacts in FT-STS 6 on previous work describing the effect of real-space oscillations of T (zs) in one-dimenstion [31, 39, 41, 42, 43, 44] and on molecules in two-dimensions. [40] As a basis for the expected FT-STS dispersion, T-matrix simulations described in detail elsewhere [17] were used to model scattering of the Ag(111) surface state. These scattering simulations provided a theoretical density of states in q-space, ρ(q,E), which was Fourier transformed into a real space to give ρ(r,E) and used to calculate T (zs), I(x, y, z, V ), and dI/dV under different set point conditions from the analytic expressions derived above. T (zs) and dI/dV were then Fourier transformed back into q-space and compared with the original density of states, ρ(q,E). Figure 5 contrasts each case and compares the simulation to the experimental data. The agreement between theory and experiment is qualitatively very good and shows that the acquisition mode dependent features are related to variations in T (zs). The strongest artifacts are introduced for the constant current maps, while the best match with the underlying LDOS is acquired by a grid with a set point bias below the onset of the surface state. 4. Discussion & Conclusions As can be seen from the simulations, the dominant factor yielding different results for different measure- ment conditions arises from the transmission function T (zs) term. In the combined dI/dV simulation of the FT-STS dispersion one can see the main features pre- viously described above: for grids with a Vs above EF there is a clear vertical, non-dispersing line above 2kF , for grids with a Vs below the onset of the band there is only a weak very broad non-dispersing feature in T (zs) with little influence on the S(qr, E), and for the series of maps with constant current feedback at each en- ergy there is a dispersing feature which crosses 2kF at EF as well as an overall increase in intensity near EF arising from strong variations in T (zs). The positions of the features observed in a grid measurement with Vs = 100 meV and a series of constant current and con- stant height maps are shown in Figure 6 to summarize the potential artifacts. As can be seen, all three con- sistently reproduce the parabolic intraband scattering dispersion. The grid produces a vertical artifact fea- ture, but only the constant current maps generate a dispersing artifact feature. We have demonstrated, through a combination of measurements in different acquisition modes, and simulations of the expected FT-STS patterns for these modes, that artifact features in FT-STS derived dispersions can occur that depend on the set- point conditions used to stabilize the tip height. The simulations show that this arises from spatial Figure 6. Comparison of all features from constant current map, constant height map, and grid S(qr, E) peak positions illustrating the surface state scattering and set point features. For all three measurement modes the surface state peaks were obtained from Lorentzian fits and in each case the data agrees well with a free electron model of surface state scattering. Set point artifacts were fit with Gaussian functions for both constant current maps and grid data, while the constant height data showed no additional features. Fits of the set-point effect feature for the grid become unreliable close to where this feature crosses the surface state feature and have been omitted here. Only the constant current maps show a dispersing set point peak. modulations in the transmission function due to variations in zs at each (x,y) pixel that are dependent on the set-point conditions. This effect is most pronounced, and most concerning, for measurements acquired by taking dI/dV maps with a simultaneous constant current feedback at each energy, as this produces a relatively strong, dispersing feature. Dispersing features with similar q-dependence have been observed on (111) noble metal surfaces using constant current maps and attributed to a number of different sources. Petersen et al. [23] first reported a secondary scattering ring and ascribed this to scattering across a neck in the bulk Fermi surface in measurements of Au(111) and Cu(111). Schouteden et al. [25], performed further measurements of the Au(111) surface state and attributed the lack of dispersion of the bulk band above EF to inelastic electron relaxation rates. Most recently, Sessi et al. [26] demonstrated that the secondary dispersion is not compatible with the position of the bulk bands in Au(111), Cu(111), or Ag(111) and instead attributed the secondary features to an acoustic surface plasmon dispersion. Our data shows that a feature following this same dispersion arises from an artifact of the constant current measurement mode. We note that in constant height maps this set point effect due to modulation of the tunnel barrier is not present. However, for sufficiently high tunnelling 00.10.20.30.40.5−80−60−40−20020406080100Energy (meV) Constant Current Surface StateConstant Current Set Point EffectGrid Surface StateGrid Set Point EffectConstant Height Surface StateFree Electron ModelDispersing artifacts in FT-STS 7 current Equation 7 also predicts that a secondary dispersing feature should appear arising from the second term, even in constant height map data. Although we did not observe features in constant height maps related to this on Ag(111), this is a possible explanation for the secondary dispersing features observed in constant height maps on Cu(111) by Sessi et al. it should be linearly dependent on the tunnelling current, thus providing a way to test whether a secondary dispersing feature in constant height maps is caused by the physics of the tunnel junction or many-body effects in the sample. If so, [26]. Lastly, we note that for grid measurements a choice of stabilization bias below the onset of the band showed a very weak influence with no distinct features, providing a way to avoid these effects without resorting to demanding constant height measurements, or a more elaborate program of returning the tip to the same location to reset the height for each measurement pixel, as has been done for AFM measurements [45]. These results urge caution in the field; features in QPI require careful consideration, and artifacts can arise depending on the measurement mode that may obscure or masquerade as physical processes in the sample. Acknowledgments The authors thank D. Bonn for valuable discussions. This work was supported by the NSERC Discovery Grant Program (No. 402072-2012), NSERC CREATE Program (No. 414110-2012), UBC, the Canada Research Chairs Program (S. A. Burke), and the NSERC PGS program (A. J. Macdonald). References [1] McElroy K, Simmonds R W, Hoffman J E and Lee D H 2003 Nature 422 592 -- 596 [2] He Y, Yin Y, Zech M, Soumyanarayanan A, Yee M M, Williams T, Boyer M C, Chatterjee K, Wise W D, Zeljkovic I, Kondo T, Takeuchi T, Ikuta H, Mistark P, Markiewicz R S, Bansil A, Sachdev S, Hudson E W and Hoffman J E 2014 Science 344 608 -- 611 [3] Huang D, Song C L, Webb T A, Fang S, Chang C Z, Moodera J S, Kaxiras E and Hoffman J E 2015 Physical Review Letters 115 017002 [4] Fujita K, Hamidian M H, Edkins S D, Kim C K, Kohsaka Y, Azuma M, Takano M, Takagi H, Eisaki H, Uchida S i, Allais A, Lawler M J, Kim E A, Sachdev S and Davis J C S 2014 Proceedings of the National Academy of Sciences 111 E3026 -- E3032 [5] Chi S, Johnston S, Levy G, Grothe S, Szedlak R, Ludbrook B, Liang R, Dosanjh P, Burke S A, Damascelli A, Bonn D A, Hardy W N and Pennec Y 2014 Physical Review B 89 104522 [6] Chuang T M, Allan M P, Lee J, Xie Y, Ni N, Bud'ko S L, Boebinger G S, Canfield, P C and Davis J C 2010 Science 327 181 -- 184 [7] Okada Y, Dhital C, Zhou W, Huemiller E D, Lin H, Basak S, Bansil A, Huang Y B, Ding H, Wang Z, Wilson S D and Madhavan V 2011 Physical Review Letters 106 206805 [8] Zeljkovic I, Okada Y, Huang C Y, Sankar R, Walkup D, Zhou W, Serbyn M, Chou F, Tsai W F, Lin H, Bansil A, Fu L, Hasan M Z and Madhavan V 2014 Nature Physics 10 572 -- 577 [9] Beidenkopf H, Roushan P, Seo J, Gorman L, Drozdov I, Hor Y S, Cava R J and Yazdani A 2011 Nature Physics 7 939 -- 943 [10] Roushan P, Seo J, Parker C V, Hor Y S, Hsieh D, Qian D, Richardella A, Hasan M Z, Cava R J and Yazdani A 2016 Nature 460 1106 -- 1109 [11] Sessi P, Reis F, Bathon T, Kokh K A, Tereshchenko O E and Bode M 2016 Nature Communications 5 5349 [12] Kohsaka Y, Kanou M, Takagi H, Hanaguri T and Sasagawa T 2015 Physical Review B 91 245312 [13] Zhang T, Levy N, Ha J, Kuk Y and Stroscio J A 2013 Physical Review B 87 115410 [14] Leicht P, Tesch J, Bouvron S, Blumenschein F, Erler P, Gragnaniello L and Fonin M 2014 Physical Review B 90 241406 [15] Rutter G M, Crain J N, Guisinger N P, Li T, First P N and Stroscio J A 2007 Science 317 219 -- 222 [16] Mallet P, Brihuega I, Bose S, Ugeda M M, G´omez- Rodr´ıguez J M, Kern K and Veuillen J Y 2012 Physical Review B 86 045444 [17] Grothe S, Johnston S, Chi S, Dosanjh P, Burke S A and Pennec Y 2013 Phys. Rev. Lett. 111(24) 246804 [18] Allan M P, Lee K, Rost A W, Fischer M H, Massee F, Kihou K, Lee C H, Iyo A, Eisaki H, Chuang T M, Davis J C and Kim E A 2014 Nature Physics 11 177 -- 182 [19] Allan M P, Rost A W, Mackenzie A P, Xie Y, Davis J C, Kihou K, Lee C H, Iyo A, Eisaki H and Chuang T M 2012 Science 336 563 -- 567 [20] Inoue H, Gyenis A, Wang Z, Li J, Oh S W, Jiang S, Ni N, Bernevig B A and Yazdani A 2016 Science 351 1184 -- 1187 ISSN 0036-8075 [21] Sprunger P T, Petersen L, Plummer E W, Laegsgaard E and Besenbacher F 1997 Science 275 1764 -- 1767 [22] Hofmann P, Briner B G, Doering M, Rust H P, Plummer E W and Bradshaw A M 1997 Phys. Rev. Lett. 79(2) 265 -- 268 [23] Petersen L, Laitenberger P, Laegsgaard E and Besenbacher F 1998 Phys. Rev. B 58(11) 7361 -- 7366 [24] Petersen L, Sprunger P T, Hofmann P, Laegsgaard E, Briner B G, Doering M, Rust H P, Bradshaw A M, Besenbacher F and Plummer E W 1998 Phys. Rev. B 57(12) R6858 -- R6861 [25] Schouteden K, Lievens P and Van Haesendonck C 2009 Phys. Rev. B 79(19) 195409 [26] Sessi P, Silkin V M, Nechaev I A, Bathon T, El-Kareh L, Chulkov E V, Echenique P M and Bode M 2015 Nature Communications 6 8691 [27] Steinbrecher M, Harutyunyan H, Ast C R and Wegner D 2013 Physical Review B 87 245436 [28] Vishik I M, Nowadnick E A, Lee W S, Shen Z X, Moritz B, Devereaux T P, Tanaka K, Sasagawa T and Fujii T 2009 Nat Phys 5 718 -- 721 [29] Paniago R, Matzdorf R, Meister G and Goldmann A 1995 Surface Science 331-333, Part B 1233 -- 1237 ISSN 0039-6028 proceedings of the 14th European Conference on Surface Science [30] Paniago R, Matzdorf R, Meister G and Goldmann A 1995 Surface Science 336 113 -- 122 ISSN 0039-6028 [31] Li J, Schneider W D and Berndt R 1997 Phys. Rev. B 56(12) 7656 -- 7659 [32] Li J, Schneider W D, Berndt R, Bryant O R and Crampin S 1998 Phys. Rev. Lett. 81(20) 4464 -- 4467 [33] B´urgi L, Brune H, Jeandupeux O and Kern K 2000 Journal of Electron Spectroscopy and Related Phenomena 109 Dispersing artifacts in FT-STS 8 33 -- 49 [34] B´urgi L, Petersen L, Brune H and Kern K 2000 Surface Science 447 L157L161 [35] Kliewer J, Berndt R, Chulkov E V, Silkin V M, Echenique P M and Crampin S 2000 Science 288 1399 -- 1402 ISSN 00368075, 10959203 [36] Eiguren A, Hellsing B, Reinert F, Nicolay G, Chulkov E V, Silkin V M, Hufner S and Echenique P M 2002 Phys. Rev. Lett. 88(6) 066805 [37] Nicolay G, Reinert F, Hufner S and Blaha P 2001 Phys. Rev. B 65(3) 033407 [38] Reinert F, Nicolay G, Schmidt S, Ehm D and Hufner S 2001 Phys. Rev. B 63(11) 115415 [39] Hellenthal C, Heimbuch R, Sotthewes K, Kooij E and Zandvliet H 2013 Phys. Rev. B 88(3) 035425 [40] Ziegler M, N´eel N, Sperl A, Kroger J and Berndt R 2009 Phys. Rev. B 80(12) 125402 [41] Koslowski B, Dietrich C, Tschetschetkin A and Ziemann P 2007 Phys. Rev. B 75(3) 035421 [42] Hormandinger G 1994 Phys. Rev. B 49(19) 13897 -- 13906 [43] Ukraintsev V A 1996 Phys. Rev. B 53(16) 11176 -- 11185 [44] Pronschinske A, Mardit D J and Dougherty D B 2011 Phys. Rev. B 84(20) 205427 [45] Mohn F, Gross L and Meyer G 2011 Applied Physics Letters 99 053106
1703.00110	1	1703	2017-03-01T02:47:07	Mach-Zehnder interferometry using spin- and valley-polarized quantum Hall edge states in graphene	[ "cond-mat.mes-hall" ]	Confined to a two-dimensional plane, electrons in a strong magnetic field travel along the edge in one-dimensional quantum Hall channels that are protected against backscattering. These channels can be used as solid-state analogues of monochromatic beams of light, providing a unique platform for studying electron interference. Electron interferometry is regarded as one of the most promising routes for studying fractional and non-Abelian statistics and quantum entanglement via two-particle interference. However, creating an edge-channel interferometer in which electron-electron interactions play an important role requires a clean system and long phase coherence lengths. Here we realize electronic Mach-Zehnder interferometers with record visibilities of up to 98% using spin- and valley-polarized edge channels that co-propagate along a PN junction in graphene. We find that inter-channel scattering between same-spin edge channels along the physical graphene edge can be used to form beamsplitters, while the absence of inter-channel scattering along gate-defined interfaces can be used to form isolated interferometer arms. Surprisingly, our interferometer is robust to dephasing effects at energies an order of magnitude larger than observed in pioneering experiments on GaAs/AlGaAs quantum wells. Our results shed light on the nature of edge-channel equilibration and open up new possibilities for studying exotic electron statistics and quantum phenomena.	cond-mat.mes-hall	cond-mat	Mach-Zehnder interferometry using spin- and valley-polarized quantum Hall edge states in graphene Di S. Wei,1 T. van der Sar,2 J. D. Sanchez-Yamagishi,2, 3 K. Watanabe,4 T. Taniguchi,4 P. Jarillo-Herrero,3 B.I. Halperin,2 and A. Yacoby2 1John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts 02138 2Department of Physics, Harvard University, Cambridge, Massachusetts 02138 3Department of Physics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139 4Advanced Materials Laboratory, National Institute for Materials Science, Tsukuba, Ibaraki 305-0044, Japan Confined to a two-dimensional plane, electrons in a strong magnetic field travel along the edge in one-dimensional quantum Hall channels that are protected against backscattering. These channels can be used as solid-state analogues of monochro- matic beams of light, providing a unique plat- form for studying electron interference. Electron interferometry is regarded as one of the most promising routes for studying fractional and non- Abelian statistics and quantum entanglement via two-particle interference. However, creating an edge-channel interferometer in which electron- electron interactions play an important role re- quires a clean system and long phase coherence lengths. Here we realize electronic Mach-Zehnder interferometers with record visibilities of up to 98% using spin- and valley-polarized edge chan- nels that co-propagate along a PN junction in graphene. We find that inter-channel scattering between same-spin edge channels along the phys- ical graphene edge can be used to form beam- splitters, while the absence of inter-channel scat- tering along gate-defined interfaces can be used to form isolated interferometer arms. Surpris- ingly, our interferometer is robust to dephas- ing effects at energies an order of magnitude larger than observed in pioneering experiments on GaAs/AlGaAs quantum wells. Our results shed light on the nature of edge-channel equili- bration and open up new possibilities for study- ing exotic electron statistics and quantum phe- nomena. INTRODUCTION Electron interference plays a central role in mesoscopic physics [1 -- 3] and is regarded as one of the most promis- ing routes for studying fractional and non-Abelian statis- tics [4, 5] and quantum entanglement via two-particle interference [6, 7]. Quantum Hall edges form excel- lent building blocks for electron interferometers as they are single-mode channels that are protected from inter- channel scattering by their quantum degrees of freedom such as spin [1, 8]. Furthermore, they can be positioned via electrostatic gating and coupled at target locations that act as beamsplitters [2, 3]. Graphene may provide an advantage compared to conventional GaAs edge-channel interferometers [9 -- 12] as the absence of a band gap al- lows the creation of hole- and electron-like edge channels that naturally meet, co-propagate, and separate at gate- defined PN interfaces [13, 14]. Moreover, the additional valley degree of freedom and the associated unique nature of graphene quantum Hall states [15, 16] open up new op- portunities for addressing long-sought goals of electron interferometry such as the observation of non-Abelian statistics [17]. In addition, the valley isospin provides new possibilities for controlling inter-channel scattering [18], a requirement for creating edge-channel interferom- eters. However, even though graphene PN junctions in the quantum Hall regime have been studied extensively [8, 13, 14, 19 -- 23], creating an edge-channel interferom- eter using spin- and valley-polarized edge channels has remained an outstanding challenge. In the paradigmatic electronic interferometer -- the Mach-Zehnder interferometer (MZI) [3] -- a beam of elec- trons is split into two paths by a beam splitter and re- combined at a second beam splitter. Here, we engineer MZIs consisting of same-spin, opposite-valley quantum Hall edge channels that co-propagate along a PN junction in graphene. Using magnetic and trans-junction electric fields, we can tune into a regime in which either one or both pairs of the same-spin edge channels belonging to the zeroth Landau level (zLL) form MZIs that coher- ently mediate the cross-junction transport (Fig. 1A). We find that these channels can be well isolated from those belonging to other Landau levels (LLs), enabling us to study a target interferometer over a large range of elec- tric fields and tune into regimes with visibilities as high as 98%. By studying PN interfaces of different lengths, we show that the interferometer beamsplitters are lo- cated where the PN interface meets the physical graphene edges, which we attribute to strong inter-valley scattering at the physical graphene edge and the absence of inter- valley scattering along the gate-defined edge. We inde- pendently verify this conclusion using a device in which we can tune the number of edge channels co-propagating along either a physical or gate-defined edge. 2 ular magnetic field B, and define two regions of different charge densities nT and nB using a bottom gate that af- fects both nT and nB and a top gate that affects only nT (Fig. 1B). The number of edge channels in these regions is given by the filling factors νT,B = (h/eB)nT,B, where e is the electron charge and h is Planck's constant. The observation of integer quantum Hall steps in a measure- ment of the two-terminal conductance at B = 4 T in the regime where νT > 0 and νB > 0 confirms that the spin- and valley-degeneracy is lifted (Supplementary Fig. 1). Next, we create a PN junction by tuning into the regime where νT < 0 and νB > 0 and study which edge channels mediate charge transport across the junction. When we measure the conductance g as a function of νT < 0 and νB > 0 at B = 4 T (Methods), we observe four regions with distinct ranges of conductance values, as well as the first indications of conductance oscillations (Fig. 1C). In region I, the conductance of the junction is near zero, which we attribute to the situation depicted in the top panel of Fig. 1A (where νB = 1 and νT = −1). Here, one N-type spin-down and one P-type spin-up edge channel co-propagate along the junction. As these chan- nels have opposite spin, inter-channel scattering is sup- pressed [8]. When we cross from region I into region II, we begin to observe transport across the junction. We at- tribute this to an additional spin-up edge channel having entered on the N-side (so that νB = 2 and νT = −1) and that electrons in this channel can scatter into the spin-up channel on the P-side (see middle panel in Fig. 1A). The observed conductance ranges approximately between 0 and e2/h, consistent with one pair of edge channels me- diating transport across the junction. Similarly, in region III we obtain the situation in which νB = 1 and νT = −2, and we attribute the observed conductance to scattering between the two spin-down edge channels. Strikingly, in region III the conductance does not change notably as we keep adding edge channels on the P-side (going to νB = 1 and νT < −2). We conclude that these additional chan- nels do not contribute to the trans-junction conductance, presumably because they belong to a higher LL which makes them spatially too distant from the PN interface. Crossing into region IV (νB ≥ 2 and νT ≤ −2), we ob- serve that the average conductance increases and ranges between 0 and 2e2/h. We attribute this to two pairs of same-spin edge channels mediating transport across the junction. Again, we see no sign of edge channels belong- ing to higher LLs entering the system and contributing to the trans-junction conductance. We conclude that the edge channels belonging to the zLL mediate the trans- junction conductance, well isolated from edge channels belonging to higher LLs. The relative isolation of the edge channels that belong to the zLL allows us to study a target pair of edge chan- nels over a large range of filling factors. As we increase the magnetic field to B = 9 T and concentrate on region III, in which νB = 1 and νT ≤ −2, we observe a striking FIG. 1. Creating a Mach-Zehnder interferometer us- ing spin- and valley-polarized quantum Hall edge channels. (A) Schematic illustration of the formation of Mach-Zehnder interferometers (MZIs) at a graphene PN junc- tion. Green and purple denote quantum Hall edge channels of opposite spin. Top panel: at (νB, νT)=(1,-1), where νB (νT) is the filling factor in the N- (P-) region, two edge channels run along the interface. Their opposite spin suppresses inter- channel scattering. Middle panel: at (νB,νT)=(2,-1), a pair of spin-up edge channels forms a MZI. Inter-channel scattering occurs at the ends of the junction as indicated by dotted lines. Bottom panel: at (νB,νT)=(2,-2), two pairs of same-spin edge channels form two MZIs. (B) Device 1: an edge-contacted monolayer graphene flake encapsulated in hexagonal boron nitride (hBN). The top gate (Au) and bottom gate (graphite) define the PN junction (P: red color, N: blue color). The top (bottom) hBN gate dielectric is 20 (30) nm thick. The top gate is contacted by a lead that runs over a bridge fabricated from hard-baked PMMA to avoid shorting to the graphene flake. The back gate (Si) is used to strongly increase the p- doping of the graphene leading up to right lead and reduce the contact resistance. The SiO2 back-gate dielectric is 285 nm thick. (C) Two-terminal conductance of device 1 in the PN regime at B = 4 T. We distinguish four regions (dashed boxes). Region I corresponds to (νB,νT) = (1,-1). Region II corresponds to νT = −1 and νB ≥ 2. Region III corresponds to νB = 1 and νT ≤ −2. Region IV corresponds to νB ≥ 2 and νT ≤ −2. RESULTS Constructing a Mach-Zehnder interferometer in a graphene PN junction To construct a MZI of spin- and valley-polarized edge channels, we use a hexagonal boron nitride (hBN) encap- sulated monolayer of graphene (Fig. 1B and Methods). We tune into the quantum Hall regime using a perpendic- 3 pattern of conductance oscillations (Fig. 2A) whose key features such as shape and periodicity depend on both νB and νT. These oscillations cannot be explained by semi-classical snake states or similar low-field phenom- ena [24 -- 27] since in our device electron transport is me- diated by quantum Hall edge channels. Instead, as we will further argue below, the well-defined periodicity of these oscillations indicates that scattering between the two edge channels that mediate the cross-junction trans- port occurs at only two points along the junction. These points form the beamsplitters that define our MZI. Its conductance, in units of e2/h, is given by g = r1t22 + t1r22 + 2t1t2r1r2cos(φ + φ0) (1) where ti (ri) is the transmission (reflection) amplitude of the i-th beamsplitter, with ri2 + ti2 = 1. The phase φ = 2πBA arises from the Aharonov-Bohm effect, where Φ0 = h/e is the flux quantum, A the effective area en- closed by the two edge channels, and φ0 an (unknown) phase associated with the beamsplitters. Φ0 Since the measurement in Fig. 2A is performed at a fixed magnetic field, we attribute the conductance os- cillations to a changing distance between the two edge channels and a resulting changing flux through the inter- ferometer. We can analyze the charge-density-dependent locations of these channels by determining where the two corresponding exchange-split Landau sublevels cross the Fermi energy, using a simple model for the spatial de- pendence of the sublevel energy (Fig. 2B-D and Sup- plementary Note 1). This model indicates that as the charge densities increase (from Fig. 2B to Fig. 2C), the edge-channel separation decreases. Furthermore, when the charge density is small (large) on a particular side of the junction, the edge-channel separation is relatively sensitive (insensitive) to the charge density on that side of the junction (Fig. 2D). Figure 2E shows that this model reproduces the key features of the data in Fig. 2A. Fur- ther data in the νB ≥ 2 and νT ≤ −2 regime, in which two MZIs act simultaneously (as depicted in the bottom panel of Fig. 1A), are shown in Supplementary Fig. 2. Beamsplitter characteristics The visibility of the oscillations in a MZI depends on the phase coherence and the transmission characteristics of the beamsplitters. We analyze the range of visibilities observed in the measurement shown in Fig. 2A by divid- ing the measurement range into a grid and calculating the local visibility V = (gmax − gmin)/(gmax + gmin), with gmax and gmin the maximum and minimum conductance within each block (Fig. 2F). In Fig. 2G we plot the re- sulting experimental cumulative probability distribution function that indicates the probability of finding a visi- bility greater than x. This distribution corresponds well FIG. 2. Characterization of a single Mach-Zehnder in- terferometer. (A) Two-terminal conductance of device 1 at B = 9 T, over a range of filling factors corresponding to a single interferometer at the PN junction. (B) Modeling the charge-density dependence of the distance between the edge channels that form a MZI. The red and blue shading illus- trates the spatial variation of the charge density close to the PN junction. The green line illustrates the spatial variation of the energies of the exchange-split ν = 1 and ν = -2 Lan- dau sublevels. The edge channels are located at the positions where these sublevels intersect the Fermi energy. The dis- tance between the edge channels determines the flux through the interferometer. Far from the PN junction, where the low- est LL is completely empty (ν = -2) or completely full (ν = 2), the exchange splitting Uex vanishes. Near the PN junction, however, the electronic ground state can develop an imbal- ance in the valley occupation, leading self-consistently to a non-zero Uex. (C) Increasing the electron and hole densi- ties decreases the distance between the edge channels. (D) A strong imbalance between the electron and hole densities. (E) Simulation of the two-terminal conductance as a function of filling factors based on the model sketched in (B-D). (F) Lo- cal visibility of the conductance oscillations observed in (A). The grey dashed box indicates where the visibility was not extracted due to non-resolved oscillations. (G) Data points: experimentally determined probability of finding a visibility greater than x, extracted from the color plot in (F). Blue line: theoretical prediction based on MZIs with beamsplitters de- scribed by random scattering matrices (Supplementary Note 2). to a theoretical prediction that is based on the assump- tion that the incoming and outgoing channels of each of the two beamsplitters of the MZI are connected by ran- dom U (2) matrices in valley space (Supplementary Note 2). Remarkably, in several regions of the conductance map (Fig. 2A) we find visibilities as high as 98%, indicat- ing near-perfect phase coherence along the PN interface. Additionally, in some regions the conductance oscillates nearly between 0 and e2/h, indicating nearly 50/50 beam splitters. Dependence of the Mach-Zehnder interference on magnetic field and DC voltage bias Next, we tune to a region of high visibility and study the conductance as a function of B and a DC voltage bias VDC (Fig. 3A). We observe that the visibility stays near-unity for VDC < 0.5 mV (Fig. 3B), and decreases at larger VDC, which may be due to thermal averaging or electron-electron interactions [3, 12]. For the 8 to 9 T field range of Fig. 3C, measurements at VDC = 0 show a constant oscillation period ∆B, which is consistent with an assumption that the area enclosed by the interferome- ter is constant and given by A = Φ0 ∆B . Subject to this as- sumption, we determine an edge-channel separation of 52 nm. Oscillations with VDC are also observed (Fig. 3D) in- dicating a bias-dependent edge-channel separation, which may be a result of a bias-induced electrostatic gating ef- fect [12]. We note that at larger filling factors we see multiple frequencies, changing frequency with field, and lobe structures, which have previously been attributed to Coulomb interactions in GaAs devices [9, 11, 12] (Sup- plementary Fig. 3). We leave the analysis of these effects to a future study. Varying the length of the PN interface To confirm that the beamsplitters are located where the PN interface meets the physical graphene edges, we measure the MZI oscillation frequency as a function of the interface length. We use device 2 (Fig. 4A-B), which has five top gates (TG1 to TG5) of varying lengths that we can address individually in two-terminal conductance measurements by using the appropriate leads. Using top and bottom gates to control the filling factors in the top- gated and non-top-gated regions, νT and νB respectively, we can tune into a regime where νT < 0 and νB > 0 to create an NPN configuration with two PN junctions in series (Supplementary Fig. 4). When we measure the two-terminal conductance at B = 8 T as a function of νT and νB (Fig. 4C and Supplementary Fig. 4), we recognize the regions corresponding to zero, one, and two pairs of same-spin edge channels mediating transport across the PN junctions, as discussed above for the measurement 4 FIG. 3. Mach-Zehnder oscillations as a function of magnetic field and DC voltage bias. (A) Two-terminal differential conductance as a function of magnetic field B and DC voltage bias VDC at (νB,νT)=(1,-2), for which only one interferometer is formed at the PN interface. (B) Visibility of the conductance oscillations shown in (A) as a function of DC bias.(C) Conductance oscillations with B at zero DC bias corresponding to the red dotted line in (A). From the period ∆B = 66 mT we calculate the distance between edge states to be 52 nm, assuming that the distance between the beamsplitters is given by the 1.2 µm width of the device. (D) Line trace corresponding to the purple dotted line in (A) showing oscillations with respect to VDC. νB + √ in Fig. 1C and further analyzed in Supplementary Note 3. In addition, we observe clear conductance oscillations, of which we expect the frequencies to reflect the gate lengths. To analyze these frequencies, we focus on the limit νB,T >> 1 in which the edge-channel separation √−νT) (Supplementary Note 4). We √ and the associated Aharanov-Bohm flux are expected to vary as ∼ 1/( √−νT) and plot the conductance data against 1/( use a Fourier transform to determine the frequency spec- trum (Supplementary Fig. 5). Normalizing the frequency axis to the average length of TG5, we find peaks at loca- tions that correspond reasonably well to those expected based on the lengths of the different gates (Fig. 4D). We conclude that the beamsplitters are located where the PN interfaces meet the physical graphene edge. Remarkably, it follows that each oscillation corresponds to a minute change in the edge-channel separation: for example, for the L=1.2 µm gate length of device 1, this change equals BL = 3.7 A. Φ0 νB + Edge-channel equilibration along gate-defined and physical edges Finally, we demonstrate the absence of inter-channel scattering along a gate-defined edge and the full equili- bration of same-spin edge channels running along a phys- ical edge. We use device 3 (Fig. 5A-D), which has two top gates that determine the number of edge channels running from the left to the right lead, and a top gate 5 (referred to as the side gate) that determines which frac- tion of the edge channels in the central region travel along the lower physical edge instead of along the side-gate- defined edge. We first confirm the presence of robust broken-symmetry quantum Hall states (Supplementary Fig. 6C). We then apply a bias VIN between the left and top lead, and measure the potential at the right lead (VOUT) as a function of the side-gate filling factor (Fig. 5E). Edge-channel equilibration in the central re- gion should reduce the chemical potential at the right lead below that of the input lead. The precise match be- tween data and model (described in Supplementary Note 5) in Fig. 5E clearly demonstrates that edge channels do not equilibrate along the side-gate-defined edge, while they do equilibrate along the physical edge provided they have the same spin [8]. We note that we observe no MZI oscillations as we sweep the magnetic field between 8.9 and 9 T (Supplementary Fig. 6D-F), presumably because there are no locations acting as beamsplitters as the edge channels do not meet at a physical edge before and after co-propagating along a gate-defined edge. DISCUSSION The experiments presented here demonstrate a robust method of engineering a high-visibility MZI in a graphene QH system. This opens up the possibility of a variety of interferometry experiments and grants us the diagnostic capabilities of measuring sub-nanometer shifts in edge- channel separation. In our experiments we observe trans- port across the insulating ν = 0 state, which is expected to be in a canted antiferromagnetic (CAF) phase in bulk graphene [28, 29]. The fact that we find that spin polar- ization is well preserved in our samples suggests that the CAF phase may be suppressed in a narrow PN junction in favor of a state where spins are fully polarized along the direction of the magnetic field. METHODS Sample Fabrication FIG. 4. Gate-length dependence of the Mach-Zehnder oscillations. (A) Optical microscope image of device 2: an edge-contacted, hBN-encapsulated monolayer of graphene with five top gates of different lengths. The top-gate dielec- tric (hBN) is 17 nm thick. The bottom hBN layer is 16 nm thick. The back-gate dielectric (SiO2) is 285 nm thick. Leads (L1-L6) are yellow. Top gates (TG1-TG5) are orange. Using a top and back gate, we induce an NPN charge configuration with two PN junctions and their associated MZIs connected in series. (B) AFM image of device 2. The graphene is in- dicated by the dashed white line. Top gates are outlined in green, leads in yellow, etched regions in blue. The lengths of both sides of each top gate are indicated in micrometers. (C) Two-terminal conductance measured across top gate 1 (TG1) using leads L1 and L2 at B = 8 T. Region I corre- sponds to (νB, νT) = (-1,1). Region II corresponds to νT = -1 and νB ≥ 2. Region III corresponds to νT ≤ −2 and νB = 1. Region IV corresponds to νB ≥ 2 and νT ≤ −2. In- set: close-up of (B) showing the top gate and the two leads used in this measurement. The edge channels are indicated by black lines. (D) Frequency spectrum of the conductance oscillations for all top gates. The x-axis is normalized to the length of TG5. The expected frequencies for each gate are indicated by the black dashed lines. All devices were fabricated on doped Si chips with a 285 nm layer of SiO2 that acted as a dielectric for the Si back gate. Graphene was mechanically exfoliated from bulk graphite obtained from NGS Naturgraphit GmbH using 1009R tape from Ultron Systems and subsequently encapsulated in hexagonal boron nitride (hBN) using a dry transfer process [30]. For device 1, we placed the resulting stack on a graphite bottom gate. Before the first metal deposition step, we annealed the devices in vacuum at 500◦C to improve device quality. We then created top gates using electron-beam lithography and thermal evaporation of Cr/Au. To fabricate edge- contacts to the graphene in device 1 without shorting to the graphite bottom gate, we selectively etched the stack down such that the bottom hBN flake remained and protected the graphite while simultaneously expos- ing the graphene flake. To fabricate edge-contacts to the graphene in devices 2 and 3, we etched through the entire hBN/graphene stack. We then created edge contacts by thermally evaporating Cr/Au while rotating the sample using a tilted rotation stage. Finally, we etched the de- vices into the desired geometry by reactive ion etching 6 FIG. 5. Absence of equilibration between edge channels running along a gate-defined edge. (A-D) Schematic of device 3: an edge-contacted, hBN-encapsulated monolayer of graphene with two top gates and a side gate. An AFM image shows the top gates (TG1 and TG2, false-colored purple) on top of the hBN-encapsulated graphene flake. Yellow (green) indicates the leads (side gate). The progression of panels (A-D) illustrates the changing locations of the edge channels in the central region as the filling factor under the side gate is tuned from νS = 0 to νS = 3, while the regions under TG1 and TG2 are kept at νT = 1 and the rest of the device is kept at νB = 4. The circulating edge states near the contacts are omitted for clarity. (E) Voltage measured at the right contact as a function of the side-gate voltage VS that tunes the side-gate filling factor νS, for νT = 1, νT = 2, and νT = 3 as indicated by the insets. The data (blue) is an average of a set of traces taken at different magnetic fields between 7.9 and 8 T (Supplementary Fig. 6). The red line indicates the expected values given by a model that assumes no equilibration along the gate-defined edge and full equilibration between same-spin channels along the physical edge, taking into account the independently measured contact resistances (Supplementary Note 5). The top trace corresponds to the sequence depicted in (A-D). in O2/CHF3 using a PMMA/HSQ bilayer of resist (pat- terned by electron-beam lithography) as the etch mask. Measurement Our measurements were performed in a Leiden dry dilution refrigerator with a base temperature of 20 mK. Measurements of differential conductance were per- formed using a lock-in amplifier with an AC excitation voltage of 10 µV at 17.77 Hz. All measurements of differ- ential conductance were corrected for contact/line resis- tances, which were independently determined by lining up the robust ν = 2 quantum Hall conductance plateau with 2e2/h. We estimated all filling factors based on a parallel-plate capacitor model with a correction to ac- count for quantum capacitance (Supplementary Note 6). [1] Datta, S. Electronic Transport in Mesoscopic Systems (Cambridge Univ. Press, 1995). [2] Chamon, C. C., Freed, D. E., Kivelson, S. A., Sondhi, S. L. & Wen, X. G. Two point-contact interferometer for quantum Hall systems. Phys. Rev. B 55, 2331 -- 2343 (1997). [3] Ji, Y., Chung, Y., Sprinzak, D., Heiblum, M. & Mahalu, D. An electronic Mach-Zehnder interferometer. Nature 422, 415 -- 418 (2003). [4] Law, K. T., Feldman, D. E. & Gefen, Y. Electronic Mach-Zehnder interferometer as a tool to probe fractional statistics. Phys. Rev. B 74, 045319 (2006). [5] Feldman, D. E. & Kitaev, A. Detecting non-abelian statistics with an electronic Mach-Zehnder interferom- eter. Phys. Rev. Lett. 97, 186803 (2006). [6] Samuelsson, P., Sukhorukov, E. V. & Buttiker, M. Two- particle Aharonov-Bohm effect and entanglement in the electronic Hanbury Brown-Twiss setup. Phys. Rev. Lett. 92, 026805 (2004). [7] Yurke, B. & Stoler, D. Bell's-inequality experiments us- ing independent-particle sources. Phys. Rev. A 46, 2229 -- 2234 (1992). [8] Amet, F., Williams, J. R., Watanabe, K., Taniguchi, T. & Goldhaber-Gordon, D. Selective equilibration of spin- polarized quantum Hall edge states in graphene. Phys. Rev. Lett. 112, 196601 (2014). [9] Neder, I., Heiblum, M., Levinson, Y., Mahalu, D. & Umansky, V. Unexpected behavior in a two-path electron interferometer. Physical Review Letters 96, 1 -- 4 (2006). [10] Neder, I. et al. Interference between two indistinguishable electrons from independent sources. Nature 448, 333 -- 337 (2007). [11] Roulleau, P. et al. Finite bias visibility of the electronic Mach-Zehnder interferometer. Physical Review B 76, 161309 (2007). [12] Bieri, E. et al. Finite-bias visibility dependence in an electronic Mach-Zehnder interferometer. Phys. Rev. B 79, 245324 (2009). [13] Abanin, D. A. & Levitov, L. S. Quantized transport in graphene p-n junctions in a magnetic field. Science 317, 641 -- 643 (2007). [14] Williams, J. R., Dicarlo, L. & Marcus, C. M. Quantum Hall effect in a gate-controlled p-n junction of graphene. Science 317, 638 -- 640 (2007). [15] Castro Neto, A. H., Guinea, F., Peres, N. M. R., Novoselov, K. S. & Geim, A. K. The electronic prop- erties of graphene. Rev. Mod. Phys. 81, 109 -- 162 (2009). [16] Young, A. F. et al. Spin and valley quantum Hall fer- romagnetism in graphene. Nature Physics 8, 550 -- 556 (2012). [17] Zibrov, A. et al. Robust nonabelian ground states and continuous quantum phase transitions in a half filled bilayer graphene Landau level. Preprint at https://arxiv.org/abs/1611.07113 (2016). [18] Tworzyd(cid:32)lo, J., Snyman, I., Akhmerov, A. R. & Beenakker, C. W. J. Valley-isospin dependence of the quantum Hall effect in a graphene p-n junction. Phys. Rev. B 76, 035411 (2007). [19] Ozyilmaz, B. et al. Electronic transport and quantum Hall effect in bipolar graphene p-n-p Junctions. Phys. Rev. Lett. 99, 166804 (2007). [20] Velasco, J., Liu, G., Bao, W. & Ning Lau, C. Electrical transport in high-quality graphene pnp junctions. New Journal of Physics 11, 095008 (2009). [21] Matsuo, S. et al. Edge mixing dynamics in graphene p-n junctions in the quantum Hall regime. Nature Commu- nications 6, 8066 (2015). [22] Machida, T. et al. Edge-channel transport of Dirac fermions in graphene quantum Hall junctions. Journal of the Physical Society of Japan 84, 1 -- 9 (2015). [23] Tovari, E. et al. Gate-controlled conductance enhance- ment from quantum Hall channels along graphene p-n junctions. Nanoscale 8, 19910 -- 19916 (2016). [24] Williams, J. R. & Marcus, C. M. Snake states along graphene p-n junctions. Phys. Rev. Lett. 107, 046602 (2011). [25] Rickhaus, P. et al. Snake trajectories in ultraclean graphene pn junctions. Nature Communications 6, 6470 (2015). [26] Taychatanapat, T. et al. Conductance oscillations in- duced by ballistic snake states in a graphene heterojunc- tion. Nature Communications 6, 6093 (2015). [27] Overweg, H. et al. Oscillating magnetoresistance in graphene p-n junctions at intermediate magnetic fields. Preprint at https://arxiv.org/abs/1612.07624 (2016). [28] Kharitonov, M. Phase diagram for the ν = 0 quan- tum hall state in monolayer graphene. Phys. Rev. B 85, 155439 (2012). [29] Young, A. F. et al. Tunable symmetry breaking and helical edge transport in a graphene quantum spin Hall state. Nature 505, 528 -- 32 (2014). [30] Wang, L. et al. One-dimensional electrical contact to a two-dimensional material. Science 342, 614 -- 7 (2013). [31] Zhang, Y., Tan, Y.-W., Stormer, H. L. & Kim, P. Ex- perimental observation of the quantum Hall effect and 7 Berry's phase in graphene. Nature 438, 201 -- 204 (2005). [32] McClure, D. T. et al. Edge-state velocity and coherence in a quantum Hall Fabry-P´erot interferometer. Phys. Rev. Lett. 103, 206806 (2009). ACKNOWLEDGEMENTS General We acknowledge helpful discussions and feedback from A. Kou, B.E. Feldman, D. Nandi, K. Shain, P. Kim, A. Stern, M. Heiblum, C. Schonenberger, K. Wang, S.L. Tomarken and S.P. Harvey. We also thank G.H. Lee, J.I.J. Wang, J.Y. Luo and X. Liu for help with fabrica- tion. Funding This work was primarily supported by the US DOE, BES Office, Division of Materials Sciences and Engineer- ing under Award de-sc0001819 (D.S.W., J.D.S.Y., P.J.H. and A.Y.). D.S.W. acknowledges partial support from the National Science Foundation Graduate Research Fel- lowship under Grant No. DGE1144152. B.I.H. acknowl- edges support from the STC Center for Integrated Quan- tum Materials, NSF grant DMR-1231319. K.W. and T.T. acknowledge support from the Elemental Strategy Initiative conducted by the MEXT, Japan and JSPS KAKENHI Grant Numbers JP26248061, JP15K21722 and JP25106006. This work was performed in part at the Harvard University Center for Nanoscale Systems (CNS), a member of the National Nanotechnology Coordinated Infrastructure Network (NNCI), which is supported by the National Science Foundation under NSF ECCS award no. 1541959. Author Contributions D.S.W., T.S., B.I.H. and A.Y. conceived and designed the experiments. D.S.W. fabricated the devices. D.S.W., T.S., J.D.S.Y. and A.Y. performed the experiments. D.S.W., T.S., J.D.S.Y., P.J.-H., B.I.H. and A.Y. ana- lyzed the data, developed the models, and wrote the pa- per. K.W. and T.T. provided the hexagonal boron nitride crystals used in the devices. CORRESPONDENCE Correspondence be should als [email protected]). and requests for addressed to D.S.W. materi- (email: SUPPLEMENTARY INFORMATION Supplementary Note 1. Modeling the distance between the edge channels forming a Mach-Zehnder interferometer In this note we describe the model used for the calcu- lations shown in Fig. 2E of the main text, which shows the conductance of a PN junction as a function of fill- ing factors to the left and right of the PN junction, νB and νT, respectively. As discussed in the main text, we attribute the oscillations to a changing flux enclosed by the two edge channels forming a Mach-Zehnder interfer- ometer. At fixed magnetic field, this change is caused by a change in distance between the edge channels. We analyze the locations of the two interferometer-forming edge channels by determining where the corresponding exchange-split Landau sublevels cross the Fermi energy, using a simple model for the spatial dependence of the sublevel energy described by E±(x) = µT − µB 2 g(x) + µT + µB 2 ± Uex(x) 2 (S1) general µ =(cid:82) ν where µB(T) is the chemical potential to the left (right) of the junction, Uex(x) is the exchange splitting, and g(x) is a function that smoothly changes from -1 to 1 across the PN junction over a distance W that is similar to the dis- tance of the graphene to the gates. For simplicity, from now on we neglect the spatial dependence of Uex. To link the chemical potentials to filling factors, we note that in δν(cid:48) dν(cid:48), which increases stepwise as Lan- dau levels are filled. For simplicity, we make the approx- imation µ = EC F and vF is the Fermi velocity [31]. Defining f±(x) = 4E±(x) , and limiting ourselves to the case of νT < 0 and νB > 0 relevant in our PN measurements, we get 2 sgn(ν)(cid:112)ν, where EC =(cid:112)2eBv2 EC δµ 0 f±(x) = −( √−νT + √ νB)g(x)+ √ νB−√−νT±uex (S2) where we have defined uex = 4Uex EC edge channel is then obtained by solving f (x) = 0, so that the distance between two edge channels is given by . The location of an (cid:20)√ νB − √−νT + uex √−νT + √ νB ∆x = g−1 (cid:20)√ (cid:21) −g−1 (cid:21) νB − √−νT − uex √−νT + √ νB (S3) For g(x) we use a logistic function of width W : g(x) = This has the inverse 2 1 + e− X W − 1. (cid:20) 1 + y 1 − y (cid:21) . g−1(y) = W ln (S4) (S5) Combining Eqs. S3 and S5, we find the distance between the two edge channels (cid:20)√ √ ∆x = W ln νB + uex/2 νB − uex/2 √−νT + uex/2 √−νT − uex/2 8 (S6) (cid:21) To calculate the plot of Fig. 2E, we now assume a MZ interferometer with 50/50 beam splitters and calculate the conductance using (cid:18) (cid:19) BL∆x Φ0 g = 0.5 + 0.5cos 2π (S7) In Fig. 2E, we used B = 9 T, L = 1.2 µm, Φ0 = h/e, W = 52 nm and, to qualitatively resemble the data, uex=0.4. Supplementary Note 2. Random scattering model for a MZ interferometer at a graphene PN junction In this note we provide a background discussion of the electronic wavefunctions corresponding to the edge channels that form an interferometer along the PN junction. We then analyze the probability to find a particular visibility of the Mach-Zehnder conductance oscillations for an interferometer that has beamsplitters described by random transmission/reflection matrices. The resulting cumulative probability distribution func- tion for the visibility is shown as the theoretical curve in Fig. 2G of the main text. of the case consider Straight junction of infinite length Let us first an infinite, translationally-invariant PN junction aligned with the y-axis. The overall problem can be formulated in terms of a solution to the time-independent single- particle Schrodinger equation at an energy E equal to the Fermi energy. We may write this solution in the form ψ(x, y, τ ), where τ is a valley index, and we consider only one spin state. We work in a gauge where the vector potential is parallel to the junction. Then, if there are just two edge states at the junction, we may write, in the vicinity of the junction ψ(x, y, τ ) = c1eik1yΦ1(x, τ ) + c2eik2yΦ1(x, τ ) (S8) where kj, for j = 1,2, are the two eigenvalues of the translation operator, Φj are the corresponding eigenvec- tors, and cj are arbitrary constants. It will be convenient to choose the normalizations of Φj so that Φj2 = 1, (S9) (cid:90) ∞ uj dx −∞ (cid:88) τ where uj = (dkj/dE)−1 is the velocity of the edge mode. In the simplest model that we have in mind, the func- tions Φj will have the approximate form Φj(x, τ ) ≈ χj(τ )exp[−(x − xj)2/(2l2 B)], (S10) where lB is the magnetic length, χj is the center of grav- ity of the state j, and the spinor χj is a function of the valley index τ . The separation ∆x = x2 − x1 may be written as ∆x = Uex/V (cid:48), (S11) where V (cid:48) is the gradient of the electrostatic potential and Uex is the exchange splitting, which we assume to be constants in the vicinity of the p-n junction. The separation ∆x is related to the momentum difference k2− k1 by ∆x = k2 − k1l2 B (S12) The exchange splitting in Eq. (2.4) results from a term in the Hartree-Fock Hamiltonian of form: Hex = n · (cid:126)τ , Uex 2 (S13) where n is a three-component unit vector and (cid:126)τ are the three Pauli matrices, acting on the valley index τ . In an approximation where one neglects valley-dependent electron-electron interactions, which only occur when two electrons are very close together, there is nothing to pick out one particular orientation of n over another. Never- theless, the exchange splitting may be greater than zero. The choice of n at a particular PN junction will then de- pend on small symmetry breaking terms, which we will not attempt to predict, and it could vary as one moves along the junction. The spinors Φj are the eigenstates of n · (cid:126)τ . The magnitude of the exchange splitting must be de- termined self-consistently, based on the local difference in the occupation of the two valley states. On the micro- scopic level, we would expect that the exchange potential is not actually a constant over the range of the width of the PN junction, so the quantity Uex in the above equa- tions should be taken as an average over a region covered by the wave functions Φj. It is straightforward to generalize the discussion of a translationally invariant junction to a situation where properties of the junction, including its orientation, may vary adiabatically along its length. (Adiabatically means that the distance scale for changes along the length of the junction should be large compared to (k2 − k1)−1.) We choose a gauge where the vector potential is always oriented parallel to the junction at the position of the junction. We may now write ψ(x, y, τ ) = c1eiϕ1(y)Φ1(x, τ ) + c2eiϕ2yΦ2(x, τ ) (cid:90) y ψj(y) = kj(y(cid:48))dy(cid:48) (S14) (S15) 0 where y is the distance along the edge and x is measured in the local perpendicular direction. The fact that the 9 magnitudes cj are independent of y is a consequence of conservation of current and our choice for the normal- ization of Φj. Junction connected to sample edges Now we consider a PN junction of length L, connected to the sample boundaries at its two ends as depicted in Fig. 1A. We assume that there is a single chiral edge state at any segment of the boundary, and we assume that the edge states at y = 0 flow into the junction while the edge states at y = L flow away from the junction. The portions of the wave function ψ incident at y = 0 may be characterized by complex amplitudes a1 and a2, such that a12 and a22 are, respectively, the currents incident from the left and from the right. Similarly, we may char- acterize the outgoing wave function by amplitudes b1 and b2 for electrons moving to the right and left, respectively, away from the end of the junction at y = L. The ampli- tudes cj for the wave function along the PN junction will be related to the amplitudes aj by a 2 x 2 unitary matrix S(0), whose form will depend on details of the sample in the region where the junction meets the edge. Similarly, we may define a matrix S(L), which relates the outgoing amplitudes bj to the amplitudes cjeiϕj (L) at the end of the PN junction. The outgoing amplitudes bj will then be related to the incoming amplitudes aj by a matrix N , which we may write as N = S(L)M S(0), (S16) where M is a diagonal matrix with elements Mjj(cid:48) = δjj(cid:48)eiϕj (L). Suppose that there is incident beam imping- ing on the junction from the left, so that a2 = 0. Defin- ing the transmission coefficient T as the probability for an electron to wind up in the right-moving state after leaving the junction, we see that T = N112 (S17) Let nα be the unit vector on the Bloch sphere that corresponds to the two-component unit vector aj ≡ S(0) j,1 , and let nβ be the unit vector on the Bloch sphere that corresponds to the two-component unit vector β ≡ . Let Rϕ by the O(3) matrix that repre- S(L)(cid:17)−1(cid:105) (cid:104)(cid:16) j,i sents a rotation by an angle ϕ about the z-axis, where (cid:90) L ϕ = ϕ2(L) − ϕ1(L) = (k2 − k1)dy (S18) Then we have 0 1 + nβ · Rϕ · nα 2 T = (S19) If we describe the vectors nα and nβ by their polar coordinates, (θ, ϕ), and define ϕ0 = ϕβ − ϕα then we obtain T = C + Dcos(ϕ + ϕ0) (S20) 1 + cosθαcosθβ sinθαsinθβ 2 2 C = , and D = 2 and r22 = cos2 θβ (S21) where we recognize Eq. 1 of the main text with t12 = cos2 θα 2 . If the length L is large, the phase ϕ will change by a large amount when we vary the magnetic field by an amount that is still too small to affect the other parameters in the above equation. Thus T will oscillate between maximum and minimum values given by Since this constraint forces x < ( we can replace the constraints (Eqn. S27) by 2 )1/2, when is small 10 − 1 < y < 1 The integral is now simple to carry out, giving us P () ≈ dy(1 − y2)(/2)1/2 = (cid:18) 8 (cid:19)1/2 9 (S30) (S31) (cid:90) 1 −1 Tmax = C + D and Tmin = C − D (S22) Supplementary Note 3. Conductance of two Mach-Zehnder interferometers in series The probability P () is then given by: k=0 If we define the visibility by V = Tmax − Tmin Tmax + Tmin then we find V = D C = sinθαsinθβ 1 + cosθαcosθβ (S23) (S24) Random scattering model Since the matrices S(0) and S(L) depend on details that we do not know how to calculate, we will consider a model in which the matrices are random matrices in the group U (2). In this case, the unit vectors nα and nβ will be randomly distributed on the Bloch sphere. We wish to calculate the probability P () that the visibility V is greater than 1 - . It is useful to change variables to x = cosθα + cosθβ 2 , y = cosθα − cosθβ 2 A() P () = (S26) where A is the area of the x − y plane that satisfies the constraints 2 − 1 < x < 1,−1 < y + x < 1,−1 < y − x < 1, (S27) C − D = > 1 + x2 − y2 − [(1 − y2 − x2 − 2xy)(1 − y2 − x2 + 2xy)](1/2) C 1 + x2 − y2 (S28) Ideally, we should compute A() numerically, which we do to calculate the theoretical curve displayed in Fig. 2G. However, we can make an analytic approximation, which should be valid in the limit of small . In this limit, we can expand the right hand side of Eq. S28 and replace it by the constraint > 2x2 (1 − y2)2 (S29) In this note, we analyze the conductance values ob- served in the different regions indicated in Fig. 4C of the main text. As discussed in the main text, we at- tribute the conductance observed in regions II and III to the presence of one interferometer at each of the PN junctions, and the conductance in region IV to the pres- ence of two interferometers at each of the PN junctions. We thus analyze the expected conductance of MZ inter- ferometers that are connected in series, as depicted in Supplementary Figure 4A. We first assume that we have only one interferometer at each of the PN junctions (corresponding to regions II and III in Fig. 4C). We also assume that phase coherence is lost in the region between the two interferometers, as the edge channels run along several microns of vacuum edge. The transmission through the two interferometers is given by (S25) T = T1T2 [(1 − T2)(1 − T1)]k = T1T2 ∞(cid:88) 1 1 − (1 − T2)(1 − T1) 1 + 1 T2 − 1 (S32) = 1 T1 where Ti is the transmission probability through interfer- ometer i. Recalling Supplementary Eq. S20, the trans- mission through a MZ interferometer is given by (cid:16) (cid:17) Ti = Ci + Dicos ϕ(i) + ϕ(i) 0 (S33) Likewise, the transmission through an NPN device with two interferometers (that are independent because of their opposite spin) at each PN interface is given by T = T ↑ + T↓ 1 + 1 ↑ T 2 1 ↑ T 1 = + − 1 1 + 1 ↓ T 2 1 ↓ T 1 − 1 (S34) To compare the average conductance observed in regions II, III, and IV of Fig. 4C of the main text to expectations based on this model, we now assume 50/50 beam splitters and average Eqs. S32 and S34 over ϕ(i=1,2). The results are shown in Supplementary Figure 4C. We see a reason- able agreement. However, we note that it is unclear why one should expect 50/50 beamsplitters. Supplementary Note 4. Analyzing the gate-length dependence of the Mach-Zehnder oscillation frequencies observed in the NPN measurements on device 2. To determine the location of the beamsplitters of our Mach-Zehnder interferometers, we analyze the frequency of the conductance oscillations in the top gate/back gate sweeps of our NPN device (device 2) for each of the five top gates. As discussed in the main text, we analyze the frequency of the oscillations observed in the NPN con- ductance data such as those shown in Fig. 4C by focus- ing on the region of large filling factors (νB,T (cid:29) 1). In this regime, we can linearize the function g(x) in νB − √−νT), the edge-channel separation varies ap- √ Eq. S3, so that it follows that for a given value of ( proximately as ∆x ∝ √ . 1 νB+ √−νT √ √ √ νB + 1 νB+ and y = ( −1 √−νT νB+ As such, the frequency of the Mach-Zehnder oscilla- √−νT . tions should be constant as a function of Therefore, for each top gate, we take the NPN conduc- √−νT) (Sup- tance data (such as the data shown in Fig. 4C) and plot it against x = plementary Figure 5). We then divide the measurement range as indicated by the boxes, focusing on the limit νB,T (cid:29) 1 (i.e, focusing on the right-hand side of the plots), and calculate the absolute value of the Fourier transform with respect to the x-coordinate for all data traces within each box. For each box we average these Fourier transforms over the y-coordinate. In order to convert from frequency to gate length, we multiply the x- coordinates in each box by a scale factor, which depends on the y-coordinate but is the same across all gates for all boxes with the same y-coordinate. The scale factors are chosen so that for TG5, the peak of the Fourier transform occurs at the gate length 22 µm for each value of y. Fi- nally, we average the resulting frequency-axis-normalized spectra (1 for each box) over all boxes. The results are plotted in Fig. 4D of the main text. is no edge-channel equilibration in the central region and under the assumption that the resistances of the leads are zero. 11 To calculate the expected VOUT, we start by assuming an infinite input impedance of our voltmeter, allowing us to relate VOUT to the chemical potentials of the edge channels arriving at the right lead as ↓ ↓ Tµ OUT ↓ T ↑ ↑ OUT + ν Tµ ↑ T + ν VOUT = − ν (S35) ν ↑ T and ν ↓ T are the number of spin-up and spin- where ν down edge channels under the left and the right top gate, ↑ ↓ ↓ with νT = ν OUT and µ T, and µ OUT are the chemical potentials of these channels when they arrive at the right lead. ↑ T +ν We now assume that edge channels only equilibrate if they have the same spin and run along the physical graphene edge. The number of spin-up and spin-down edge channels running along the lower physical edge in ↓ the central region is given by ν S, respectively, with νS = ν ↑ S and ν ↑ S + ν ↓ S. We thus expect ↑,↓ S − ν ↑,↓ S νS ≤ νB ↑,↓ T µIN + (ν ν (cid:17) ∧(cid:16) ↑,↓ ↑,↓ S > ν T (cid:16) ν ν ↑,↓ T )µG ↑,↓ µ OUT = for (cid:17) (cid:17) (cid:17) (S36) (S37) (S38) and and ↑,↓ T µIN + (ν ν ↑,↓ B − ν ↑,↓ B ↑,↓ T )µG ↑,↓ µ OUT = ν (cid:16) for ↑,↓ ↑,↓ S > ν T ν νS > νB (cid:17) ∧(cid:16) (cid:16) ↑,↓ µ OUT = µIN for ↑,↓ S ≤ ν ν ↑,↓ T Supplementary Note 5. Gate-defined equilibration studies. In this note, we describe how we derive the expected equilibration curves shown in Fig. 5E of the main text (red lines), which are based on the assumption that edge- channels only equilibrate if they have the same spin and run along the physical graphene edge. In the measure- ments of Fig. 5E we apply a voltage VIN to the left lead, ground the top lead (VG = 0), and measure the voltage VOUT at the right lead. Our goal is to calculate VOUT as a function of the side-gate filling factor νS. We keep the filling factor under the left and right top gate equal, calling it νT. Furthermore, we work in the regime where νB > νT ≥ 1, where νB is the filling factor in the non- top-gated regions, so that we expect VOUT = VIN if there where µG and µIN are the chemical potentials of the edge channels emerging from the top and left lead respectively. Equations S36 to S38, substituted into Eq. S35, describe the expected equilibration curves shown in Fig. 5E of the main text (red lines). However, to get those curves we need to include the effect of the non-zero resistances of the left and top lead, RIN and RG respectively. The lead resistances are non-zero because of RC filters, wires, and contact resistance. From current conservation at the left and top lead, we get µIN = −VIN νTRG + RQ νT (RIN + RG) + RQ (S39) and µG = −VIN νTRG νT(RIN + RG) + RQ . (S40) 12 Experimentally, we determine the resistances of the leads by setting νB = νT = 2 and measuring the quantum Hall resistance plateau using the left and the top lead. The value of this plateau is given by R = RQ/2 + RIN + RG, with RQ = h/e2, allowing us to extract RIN + RG. We then assume RIN = RG yielding RIN = RG = 4.4kΩ. Using Eqs. S35 to S40, we obtain the traces plotted in Fig. 5E of the main text. ing both the top and bottom gate. We locate the cen- ter of the νB = 0 plateau, and if it is shifted from VB = 0 by VB,off we take this into account by substi- tuting VB → VB − VB,off in our equation for nB. We use a similar procedure for nT. As described in the main text, we then calculate the filling factor ν(T,B) using the equation ν(T,B) = (h/eB)n(T,B) where B is the magnetic field, and h is Planck's constant. Supplementary Note 6. Calculating charge densities and filling factors from gate voltages. In this note we describe how we obtain the filling fac- tor values displayed on the axes of our plots. First, we estimate the charge density nB in the non-top-gated re- gion using a simple parallel-plate capacitor model via the equation nB = 0BVB/(dBe), where 0 is the vacuum permittivity, B is the dielectric constant of the back- gate dielectric (either hBN or SiO2), VB is the applied back-gate voltage, dB is the thickness of the back-gate dielectric, and e is the electron charge. To calculate the charge density nT in a top-gated region, we use the equa- tion nT = nB + 0TVT/(dTe), where T is the dielectric constant of the hBN top-gate dielectric, VT is the applied top-gate voltage, and dT is the thickness of the top-gate dielectric. To determine the precise location of the charge neu- trality point, we measure the conductance while sweep- Due to quantum capacitance effects, these filling fac- tors do not precisely correspond with the positions where we see robust plateaus in the quantum Hall regime. To correct for this, we tune to an NN(cid:48) configuration (defined by νT > 0 and νB > 0) where we are able to identify the locations of particular filling factors by the observation of robust quantum Hall plateaus. We line up the fill- ing factor on the axis with the corresponding plateau by multiplying by a constant factor Cq that accounts for the quantum capacitance. Then, to determine the loca- tion of the filling factors in the NP regime, we assume that the quantum Hall plateaus induced on the electron side are of the same size as the quantum Hall plateaus induced on the hole side, and multiply the filling factor obtained from the parallel-plate capacitor model by the same Cq factor. We note that exchange-split levels often span a smaller density range than those separated by a cyclotron energy gap (i.e. the ν = 1 plateau is smaller than the ν = 2 plateau), introducing a small uncertainty into this procedure. 13 FIG. S1. Characterization of device 1 in the regime where νB > 0 and νT > 0 (which we call the NN(cid:48) regime). (A) Optical microscope image of device 1: a triple-gated, hexagonal boron nitride-encapsulated monolayer of graphene. This image corresponds to the schematic in Fig. 1B of the main text. The encapsulated graphene is outlined by the white dashed lines. We tune the filling factor νT under the top gate using both the graphite bottom gate and the Au/Cr top gate. We tune the filling factor νB in the region to the left of the top gate using the bottom gate only. The device sits on a 285 nm SiO2/Si global back gate that we use to strongly dope the graphene leading up to the right lead, thus reducing the contact resistance. A bridge of hard-baked PMMA supports the lead contacting the top gate and prevents shorting of this lead to the graphene. (B) A schematic illustration of the edge states present in the system in the NN(cid:48) regime, with (νB, νT) = (1,3) as an example. In the NN(cid:48) regime, the conductance is given by min(νB, νT). (C) Two-terminal conductance in the NN(cid:48) regime. The Si back gate is set to 60 V. (D) Line trace corresponding to the dotted purple line in (B). The observation of conductance quantization in steps of e2/h confirms that the spin and valley degeneracy is fully lifted in the zLL. 14 FIG. S2. Two-terminal conductance of device 1 in the PN regime (in which νB > 0 and νT < 0) at B = 4 T and large filling factors. (A) We observe two hyperbola-shaped sets of conductance oscillations (the red dotted lines guide the eye through the centers of these hyperbolas). As discussed in the main text, our data indicates that these Mach-Zehnders are formed by the two pairs of same-spin edge channels belonging to the zLL. We observe that the conductance oscillates approximately between 0 and 2e2/h, even at large filling factors, indicating that two Mach-Zehnder interferometers mediate transport across the PN junction even when there are many edge channels in the system. (B) Schematic that depicts edge channels belonging to higher LLs that do not communicate across the junction or with the zLL, presumably because of their larger spatial separation. 15 FIG. S3. The effect of a DC bias on the differential conductance of a PN junction. (A) Two-terminal conductance of device 1 as a function of magnetic field B and DC voltage bias VDC. At B = 8 T, we have νB = 1 and νT = 2. Different spatial derivatives of the energies of the two Landau sublevels forming an interferometer (see e.g., Fig. 2D of the main text) can lead to an energy-dependent inter-edge-channel distance ∆x, which results in a differential conductance that depends on VDC: if VDC is applied asymmetrically, as in our measurements (with the chemical potential of the left channel raised to VDC and that of the right channel remaining at 0V), the differential conductance is given by g ∼ cos [ 2πBL ∆x(VDC)]. In this case it is clear that a change in ∆x caused by a change in VDC can be compensated for by a change in B, consistent with the diagonal stripes of constant differential conductance observed in the region around B = 8 T and in Fig. 3A of the main text. If the bias is somehow symmetrized, due to e.g. electron-electron interactions [32], the chemical potential of the left (right) channel equals VDC )]. This may lead to more complex behavior such as the checkerboard patterns observed around B = 6 T [32]. We note that a bias-dependent electrostatic gating effect may also change the inter-channel distance [12] and correspondingly lead to a bias-dependent differential conductance. ), and correspondingly g ∼ cos[ 2πBL −VDC −VDC ∆x( VDC 2 )]+ cos[ 2πBL Φ0 ∆x( Φ0 Φ0 ( 2 2 2 16 FIG. S4. Analyzing the average conductance observed in NPN measurements on device 2. (A) Schematic of two PN junctions in series (MZ1 and MZ2), as formed in our NPN device. µin is the chemical potential of the edge entering the first interferometer and µout is the chemical potential exiting the second interferometer. (B) Two-terminal conductance measurement as a function of back-gate and top-gate filling factors νB and νT, measured across top gate 1. The red (black) dashed box indicates a region with one (two) edge channel(s) in the top-gated region. (C) The red (black) data corresponds to the measured conductance within the red (black) dashed box in (B), averaged over νT. The dashed lines indicate the expected average conductance corresponding to 0, 1, or 2 interferometers formed at each of the PN interfaces, assuming 50/50 beamsplitters as discussed in Supplementary Note 3. 17 FIG. S5. Analyzing the gate-length dependence of the Mach-Zehnder oscillation frequencies observed in NPN devices. (A-E) The conductance maps for all top gates (TG1-TG5) at B = 8 T, plotted in a transformed coordinate system. We address each top gate individually by using the appropriate leads. An image of the device is shown in Fig. 4A of the main text. The lengths of the top gates are shown in Fig. 4B of the main text. Note that top gate 5 is the longest and correspondingly shows the fastest conductance oscillations. The boxes indicate the regions in which we take Fourier transforms of the data to compare the frequency of the observed Mach-Zehnder oscillations between the different gates, resulting in Fig. 4D of the main text (see Supplementary Note 4). 18 FIG. S6. Device 3: verifying the presence of broken-symmetry quantum Hall states and measurements of edge channel equilibration as a function of magnetic field. (A) Optical image of device 3. The dotted line outlines the graphene. (B) AFM image of the clean, hBN-encapsulated graphene flake used for device 3. (C) Two-terminal conductance measured across top gate 1 (TG1) using the left and middle lead, as a function of the filling factor νTG1 under TG1 and the filling factor νB in the non-top-gated region. Conductance plateaus that are present for all integers from ν = 1 to ν = 6 confirm that the spin and valley degeneracy of the Landau levels is lifted. These plateaus are also present in a similar measurement across TG2 (not shown). (D-F) Equilibration measurements as a function of B and the side-gate voltage VS, as described in the main text. The two top gates are set at νT=1 for (D) νT=2 for (E) and νT=3 for (F)
1006.0901	2	1006	2010-09-22T23:58:00	Dirac Point and Edge States in a Microwave Realization of Tight-Binding Graphene-like Structures	[ "cond-mat.mes-hall" ]	We present a microwave realization of finite tight-binding graphene-like structures. The structures are realized using discs with a high index of refraction. The discs are placed on a metallic surface while a second surface is adjusted atop the discs, such that the waves coupling the discs in the air are evanescent, leading to the tight-binding behavior. In reflection measurements the Dirac point and a linear increase close to the Dirac point is observed, if the measurement is performed inside the sample. Resonances due to edge states are found close to the Dirac point if the measurements are performed at the zigzag-edge or at the corner in case of a broken benzene ring.	cond-mat.mes-hall	cond-mat	Dirac Point and Edge States in a Microwave Realization of Tight-Binding Graphene-like Structures U. Kuhl,1, 2 S. Barkhofen,1 T. Tudorovskiy,1 H.-J. Stockmann,1 T. Hossain,2 L. de Forges de Parny,2 and F. Mortessagne2 1Fachbereich Physik, Philipps-Universitat Marburg, Renthof 5, 35032 Marburg, Germany 2Laboratoire de Physique de la Mati`ere Condens´ee, CNRS UMR 6622, Universit´e de Nice Sophia-Antipolis, 06108 Nice, France (Dated: October 16, 2018) We present a microwave realization of finite tight-binding graphene-like structures. The structures are realized using disks with a high index of refraction. The disks are placed on a metallic surface while a second surface is adjusted atop the discs, such that the waves coupling the disks in the air are evanescent, leading to the tight-binding behavior. In reflection measurements the Dirac point and a linear increase close to the Dirac point is observed, if the measurement is performed inside the sample. Resonances due to edge states are found close to the Dirac point if the measurements are performed at the zigzag-edge or at the corner in case of a broken benzene ring. PACS numbers: 42.70.Qs, 73.22.-f, 71.20.-b, 03.65.Nk Due to its electronic shell structure carbon forms struc- tures such as diamonds, fullerenes and nanotubes, which all have very specific and fascinating properties, mechan- ically as electronically. Another realization is graphene, a one-atom-thick allotrope of carbon, which has a structure of the honeycomb lattice, i. e. two combined triangular lattices. A common feature of all sheet-like carbon struc- tures including graphene is sp2 hybridization leaving one pz orbital for the bonding. This is the starting point for the tight-binding approximation used in many theoreti- cal descriptions of graphene including the disseminating works by Wallace1 and Semenoff.2 Due to the specific symmetry of the lattice the gapless band structure has a conical singularity at two k vectors,1 called nowadays the Dirac points. The name comes from the fact that the reduced equation around the Dirac points corresponds to the Dirac equation of a massless particle resulting in a band structure of the form of two cones touching at the tips.3,4 Recently it was shown that this single layer can be realized and investigated experimentally.5,6 The main ingredient for this relation is the symmetry. Thus a realization via a photonic crystal7,8 was proposed.9 In addition in two microwave experiments signatures of the Dirac point had been found in the transmission in an triangular set-up.10,11 Both experiments correspond to open scattering arrangements. Graphene, on the other hand, corresponds to a closed system because of the tight- binding between nearest neighbors preventing an escape to the outer world. We consider the aspect of symme- try and the aspect of tight-binding with mainly nearest neighbor interaction equally important.12 This is realized by a microwave experiment using dielectric disks with a high index of refraction in a hexagonal lattice. The disks are weakly coupled by evanescent modes as will be ex- plained later. So far microwave experiments have been used to show different kinds of effects originally coming from con- densed matter physics such as the Hofstadter butterfly,13 FIG. 1. (color online) Spectrum of a single disk for the TE antenna (dark, black) and the TM antenna (light, yellow). The TE1 resonance of the disc, which is used for further in- vestigations is marked. The inset shows the TE antenna with a single disk on the metallic copper plate. The top plate is not shown. localization in disordered systems,14,15 and transport in case of correlated disorder.16 In the proposed microwave setup it is feasible to in- vestigate defects, edges, disorder etc., as the realized "graphene" is a macroscopic object. Thus it is possible to study problems experimentally which are not accessible in real graphene at the moment. Additionally one can have a look into the system and study e. g. the density of states (DOS) or transport properties as well. In this paper we will concentrate on the local DOS (LDOS) and its characteristics at edges or corners. As a starting point we investigate a single disc, then the coupling of two of them and proceed via a benzene ring to graphene flakes. We measure the reflection S11 of an antenna situated close to a disk of rd=4 mm radius, hd=5 mm height and an index of refraction nd ≈ 6. The antenna is bended (see inset of Fig. 1) to excite both transverse magnetic (TM) and transverse electric (TE) modes whereas a standard dipole antenna can only ex- cite the TM modes. The disk is placed on a metallic plate (see inset of Fig. 1), a second plate is covering the 2 FIG. 3. (color online) Spectrum of a single benzene cell. The dotted line corresponds to a spectrum of a perturbed benzene cell, where the lifting of the degeneracy can be seen. tween the discs. Since in reality there is a gap between the top plate and the discs, this can be correct only ap- proximately. But still we use Eq. (1) with an effective γ to be determined from the experiment. A direct coupling to the TM0 and TM1 modes is strongly suppressed due to the continuity conditions of the fields. From Eq. (1) we obtain for the coupling between two disks K0(cid:18)γ(cid:20)rd + κ(d) = κ0(cid:12)(cid:12)(cid:12)(cid:12) 2 (2) d 2(cid:21)(cid:19)(cid:12)(cid:12)(cid:12)(cid:12) where d is the distance between the disks (d=0 means the disks are touching) and κ0 is a constant. The frequency splitting is given by ∆ν(d) = q4κ(d)2 + (∆ν0)2, (3) where ∆ν0 is taking into account a small eigenfrequency difference of the disks and the differing coupling of the antenna to the discs, which actually dominates. The res- onance splitting is shown in Fig. 2 exhibiting an agree- ment with the predicted behavior. Thus each disk brings in just one bound state coupled evanescently to its neigh- bors, exactly the situation found in a tight-binding sys- tem. As a next test of the tight-binding approximation, we performed a measurement with three disks in a row as a function of the distance and found the coupling constant to be consistent with the two disk measurement. The next-nearest-neighbor coupling showed up to be about 6% of the nearest neighbor coupling. As graphene is made of coupled benzene cells we investigated the ben- zene cell (see inset of Fig. 3) next. In Fig. 3 the spectrum (dark solid line) is shown. Benzene has dihedral symme- try D6h with two singlets and two doublets, where the singlets have the extremal energies. Correspondingly we observe four resonances. The degeneracy of the doublets can be lifted by a perturbation, e. g., by moving one disk (see dotted line). Again we performed a measurement varying the distance and found a good agreement with the coupling constant obtained by the two disk measure- ment. FIG. 2. (color online) Splitting of the eigenfrequencies of two disks as a function of the distance caused by evanescent cou- pling. The experimental data (crosses) can be described by a Bessel function K0 (see Eqs (2) and (3)) where κ0 ≈ 1.82 GHz and γ ≈ 0.288 mm−1 and a small offset ∆ν0 ≈ 0.022 GHz. The inset shows the spectrum and splitting for d=3.3 mm. whole setup at height h = 16 mm. The same disks have been used in Ref.15 to find localization in disordered mi- crowave structures. In contrast to previous experiments the top plate is not touching the disks but it is 11 mm above the disc. The distance is chosen such that the reso- nance is still sharp but the eigenfrequency is not sensitive on small deviations of the distance to the top plate. In Fig. 1 the measured spectrum, more precisely 1 − S112, of a single disk is shown. The dark black curve presents the spectrum measured by the antenna shown in the in- set coupling both to the TE and TM modes whereas the light curve was measured with a pure dipole antenna only coupling to the TM mode. The TE01-resonance which is used solely for the further investigations is marked. For all measurements we took care that the resonances below and above the investigated frequency range were still separated from the frequency range of interest. The eigenfrequency of the TE resonance decreased with in- creasing height of the plate, demonstrating that it is a TE01-resonance of the disc. Now we investigate the coupling of two discs. Be- tween the disks the TE1 mode can be described approx- imately by only a z component of the magnetic field ~B = (0, 0, Bz) and perpendicular components of the elec- tric field. Let us assume for the moment that both the bottom and top plates touch the disks from above and below. Then the z dependence can be separated, yielding for the lowest TE-mode close to an individual disk r > rd Bz(x, y, z, k) = B0 ×(cid:26) sin(cid:0) π nh z(cid:1) J0(k⊥r) r < rd α sin(cid:0) π h z(cid:1) K0(γr) where k⊥ = qk2 −(cid:0) π , γ = q(cid:0) π nh(cid:1)2 h(cid:1)2 − k2, and r is the distance from the center of the disc. J0 and K0 are Bessel functions and α is a constant to be determined from the continuity equations at the surface. At the resonance fre- quency of the disk the wave number for r > rd is thus purely imaginary leading to an evanescent coupling be- (1) 3 FIG. 4. (Color online) Spectrum of a graphene flake, where the spectrum was averaged over three different flake forms and different interior antenna positions. The dashed vertical line marks the resonance position of a single disk corresponding approximately to the Dirac point. The dotted horizontal line corresponds to the background. The dashed lines give a guide to the eye to see the linear increase of the DOS. Before we come to the final measurements we have to establish a connection between our measured quantity, the reflection S11, and the DOS or more precisely the LDOS. The starting point is the relation between the Green's function and the local density of states L(E) = − 1 π ImG.17 Here we illustrate a short hand argument the connection between the reflection and the LDOS. The total current through the antenna is given by T = 1 − S112 = H ds j, where j = ImΨ∗∇Ψ is the current density. The wave function of the field excited by a point-like antenna at R is proportional to the Green's function, ψ ∝ G(R, E),18 I ds j = 1 2 I ds (Ψ∇Ψ∗ − Ψ∗∇Ψ) = −B(R; E)2Im G(R; E). (4) where the proportionality coefficient B(R; E) depends on the details of the coupling. Thus finally we find 1 − S11(E)2 = πB(R; E)2L(E), (5) where L(E) is the LDOS of the unperturbed system. To take into account the perturbation by the measurement antenna, in the correct derivation the Green's function G has to be replaced by a properly renormalized Green's function ξβ, where β corresponds to the scattering length of the antenna.18,19 The coefficient B(R; E)2 is defined by the antenna properties which are slowly varying with energy and thus can be set to B(R; ED)2 in the vicinity of the Dirac point ED = k2 D. With L(E) ∼ E − ED ∼ k − kD we expect 1 − S11(E)2 ∼ k − kD. (6) Now we can proceed to the discussion of our exper- imental findings for the graphene layers. We generated different graphene flakes and averaged over flakes and an- tenna positions. By this averaging procedure the LDOS FIG. 5. Spectrum of a graphene flake, where the antenna is placed at the zigzag boundary. The inset shows the actual lattice and the cross marks the position of the measuring an- tenna. The dashed vertical line is the resonance position of a single disc. is turned into the DOS. The result is shown in Fig. 4. At the frequency corresponding to the eigenfrequency a single disk (vertical dotted line) a spectral gap is ob- served which corresponds to the Dirac point in an infi- nite system. Additionally a symmetric linear increase in the DOS is found close to the Dirac point. The overall decrease in 1 − S112 with frequency has its origin in the fact that the measuring antenna is not placed at a disk center, but somewhere in between the discs, leading to a frequency dependent decrease in B at the antenna due to Eq. (1). Now let us have a look what happens if we move the antenna to the outside, e. g. to the zigzag edge. This measurement is presented in Fig. 5. A state close to the Dirac point is observed, which cannot be seen in Fig. 4, even though it is the same system. As the measured quantity is related to the LDOS and not the DOS it is possible to test the states locally and not only globally. We also moved the antenna to the armchair edge and did not observe a state close to the Dirac point. This is in accordance with the findings for a zigzag-edged in- finitely long graphene ribbons where the effect of edge states was discussed,4,20 and where the localization of the edge states was estimated to about one to two atomic lay- ers. If we move the antenna to the lower left corner (see inset Fig. 5) where there is an isolated disk not contained in a full benzene ring, again a single state is observed at the Dirac point. By removing the disk at the corner the state disappears. To interpret our experimental findings we additionally performed numerical calculations by determining eigen- values and eigenvectors of the Hamiltonian H = ED · 1 + κC, (7) where ED is the gap energy, κ the coupling constant, and C the connectivity matrix with elements Cnm = 1 or 0 for nearest and non-nearest neighbors, respectively. The resulting DOS is shown in Fig. 6. The histogram corre- sponds to the experimental findings presented in Fig. 5, 4 can not exist, whereas in the experiment the existing edge states are not observable, due to their small amplitude inside the flake. Finally by removing the corner disk leaving the lattice only with complete benzene cells the corner state did vanished, again in accordance with the experimental findings. In this paper we presented the first microwave real- ization of graphene in the tight-binding approximation. Using high index of refraction disks inside two metal- lic plates, where the disks resonances are only coupled evanescently, a tight-binding realization became possi- ble. The Dirac point and a linear increase in the DOS close to it have been found as well as edge and corner states. It was essential for this approach to conceive a system where each disk brings in just one state coupling evanescently with its neighbors. This is a qualitatively new feature for the microwave experiments as a whole, enabling a new class of experiments in the context of molecular orbital theory, going by far beyond the specific example of graphene studied in the present work. The results for benzene shown in Fig. 3 illustrate for a text- book example the feasibility of the method. Due to the flexibility of the set-up and the closeness of the system additionally graphene like systems with disorder, defects, graphene quantum dots21 or even Dirac oscillators22 can be investigated. By introducing different types of disks with different resonance frequencies it is also possible to investigate systems with different atom per unit celllike boron-nitride, corresponding to a Dirac equation for par- ticles with masses. This work was supported by the Deutsche Forschungs- gemeinschaft via an individual grant and the Forscher- gruppe 760: Scattering systems with complex dynamics. U. K. and S. B. thank the LPMC at Nice for the hospi- tality during different long term visits and the University of Nice and the F´ed´eration Doblin for financial supports. We thank M. Miski-Oglu, T. Seligman, and E. Sadurn´ı for helpful discussions. Towards Nanoscale Photonic Devices (Springer, Berlin, 2005). 9 R. A. Sepkhanov, J. Nilsson, and C. W. J. Beenakker, Phys. Rev. B, 78, 045122 (2008). 10 S. R. Zandbergen and M. J. A. de Dood, Phys. Rev. Lett., 104, 043903 (2010). 11 S. Bittner, B. Dietz, M. Miski-Oglu, P. O. Iriarte, and F. Schafer, Phys. Rev. B, 82, 014301 A. Richter, (2010). 12 C. Bena and G. Montambaux, New J. of Physics, 11, 095003 (2009). 13 U. Kuhl and H.-J. Stockmann, Phys. Rev. Lett., 80, 3232 FIG. 6. (color online) DOS from a numerical calculation of the systems shown in the top right corner (histogram). Ad- ditionally the right inset shows an grey scale indication of an edge state found close to the Dirac frequency. In the left inset an typical eigenfunction of an edge state is presented, where the disk in the lower left corner was removed. The crosses cor- respond to a system with periodic boundary conditions taking 178 disks into account. where the edge states are strongly observed but inter- nal states are suppressed. One observes a peak of the DOS at the Dirac point. Analyzing the wave functions at this energy we found only edge states similar to the one presented in the left inset of Fig. 6 as well as a single corner state observed if an individual disk not part of a full benzene cell is added (see right inset). By moving the measurement antenna toward the center of the system, corner or edge states cannot be seen as their probability inside is very small and the measured quantity is related to the LDOS. In addition we performed the calculation with periodic boundary conditions. Now the edge states have disappeared, and the triangular gap at the Dirac point is clearly seen (crosses in Fig. 6). This is similar to the situation of an internal antenna, i. e., it corresponds to Fig. 4, where the Dirac point is found as well. In the numerics with periodic boundary conditions edge states 1 P. R. Wallace, Phys. Rev., 71, 622 (1947). 2 G. W. Semenoff, Phys. Rev. Lett., 53, 2449 (1984). 3 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). 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 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). 6 A. K. Geim and K. S. Novoselov, Nature Materials, 6, 183 (2007). (1998). 7 C. M. Soukoulis, ed., Photonic Band Gaps and Localiza- tion, NATO ASI Series B: Physics, Vol. 308 (Plenum Press, New York, 1993). 8 J. M. Lourtioz, H. Benisty, V. Berger, J. M. Gerard, and A. Tchelnokov, Photonic Crystals - D. Maystre, 14 A. A. Chabanov, M. Stoytchev, and A. Z. Genack, Nature, 404, 850 (2000). 15 D. Laurent, O. Legrand, P. Sebbah, C. Vanneste, and F. Mortessagne, Phys. Rev. Lett., 99, 253902 (2007). 16 U. Kuhl, F. M. Izrailev, and A. A. Krokhin, Phys. Rev. Lett., 100, 126402 (2008). 17 E. Akkermans and G. Montambaux, eds., Mesoscopic Physics of electrons and photons (Cambridge University Press, Cambridge, 2006). 18 T. Tudorovskiy, U. Kuhl, and H.-J. Stockmann, "Singular statistics revised," Preprint (2009), arXiv:0910.3079. 19 T. Tudorovskiy, R. Hohmann, U. Kuhl, and H.-J. Stockmann, J. Phys. A, 41, 275101 (2008). 20 F. Munoz-Rojas, D. Jacob, J. Fern´andez-Rossier, and J. J. Palacios, Phys. Rev. B, 74, 195417 (2006). 21 F. Libisch, C. Stampfer, and J. Burgdorfer, Phys. Rev. B, 79, 115423 (2009). 22 E. Sadurn´ı, T. Seligman, and F. Mortessagne, New J. of Physics, 12, 053014 (2010). 5
1302.1411	1	1302	2013-02-06T15:51:43	Evolution of variable range hopping in strongly localized two dimensional electron gas at NdAlO3/SrTiO3 (100) heterointerfaces	[ "cond-mat.mes-hall" ]	We report evolution of the two-dimensional electron gas behavior at the NdAlO3/SrTiO3 heterointerfaces with varying thicknesses of the NdAlO3 overlayer. The samples with a thicker NdAlO3 show strong localizations at low temperatures and the degree of localization is found to increase with the NdAlO3 thickness. The T -1/3 temperature dependence of the sheet resistance at low temperatures and the magnetoresistance study reveal that the conduction is governed by a two-dimensional variable range hopping mechanism in this strong localized regime. We attribute this thickness dependence of the transport properties of the NdAlO3/SrTiO3 interfaces to the interface strain induced by the overlayers.	cond-mat.mes-hall	cond-mat	Evolution of variable range hopping in strong ly localized two dimensional electron gas at NdAlO3/SrTiO3 (100) heterointerfaces A. Annadi,1 ,2 A. Putra,1,3 A. Sriva stava,1,2* X. Wang,1,2* Z. Huang,1 ,2* Z.Q . Liu,1,2* T. Venkatesan,1,2,3 and Ariando1 ,2 ,a 1NUSNNI-Nanocore, National University of Singapore, Singapore 117411, Singapore 2Department of Physics, National University of Singapore, Singapore 117542, Singapore 3Department of Electrical and Computer Engineering, National University of Singapore, Singapore 117576, Singapore a)[email protected] We report evolution of the two -dimensional electron gas behavior at the NdAlO3 /SrTiO3 heterointerfaces with varying thicknesses of the NdAlO3 overlayer. The samples with a thicker NdAlO3 show strong localizations at low temperatures and the degree of localizat ion is found to increase with the NdAlO3 thickness. The T -1/3 temperature dependence of the sheet resistance at low temperatures and the magnetoresistance study reveal that the conduction is governed by a two-dimensional variable range hopping mechanism in this strong localized regime . We attribute this thickness dependence of the transport properties of the NdAlO3/SrTiO3 interfaces to the interface strain induced by the overlayers. 1 The discovery of the two dimensional electron ga s (2DEG ) at the LaAlO3/SrTiO3 oxide hetero interface1 has made this material system a primary candidate for applications in oxide electronics. This system ha s further triggered the interest of the community as novel phenomena, such a s superconductivity and magnetism,2-5 were observed at this interface and fur thermore these properties were shown to be sensitive to the external parameters such as O2 deposition pressure6 and strain effects7 at the interface. One of the character istic features of the 2DEG is the overlayer thickness dependence8,9 of its conductivity, where a thickness dependent metal to insulator transition (MIT) transition wa s observed at a critical thickness of LaAlO3 overlayer. Furthermore, a weak to strong localization 10,11 and magnetic Kondo behavior6 were repor ted in thicker samples, suggesting that a variety of conduction mechanisms were operating at these interfaces. Moreover the superconductivity at these interfaces wa s also shown to be sensitive to the thickness2 of the over layer where the localization is suggested to be minimum. Very recently, we repor ted the formation of 2DEG at various polar /nonpolar oxide interfaces12 showing a stronger localization of 2DEG compared to that at the LaAlO3/SrTiO3 interfaces. This stronger localization can possibly give rise to different transport properties, and thu s its dependence on the overlayer thickness is very crucial to be investigated . In this paper we investigate the transport properties of the 2DEG formed at the NdAlO3 /SrTiO3 heterointerface with NdAlO3 overlayer thicknesses of 6, 12 and 16 unit cells (u.c) grown on TiO2 terminated13 SrTiO3 (100) substrates by pulsed La ser deposition (PLD) . T he ba sic difference between NdAlO 3/SrTiO3 and LaAlO3/SrTiO3 interfaces is that the former one offers larger la ttice mismatch of 3.6%. The parameters used for PLD growth process are as follow s: 780 oC substrate temperature, oxygen pressure of 1x10-3 Torr and energy density of the la ser ( = 248 nm) of 1.3 - 1.4 Jcm-2 , with a pulse repetition rate of 1 Hz. During the film growth , the thickness of the 2 NdAlO3 is monitored by in-situ Reflection High Energy Electron Diffraction (RHEED) as shown in Fig. 1a. Figure 1b shows the Atomic Force Microscopy (AFM) topography image a fter the deposition of NdAlO 3/SrTiO3 sample showing well defined step flow surfaces. The RHEED oscillations and the preserved step and terrace stru cture shown by AFM imply the layer -by-layer growth of the overlayers. The electr ical connections to the heterointerface are made by Al wire bonding and the electrical mea surements were performed by Quantum Design Physical Properties Mea surement System (PPMS). Sheet resistance (Rs), carrier density (ns) and mobility ( were obtained by using the Van der Pauw mea surement geometry. Figure 2a shows the Rs variation with temperature for NdAlO3 /SrTiO3 interfaces of var iou s thicknesses. The 6 u.c sample shows a typical metallic behavior (albeit a weak upturn at low temperature) , wherea s thicker samples (12 and 16 u.c) show a metallic behavior (300-70 K) but with strong upturns in Rs with resistance minima at about 70 K. The emergence of the upturns with overlayer thickness observed here is in agreement with previou s repor t s on LaAlO3/SrTiO3.9 However, a significant observation can be made for the low temperature transport for the thicker NdAlO3/SrTiO3 samples, i.e. below 15 K the variation in Rs is more distinct indicating a different type of transpor t mechanism operating at low temperatures. Figure 2b shows the temperature dependenence of ns and  for the corresponding samples. For the thicker samples the ns decrea ses in the temperature range of 300 -20 K, can be attributed to the localization of activated charge carr iers at the interface. The ns in this range can be fit with a simple Arrhenius equation of the form Ae -Ea/T, where Ea is the thermal activation energy for the activated charge carriers. The Ea is of the order of 12 meV which is larger compared to 6 meV for the LaAlO3/SrTiO3 interfaces.14 The large activation and strong localizations observed in 3 thicker samples would be due to the large structural distortions present in these systems arising from the larger interface strain present. Further , there is an increa se in ns at low temperatures below 20 K down to 2 K, while simultaneously an increa se in Rs is observed. For the temperatures below 20 K the increa se in both the Rs and ns implies that the transport mechanism cannot only be simple band (Drude) conduction and indicates that other type of transport mechanism is in operation at low temperatures. Contrary to this, for the 6 u.c sample there is no such carrier activa tion for ns with temperature. Meanwhile the  for the 6 u.c sample increa ses monotonically from 9 cm2V-1s-1at 300 K to 900 cm2V-1s-1 at 2 K. The relatively high mobility of the 6 u.c sample would be favorable for the further studies of superconductivity. On the other hand, for the thicker samples (12 and 16 u .c) the  at 300 K is about 5 cm2V-1s-1 and it drops significantly towards the low temperatures, which is further evidence for strong localization in these samples. The low temperature behavior of Rs, ns and  confirms the strong localization of carriers a ssociated with a change of transpor t mechanism at low temperatures. In general, the observed low temperature sharp upturns in Rs can arise from variou s transport mechanisms namely, Variable Range Hopping (VRH) due to strong localiza tions, a magnetic Kondo scattering and thermal activation. In order to further verify the transport mechanism in our system, the low temperature variation of Rs is analyzed. The observed strong localization of 2DEG at the interface suggests that the most likely mechanism would be the VRH, which is a transport mechanism in which the conduction takes place through a hopping of carriers between localized states. The variation of resistance in the VRH regime can be described by the equation15 R (T)=R0 exp (T0/T)1/(n+1), here n is the dimensionality of the system. Figure 3 shows the corresponding 2D VRH (n = 2) fit for the exper imental data a t low temperatures (40-2 K) . 4 Clearly, the experimental data below 20 K fits well with the VRH formula and moreover the T-1/3 dependence (n = 2) confirms the 2D nature of the conducting channel. The crossover regime between the metallic and VRH (70 -20 K) follows roughly a logarithmic temperature dependence, indicating a weak localization behaviour. On the other hand, the upturns in Rs could or iginate from magnetic Kondo effect following the dependence given by R(T)=R0 ln (T/Te f). However this dependence does not fit our data , indica ting that Kondo scattering is not the governing mechanism here. Moreover the non saturating trend of Rs even up to 2 K also further confirms this. For the thicker samples the ns at 20 K are of the order of ~1012 cm-2 which is one order of magnitude lower than that in the thin 6 u.c sample (Fig . 2b). Hence, it seems that there could be a critical ns at these interfaces below which the transpor t mechanism converts to VRH in the strong localiza tion regime. The reduced ns seen in thicker samples could be due to the large amount of interface strain created by thicker NdAlO3 over layers. The VRH transport mechanism can be further evaluated by magneto transport mea surements, since in the hopping transport the carr ier conduction is quite sensitive to magnetic fields. Magneto-resistance (MR) mea surements performed on the 12 u.c NdAlO 3/SrTiO3 sample at variou s temperatures with different current to magnetic field orientations are shown in Figure 4a which shows the out of plane MR (magnetic field is perpendicular to the current and sample surface) mea sured at various temperatures. The MR is negative at 2 K with -30 % at 9 T. The negative MR is indeed one of the strong signatures for VRH type transport.16, 17, 18 Further, the negative MR diminishes with increa se in temperature and turns to positive MR for higher temperatures. The negative MR reaches a minimum at about 20 K which is exactly the temperature range where the changes in trend are simultaneously seen in Rs and ns. The or igin of 5 negative MR in VRH is attr ibuted to the disruption of the interference effects between forward scattering events in the presence of magnetic field. By considering the interference among the hopping paths between two sites at a distance RM apart, where RM is the optimum hopping length, it is shown that this interference a ffects the hopping probability between these two sites . Inset of Fig. 4a shows the temperature dependence of MR in the negative MR regime at 9 T which scales approximately as T-1 a s expected in the ca se of 2D VRH . 19 The temperature dependence of hopping length (RM) ba sically determines the MR dependence on temperature. Further, the positive MR observed above 20 K ar ises from Lorentz scattering of electrons in presence of magnetic field, which has been extensively studied for the 2DEG LaAlO3/ SrTiO3 interface.10, 20, 21 Figure 4b shows the plot of MR a s a function of magnetic field, where it shows linear B dependence at high magnetic fields and B2-dependence for low fields (shown in inset), which is in accordance with theory and exper iments in VRH regime.19, 22 To exclude the orbital contr ibutions to MR in VRH regime, measurements were further performed in in-plane mode (current and magnetic field are parallel and in the plane of the 2DEG) which is shown in Fig 4 c. In this ca se, it can be seen that MR is negative for the temperatures below 20 K and diminishes and turns to zero for the temperatures above 15 K. The absence of MR above 20 K suggests the absence of Lorentz scattering of electrons in this in-plane geometry. Figure 4d shows the angle dependence of Rs mea sured at 2 K and 9 T. Here 0 degree corresponds to in plane mode and 90 degree corresponds to out of-plane mode. It can be observed that the Rs show large anisotropy between in-plane and out of-plane modes, which is due to the confinement effect of the 2DEG. These MR mea surements further confirm that the transpor t mechanism at low temperatures for the thicker samples is governed by VRH and this VRH regime starts to appear about 20 K which is evidenced by the change in the MR (from positive to negative) . Generally, the hopping 6 transport favors over other mechanisms when a considerable amount of loca lization is present in the system, the carrier confinement would also be crucial in these 2D systems where the absence of lateral transport can further favors the hopping transport. Finally, we comment on the unu sual carrier density recovery at low temperatures. The coincidence of the critica l temperature with the evolution of VRH regime, where the carrier density increa ses a s the tempera ture decrea ses suggests that it may relate to hopping transport. Thu s we suggest that the overall carr ier density behavior could be understood by considering two regimes of transpor t mechanisms where initially the carrier density decrea ses with decreasing temperature due to carrier activation and once the a cr itical lower carr ier density is approached then the carr ier density starts to increase due to var iable range hopping process. The way we can interpret this a s in the hopping transport, some of the localized electrons would become electrically mobile in hopping between nearest neighbour sites, thu s enhances the mobile carr ier density in this regime. In summary, we have shown the thickness dependence of localization effects of 2DEG at NdAlO3/SrTiO3 interfaces. The emergence of 2D VRH is observed at low tempera tures at about 20 K in thicker samples. The VRH transpor t mechanism is further confirmed by a deta iled study of magneto resistance. Our results further empha size the dominant role of overlayer thickness in controlling the transport proper ties of 2DEG at these interfaces through the interface strain effects. Moreover the study presented in this report can certainly a ssist to distinguish the localization from other mechanisms which is very impor tant in understanding the physics of electron transports in two dimensional oxide systems . 7 Acknowledgement We thank the National Research Foundation (NRF) Singapore under the Competitive Research Program (CRP) “Tailoring Oxide Electronics by Atomic Control” NRF2008NRF-CRP002-024, National University of Singapore (NUS) cross -faculty grant and FRC (ARF Grant No. R -144 - 000-278-112) for financial suppor t. 8 References 1A. Ohtomo, and H . Y. Hwang, Nature 427, 423 (2004). 2N. Reyren, S. Thiel, A. D. Caviglia, L. Fitting Kourkoutis, G. Hammerl, C . Richter, C. W. Schneider , T. Kopp, A. -S. Ruetschi, D . Jaccard, M. Gabay, D. A. Muller, J . -M. Triscone, and J. Mannhart, Science 317, 1196 (2007). 3Ar iando, X. Wang, G. Ba skaran, Z. Q . Liu, J. Huijben, J. B . Y i, A. Annadi, A. Roy Barman, A. Rusydi, Y . P. Feng, J . Ding, H . Hilgenkamp, and T. Venkatesan, Nat. Commun. 2, 188 (2011). 4Julie A. Ber t, B . Kalisky, C. Bell, M. Kim, Y. Hikita, and H. Y. Hwang, Na ture Phys . 7, 767 (2011). 5Lu Li, C. Richter, J . Mannhar t, and R. C. Ashoori, Nature Phys .7, 762 (2011). 6A. Brinkman, M. Huijben, M. Van Zalk, J. Huijben, U. Zeitler , J. C. Maan, W. G . Van Der Wiel, G. R ijnders, D. H. A. Blank, and H. H ilgenkamp, Nature Ma ter . 6, 493 (2007). 7C. W. Bark, D. A. Felker , Y. Wang, Y. Zhang, H . W. Jang, C . M. Folkman, J. W . Park, S. H. Baek, H . Zhou, D. D. Fong, X. Q . Pan, E. Y. T symbal, M. S. Rzchowski, and C . B. Eom, PNAS 108, 4720 (2010). 8S. Thiel, G . Hammerl, A. Schmehl, C. W. Schneider, and J. Mannhart, Science 313, 1942 (2006). 9C. Bell, S. Hara shima, Y. Hikita, and H. Y . Hwang, Apl. Phy. Lett. 94, 222111 (2009). 9 10F. J . Wong, R . V. Chopdekar, and Y . Suzuki, Phy. Rev. B 82, 165413 (2010). 11T. Hernandez, C . W. Bark, D. A. Felker , C . B. Eom, and M. S. Rzchowski, Phy. Rev. B 85, 161407 (R) (2012). 12 A. Annadi, A. Putra, Z. Q. Liu, X. Wang, K. Gopinadhan, Z. Huang, S. Dha r, T. Venkatesan, and Ariando, Phy. Rev. B 86, 085450 (2012). 13M. Kawa saki, K. Takahashi, T. Maeda, R. T suchiya, M. Shinohara, O . Ishiyama , T. Yonezawa, M. Yoshimoto, and H. Koinuma, Science 266,1540 (1994). 14M. Huijben, G. Rijnders, D. H. A. B lank , S. Bals, S. V. Aer t, J . Verbeeck, G. V. T endeloo, A. Brinkman, H. Hilgenkamp, Nature Mater. 5, 556 - 560 (2006). 15N. F. Mott: Conduction in glasses containing transition metal ions. J . Non -Crystal. Solids 1,1 (1968). 16V. I. Nguyen, B. Z. Spivak and B . I . Shklovskii, Pis'ma Zh. Ek sp. Teor. Fiz. 41, 35 (1985) JETP Lett. 41, 42 (1985). 17V. I . Nguyen, B . Z. Spivak and B. I. Shklovskii, Zh. Ek sp. T eor . Fiz. 89, 1770 (1985) Sov. Phys. JETP 62, 1021 (1985). 18U. Sivan, O. Entin-Wohlman, and Y. Imry, Phys. Rev. Lett. 60, 1566 (1988). 19W. Schirmacher, Phys. Rev. B 41 2461 (1990). 20X. Wang, W. M. Lü, A. Annadi, Z. Q. Liu, K. Gopinadhan, S. Dhar, T. Venkatesan, and Ariando. Phy. Rev. B. 84, 075312 (2011). 21M. Ben Shalom, C . W. Tai, Y . Lereah, M. Sachs, E. Levy, D. Rakhmilevitch, A. Palevski, and Y. Dagan, Phy. Rev. B 80, 140403(R) (2009). 22K. Schlenga, H. Bach and K. Westerholt. Physica C 221 , 161 (1994). 10 Figure captions: FIG. 1. (a) RHEED oscillations obtained during the growth of the NdAlO3/SrTiO3 (100) samples (b) AFM image of the NdAlO3/SrTiO3 sample showing the step flow surface indicating the layer by layer growth of the NdAlO3 films. FiG. 2. Temperature dependence of (a) Sheet resistance, Rs, (b) Charge carrier density , ns, and mobility,  for NdAlO3/SrTiO3 samples with different thicknesses. FIG. 3. The ln(Rs) vs. (1/T)1/3 graph for12 and 16 uc NdAlO 3/SrTiO3 samples, a 2D variable range hopping (VRH ) fit. FIG. 4. (a) Out-of-plane MR mea sured at different temperatures for 12 uc NdAlO3/SrTiO3 sample. Inset: scaling of MR a t 9 T with temperature for negative MR part. (b) MR (Out of plane) mea sured at 2 K with magnetic field showing linear variation at high magnetic fields. Inset: B2 dependence for low magnetic fields. (c) In-plane MR mea sured at different temperatures. (d) Angle dependence of Rs at 2 K and 9 T . 11 Annadi et al., FIG. 1 12 Annadi et al., FIG. 2 13 11010010121013101410151016 Temperature (K)10-410-2100102104 (cm2 V-1s-1) ns (cm-2) (b)110100100101102103104105106 Rs ()Temperature (K) 6 u.c 12 u.c 16 u.c(a) Annadi et al., FIG. 3 14 0.30.40.50.60.70.8810121416 12 u.c 16 u.cln (Rs )(1/T)1/3 Annadi et al., FIG. 4 15
1101.5517	1	1101	2011-01-28T11:50:05	THz lasing in a polariton system: Quantum theory	[ "cond-mat.mes-hall", "quant-ph" ]	We study the laser regime of terahertz (THz) emission from a semiconductor microcavity in the strong coupling regime, where optical transitions between upper and lower exciton-polariton modes are allowed due to the mixing of the upper mode with one of the dark exciton states. Using a system of master-Boltzmann equations describing both polariton modes and the THz mode, we calculate the first and second order coherences and the spectral shape of THz emission. This analysis shows that THz lasing in microcavities is possible provided that the system is embedded in a good THz cavity, that the optical (polariton) lasing condition is fulfilled and if the depletion of the upper polariton mode due to acoustic phonon assisted relaxation processes is reduced. This latter condition is likely to be realised in pillar microcavities, which seem to be the most suitable candidates for realisation of a THz laser.	cond-mat.mes-hall	cond-mat	THz lasing in a polariton system: Quantum theory School of Physics and Astronomy, University of Southampton, SO17 1BJ, Southampton, United Kingdom Elena del Valle and Alexey Kavokin (Dated: March 31, 2021) 1 1 0 2 n a J 8 2 ] l l a h - s e m . t a m - d n o c [ 1 v 7 1 5 5 . 1 0 1 1 : v i X r a We study the laser regime of terahertz (THz) emission from a semiconductor microcavity in the strong cou- pling regime, where optical transitions between upper and lower exciton-polariton modes are allowed due to the mixing of the upper mode with one of the dark exciton states. Using a system of master-Boltzmann equations describing both polariton modes and the THz mode, we calculate the first and second order coherences and the spectral shape of THz emission. This analysis shows that THz lasing in microcavities is possible provided that the system is embedded in a good THz cavity, that the optical (polariton) lasing condition is fulfilled and if the depletion of the upper polariton mode due to acoustic phonon assisted relaxation processes is reduced. This latter condition is likely to be realised in pillar microcavities, which seem to be the most suitable candidates for realisation of a THz laser. Realisation of efficient terahertz (THz) radiation sources and detectors is one of the important objectives of modern applied optics [1, 2]. THz radiation based methods have a lot of potential applications in biology, medicine, security and non-destructive in-depth imaging. Also, wireless data transfer utilising THz radiation could provide higher transfer rates for in-door short distance or high altitude communications. To allow such applications of THz radiation, the creation of cheap, reliable, scalable and portable emitters is extremely important. None of the existing ones so far satisfies all the application requirements. For example, the emitters based on nonlinear-optical frequency down-conversion, gas THz laser, vacuum tube and systems based on short-pulse spectroscopy are bulky, expensive, and power consuming. Various semicon- ductor devices based on intersubband optical transitions [3] are compact but have a limited wavelength adjustment range, have low quantum efficiency and require cryogenic cooling. Among the factors which limit the efficiency of semiconduc- tor THz sources is the short lifetime of the electronic states in- volved (typically, fractions of a nanosecond) compared to the time for spontaneous emission of a THz photon (typically mil- liseconds). The methods of reducing this mismatch include the use of the Purcell effect in THz cavities or the cascade effect in quantum cascade lasers [4 -- 6] (QCL). Nevertheless, till now the QCL in the spectral region about 1THz remains costly and short-lived and still show a quantum efficiency of less than 1%. Moreover, so far there are no commercially available reliable compact and cheap detectors of THz radia- tion which are in great need for information communication technologies. Recent studies of strong coupling intersubband microcav- ities have shown the possibility of stimulated scattering of intersubband polaritons [7]. Very recently, it was proposed to generate THz radiation in semiconductor microcavities in the regime of exciton-polariton lasing [8]. The quantum ef- ficiency of this THz source is governed by population of the final polariton state, which may be tuned over a large range by means of the optical pumping. In the strong coupling regime in a microcavity [9], the dispersion of exciton-polaritons is described by two bands both having minima at zero in-plane wave vector k. At k = 0, the energy splitting between the two branches is of the order of several meV, which makes this system attractive for THz applications (1THz corresponds to FIG. 1. Scheme for the THz laser in a microcavity embedded in a THz cavity. Upper and lower polariton branches (dispersion rela- tion) are depicted in blue and pink, respectively, with the mechanisms considered for all the modes (a, b, c): decay (γ, γc), non-resonant continuous pump (electronic injection) on the upper branch (P) and upper to lower polariton relaxation via exciton reservoir (γab). THz photons are emitted (absorbed) during the upper to lower (lower to upper) polariton conversion at rate G. 4meV). Stimulated scattering of exciton polaritons into the lowest energy state leads to so-called polariton lasing, re- cently observed in GaAs [10] and GaN [11] based microcav- ities. If the scattering from the upper to the lower polariton branch were accompanied by the emission a photon, polari- ton lasers would emit THz radiation, and this emission would be stimulated by the population of the lowest energy polariton state. However, this process is a priori forbidden since an opti- cal dipole operator cannot directly couple the polariton states formed by the same exciton state. This obstacle can be re- moved if one of the polariton states of interest is mixed with an exciton state of a different parity, say, the e1hh2 exciton state formed by an electron at the lowest energy level in a quantum well (QW) and a heavy hole at the second energy level in the QW [8]. This state is typically a few meV above the exciton ground state, e1hh1. Nevertheless, by an appropriate choice of the QW width and exciton-photon detuning in the micro- cavity, the state can be brought into resonance with the lowest b = lower polaritons c = THz photons € γc€ γ€ γ€ P€ γab€ Ga = upper polaritons energy upper polariton state. Being resonant, the two states can be easily hybridised by any weak perturbation, such as, e.g., a built-in or applied electric field. The optical transition between such a hybridised state and the lowest e1hh1 exciton polariton state is allowed. In this Letter, we describe theoretically the quantum prop- erties of THz light generated by polariton lasers according to the above scheme. We consider the model system shown in Fig. 1. It consists of a microcavity with embedded quantum wells operating in the strong coupling regime, and placed in- side a THz cavity. Together with the wave-guiding effect of the microcavity structure, adding the lateral THz cavity would achieve an effective 3D confinement of the THz mode, giving rise to enhancement of spontaneous emission rate through the Purcell effect [12 -- 14]. We will show that in the polariton las- ing regime, this device also operates as a THz laser. Moreover, polariton lasing may be triggered by an external THz radia- tion which would stimulate scattering of polaritons between the upper and lower modes in the microcavity. Therefore, the proposed device could also operate as a detector of THz radi- ation. The two polariton branches at k = 0 are modelled by two Bose fields with annihilation operators a (upper) and b (lower). Both modes (being at resonance) lose particles at the same rate γ. This rate provides the units of the problem (for standard microcavities, γ ≈ 0.1ps−1). In Fig. 1, the sys- tem is sketched together with the relevant parameters. Only one of the modes, a, receives incoherently polaritons at a rate P due to electronic injection. Upper polaritons can convert into lower polaritons by emitting a THz photon (with oper- ator c) at a rate G. Energy is conserved in this process as Ec = Ea − Eb. The opposite process happens with the same probability G since the THz photons, once emitted, remain in the cavity long enough to make the transformation reversible. The THz cavity is not perfect and does, however, lose particles at a rate γc. There is a last relaxation process that must be in- cluded: the conversion of upper polaritons into lower ones via the exciton reservoir at rate γab, through irreversible phonon emission. This process is important in planar microcavities and is unfavorable for THz generation, as will be shown be- low. Fortunately, the acoustic phonon induced depletion of the upper polariton mode may be strongly suppressed in micropil- lars [10]. All these incoherent processes can be described as Lindblad terms (L ) in the master equation of the total density matrix of the system, ρ: ∂tρ = (cid:104)γ 2 (cid:17) (cid:16)La + Lb (cid:16)L + γc 2 (ab†c†) + L G 2 Lc + La† + γab 2 L (ab†) (1a) (cid:17)(cid:105) P 2 ρ , + (a†bc) (1b) where LOρ ≡ 2OρO† − O†Oρ − ρO†O. All the irreversible processes are gathered in Eq. (1a), that is, decay, non-resonant pump and relaxation processes. Eq. (1b) represents the non- linear processes of upper to lower polariton conversion and THz photon emission (and vice versa) similarly to the evapo- rative cooling processes in an atom laser [15 -- 17]. The Hilbert space of the three modes is spanned by the number states { n,m,r(cid:105) with n,m,r ∈ N} the number of upper 2 FIG. 2. Relevant quantities in the steady state as a function of pump, for the THz field (x = c in blue, circles), the upper polaritons (x = a, in purple, squares) and the lower polaritons (x = b in brown, rhom- bus). (a) Average populations nx. (b) Second order coherence func- tions g(2) (d) x Linewidth of the THz spectrum of emission. Parameters are: G = γ, γab = 0 and γc = 0.1γ. (c) Cross correlation functions g(2) ac and g(2) ab . bc , g(2) . polaritons, lower polaritons and THz photons, respectively. The occupation of these states described by the probability function P[n,m,r] ≡ (cid:104)n,m,r ρ n,m,r(cid:105) follows the set of cou- pled quantum Boltzmann master equations [18]: + γ γ(n + m) + γcr + P(n + 1) ∂tP[n,m,r] = −(cid:110) (cid:111)P[n,m,r] (cid:111) (cid:110) (n + 1)P[n+1,m,r] + (m + 1)P[n,m+1,r] (cid:111) (cid:110) + γc(r + 1)P[n,m,r+1] + PnP[n−1,m,r] (cid:111) (cid:110)P[n+1,m−1,r−1] − P[n,m,r] (n + 1)mP[n+1,m−1,r] − n(m + 1)P[n,m,r] (cid:110)P[n−1,m+1,r+1] − P[n,m,r] (cid:111) + Gn(m + 1)(r + 1) + G(n + 1)mr + γab . (2) x = (cid:104)x†x†xx(cid:105)/n2 In what follows, we solve these equations numerically in the steady state of the system. Then, any average single-time quantity can be computed as (cid:104)O(cid:105) = Tr(ρO) = ∑n,m,r P[n,m,r](cid:104)n,m,r O n,m,r(cid:105). The relevant quantities characterizing each mode (x = a,b,c), that we use to study lasing emission and correlations are: the average occupation numbers nx = (cid:104)x†x(cid:105); the second order temporal correlation function at zero delay, g(2) x, that is 2 for a ther- mal and 1 for a Poissonian distribution; the cross correlation xy = (cid:104)x†xy†y(cid:105)/(nxny). function between two modes x and y, g(2) These quantities are plotted in Fig. 2(a,b,c) as a function of the pumping rate P. Let us explain the choice of parameters of this figure, that brings the system in the THz lasing regime. The first condition for THz lasing is the accumulation of a large population of THz photons (nc (cid:29) 1), for which a strong nonlinearity and a good THz cavity are required. Further- c → 1). more, this should also lead to a coherence buildup (g(2) Assuming that other destructive issues (as pure dephasing) are negligible in the system, the efficiency of such coher- ence buildup is further enhanced by the irreversibility of the polariton-to-photon conversion [17]. The conversion should happen effectively mainly in the polariton-to-photon sense so that the formation of correlations is not disrupted by the in- verse process. However, if, after one upper polariton has transformed into a THz photon plus a lower polariton, these two last remain for a long time in the cavity, the inverse pro- cess is more likely to happen. Given that, at the same time, we wish the THz population to remain high, the only way to suppress the photon-to-polariton process is that lower polari- tons do not remain for long times in the cavity, that is, we need γ (cid:29) γc. Therefore, we have chosen G = γ and γc = 0.1γ, which allows for an accumulation of photons faster than of po- laritons, as shown in Fig. 2(a). For the moment, we have also neglected the competing process through the exciton reservoir (γab = 0) that does not produce THz radiation and clearly hin- ders lasing. In these conditions, Fig. 2(a) shows a nonlinear increase of the THz population (circles) which is accompanied c → 1 in Fig. 2(b). Note that there is not by the expected g(2) a sharp threshold for the lasing emission. The THz photons gradually and monotonously increase and acquire coherence. One can in any case define the threshold as the pump where nc = 1, that is, P ≈ 0.22γ in this case. To link the present full-quantum master equation approach with the kinetic model [8], we derive from Eq. (1) the steady state equations for the average populations, that are of the type of a Boltzmann source and sink rate equation, i.e., nx = Peff x /Γeff x bc nbnc ab nb + g(2) ab nb) + G(1 + g(2) (4a) ac nc) (4b) (4c) (3) for x = a,b,c. This provides us with the effective pumping and intensity decay rates for the three modes: a = P + Gg(2) Peff a = γ − P + γab(1 + g(2) Γeff b = [γab + G(1 + g(2) Peff b = γ − γabg(2) Γeff c = G(1 + g(2) Peff c = γc + G(g(2) Γeff (4f) The semiclassical rate equations of Ref. [8] are recovered by setting all cross correlation functions to unity. Solving exactly these rate equations, one obtains approximated formulas for the populations that, in turn, provide analytical expressions for all the effective parameters above. The expressions are too , Prate lengthy to be given here but we still note them as nrate and Γrate ac nc)]na ab na + G(g(2) ab nb)na bc nb − g(2) for later comparison. bc nc − g(2) (4d) (4e) ac na) . ab na) x In our full quantum derivation, cross correlation functions are computed numerically and self-consistently and can be, in general, larger (resp. smaller) than unity, showing bunching (resp. antibunching) of joint emission in the modes x and y. This is shown in Fig. 2(c), where cross emission between upper polaritons and THz photons is antibunched (squares, g(2) ac < 1), as the destruction of the first one implies the cre- ation of the second ones. This also tends to be the case be- tween upper and lower polaritons, however, in this case due to x x 3 FIG. 3. Relevant THz quantities in the steady state as a function of pump: nc (first line), g(2) (second line) and Γc (third line). The first c column (plots a, b, c) corresponds to fixing γc = 0.1γ and increasing γab in the sense indicated by the arrow (from 0 to 0.5 in steps of 0.05). The second column (plots d, e, f) corresponds to fixing γab = 0 and increasing γc in the sense indicated by the arrow (from 0.1 to 0.9 in steps of 0.1). the large difference nb (cid:28) nc, they have become independent ab ≈ 1). The cross emission of lower polaritons (rhombus, g(2) and THz photons is bunched (circles, g(2) bc > 1) for the same reason: they are produced at the same time and their emission tends to be simultaneous. At very high pumps all the emis- sions become statistically independent (g(2) xy → 1). Another typical manifestation of lasing is line narrowing of the luminescence emission, that is, the increasing lifetime of the photons emitted by the lasing mode: in the incoher- ent regime, the linewidth corresponds to the inverse lifetime of a single (incoherent) photon, whereas in the lasing regime, THz photons form a collective state with a lifetime affected by the total population. The effective inverse lifetime of the THz mode is given approximately by Eq. (4f) in the limit of inde- pendent classical modes. Γeff c decreases appreciably when G bc nb − g(2) is large and g(2) ac na decreases with pumping. In the ac ≈ 1 and na, nb are small, the main case of Fig. 2, where g(2) element that determines the linewidth is the cross correlation function g(2) bc , which is large (bunched) but strongly decreasing with pump. To check that this process is accompanied by the expected line narrowing of the emission, predicted by Eq. (4f) in the regime of Fig. 2, we compute the photoluminescence spec- trum of THz emission. It is defined, in the steady state, as 0 (cid:104)c†(0)c(τ)(cid:105)eiωτdτ and is computed (with the S(ω) ∝ ℜ(cid:82) ∞ quantum regression formula) from the two-time correlator: (cid:104)c†(0)c(τ)(cid:105) = ∑ n,m,r √ rW[n,m,r](τ) (5) in terms of W[n,m,r], that follow the same equation as the off- diagonal density matrix elements (cid:104)n,m,r ρ n,m,r−1(cid:105), with √ rP[n,m,r]. The required equations initial values W[n,m,r](0) = read: ∂τ W[n,m,r] = −(cid:110) (cid:110) + γ (cid:111)W[n,m,r] (cid:111) (cid:111) γ(n + m) + γc(r− 1 2 ) + P(n + 1) (n + 1)W[n+1,m,r] + (m + 1)W[n,m+1,r] (cid:110) + γc (n + 1)mW[n+1,m−1,r−1] − n(m + 1)W[n,m,r] (cid:112)r(r + 1)W[n,m,r+1] + PnW[n−1,m,r] (cid:110)(cid:112)r(r− 1)W[n+1,m−1,r−1] − (r− 1 (cid:110)(cid:112)r(r + 1)W[n−1,m+1,r+1]− (r + 2 )W[n,m,r] )W[n,m,r] 1 2 (cid:111) (cid:111) . (6) + γab + G(n + 1)m +Gn(m +1) The THz spectrum of emission consists of only one peak. The associated linewidth is extracted and plotted in Fig. 2(d). We find that, although not quantitatively exact, the trend (narrow- ing) follows remarkably that predicted by the effective THz intensity decay rate, Eq. (4f). Finally, in Fig. 3, we analyse how two key factors influence the quality of the THz lasing. Still as a function of pump, we show how THz population (first row), coherence g(2) (second c row) and its spectral linewidth, Γc, (third row) are spoiled as, on the one hand, incoherent relaxation from upper to lower polaritons through the reservoirs (i.e., not through the THz mode) is increased (left column), and, on the other hand, as the THz cavity Q factor is decreased (right column). The sharpest feature to tell when lasing disappears is the slope of the linewidth curve, that becomes positive at all pumps. This happens at γab ≈ 0.2γ in the first case and at γc ≈ 0.7γ in the second. These are the minimum values required to achieve the THz lasing with all its features. The analytical approxi- mated expression for the linewidth, Γrate , is only valid for a bad THz cavity, when the system is in the "weak coupling regime" (γc > G, γ) with the THz mode (see Fig. 4). Here, the rate equations, that neglect quantum correlations, give a good description of the dynamics. With better systems able to enter the lasing regime, one must solve the full quantum problem, including one- and two-time correlators, in order to retain important qualitative features of the linewidth such as line narrowing. c 4 To conclude, our analysis shows that although an efficient THz lasing is indeed possible in optically pumped polariton systems, it requires a special care in the design of the struc- ture. One must ensure a good THz cavity and small deple- tion of the k = 0 state of the upper polariton mode due to the phonon-assisted relaxation of polaritons from this state to the states of the lower polariton mode characterised by large in- plane wavevectors k. The latter obstacle is the most serious as upper polariton depletion is usually very fast in planar cavi- ties. This can be circumvented by the use of pillar microcavi- ties [10], where a full three-dimensional photonic confinement is achieved. In pillars, the spectrum of exciton-polaritons con- sists of discrete states, so that there is no reservoir of states FIG. 4. Difference between the numerically computed THz emis- sion linewidth (Fig. 3c,f) and the analytical approximated formula c − Γc)/Γc. Pumping is obtained from the rate equations: 100(Γrate fixed at P = 0.451γ. with large wavevectors. Optical transitions between exciton polariton states in the pillars are usually forbidden, but if one of these states is brought into resonance with the e1hh2 or e2hh1 exciton state, the transition between this state and po- lariton states produced by coupling of e1hh1 excitons with light becomes possible with emission of a THz photon. Also, a pillar microcavity is certainly a more appropriate structure to be fitted in still another cavity, the THz one, needed for THz lasing. Acknowledgments -- We acknowledge fruitful discussions with M. Kaliteevski, K. Kavokin, I. Shelykh and F. P. Laussy. This research is supported by the Newton International Fel- lowship program. [1] D. Dragoman and M. Dragoman, Progr. in Quant. Electronics World 43, 90 (2007). [2] G. Davies and E. Lienfield, Physics World 17, 37 (2004). [3] Q. Hu, B. Williams, S. Kumar, H. Callebaut, S. Kohen, and J. Reno, Semicond. Sci. Technol. 20, S228 (2005). [4] R. F. Kazarinov and R. A. Suris, Sov. Phys. Semiconductors 5, 28, 1 (2004). 707 (1971). [7] S. De Liberato and C. Ciuti, Phys. Rev. Lett. 102, 136403 (2009). [8] K. Kavokin, M. Kaliteevski, R. Abram, A. Kavokin, S. Sharkova, and I. Shelykh, Appl. Phys. Lett. 97, 201111 (2010). [9] A. Kavokin, J. J. Baumberg, G. Malpuech, and F. P. Laussy, [5] J. Faist, F. Capasso, D. L. Sivco, C. Sirtori, A. L. Hutchinson, Microcavities (Oxford University Press, 2007). and A. Y. Cho, Science 264, 553 (1994). [6] E. Normand, I. Howieson, and M. T. McCulloch, Laser Focus [10] D. Bajoni, P. Senellart, E. Wertz, I. Sagnes, A. Miard, A. Lemaıtre, and J. Bloch, Phys. Rev. Lett. 100, 047401 (2008). 40%30%20%10%0%[11] J. Levrat, R. Butte, T. Christian, M. Glauser, E. Feltin, J.-F. Carlin, N. Grandjean, D. Read, A. V. Kavokin, and Y. Rubo, Phys. Rev. Lett. 104, 166402 (2010). [12] E. M. Purcell, Phys. Rev. 69, 681 (1946). [13] J.-M. Gerard and B. Gayral, Journ. Lightwave Technol. 17, [14] Y. Todorov, I. Sagnes, I. Abram, and C. Minot, Phys. Rev. Lett. 2089 (1999). 99, 223603 (2007). [15] M. Holland, K. Burnett, C. Gardiner, J. I. Cirac, and P. Zoller, Phys. Rev. A 54, R1757 (1996). Rev. Lett. 93, 016402 (2004). [16] F. P. Laussy, G. Malpuech, A. Kavokin, and P. Bigenwald, Phys. [17] E. del Valle, D. Sanvitto, A. Amo, F. P. Laussy, R. Andr´e, C. Tejedor, and L. Vina, Phys. Rev. Lett. 103, 096404 (2009). [18] C. W. Gardiner and P. Zoller, Phys. Rev. A 55, 2902 (1997). 5
1201.6329	2	1201	2012-05-22T13:54:20	Theory of STM junctions for \pi-conjugated molecules on thin insulating films	[ "cond-mat.mes-hall" ]	A microscopic theory of the transport in a scanning tunnelling microscope (STM) set-up is introduced for \pi-conjugated molecules on insulating films, based on the density matrix formalism. A key role is played in the theory by the energy dependent tunnelling rates which account for the coupling of the molecule to the tip and to the substrate. In particular, we analyze how the geometrical differences between the localized tip and extended substrate are encoded in the tunnelling rate and influence the transport characteristics. Finally, using benzene as an example of a planar, rotationally symmetric molecule, we calculate the STM current voltage characteristics and current maps and analyze them in terms of few relevant angular momentum channels.	cond-mat.mes-hall	cond-mat	Theory of STM junctions for π-conjugated molecules on thin insulating films Sandra Sobczyk, Andrea Donarini,∗ and Milena Grifoni Institut fur Theoretische Physik, Universitat Regensburg, 93040 Regensburg, Germany (Dated: November 2, 2018) A microscopic theory of the transport in a scanning tunnelling microscope (STM) set-up is in- troduced for π-conjugated molecules on insulating films, based on the density matrix formalism. A key role is played in the theory by the energy dependent tunnelling rates which account for the coupling of the molecule to the tip and to the substrate. In particular, we analyze how the geo- metrical differences between the localized tip and extended substrate are encoded in the tunnelling rate and influence the transport characteristics. Finally, using benzene as an example of a planar, rotationally symmetric molecule, we calculate the STM current voltage characteristics and current maps and analyze them in terms of few relevant angular momentum channels. PACS numbers: 85.65.+h, 68.37.Ef, 73.63.-b I. INTRODUCTION Scanning tunnelling microscopy (STM) is an impor- tant tool for imaging surface structures and for studying the electronic properties of individual molecules since its introduction by Binnig and Rohrer1,2. Various authors have developed theories of STM3 -- 8,10 -- 15, among those the famous ones published by Tersoff and Hamann4,8,9 in the 1980s. Their work is the basic theory used to ex- plain STM images without atomic resolution16, i.e. STM images with characteristic feature sizes of ≥ 1nm, for ex- ample the scattered waves of surface states, as well as ad- sorbates, defects and substitution atoms on the surface17. Tersoff and Hamann showed that those experiments, as those on reconstructed Au surfaces, may have a simple explanation. In their articles the tip was modeled as a spherical potential well of radius R = 9 A, taking the s- wave solution of the macroscopic Schrodinger equation to describe the electronic tip-state. With Bardeen's pertur- bation theory of tunnelling18, they showed that the STM image is approximately the Fermi-level local density of states (LDOS) contour of the sample at the center of the sphere. Though the Tersoff-Hamann approach can- not be used to explain famous STM experiments that show atomic resolution, because it ignores the detailed structure of the tip wave functions. For true atomic res- olution, for which the length scale is much smaller than one nanometer, the convolution of tip states and sam- ple states must be taken into account19. Chen presented an extension of the Tersoff-Hamann theory that implies more detailed tip-models and allows to interpret higher resolution STM images11,20,21. Several other authors sug- gested that atomic resolution demands small tip-sample distances6,10,22, which are not fully described within the Bardeen tunnelling theory18. In fact the majority of the STM studies of single molecules, in experiment and in theory, has so far been limited to molecules on metals or semiconductors. In these cases the electronic properties of an individual molecule are strongly perturbed by the presence of the substrate electrons. In order to understand the electronic properties of an individual molecule, an electronic decou- pling from the supporting substrate is desirable. Hence, in the seminal experiments [23,24], STM measurements have been performed on molecules on insulating films having a thickness of only few atomic layers. The layer is in turn grown on top of a metallic substrate. This set- up allows to electronically decouple the molecule from the metallic surface, so that electronic properties of in- dividual molecules can be studied. At the same time the electrons can still tunnel through the insulating films, fa- cilitating imaging with the low-temperature STM at a low tunnelling current. In this work we present an STM theory that enables to study the transport properties of individual π-conjugated molecules in the latter STM configuration. We model the device with a double-barrier tunnelling set-up, and treat its dynamics in the sequential tunnelling limit via a density matrix approach. We show that the geometri- cal aspects in the coupling to the substrate and the tip, results into significantly different, energy dependent tun- nelling rates. Using benzene as an example, we calculate current voltage characteristics and constant height cur- rent maps for different biases and substrate work func- tions, thus simulating STM images with atomic resolu- tion. Due to the rotational symmetry of the benzene molecule we express the theory in the angular momen- tum basis, and we prove that the tunnelling dynamics from/to the extended substrate is described by angular momentum channels. Vice versa, the localized tip mixes, in the tunnelling events, the angular momentum states of the molecule. This mixing produces, for specific sub- strate work functions, negative differential conductance and current blocking also detectable in the topography of the STM surface plots. Both the Pauli and the generalized master equation have been repeatedly used in the modelling of STM junctions25 -- 30. Nevertheless, to our knowledge, STM junctions with a thin insulating layer have not been sys- tematically studied within the framework of the general- ized master equation. This paper is outlined as follows: in section II we present a general transport theory for π-conjugated molecules in the STM set-up. We introduce the model Hamiltonian of the system and provide a detailed analysis of the tunnelling dynamics in terms of energy dependent tunnelling rates. In section III we apply the theory to a benzene molecule. The corresponding current voltage characteristics and current maps are discussed in section IV. Finally, conclusions and remarks are presented in sec- tion V. a) II. LOW ENERGY THEORY OF STM ON INSULATING LAYERS b) z0 0 d ztip 2 z x y zend z T εF T +ε0 T ε0 T Φ0 tip A. Hamiltonian and tunnelling amplitudes A scanning tunnelling microscopy (STM) set-up with a thin insulating film involves the STM tip, the substrate and the molecule (Fig. 1a), weakly coupled to each other. Therefore we can describe the whole system by the total Hamiltonian H = Hm + Hsub + Htip + Htun . (1) The first term gives the Hamiltonian of an arbitrary π- conjugated molecule. We assume that only the π-orbitals contribute to transport. Thus, to each atom is assigned only one orbital (the 2pz orbital orthogonal to the plane of the molecule), while the entire σ backbone is included only via the parametrization of the Hamiltonian for the π-conjugated electrons. The latter, written in the atomic basis, is a simplified version of the Pariser-Parr-Pople (PPP) Hamiltonian31,32, expressed in terms of the non- interacting Huckel-Hamiltonian33 and a constant inter- action term: Hm =Xασ + aαd†ασdασ + Xα6=βσ U (N − N0)2 , 1 2 bαβd†ασdβσ+ (2) where d†ασ creates an electron of spin σ in the pz-orbital of the atom α, and α = 1, ..., M runs over the M atoms of the molecule. The hopping energies bαβ are assigned using the Slater-Koster method34 with atomic param- eters and geometrical configurations obtained from the literature. The on-site energy for the atom α is de- noted by aα and can also vary from atom to atom. Fi- nally, the constant interaction model35 assumes that the Coulomb interaction between the electrons is parameter- ized by a constant capacitance C, what is finally defin- ing the Coulomb interaction U = e2 2C , where e is the charge quantum. This model also assumes that the dis- crete single-particle energy spectrum is unaffected by the interactions. Finally, N =Pασ d†ασdασ counts the num- ber of π-electrons in the molecule which is N0 for the neutral case. The simplicity of the Hamiltonian for the molecule pre- sented here allows to carry out most of the calculations (specifically the ones relative to benzene presented in sec- tions III and IV) at an analytic level since the many-body S +ε0 εF S S ε0 S Φ0 LUMO HOMO substrate molecule FIG. 1: (Color online) Panel (a) - Sketch of the investigated STM set-up. A π-conjugated molecule, here benzene, is sep- arated by a metal substrate (yellow) only through an ultra- thin insulating film (red). A bias voltage is applied between the substrate and the tip. Panel (b) - Schematic illustration for the sum of the potentials of the substrate, the molecule and the tip v = vsub + vm + vtip, along the z direction. We choose the energy of the vacuum between the molecule and the tip, as well as the energy of the tunnelling barrier be- tween molecule and substrate to be zero. The energies at the bottom of the conduction band of tip and substrate are 0 = −ΦS/T εS/T are the Fermi energies mea- sured from the band bottom and ΦS/T are the work functions for the tip and the substrate. The work functions are shifted by the applied bias voltage. , where εS/T −εS/T F 0 F 0 eigenstates of the interacting Hamiltonian coincide, in this case, with the ones of the non interacting one. Nev- ertheless, the transport theory is not affected by the par- ticular choice of the Hamiltonian for the molecule and the transport characteristics remain qualitatively the same for the different models, as far as the symmetry of the states is preserved. We consider the tip and and the substrate as reservoirs of non interacting electrons. In particular, we describe the metallic substrate as a potential well (see Fig. 1b) with no confinement in the x and y direction. The asso- ciated Hamiltonian Hsub reads Hsub =X~kσ εS ~k c† S~kσ cS~kσ, (3) 0 + 2~k2 = εS 2m with ~k = (kx, ky, kz) and c† where εS ~k S~kσ creates an electron of momentum ~k and spin σ in the substrate and z0 is the z extension of the substrate (see Fig. 1). The continuous choice also for the z component of the momentum is justified in the limit z0 ≫ λF where λF is the Fermi wave length of the substrate. Only bound states (εS ~k explicit wave function is given in the Appendix A. < 0) are considered in the calculation and their An analogous shallow square potential for the z direc- tion describes the metallic tip. A parabolic confinement in the x and y direction is though added to the model to simulate the spatial localization of the tip states. The tip Hamiltonian reads: Htip =Xkz σ kz c†T kz σcT kzσ , εT (4) = εT 0 + ω + 2k2 2m and c†T kzσ creates an electron where εT kz with momentum kz, spin σ, and in the ground state with respect to the lateral confinement. z We are confident that the particular choice of the con- finement for the tip Hamiltonian is not crucial for the results. Nevertheless, as it has already been theoretically predicted11 and experimentally observed36, the symme- try of the tip is very important. We will restrict in this work to tip wave functions which are rotationally invari- ant with respect to an axis perpendicular to the surface of the substrate. The last term of Eq. (1) is the tunnelling Hamiltonian. It contains two parts: one for the substrate-molecule tun- nelling, the other for the tip-molecule tunnelling: Htun = Xχkiσ tχ kic†χkσdiσ + h.c. . (5) The index i denotes the molecular orbital, i.e. the lin- ear combination of the atomic pz orbitals introduced in Eq. (2), χ = S, T indicates the substrate or the tip and we have introduced the general label k indicating the orbital quantum numbers of both the leads with the identifica- tion k = ~k for the substrate and k = kz for the tip. The coefficient tχ ki is the tunnelling amplitude that contains all the geometrical information about the tunnelling pro- cesses. Denoting by h = p2 2m + vm + vsub + vtip the single particle Hamiltonian for an electron in the STM set-up, we define this amplitude by tχ ki := hχkσhiσi , (6) where χkσi and iσi are eigenstates of the reservoir χ and of the molecule, respectively. The kinetic energy of the electron is given by p2 2m . The molecule, tip and substrate potentials are denoted by vm, vtip and vsub, respectively. The z-dependence of the total potential v = vm + vsub + vtip is schematically shown in Fig. 1b. It is the sum of three potential wells, for the substrate, molecule and tip where εχ 0 < 0 defines the bottom of the conduction band and εχ F < 0 are the Fermi energies. For the tunnelling amplitudes, it follows: 0 + εχ p2 2m + vm tχ ki = hχkσ = εihχkσiσi = εiXα {z =hmol } iσi + hχkσvsub + vtipiσi } hχkσασihασiσi , {z ∼0 (7) 3 satisfies where hmol is the non-interacting single-particle Huckel- Hamiltonian that the eigenvalue equation hmoliσi = εiiσi. The key observation to understand why the matrix ele- ment hχkσvsub + vtipiσi can be neglected while the con- tribution hχkσvmiσi containing the molecular potential should be retained is the larger penetration length of the lead wave function, with respect to that of the molec- ular orbital, into the barrier region separating the lead and the molecule. This difference implies in fact that the relevant integration region for the matrix element hχkσvsub + vtip + vmiσi is shifted towards the molecule. Consequently the kinetic energy contribution should be complemented by the one of the molecular potential. For systems characterized by states with comparable pene- tration lengths instead, the relevant integration region is in the tunnelling barrier and the kinetic energy yields the dominant contribution. The different penetration lengths for the lead and molecule wave functions is justified as follows. First, the spatial extension of the valence orbitals is larger for the metallic atoms of the lead than for the ones in the con- jugated molecule. Moreover, the states in the lead which dominate the tunnelling have no nodal planes perpendic- ular to the molecular surface (low kk) while the HOMO and LUMO states of a conjugated molecule have usu- ally several nodal planes perpendicular to the plane of the molecule. These perpendicular nodal planes are as- sociated to a destructive interference between the atomic wave functions which implies that the higher the num- ber of nodal planes, the shorter is the extension of the molecular orbital in the direction perpendicular to the molecular plane. Notice that the energy of the vacuum between the molecule and the tip has been set to zero. Likewise we also set to zero the top of the tunnelling barrier between the molecule and substrate, corresponding to the thin insulating layer. The theory is not affected though by a different value of the potential in the barrier regions as far as the latter is spatially uniform if compared with the product of the lead and molecule wave functions in the same region. A more precise description of the lead po- tential would in first approximation just lead to a renor- malization in Eq. (7) of the orbital energy εi. In the last step of Eq. (7) we added the completeness 1 =Pα ασihασ, where ασi is the pz-state of the atom α, thus showing that the wanted matrix element can be expressed in terms of the overlap hχkσασi of the lead and the pz-orbital and the basis transformation hασiσi from the molecular to the atomic orbital. Finally, we obtain for the tunnelling amplitudes: tT ki = εiXα ki = εiXα tS OT (kz, ~Rtip − ~Rα)hασiσi , e−i~k· ~RαOS(~k)hασiσi , (8) (9) TABLE I: Parameters αi, βi used for the Gaussian pz-orbitals B. Tunneling dynamics 4 2 1 i αi(cid:2) 1 A2 (cid:3) 0.368 1.113 4.997 βi(cid:2) 1 A5/2 (cid:3) 0.502 1.438 2.620 3 where ~Rα and ~Rtip are the position of the atom α and of the tip, respectively. The overlaps Oχ are given ex- plicitly in the Appendix A and are calculated using the pz-orbital37: pz(~r− ~Rα) = h~rαGi = nGXi βi (~r− ~Rα)· ez e−αi~r− ~Rα2 , (10) where ez is the versor in the direction perpendicular to the molecular plane, the coefficient nG assures normaliza- tion and the parameters αi and βi, that we show in table I for the specific case of a carbon atom, define the gaussian representation for a Slater type orbital commonly used in DFT calculations38,39. Analogous parametrizations are available also for other atoms and allow a straightforward application of the model to generic planar π-conjugated molecules. The overlap functions of the substrate and the tip are qualitatively different since they reflect the different geometries of the corresponding contacts. The plane wave description of the electrons in the substrate implies that in Eq. (9) the position of the atom ~Rα only appears in the phase factor as a scalar product with the component of the momentum parallel to the substrate, ~k. Additionally we obtain a function that only depends on the electron's momentum ~k in the substrate and on the thickness of the insulating barrier. This particular form already suggests that the tunnelling between the sub- strate and the molecule is not an incoherent collection of tunnelling events happening in correspondence to the dif- ferent atoms since their position is recorded in the phase of the tunnelling amplitude. Some of the consequences of this spatial coherence will appear more clearly in sec- tion III where we analyze the special case of a benzene STM junction. The overlap function for the tip is more complex. Due to the cylindrical symmetry of the tip and atomic orbital with respect of their rotational axes, we can only further conclude that only the modulus of the component of ~Rtip − ~Rα parallel to the molecular plane influences the tunnelling (see Appendix A). Our method of choice to treat the dynamics in the regime of weak coupling between system and leads is the Liouville equation method. A detailed discussion and derivation of the equations of motion for the reduced den- sity operator of the system can be found e.g. in40,41; we will give here only a short overview adapted to the STM set-up. We start from the Liouville equation for the total den- sity operator ρ(t) of the whole system consisting of the molecule, the tip and the substrate. Using the interac- tion picture and treating the tunnelling Hamiltonian (5) as a perturbation we get: i dρI (t) dt = [H I tun(t), ρI (t)] , (11) where the subscript I indicates the use of the interaction picture. Since we are not interested in the microscopic state of the leads, we focus on the time evolution of the reduced density matrix (RDM) σ = TrS+T{ρ(t)}, which is formally obtained by taking the trace over the unob- served degrees of freedom of the tip and the substrate. The equation of motion for the RDM reads to lowest non- vanishing order in the coupling to the substrate and the tip42 σ = − i [Hm, σ] − i [Heff , σ] + Ltunσ := Lσ. (12) The first term of this so called generalized master equa- tion (GME) gives the coherent evolution of the system in absence of the substrate and the tip. In the secular ap- proximation we only keep coherences between degenerate states and thus this term vanishes40. The commutator with Heff includes the normalization of the coherent dy- namics introduced by the couplings to the leads. Finally, the operator Ltun describes the sequential tunnelling pro- cesses. The sum of these three contributions defines the Liouville operator L. Let us concentrate first on the tunnelling processes oc- curring in the system. The corresponding contribution to the master equation, projected into the subspace of N -particles and energy E reads: LtunσN E = − ij(E − Hm)f−χ (E − Hm)djτ + djτ Γχ ij(Hm − E)f + ij(E′ − E)σN +1E′ χ (Hm − E)d†iτi σN E + h.c.o f−χ (E′ − E)d†iτiPN E (13) 1 2Xχτ Xij nPN Ehd†iτ Γχ +Xχτ XijE′ PN Ehd†iτ Γχ ij(E − E′)σN−1E′ χ (E − E′)djτ + djτ Γχ f + where σN E := PN EσPN E being PNE :=Pl N ElihN El the projection operator on the subspace of N particles and energy E, and l the additional quantum number that distinguishes between degenerate states. Moreover, f + χ (x) is the Fermi function for the lead χ, f + χ (x) := f (x − µχ), and f−χ (x) := 1 − f + χ (x). The terms pro- portional to f + χ (x) describe in (13) tunnelling events to the molecule, while the tunnelling out of the molecule is associated to f−χ (x). Finally µχ stands for the electro- chemical potentials of the substrate or the tip. They are defined via the applied bias voltage as µS = µ0 + (1 − c)eVb, µT = µ0 − c eVb and consequently eVb = µS − µT , with the electron charge e, the equilibrium potential µ0 and the coefficient c governing the relative bias drop at the tip and the substrate. A symmetrical potential drop is obtained for c = 1/2, while for c = 1 the bias drops completely at the tip-molecule interface. Finally µ0 = −Φ0 relates the equilibrium chemical potential to the work function and, in equilibrium, the work func- tions of the two leads are assumed equal. Beside the Fermi function, the tunnelling rates are characterized by the geometrical component: Γχ ij (∆E) = ki)∗ tχ (tχ kj δ(εχ k − ∆E) . (14) 2π Xk The argument ∆E of the rate Γχ ij is the energy difference EN +1 − EN of the many body states involved in the tun- nelling process, sometimes written in Eq. (13) in terms of the operator Hm. Notice that the rate Γχ ij vanishes if ∆E > 0 since we restrict the Hilbert space of the leads to the bound states i.e. εk < 0. The quantity Γχ ij plays a central role in the theory and in the following section we will discuss its calculation in detail for the tip and the substrate case using the example of a benzene molecule. A natural expression for the current operators is ob- tained in terms of the time derivative of the reduced den- sity matrix: hIsub + Itipi =XN E Tr(cid:8)N σN E(cid:9) , (15) where Isub/tip are the current operators calculated for the substrate and the tip interfaces. Conventionally we assume the current to be positive when it increases the charge on the molecule. Thus, in the stationary limit, hIsub + Itipi is zero. The stationary current is obtained as the average: hIsubi = Tr {σstatIsub} = −hItipi , (16) where σstat = limt→∞ σ(t) is the stationary density op- erator that can be found from σstat = Lσstat = 0 , (17) where L is the Liouville operator. Finally, by following exactly the procedure given in [41], we find the explicit expressions for the current operators: 5 where the energy renormalization terms, present in the GME, do not appear. Since the tunnelling changes the number of electrons on the molecule, the latter behaves as an open system and it is useful to introduce the operator H′m = Hm − µ0N where N counts the number of electrons on the molecule. For example, at zero temperature and zero bias the equi- librium is reached when the molecule is in the ground state of H′m and not of Hm. As we have already shown elsewhere43, also the non-equilibrium conditions for transport can be better understood in terms of the spectrum of H′m. For this reason in Figs. 3 and 4 the geometrical part of the rates is plotted as a function of ∆E′ := ∆E − µ0. III. THEORY APPLIED TO BENZENE The molecular orbitals of benzene are also eigenfunc- tions of the projection l of the angular momentum along the main rotational axis, which we assume to be the z- axis. Therefore, the basis transformation that occurs in Eq. (7) reads for a benzene molecule hασlσi = 1 √6 ei 2π 6 αl (19) and the corresponding single particle eigenenergies εl, oc- curring in the Eqs. (8) and (9) for the tunnelling ampli- tudes, read: εl = a + 2b cos(cid:18) 2π 6 l(cid:19) . (20) For a benzene molecule the possible values of the an- gular momentum quantum number l are 0 ,±1 ,±2 , 3 corresponding to the energy level scheme of the Huckel Hamiltonian shown in Fig. 2. Since the Hamiltonian is invariant under the discrete rotations of angles nπ/3 with n ∈ Z, the same quantum numbers also label the many-body eigenstates of the benzene molecule, irrespec- tive of the complexity of the description of the Coulomb interaction41. All the single particle states show a twofold spin degeneracy but only few states possess an additional twofold orbital degeneracy. The latter is essential for the explanation of the transport features of benzene within an STM experiment. A. The substrate-molecule tunnelling rates Iχ = XN Eσij ij(Hm − E)f + PN E(cid:20)djσΓχ χ (Hm − E)d†iσ ij (E − Hm)f−χ (E − Hm)djσ(cid:21)PN E , −d†iσΓχ (18) Let us start with a detailed discussion of the substrate- molecule tunnelling rate. To perform the sum over the momenta ~k in Eq. (14) we transform it into energy inte- E a-2b a-b a-b a+b a+b a+2b a l=3 l=+ - 2 l=+- 1 l=0 FIG. 2: Energy levels of the Huckel Hamiltonian and the corresponding values of the angular momentum l. grals, using the definitions ε := 2~k2 2m and εz := 2k2 2m : z , m X~k 2 Z 2π r m =X~k Xkz dϑZ εS 2 Z εS 1 0 0 0 F +ΦS 0 F +ΦS 0 dε , dεz 1 √εz , → S X~k Xkz → z0 (21) (22) (23) where the volume V = z0S is canceled out in the ther- modynamic limit by the normalization of the orbitals which define the overlap function. Moreover we observe that Eq. (14) requires the calculation of the product ~kl′ = ~kl(cid:17)∗ tS (cid:16)tS εlεl′Xαα′ OS(~k)2hασlσihα′σl′σie+i~k·( ~Rα− ~Rα′ ) . (24) function in Eq. (24) as We write the exponential e+i~k·( ~Rα− ~Rα′ ) = e+i~k ~Rα− ~Rα′ cos ϑ and the equation fi- nally becomes (cid:16)tS ~kl(cid:17)∗ tS ~kl′ = 1 6 εlεl′Xγ e−i 2π 6 l′γXα e−i 2π 6 α(l−l′) (25) +iq 2m 2 ε∆ ~Rγ cos ϑ × e OS(ε, εz)2 , 0 3 ΓS F +ΦS π2 4 m ll′ (∆E) = δll′ over ε yields and the fact that Pγ ei 2π 6 lγ = Pγ e−i 2π l Z εS 2 √2 ε2 (cid:0)∆E − εz − εS 0 , εz)2 0(cid:1) Θ (εz − ∆E) . ×Xγ × OS(∆E − εz − εS × Θ(cid:0)∆E − εz − εS J0 r 2m dεz 0 V √εz 0(cid:1)∆ ~Rγ! e−i 2π 6 lγ (28) 6 with J0(x) the zero-order Bessel function. Finally, using the relation e±i 2π 6 α(l−l′) = 6 δll′ , (27) Xα 6 lγ the integral The integral in Eq. (28) has to be solved numerically. The main result of the latter calculations is ΓS ll′ (∆E) = δll′ ΓS l (∆E) , (29) which ensures that tunnelling processes involving the substrate are happening through angular momentum channels because a mixing of angular momenta is not al- lowed in the substrate. We will see that this only happens for substrate-tunnelling-processes, while there is no con- servation rule for angular momenta in the tip-tunnelling case. The function ΓS l (∆E) is the geometrical rate and we plot it in Fig. 3 for different angular momenta. The rates decrease of several order of magnitudes by increas- ing the absolute value of the projection of the angular momentum l. This is the direct consequence of the de- creasing extension of the molecular orbitals in the direc- tion perpendicular to the molecular plane with increasing the number of vertical nodal planes. The lower limit of the energy axis in Fig. 3 is −εS F while the upper limit is the work function φS 0 . These limits are set by the substrate model in which only bound states 0 < ǫS of a single band are taken into account (εS k < 0). While approaching the low energy limit ∆E = −εS F both the density and the penetration length of the states in the substrate which contribute to the rate reduce, hence the turn down. On the other hand, the increasing of the density of states and of the penetration length explains the turn up at the upper energy border (∆E = φS 0 ). where we introduced α − α′ := γ , ~Rα − ~Rα′ = ∆ ~Rγ. We insert Eq. (25) in the substrate case of Eq. (14) and, after solving the integral over dϑ, we find: B. The tip-molecule tunnelling rates e+i 2π 6 α(l−l′) 2√2εlεl′Xα ΓS 3 ll′ (∆E) = π2 64 m dε Z εF +Φ0 ×Z εF +Φ0 × OS(ε, εz)2e+i 2π 0 0 V dεz √εz Xγ 6 l′γδ(εk − ∆E) , J0 r 2m 2 ε∆ ~Rγ! Let us now discuss the tunnelling events happening be- tween the tip and the molecule. To model the tip we con- sider a harmonic confinement in the x and y directions. By considering the tip to be in the ground state of the 2- dimensional harmonic oscillator, the longitudinal energy ε is fixed to be the constant ε = ω, cf. below Eq. (4). The sum in Eq. (14) thus transforms into a sum over (26) kz. Because of the relation kz =q 2m 2 εz we can replace ] z H [ S l Γ l = 0 l = ∓1 l = ∓2 l = 3 ΔE' = ΔE - μ0 [eV] (Color online) Tunneling rate ΓS l FIG. 3: describing substrate-molecule tunnelling processes for different angular momentum quantum numbers l. The thickness of the sub- strate barrier is d = 3A, while work function and Fermi energy are respectively φS 0 = 4eV and εS F = 7eV p m 2 R dεz zend−ztip √εz . Eq. (8) implies the sum by the integral: Pkz → 1 kz l(cid:1)∗ tT (cid:0)tT kz l′ = 1 6 (αl−α′l′) εlεl′e−i 2π 6Xαα′ × O∗T (kz, ~Rtip − ~Rα)OT (kz, ~Rtip − ~Rα′ ) , (30) that we insert in Eq. (14). After solving the energy inte- gral we finally find ll′ (∆E, ~Rtip) = ΓT εlεl′ e−i 2π 6 (αl−α′l′) π 62r m 2 Xαα′ 0 ) , 0 − ω zend − ztip p∆E − εT × O∗T (k, ~Rtip − ~Rα)OT (k, ~Rtip − ~Rα′ ) × Θ(∆E − ω − εT 0 )Θ(2ω − ∆E + εT where k = q 2m (31) 0 ). The occurrence of both l and l′ in the latter equation, shows that a mix- ing of angular momenta during the tip-tunnelling process takes place. Upon inspection of Eq. (31) we find some important relations obeyed by the tunnelling rate, where we use the fact that l and l′ always occur in the form l′ = ±l: 2 (∆E − ω − εT ¯l¯l )∗ ∈ R, ΓT ll = ΓT ¯l¯l = (ΓT ¯ll )∗, l¯l = (ΓT ΓT l¯l = ΓT ΓT ll = ΓT ll , (32) where we have introduced the notation ¯l ≡ −l. Thanks to the relations (32) we can rewrite the tunnelling rate as 7 l ≡ ΓT where ΓT ll , which implies the existence of an angu- lar momentum dependent phase when l 6= l′. In Fig. 4 we show the diagonal elements of the rate matrix ΓT ll′ ex- emplified for l = ±1 and l = ±2. As for the substrate, the channel l = ±1 leads to a much larger rate than the channel l = ±2. The phase in the off diagonal elements depends on the tip position ~Rtip and it is calculated as φl( ~Rtip) = arg(tT kz l). (34) In Fig. 5 we show the values acquired by the phase φl( ~Rtip) as a function of the tip position. The phase is approximately constant along the radii leaving the cen- ter of the molecule. Due to the cylindrical symmetry of the tip wave function a good approximation to the phase φl( ~Rtip) is given by: φl( ~Rtip) = lθtip, (35) where θtip is the angle describing the projection of the tip position on the molecular plane if the origin is the center of the molecule. By convention we assume θtip = 0 along the radius that intersects the position of the atom 0 of the molecule (see Fig. 5). The derivation of this simple expression for φl as well as a discussion on its limits of validity are given in Appendix C. Notice that the phase defined in Eq. (34) only depends on ~Rtip even if Oα contains the bias. Nevertheless, the tunnelling rate Eq. (33) depends on the bias via the Fermi energy. In Fig. 5 the position of the φl = 0 line is arbitrary and connected to the arbitrary choice of overall phase for the molecular orbital with angular momentum l. A different choice of the overall phase would, nevertheless simply appear as a rigid rotation of the plots. Moreover, this arbitrariness has no influence on the current voltage characteristics of the junction. In the substrate the tunnelling matrix is diagonal and proportional to the identity matrix, independent of the basis representation, see Eq. (29). In contrast, according to Eq. (33), off-diagonal elements are present in the tip- tunnelling matrix which, in the basis {li, ¯li}, reads ΓT = ΓT l 1 e−2iφl( ~Rtip) e+2iφl( ~Rtip) 1 ! . (36) An interesting effect of the localized character of the tunnelling from/to the tip can be better appreciated by switching to the basis which diagonalizes the matrix in Eq. (36). The substrate rate matrix is still proportional to the identity matrix. For the tip rate matrix we get instead: ΓT = ΓT l 2 0 0 0! . (37) ΓT ll′ = ΓT l e−iφl( ~Rtip)(l−l′)/l, (33) One diagonal element becomes zero, indicating that there are states which are coupled to the substrate but ] z H [ T Γl l = 0 l = ∓1 l = ∓2 l = 3 ΔE' = ΔE - μ0 [eV] FIG. 4: (Color online) Diagonal elements ΓT l of the tip tun- nelling rate matrix ΓT ll′ for the different angular momentum states. The rates are calculated assuming ztip − d = 3.5 A, φT 0 = 4eV , εT F = 7eV and ω = 4eV . The presence of the harmonic confinement explains also the different energy lim- its with respect to the ones of Fig. 3. The lower limit is at −εT F + ω while the upper limit is at −εT F + 2ω. 8 late the time evolution of the reduced density matrix as- sociated to Ltun and the corresponding stationary state. The stationary density matrix is block diagonal in parti- cle number, energy and spin. In particular, if we restrict the dynamics to low biases, the only relevant states en- tering the dynamics are the states 5glτi, 6g00i, and 7glτi, being the cation, neutral and anion ground states respectively. The neutral ground state is non degener- ate while the anion and cation are four times degenerate, due to the combination of the spin and orbital degenera- cies. The specific form of the stationary density matrix depends on the bias, the temperature, and the tip posi- tion. Nevertheless, due to the form of the tunnelling rate matrices, the two dimensional sub-blocks corresponding to orbitally degenerate states have always the following structure: ¯N Eg τ stat = σ A Be−2iφl( ~Rtip) Be+2iφl( ~Rtip) A ! , (38) l=1 ] Å [ y 3 π 2 1 2 1 0 0 ] Å [ y 3 l=2 π 0 0 4 5 4 5 -π -π x [Å] x [Å] FIG. 5: (Color online) Phase φl of the tunnelling rate matrix ΓT ll′ , Eq. (34). The phase is almost constant if the tip is moved along the radii outgoing from the center of the molecule. The carbon atoms are labelled by α = 0, . . . , 5. not to the tip. The decoupled state represents a blocking state, which can be populated by a tunnelling event from (to) the substrate but cannot be depopulated by a tun- nelling event to (from) the tip. The presence of blocking states is visible in the current-voltage characteristic, as we will discuss in the next section. where ¯N = 5, 7, the spin τ =↑,↓ and the parameters A, B are functions of the tip position ~Rtip and of the bias Vb (see Appendix B). This result is a posteriori not surpris- ing. The comparison of Eq. (38) with Eq. (36) reveals in fact that the density matrix and the rate matrices are diagonalized by the same basis transformation (the sub- strate rate matrix is diagonal in all bases). Thus, the form of σstat could be calculated from the observation that the dynamics of the populations and the coherences is decoupled when expressed in the eigenbasis of the rate matrices. It should be noticed that the diagonalizing ba- sis depends on the phase, see Eq. (34), which in turn depends on the tip position. Thus it is not possible to describe the system using only populations in a unique basis valid for all the positions of the tip. D. The effective Hamiltonian C. Stationary density matrix By combining now the expression for the tunnelling rates with the dynamical equation Eq. (13) we can calcu- Until now we only concentrated on the sequential tun- nelling processes in the system. We still have to discuss the imaginary term in Eq. (12) which contains the effec- tive Hamiltonian Heff. The latter is defined as: Heff = 1 2πXN EXχσ Xll′ PN Ehd†lσΓχ ll′ (E − Hm)pχ(E − Hm)dl′σ + dl′σΓχ ll′ (Hm − E)pχ(Hm − E)d†lσiPN E, (39) with and the the projector PN E part principal = Pn N EnihN En functions pχ(x) = −ReΨh 1 2 + i 2πkB T (x − µχ)i, with T being the tem- perature and Ψ the digamma function. Eq. (39) shows that the effective Hamiltonian is block diagonal in particle number and energy, exactly as the density matrix in the secular approximation. Consequently, it only influences the dynamics of the system in presence of degenerate states with corresponding subblocks larger than a mere complex number. For the sake of simplicity we will include in the following calculations only the an- ion ground states, (i.e. the spin and orbitally degenerate 7 particle ground states). Analogous arguments holds for all the other degenerate states of the molecule. If Γll′ ∝ δll′ (substrate case, see Eq. (29)), the effective Hamiltonian Heff in the 7 particle ground state subspace is proportional to the identity matrix, as can be proven from Eq. (39) remembering that Hm conserves the an- gular momentum and it is invariant under the symmetry operation that brings 7glτi into 7g¯lτi and moreover that ΓS ll = ΓS ¯l¯l. Thus, the substrate contribution to Heff triv- ially commutes with σstat. If the angular momenta l and l′ can mix, like in the tip case, Heff acquires off diagonal terms and a more detailed discussion is required. In par- ticular, the form of the off diagonal elements depend on the particular model taken to describe the interaction on the molecule. As shown in the Appendix D, within the constant interaction model, the effective Hamiltonian for the tip can be written in the form: H T eff = ωL , where ω = + 1 πh7glσd†lσ6g 0 0ih6g 0 0d¯lσ7g¯lσi × ΓT l (E7g − E6g)pT (E7g − E6g) 1 πh7glσd¯lσ8g 0 2σih8g 0 2σd†lσ7g¯lσi × ΓT l (E8g − E7g)pT (E8g − E7g) (40) (41) is the renormalization of the Bohr frequencies for the system and L = 2 1 e−2iφl( ~Rtip) e+2iφl( ~Rtip) 1 ! . (42) Hence the effective Hamiltonian H T eff commutes with the stationary density operator σstat given in Eq. (38). In conclusion, even if different from zero, the effective Hamiltonian does not contribute to the stationary dy- namics of our system because it commutes with the sta- tionary density matrix Eq. (38) calculated using only the tunnelling component of the Liouvillean. For a generic description of the Coulomb interaction on the molecule, corrections to Heff given by the 8 and 6 particle excited states should be taken into account and the form of Heff is modified. For the sake of simplicity we restrict here to the constant interaction model. More details on the derivation and the discussion on the most general case are given instead in the Appendix D. 9 ΔE'5-6 { l=+1 l=-1 { l=+2 l=-2 { ΔE'6-7 7 l=0 6 5 Particle Number ] V e [ 0 μ N - E = ' E FIG. 6: (Color online) Together with a change in the en- ergy, the transition from the 6-particle ground state to the 7-particle (5-particle) ground states is also associated with a change in the angular momentum of ∆l = ±2 (∆l = ±1). IV. I-V CHARACTERISTICS AND CURRENT MAPS OF A BENZENE MOLECULE In the following discussion of the current voltage char- acteristics and current maps we only consider the ground state transition 6g00i ↔ 7glτi or 6g00i ↔ 5glτi. In Fig. 6 we represent the corresponding energy levels as a function of the particle number for a particular choice of the work function (we assume ΦT 0 so that the chem- ical potentials are the same at Vb = 0). In the tunnelling event the molecule changes its particle number, angular momentum and energy (see Fig. 6). All these changes leave their fingerprints in the current voltage character- istics and current maps presented in Figs. 7-9. 0 = ΦS Φ0 = 5.0 eV, T = 8K: 1 0 -1 -2 -3 ] A p [ I b) Coulomb- blockade a) resonance current [pA] a) Vb = -1.688 V I [pA] -2 -1.5 -1 -0.5 ] Å [ y current [pA] b) Vb = -1.680 V I [pA] -5 -12 -10 -8 -6 -4 -2 ] Å [ y -4 -3 -2 -1 0 1 2 Vb [V] x [Å] x [Å] FIG. 7: (Color online) Current voltage characteristics and current maps associated to the neutral-anion transition. The current maps are calculated with ztip − d = 5A. Notice that the map in the Coulomb blockade region is just a rescaling of the one at resonance. In particular, the current is exponentially suppressed at small biases (the so called "in gap region" of trans- port) due to the Coulomb blockade44. The bias at which current starts to flow corresponds to a resonant condition between the chemical potential in the source (or drain) lead and the difference in the energy ∆E between the many-body states participating to the transport. For this reason the current voltage characteristics (and the associated differential conductance traces) recorded with an STM junction represent a valuable spectroscopic tool to investigate the many-body spectrum of the molecule. One has to keep in mind nevertheless that i) the reso- nant bias depends on the value of the work function of the leads, ii) the bias drops very asymmetrically at the tip and substrate interface with an associated very different amount of energy available to the molecular transition. The shift in the position of the resonance with the work function can be observed by comparing the positions of the step in the current at negative biases in Fig. 7 and 9. Φ0 = 7.0 eV, T = 8 K: ] A p [ I 80 60 40 20 0 b) resonance a) Coulomb- blockade ] Å [ y current [pA] a) Vb = 2.150 V I [pA] current [pA] b) Vb = 2.156 V I [pA] ] Å [ y 0 1 2 Vb [V] 3 4 x [Å] x [Å] FIG. 8: (Color online) Current voltage characteristics and current maps associated to the neutral-cation transition. The current maps are calculated with ztip − d = 5A. Notice that the value of the current at resonance is much higher than the one relative to the neutral-anion case (see Fig. 7). In Fig. 9 one can also observe how the same molecu- lar transition (between the neutral and anionic molecule) gives signals at different biases if triggered by a substrate (Vb > 0) or a tip (Vb < 0) tunnelling event. A larger bias (in absolute value) is needed for a substrate transition since most of the bias drop concentrates at the tip in- terface. Moreover the current signal obtained at positive bias is a peak instead of a step due to an interference blocking effect analogous to the one discussed in [41]. In the interference blocking region the system is blocked into a particular linear combination of the 7 particles ground states that can be populated from the substrate but can- not be depopulated towards the tip. The angular momentum channel involved in the trans- port depends on the difference in the angular momentum of the many-body states participating to the tunnelling events. The neutral-anion and neutral-cation transitions correspond to ∆l = ±2 and ∆l = ±1 respectively, cf. Fig. 6, thus involving different angular momentum chan- nels. Since the lower is the angular momentum of the channel the larger are the rates, the current associated to the neutral-cation transition is larger than the one of the neutral-anion one, as it can be seen by comparing the resonant currents of Figs. 7 and 8. By comparing the same figures one finds also qualitative differences in the constant heights current maps: yet another finger- print of the different states involved in the transitions. The same differences are also confirmed by the constant current images presented in Fig. 10. Finally, the current maps presented in Fig. 9 suggests that also the interference effects have a topographic sig- nature. The current map taken in the Coulomb blockade region is in fact qualitatively different from the one taken in the interference blockade. To conclude, a comparison with the widely applied Ter- soff and Hamann (TH) theory4,8,9 is compulsory. In par- ticular, for what concerns the current maps presented in Fig. 7 and Fig. 8, we do not expect qualitative dif- 10 ferences between the effectively single particle TH the- ory and our many-body approach. Yet, this is almost accidental for the following reasons: i) we decided for simplicity to describe the system using a constant inter- action model in which the many-body states are single Slater determinants; ii) the initial and final many-body states of the tunneling event (e.g the neutral and anion ground states) fix the corresponding variation of angu- lar momentum (∆l = ±2). Consequently, in the par- ticular case of benzene, only one single particle orbital contributes to the current formula given in Eq. (18). In general, though, many Slater determinants are necessary to identify a single many-body state and many molecu- lar orbitals would contribute to the transport. Moreover TH would not be able to address the interference block- ing regime and the associated current maps since it is effectively a non interacting single particle theory. Φ0 = 4.0 eV, T = 8K: ] A p [ I 1 0 -1 -2 -3 -4 -2 a) Coulomb- blockade b) interference- blockade ] Å [ y I [pA] a) Vb = 3.04 V x [Å] I [pA] b) Vb = 3.13 V ] Å [ y -1 0 1 2 3 4 Vb [V] x [Å] FIG. 9: (Color online) Current voltage characteristics and current maps associated to the neutral-anion transition. In- terestingly the current map in the interference blockade re- gion shows novel topographic features if compared with other maps involving the same states (see also Fig. 7). In the inset a zoom on the interference current peak is presented. ] ] Å [ d - Å [ d - p i t z p i t z x [Å ] [ Å ] y x [Å ] [ Å ] y FIG. 10: (Color online) Constant current topographic images. The left panel refers to the neutral-cation resonance, (φ0 = 7eV , Vb = 2.156V , I = 300pA ), the right panel, instead, to the neutral-anion resonance (φ0 = 5eV , Vb = −1.688V , I = 100pA). V. CONCLUSIONS In this paper we presented an STM transport theory sufficiently general to be applied to any device consisting of a π-conjugated molecule weakly coupled both to the substrate and the tip. While the weak tunnelling cou- pling to the tip is a natural assumption in STM exper- iments, the weak coupling to the substrate is motivated by recent STM set-ups with substrates covered by a thin insulating film23,24,36. The essentially different geometry of the STM tip and the substrate is reflected in the respective tunnelling am- plitudes, whose energy dependence induces, within a den- sity matrix approach, characteristic non-constant tun- nelling rate matrices. The latter play a central role in the Liouville operator, which determines the dynamics of the system, and in the current operator. Interestingly, for these system, due to the different pen- etration lengths of the metallic states of the tip/substrate and the molecular orbitals into the corresponding tun- nelling barriers, the tunnelling amplitudes cannot be cal- culated using the standard Tersoff and Hamann approach and an alternative method is proposed. As an application of our general results we used a ben- zene molecule that enabled us to express the theory in the basis of the angular momentum l. The explicit cal- culation of the tunnelling rate matrices in the momen- tum basis shows a fundamental difference between the tip and substrate tunnelling dynamics. The delocalized tunnelling at the substrate happens via angular momen- tum channels (diagonal tunnelling matrices) while the localized tip mixes the angular momenta (off diagonal matrices). A direct consequence of this different tunnelling sce- nario for the two leads is found in the current voltage characteristics. At voltages sufficiently large to lift the Coulomb blockade, interference blocking occurs when de- generate states participate to the transport. While the presence of degenerate states is a necessary condition for the interference, only the tip tunnelling can detect it due to its localized nature which mixes the angular momenta in the tunnelling event. Moreover, also the STM surface-images can be calcu- lated within our theory. By varying the work function of the substrate we show simulations of STM constant height current maps and constant current topographic images in which the transport is dominated either by 11 neutral-anion or neutral-cation transitions. In particular, striking is the difference in the current maps obtained in the resonant and interference blocking regime although the same many body states participate to the transport (see Figs. 7 and 9). Acknowledgements We thank prof. Jascha Repp for the fruitful discussions. Moreover, we acknowledge financial support by the DFG within the research programs SPP 1243 and SFB 689. Appendix A: Calculation of the overlap functions To calculate the tunnelling amplitude in equation (7) we need to calculate the overlap between the metal's wave function and the pz-orbital. The latter is given, in the Gaussian description by37 -- 39 βj (~r − ~Rα) · ez e−αj~r− ~Rα2 , (A1) where nG is the normalization factor which ensures h~rαGi = nGXj R d~rh~rαi2 = 1, ~Rα is the position of the atom α and ez is the versor in the direction perpendicular to the plane of the molecule. Since the overlap is calculated as a function of the quantum number k defining the lead wave function, we will call the bracket hχkσασi overlap function. a b z V 0 FIG. 11: Scheme of a 1-dimensional, finite potential well with borders a and b and depth V0. In our model both the tip and the substrate are de- scribed in the z direction as potential wells45. For future reference we report here the general expression for the eigenfunction of an arbitrary 1-dimensional potential well of depth V0 and whose borders are a and b, see Fig. 11: Ψkz (z; a, b, V0) = nz  e−κa [U sin(kza) + cos(kza)] eκz , U sin(kzz) + cos(kzz) , e+κb [U sin(kzb) + cos(kzb)] e−κz , if − ∞ < z < a if a < z < b if b < z < ∞ , (A2) where nz ensures the normalization R dzΨkz (z)2 = 1 and U = kz sin(kzb) − κ cos(kzb) kz cos(kzb) + κ sin(kzb) . substrate overlap functions. The integral is: 12 Fkz (a, b, V0, αj) =Z +∞ −∞ zΨkz (z + d; a, b, V0)e−αj z2 , (A3) where for simplicity we have omitted in F the dependence on the parameter d. The integration yields: The occurring wave number reads kz = q 2m κ = q 2m z , respectively. Due to the large size of the potential well compared to the Fermi wavelength we neglect the quantization of kz obtained by the corre- sponding eigenvalue equation. 2 V0 − k2 2 ǫz and We conclude this introductory part with the explicit calculation of an integral common to both the tip and Fkz (a, b, V0, αj) = nz 3 4α 2 j ×(e− k2 z 4αj 2Re"e−ikz d (1 + iU )"√αj(cid:18)e−αj(cid:16)a−d+ ikz 2αj (cid:17)2 − e−αj(cid:16)b−d+ ikz 2αj (cid:17)2(cid:19) ikz 2αj(cid:19)(cid:21)(cid:19)## (A4) + e κ2 4αj"−Ae+κd(cid:18)2√αje−αj(cid:16)a−d− κ + Be−κd(cid:18)2√αje−αj(cid:16)b−d+ κ 2αj (cid:17)2 2αj (cid:17)2 ikz√π 2 (cid:18)erf(cid:20)√αj(cid:18)b − d + − ikz 2αj(cid:19)(cid:21) − erf(cid:20)√αj(cid:18)a − d + 2αj(cid:19)(cid:21)(cid:19)(cid:19) 2αj(cid:19)(cid:21)(cid:19)(cid:19)#), − κ√π(cid:18)1 + erf(cid:20)√αj(cid:18)a − d − − κ√π(cid:18)1 − erf(cid:20)√αj(cid:18)b − d + κ κ where we used the abbreviations 1. Overlap molecule-substrate A = e−κa [U sin(kza) + cos(kza)] , B = e+κb [U sin(kzb) + cos(kzb)] . In equation (A4) the error function erf[ζ] with ζ ∈ C arises several times. It is defined as the integral of the normal distribution from 0 to ζ scaled such that erf[±∞] = ±1: erf[ζ] = 2 √πZ ζ 0 e−t2 dt and it is an entire function valid for real- and complex valued numbers46. Furthermore there holds 2 √πZ ζ2 ζ1 e−t2 dt = erf[ζ2] − erf[ζ1] . Both for the wave function Ψkz and the integral Fkz the tip and the substrate cases are obtained by the sub- stitutions (see also the triple-well in Fig. 1) sub :  a → z0 b → 0 V0 → −εS 0 tip :  a → ztip b → zend V0 → −εT 0 Let us consider the substrate case in which, for the sake of simplicity, we neglect in the following the spinor component of the substrate and atomic states. Accord- ing to the model given in the main text and sketched in Fig. 1, the substrate's wave function is given by hx, y, zS~ki = 1 √S e+i(kxx+kyy)Ψkz (z; z0, 0,−εS 0 ) , (A5) where kx/y/z =q 2m 2 εx/y/z and S is the area of the sur- face of the substrate on which the molecule lies. The ex- ponentials in Eq. (A5) stem from using no confinement to describe the substrate in the x and y direction and pe- riodic boundary conditions. Due to the large size of the substrate in all the three directions if compared with the Fermi wavelength λF =p2/(2mεF ) we neglect the mo- mentum quantization in all three directions. By setting the origin of the coordinate system in ~Rα and performing the Gaussian integrals in the x and y direction one easily obtains: hS~kαGi =e−i~k· ~Rα nG√S Xj k2 4αj e− πβj αj dz z Ψkz (z + d, z0, 0,−εT 0 )e−αj z2 −∞ ×Z +∞ =e−i~k· ~Rα nG√S Xj × Fkz (z0, 0,−εS k2 4αj e− πβj αj 0 , αj) := e−i~kk· ~RαOS(~k), (A6) where ~k · ~Rα = kxxα + kyyα and the integral in the z direction has been performed with the help of Eqs. (A3) and (A4). Notice the suppression of the overlap for high values of the parallel component of the momentum ~kk in the substrate wave function given by the gaussian pref- actor and also the phase factor which depends on the position of the carbon atom ~Rα and on ~kk. a Slater-type orbital47,48: Instead of using a Gaussian pz orbital we can also use h~rαSi = 2 1 a0 (cid:19) 5 2√6(cid:18) Zeff (~r − ~Rα) · eze− Zef f a0 ~r− ~Rα, (A7) where a0 = 0.53 A is the Bohr radius and Zeff is a fit- ting parameters that takes into account the screening of the nuclear potential given by the core electrons. In Fig. 12 we show the substrate-tunnelling rates for the dif- ferent benzene molecular orbitals calculated according to Eq. (28). We compare the rates obtained using Gaussian and Slater-type orbitals using a distance d = 3 A between the end of the metallic well (the substrate) and the plane of the molecule. As one can see the two results are in good agreement. The discrepancy between the two de- scriptions depends nevertheless on the distance d due to the difference in the tails of the Slater and Gaussian de- scriptions of the pz orbital. A good agreement is reached in the range of d we are interested in (d = 1 A − 6 A). 2. Overlap molecule-tip We continue with the calculation of the tip-orbital overlap. The atomic wave function is described again by the Gaussian orbitals given in Eq. (A1). The tip is modeled assuming a harmonic confinement in x and y direction, and a quantum well for the z one. The overlap reads: hx, y, zT kzi =r mω ×e− mω π Ψkz (z; ztip, zend,−εT 0 ) 2 ((x−xtip)2+(y−ytip)2) . (A8) 13 ] z H [ S l Γ l = 0 l = ∓1 l = ∓2 l = 3 ΔE' = ΔE - μ0 [eV] FIG. 12: (Color online) Tunneling-rates obtained by us- ing Slater-type orbitals (solid lines) and Gaussian orbitals (dashed lines). The rates are calculated for a substrate- molecule distance d = 3 A. In the Slater-type orbital Zeff = 2. The overlap function is a three dimensional which, in Cartesian coordinates, reads: integral βj π Xj hT kzαGi = nGr mω dyZ ∞ dxZ ∞ Z ∞ −∞ × (z − d)Ψkz (z; ztip, zend,−εT 0 ) 2 [(x−xtip)2+(y−ytip)2] , × e− mω −∞ −∞ dz e−αj[(x−xα)2+(y−yα)2+(z−d)2] (A9) where we have already set zα = d ∀α. We shift again the origin of the coordinates to the center of the pz orbital, ~Rα, and perform the gaussian integrals in the x and y direction. Moreover it is convenient to introduce new variables describing the tip-atom distance ∆x = xtip − xα, ∆y = ytip − yα. The resulting overlap function reads hT kzαGi =nGr mω 2αj +mω (∆x2+∆y2) βjπ e− mωαj π Xj αj + mω 2 × Fkz (ztip, zend,−εT := OT (kz, ~Rtip − ~Rα) 0 , αj) (A10) and concludes this section dedicated to the explicit cal- culation of the overlap functions. Appendix B: The stationary density matrix In Eq. (38) we only gave the generic form of the station- ary density matrix σstat for an orbitally degenerate sub- space. In this section we will show how to calculate it and finally give the complete result for a specific example. For the sake of simplicity we concentrate on the transitions 6g ↔ 7g, but the calculation can be easily reproduced for all other transitions. The reduced density matrix σstat for the specific subspace that we are considering is composed of a single-element sub-block associated to the 6 particle ground state, a 2 × 2 sub-block associated to the sub- space {7g + 2 ↑i, 7g − 2 ↑i} and finally a 2× 2 sub-block relative to the space span{7g + 2 ↓i, 7g − 2 ↓i}. Since we are interested in orbital (but not spin) coherences the Liouvillean is a linear operator of dimension 9 × 9. We choose the basis: 6gii,   7g ↑; +2, +2ii 7g ↑;−2,−2ii 7g ↑; +2,−2ii 7g ↑;−2, +2ii 7g ↓; +2, +2ii 7g ↓;−2,−2ii 7g ↓; +2,−2ii 7g ↓;−2, +2ii ,  (B1) ii denotes a vector in the density where the notation matrix space. We organize the tunnelling Liouvillean in the following form: (Ltun)6g 7g =  L66 L67↑ L67↓ L7↑6 L7↑7↑ 0 L7↓6 S ΓS(cid:1) is the depopulation rate T ΓT + f + L7↓7↓ (B2) 0 ,   where L66 = −4(cid:0)f + of the 6 particle ground state and the coefficients ΓS/T 14 stand for the diagonal elements of the tunnelling rates of the substrate or the tip. Moreover, the rates and the Fermi functions are calculated at the same energy δE = E7g − E6g . The other elements of the matrix (Ltun)6g 7g are matrices themselves. In particular: L67↑ = L67↓ =f−S ΓS(cid:16) 1 1 0 0(cid:17) + f−T ΓT(cid:16) 1 1 e+2iφ2 e−2iφ2 (cid:17) (B3) are the population "rates" of the 6 particle ground state starting from the states 7gl ↑i and 7gl ↓i, while L7↑6 = L7↓6 =f + S ΓS(cid:16) 1 1 0 0(cid:17)T T ΓT(cid:16) 1 1 e−2iφ2 e+2iφ2 (cid:17)T + f + (B4) are the population "rates" of the states 7gl ↑i and 7gl ↓i starting from the state 6gi. Finally L7↑7↑ = L7↓7↓ = −f−T ΓT  1 0 0 1 e+2iφ2/2 e−2iφ2/2 e+2iφ2/2 e−2iφ2/2 e−2iφ2/2 e−2iφ2/2 e+2iφ2 /2 e+2iφ2 /2 1 0 0 1   − f−S ΓS  1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1   (B5) is the depopulation "rate" of the states 7gl ↑i and 7gl ↓i towards the 6 particle ground state. The stationary solution of the Generalized Master Equation Eq. (12) is found by calculating the null space of the Liouville operator. Here we restrict ourselves to the operator Ltun describing the sequential tunnelling dy- namics. A discussion about the relevance of the com- mutator with the effective Hamiltonian is left to the last appendix. If the leads are not superconductors, non mag- netic or with parallel polarization and weakly coupled to the molecule, the stationary density matrix is block di- agonal in particle number, energy and spin. Thus, the stationary solution which corresponds to the Liouvillean given in Eq. (B2) can be cast into the form: where the 7 particle subblocks, when written in the basis {7g +2 τi,7g −2 τi}, read: σ7g↑ = σ7g↓ = Be−2iφ2 Be+2iφ2 ! (B7) A A with σ6g = A = B = f−S ΓS(f−S ΓS + 2f−T ΓT ) N , f−T ΓT f + S ΓS + f−S ΓS(f + S ΓS + f + T ΓT ) N ΓSΓT (f−S f + T − f−T f + S ) N , , (B8) σstat =  0 σ6g 0 σ7g↑ 0 0 0 0 σ7g↓   (B6) and the normalization N defined by the relation Trσstat = 1. This result is worth some further analy- sis. First of all it is interesting to notice that B = 0 only if at least one of the following conditions is satisfied T /f−T = f + i) ΓS = 0 which is never happening, ii) ΓT = 0 which holds if ~Rtip is on the main rotational axis of benzene, iii) f + S /f−S which is satisfied only in equilibrium when µT = µS. This analysis shows how the interference between states with different angular momenta is ubiqui- tous in the molecular junction. Eventually, it is easy to prove that the eigenvalues of the stationary density ma- trix are σ6g , A + B and A − B. The ratio between these eigenvalues gives a key to the physical interpretation of the stationary density matrix. In fact: A + B σ6g A − B σ6g = = T S + 2ΓT f + ΓSf + ΓSf−S + 2ΓT f−T f + S f−S = e−β(∆E−µS) , , (B9) which can be interpreted as follows: σD 7g := A − B is the occupation of the 7 particle state 7gDτi which is decoupled from the tip and coupled to the 6 par- ticle ground states only via tunnelling events happen- ing at the molecule substrate interface. For this rea- son the ratio σD 7g/σ6g is the same as the one obtained in thermal equilibrium with the substrate. On the other hand σC 7g := A + B is the population of the 7 particle state 7gCτi which can exchange particles both at the molecule-substrate and at the molecule-tip interfaces. In particular, the rate of exchange for the state 7gCτi is double than the rate of exchange of the angular momen- tum states 7glτi (see Eq. (37)). The detailed balance gives immediately the first relation in Eq. (B9). The phase of the tunnelling amplitude between a ben- zene molecular orbital and a tip state plays an important role in the calculation of the transport characteristics of the STM junction. In this section we derive the approx- imate formula describing this phase given by Eq. (35), and also its limit of validity. Due to the cylindrical sym- metry of the tip wave function, for the overlap function with the atomic wave function OT (kz, ~Rtip− ~Rα) it holds: (C1) OT (kz, ~Rtip − ~Rα) = f (kz, ztip,~rtip − ~rα), where f is a real function (see Eq. (A10) in appendix A) and we have introduced cylindrical coordinates with the origin in the center of the molecule and the z axis perpendicular to the molecular plane. Every point ~R in the space is thus described by the triplet (z, r, θ) and we fix θ = 0 along the radius intersecting the atom with α = 0 (see Fig. 5). Finally, we have defined ~r to be the projection of ~R in the plane of the molecule. It follows immediately that ~rtip − ~rα =qa2 + r2 tip − 2artip cos(θtip − θα), (C2) Appendix C: Phase of the tunnelling amplitude gnl(θtip) = 15 where a is the distance between the carbon atoms and the center of the molecule, and θα = (2π/6)α with α = 0, . . . , 5. Combining Eqs. (C1), (8), (19) and (34) we obtain: φl( ~Rtip) = arg(Xα and, consequently: f [ztip, rtip, cos(θα − θtip)]eilθα) (C3) φl( ~Rtip) − lθtip = arg(Xα f (ztip, rtip, cos φα)eilφα) , (C4) where φα = θα − θtip. If now we expand f in the Taylor series: f (ztip, rtip, cos φα) = (cos φα)n (C5) we reduce the problem to the evaluation of the functions ∞Xn=0 f (n) n! (cid:12)(cid:12)(cid:12)(cid:12)(ztip,rtip,0) gnl(θtip) =Xα [cos(θα − θtip)]neil(θα−θtip), (C6) which is easily done by means of the Euler formula for the cosine and the binomial theorem. The solution reads: 2 6 n+6a−l 2nXc∈Z n ×(cid:12)(cid:12)(cid:12)cosh π ! e−i6cθtip (n + 6c − l)i(cid:12)(cid:12)(cid:12) 2 × θ(n + 6c − l + 2)θ(n − 6c + l + 2), (C7) with θ(x) = 1 if x > 0 and 0 elsewhere. By analyzing Eq. (C7) we obtain the following general properties: i) If θtip = nπ/6, with n ∈ N, the function gnl(θtip) is real, thus Eq. (35) is exact when ~Rtip is on the planes perpendicular to the molecule passing through the center of the molecule and one of the atoms or the center and the middle point of a carbon-carbon bond. ii) gn1 = 0 if n is even and gn2 = 0 if n is odd, ∀θtip. iii) gn1 is real for n ≤ 4 and gn2 is real for n ≤ 3. The combination of the observation ii) and iii) supports the validity of Eq. (35) on the entire space. Appendix D: The effective Hamiltonian In this section we analyze the form of the effective Hamiltonian Heff introduced in Eq. (39) both in the case of a constant interaction model or a more generic model for the interaction. The discussion will always be re- stricted to the subspace spanned by the 7 particle ground states in which the effective Hamiltonian reduces to a 4 × 4 matrix whose generic element is h7glτHeff7gl′τ′i. Since [Hm, Sz] = 0 it follows immediately that Heff is diagonal in the spin quantum number. Moreover one proves the following relations: tical and ii) that the diagonal elements in each of the two subblocks are equal. In order to prove the relations given in Eq. (D1) it is useful to introduce the symmetry operations Uspin and Uorb defined as follows: h7glτHeff7gl′τi = h7gl¯τHeff7gl′ ¯τi, h7glτHeff7glτi = h7g¯l¯τHeff7g¯l¯τi, (D1) dl¯σ = UspindlσU†spin, d¯lσ = UorbdlσU†orb. (D2) which ensures i) that the two spin subblocks are iden- The proof of the first relation in (D1) is readily given: 16 h7gnτHeff7gn′τi = = ll′ (E7g − Hm)pχ(E7g − Hm)dl′σ7gn′τi ll′ (Hm − E7g )pχ(Hm − E7g )d†lσ7gn′τii ll′ (E7g − Hm)pχ(E7g − Hm)dl′ ¯σ7gn′ ¯τi 1 2πXll′χhh7gnτd†lσΓχ +h7gnτdl′σΓχ 2πXll′χhh7gn¯τd†l¯σΓχ 1 +h7gn¯τdl′ ¯σΓχ ll′ (Hm − E7g )pχ(Hm − E7g )d†l¯σ7gn′¯τii = h7gn¯τHeff7gn′¯τi, (D3) where for the second equality we have introduced the identity operators U†spinUspin before and after the oper- ators dlσ and d†l′σ. The last equality is obtained by re- placing ¯σ → σ in the sum and remembering that Γχ ll′ is independent of the spin of the electron in the lead. The second relation in (D1) is obtained in an analogous way: h7gnτHeff7gnτi = = ll(E7g − Hm)pχ(E7g − Hm)dlσ7gnτi ll(Hm − E7g )pχ(Hm − E7g )d†lσ7gnτii ll(E7g − Hm)pχ(E7g − Hm)d¯lσ7g ¯nτi 1 2πXlχ hh7gnτd†lσΓχ +h7gnτdlσΓχ 2πXlχ hh7g ¯nτd†¯lσΓχ 1 +h7g ¯nτd¯lσΓχ ll(Hm − E7g )pχ(Hm − E7g )d†¯lσ7g ¯nτii = h7g ¯nτHeff7g ¯nτi, (D4) where the first equality is obtained by removing the sum over l′ since the Hamiltonian Hm conserves the z projec- tion of the angular momentum, the second equality pro- ceeds instead by inserting the identities U†orbUorb before and after the operators dlσ and d†l′σ. Finally, in the last equality, we have redefined ¯l → l and used the symmetry property of the rate matrices Γχ For the analysis of the off diagonal elements of Heff within a single spin subblock we have to distinguish be- tween the substrate and the tip case. In the substrate case ΓS ll′ ∝ δll′ which directly implies that also the com- ponent of Heff given by the coupling to the substrate is diagonal and, due to the second relation in (D1) propor- ll = Γχ ¯l¯l. tional to the identity matrix and thus irrelevant for the dynamics of the molecule. Thus, let us concentrate on the tip contribution. It is possible to demonstrate that: h7g + 2 τH T eff7g −2 τi = Ae−2iφ2 + Be−iφ1, (D5) where we have introduced the notation H T eff to indicate the component of Heff with χ = T , φ1 and φ2 are the phases of the tunnelling amplitudes calculated in the pre- vious section. Finally, A, B ∈ R are given by A = B = 1 2πXσ hh7g 2 τd†2σ(cid:12)(cid:12)ΓT +h7g 2 τd−2σ(cid:12)(cid:12)ΓT ReXσ hh7g 2 τd†1σ(cid:12)(cid:12)ΓT +h7g 2 τd3σ(cid:12)(cid:12)ΓT 2,−2(E7g − Hm)(cid:12)(cid:12) pT (E7g − Hm)d−2σ7g −2 τi 2,−2(Hm − E7g )(cid:12)(cid:12) pT (Hm − E7g )d†2σ7g −2 τii , 13(E7g − Hm)(cid:12)(cid:12) pT (E7g − Hm)d3σ7g −2 τi 13(Hm − E7g )(cid:12)(cid:12) pT (Hm − E7g )d†1σ7g −2 τii . 1 π 17 (D6) The proof of Eq. (D5) proceeds as follows. Let us start from the definition of the off diagonal matrix element: h7g 2 τH T eff7g − 2τi = 1 2πXll′χhh7g 2 τd†lσΓχ +h7g − 2τdl′σΓχ ll′ (E7g − Hm)pχ(E7g − Hm)dl′σ7g − 2τi ll′ (Hm − E7g )pχ(Hm − E7g )d†lσ7g − 2τii . (D7) The sums over l and l′ are a priori independent and run over all possible single particle angular momenta: l, l′ = −2, . . . , 3. The angular momentum conservation of Hm implies, nevertheless, that the combinations which contribute to the sum must satisfy the condition Finally, it is not difficult to prove, starting from Eq. (31), the following properties for the elements of the rate matrix ΓT : 2 − (−2) = l − l′ (mod 6), (D8) which restricts the sum to the three pairs: ΓT ll′ = ΓT ΓT ll′ = ΓT ll′e−i(φl−φl′ ), ¯ll′ = ΓT l ¯l′. (D10) l = +2, l = +1, l = +3, l′ = −2; l′ = +3; l′ = −1. (D9) Combining Eq. (D7) with (D9) and (D10), one obtains:   h7g 2 τH T eff7g −2 τi = 1 2πXσ hh7g 2 τd†2σ(cid:12)(cid:12)ΓT + h7g 2 τd−2σ(cid:12)(cid:12)ΓT + h7g 2 τd†1σ(cid:12)(cid:12)ΓT + h7g 2 τd3σ(cid:12)(cid:12)ΓT + h7g 2 τd†3σ(cid:12)(cid:12)ΓT + h7g 2 τd−1σ(cid:12)(cid:12)ΓT 2,−2(E7g − Hm)(cid:12)(cid:12) e−2iφ2pT (E7g − Hm)d−2σ7g −2 τi 2,−2(Hm − E7g )(cid:12)(cid:12) e−2iφ2pT (Hm − E7g )d†2σ7g −2 τi 13(E7g − Hm)(cid:12)(cid:12) e−iφ1 pT (E7g − Hm)d3σ7g −2 τi 13(Hm − E7g )(cid:12)(cid:12) e−iφ1 pT (Hm − E7g )d†1σ7g −2 τi 3,−1(E7g − Hm)(cid:12)(cid:12) e−iφ1pT (E7g − Hm)d−1σ7g −2 τi 3,−1(Hm − E7g )(cid:12)(cid:12) e−iφ1pT (Hm − E7g )d†3σ7g −2 τii , (D11) from which Eq. (D5) can be easily obtained. It is now interesting to explore the different limits of Eq. (D5). In the constant interaction picture, for example, the term proportional to B vanishes. The eigenstates of the in- teracting Hamiltonian Hm coincide in fact in the con- stant interaction model with the single Slater determi- 18 nant eigenstates of the non interacting one. In practice, the 7 particle ground state 7glτi can be written as: d†lτ6g 0 0i, with 6g 0 0i = +1 Yl=−1 Yτ =↑,↓ d†lτ0i. Thus, it follows immediately that: (D12) (D13) (42). This choice is particularly interesting among all others since if θtip = nπ/3 (for example when the tip is exactly above one of the carbon atoms) the operator L defined in Eq. (42) is the generator of the discrete rota- tions around the axis passing through the center of the molecule and the carbon atom closest to the tip. Finally let us consider under which conditions the effec- tive Hamiltonian commutes with the stationary density matrix evaluated only taking into account the tunnelling dynamics. By combining Eqs. (D5) and (38) one eqsily obtains for the 7 particle ground state with spin τ : d3σ7g −2 τi = 0, d†1σ7g −2 τi = 0. (D14) [Heff , σstat] = 2iBH Bσ sin(2φ2 − φ1)σz, (D16) By inserting Eq. (D14) into the second equality in (D6) one concludes that, in the constant interaction model, the effective Hamiltonian for the 7 particle ground state has the form: (Heff )7g = K Ae+2iφ2 Ae−2iφ2 K ! , (D15) where the hermitianicity of the Heff has been used. The constant K obtained from the direct evaluation of Eq. (39) is different from the off-diagonal constant A. Nevertheless, any contribution to the N, E, Sz subblock of the effective Hamiltonian which is proportional to the unity matrix does not influence the dynamics of the sys- tem (see Eq. (12)). Thus we chose to set K = A which gives the form of the Heff given by the Eqs. (41) and where σz is the third Pauli matrix and we have intro- duced the subscripts σ and H to distinguish between the constants proceeding from the density matrix and the ef- fective Hamiltonian. In the constant interaction picture BH = 0, while Bσ = 0 if the tip is respecting the rota- tional symmetry of the molecule, i.e. ~Rtip is on the prin- cipal rotational axis of the molecule. Finally a vanish- ing condition can be obtained also from the phases when 2φ2 − φ1 = nπ. By assuming the approximate expres- sion for the phase given by Eq. (35) one gets θtip = nπ/3 which corresponds to a tip belonging to one of the verti- cal mirror planes for the molecule intersecting a carbon atom. Notice that for these special values of θtip Eq. (35) is exact. For completeness we conclude with the results regard- ing the 5 particle ground state. The effective Hamiltonian for the generic description of the Coulomb interaction reads: (Heff )5g τ = K Ae−2iφ1 + Be−iφ2 + Ceiφ1 Ae−2iφ1 + Be−iφ2 + Ceiφ1 K ! , where A, B, C ∈ R are given by A = B = C = 1 2πXσ hh5g 1 τd†1σ(cid:12)(cid:12)ΓT +h5g 1 τd−1σ(cid:12)(cid:12)ΓT ReXσ hh5g 1 τd†2σ(cid:12)(cid:12)ΓT +h5g 1 τd0σ(cid:12)(cid:12)ΓT ReXσ hh5g 1 τd†3σ(cid:12)(cid:12)ΓT +h5g 1 τd1σ(cid:12)(cid:12)ΓT 1,−1(E5g − Hm)(cid:12)(cid:12) pT (E5g − Hm)d−1σ5g −1 τi 1,−1(Hm − E5g )(cid:12)(cid:12) pT (Hm − E7g )d†1σ7g −1 τii , 20(E7g − Hm)(cid:12)(cid:12) pT (E5g − Hm)d2σ5g −1 τi 20(Hm − E7g )(cid:12)(cid:12) pT (Hm − E7g )d†2σ5g −1 τii , 31(E7g − Hm)(cid:12)(cid:12) pT (E5g − Hm)d1σ5g −1 τi 31(Hm − E7g )(cid:12)(cid:12) pT (Hm − E7g )d†3σ5g −1 τii . 1 π 1 π (D17) (D18) In close analogy with the 7 particle case, one proves that B and C vanish in the constant interaction picture and also that for θtip = nπ/3 the effective Hamiltonian commutes with the stationary density matrix calculated only considering the tunnelling dynamics. 19 ∗ Electronic address: [email protected] 1 G. Binnig, H. Rohrer, Ch. Gerber, and E. Weibel, Phys. Rev. Lett. 49, 57 (1982). 2 G. Binnig, and H. Rohrer, Surf. Sci. 126, 236 (1983). 3 N. Garcia, C. Ocal, and F. Flores, Phys. Rev. Lett. 50, 2002 (1983). 4 J. Tersoff, and D. R. Hamann, Phys. Rev. Lett. 50, 1998 (1983). 5 N. Garcia, and F. Flores, Physica B+C (Amsterdam) 127B, 137 (1984). 6 A. Baratoff, Physica B+C (Amsterdam) 127B, 143 (1984). 7 E. Stoll, A. Baratoff, A. Selloni, and P. Carnevali, J. Phys. C 17, 3073 (1984). 8 J. Tersoff, and D. R. Hamann, Phys. Rev. B 31, 805 (1985). 9 P. K. Hansma, and J. Tersoff, J. Appl. Phys. 61, R1 (1987). 10 C. Noguera, J. Microsc. 152, Part 1, 3 (1988). 11 C. J. Chen, Phys. Rev. B 42, 8841 (1990). 12 G. Doyen, E. Koetter, J. P. Vigneron, and M. Scheffler Appl. Phys. A51, 281 (1990). 13 F. Calleja, A. Arnau, J. J. Hinarejos, et al. Phys. Rev. Lett. 92, 206101 (2004). 14 J. Buker, and G. Kirczenow, Phys. Rev. B 72, 205338 (2005). (2002). 29 D. A. Ryndyk, P. D'Amico, G. Cuniberti, and K. Richter, Phys. Rev. B 78, 085409 (2008). 30 F. Santandrea, L. Y. Gorelik, R. I. Shekhter, and M. Jon- son New J. Phys. 13, 103017 (2011). 31 R. Pariser, and R. G. Parr, J. Chem. Phys. 21, 466 (1953). 32 J. A. Pople, Tran. Far. Soc. 42, 1375 (1953). 33 E. Huckel, Z. Phys. 70 (3/4), 204 (1931). 34 J. C. Slater, and G. F. Koster Phys. Rev. B 94, 1498 (1954). 35 L. P. Kouwenhoven, D. G. Austing, and S. Tarucha, Rep. Prog. Phys. 64, 701 (2001). 36 L. Gross, N. Moll, F. Mohn, A. Curioni, G. Meyer, F. Hanke, and M. Persson Phys. Rev. Lett. 107, 086101 (2011). 37 W. J. Hehre, R. F. Stewart, and J. A. Pople , J. Chem. Phys. 51(6), 2657 (1980). 38 F. Jensen, Introduction to Computational Chemistry (John Wiley & Sons, New York, (1999). 39 J. Anchell, et al., Computer code NWCHEM, Pacific Northwest National Laboratory Richland, WA 99352-0999 (1998). 40 K. Blum, Density matrix Theory and Applications 15 C. Toher, I. Rungger, and S. Sanvito, Phys. Rev. B 79, (Plenum Press, New York, (1996). 205427 (2009). 16 C. J. Chen, Introduction to Scanning Tunneling Mi- Phys. Rev. B 79, 235404 (2009). 41 D. Darau, G. Begemann, A. Donarini, and M. Grifoni, croscopy (2) Oxford University Press, New York, (2008). 17 S. U. Nanayakkara, E. C. H. Sykes, L. C. Fernandez-Torres, M.M. Blake, and P.S. Weiss, Phys. Rev. Lett. 98, 206108 (2007). 18 J. Bardeen, Phys. Rev. Lett. 6, 2 (1961). 19 W. A. Hofer, A. S. Foster, and A. L. Shluger Rev. Mod. Phys. 75, 1287 (2003). 20 C. J. Chen, Phys. Rev. Lett. 65, 448 (1990). 21 C. J. Chen, J. Vac. Sci. Technol. A 9, 44 (1991). 22 M. Herz, F. J. Giessibl, and J. Mannhart, Phys. Rev. B 68, 045301 (2003). 23 J. Repp, G. Meyer, S. M. Stojkovic, A. Gourdon, and C. Joachim, Phys. Rev. Lett. 94, 026803 (2005). 24 P. Liljeroth, J. Repp, and G. Meyer, Science 317, 1203 (2007). 25 S. Gao, M. Persson, and B. I. Lundqvist, Sol. State. Comm. 84, 271 (1992). 26 K. Stokbro, C. Thirstrup, M. Sakurai, U. Quaade, Ben Yu-Kuang Hu, F. Perez-Murano, and F. Grey, Phys. Rev. Lett. 80, 2618 (1998). 27 G. Ma and H. Guo, Chem. Phys. Lett. 317, 315 (2000). 28 V. Wong and M. Gruebele Chem. Phys. Lett. 363, 182 42 A. Donarini, G. Begemann, and M. Grifoni, Nano Lett. 9, 2897 (2009). 43 A. Donarini, G. Begemann, and M. Grifoni, Phys. Rev. B 82, 125451 (2010). χ(x, y) + v⊥ tip(x, y) = 1 sub(x, y) = 0 while vk 44 C. W. J. Beenakker Phys. Rev. B 44, 1646 (1991). 45 For the calculation of the leads wave functions we have used a the potential of the form vχ = vk χ (z) where vk 2 mω2[(x − xtip)2 + (y −ytip)2]. The choice for the tip potential as a sum of par- allel and perpendicular components greatly simplifies the calculation since it allows for the separation of variables in the Schroedinger equation but it also assigns an unphysi- cal positive potential to the vacuum region. The calculated eigenfunction represent, though, a good approximation for the ones calculated with a more realistic potential, at least close to the tip region. 46 M. Abramowitz and I. Stegun Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Ta- bles, Dover Publications, New York (1965). 47 J.C. Slater, Phys. Rev. 36, 57 (1930). 48 D. Belkic, and H.S. Taylor, Physica Scripta 39, 226 (1989).
1902.07340	1	1902	2019-02-19T23:03:56	Room-temperature giant Stark effect of single photon emitter in van der Waals material	[ "cond-mat.mes-hall", "quant-ph" ]	Single photon emitters (SPEs) are critical building blocks needed for quantum science and technology. For practical applications, large-scale room-temperature solid-state platforms are required. Color centers in layered hexagonal boron nitride (hBN) have recently been found to be ultra-bright and stable SPEs at room temperature. Yet, to scale up solid-state quantum information processing, large tuning range of single photon energy is demanded for wavelength division multiplexing quantum key distribution, where indistinguishability is not required, and for indistinguishable single-photon production from multi-emitters. Stark effect can tune the single photon energy by an electric field, which however, has been achieved only at cryogenic temperature so far. Here we report the first room-temperature Stark effect of SPEs by exploiting hBN color centers. Surprisingly, we observe a giant Stark shift of single photon more than 30 meV, about one order of magnitude greater than previously reported in color center emitters. Moreover, for the first time, the orientation of the electric permanent dipole moment in the solid-state SPE is determined via angle-resolved Stark effect, revealing the intrinsic broken symmetries at such a color center. The remarkable Stark shift discovered here and the significant advance in understanding its atomic structure pave a way towards the scalable solid-state on-chip quantum communication and computation at room temperature.	cond-mat.mes-hall	cond-mat	Room-temperature giant Stark effect of single photon emitter in van der Waals material Yang Xia1, Quanwei Li1, Jeongmin Kim1, Wei Bao1, Cheng Gong1, Sui Yang1, Yuan Wang1 and Xiang Zhang1,2† 1 NSF Nanoscale Science and Engineering Center (NSEC), 3112 Etcheverry Hall, University of California, Berkeley, California 94720, USA. 2 Materials Sciences Division, Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, California 94720, USA. * These authors contributed equally to this work. † e-mail: [email protected] Single photon emitters (SPEs) are critical building blocks needed for quantum science and technology1. For practical applications, large-scale room-temperature solid-state platforms are required2,3. Color centers in layered hexagonal boron nitride (hBN) have recently been found to be ultra-bright and stable SPEs at room temperature4. Yet, to scale up solid-state quantum information processing, large tuning range of single photon energy is demanded for wavelength division multiplexing quantum key distribution, where indistinguishability is not required, and for indistinguishable single-photon production from multi-emitters. Stark effect can tune the single photon energy by an electric field, which however, has been achieved only at cryogenic first room-temperature Stark effect of SPEs by exploiting hBN color centers. Surprisingly, we observe a giant Stark shift of single photon more than 30 meV, about one order of magnitude greater than previously reported in color center emitters7 -- 11. Moreover, for the first time, the orientation of the electric permanent dipole moment in the solid-state SPE is far5 -- 8. Here we report temperature so the 1 determined via angle-resolved Stark effect, revealing the intrinsic broken symmetries at such a color center. The remarkable Stark shift discovered here and the significant advance in understanding its atomic structure pave a way towards the scalable solid-state on-chip quantum communication and computation at room temperature. Van der Waals (vdW) materials, ranging from semi-metallic graphene12 and semiconducting transition metal dichalcogenides13 to insulating hexagonal boron nitride (hBN), have enabled remarkable scientific and technological breakthroughs over the last two decades thanks to their exceptional electronic and optical properties. Both the single materials and the heterostructures have been exploited to demonstrate appealing device applications14, such as light emitting diodes15, lasers16 and optical modulators17. While most cases deal with classical information, only a few studies have been reported in the quantum regime at liquid helium temperature18 -- 23. Recently, color centers in hBN have emerged as superb room-temperature solid-state single photon emitters (SPEs)4, which opens up the possibilities of utilizing vdW materials as a platform for room-temperature solid-state quantum information systems. They are capable to work at room temperature, and among the brightest SPEs due to its high internal quantum efficiency. Moreover, high-efficiency photon extraction can be greatly facilitated by their intrinsic layered material structures4. Consequently, millions of linearly polarized photons per second can be easily detected without additional photon extraction structures. Furthermore, their facile integration with photonic and electrical components is highly preferred for integrated on-chip quantum information systems13,14,24. One major challenge for all solid-state single photon emitters is the random variation of emission energy caused by the inhomogeneity in local environment. Such variation breaks the indistinguishability of single photons from multiple emitters, which is critically required for large scale quantum computation, such as universal linear optics25 and boson sampling26. The 2 randomness also prevents scaling up the room temperature quantum communication systems from using wavelength-division multiplexing (WDM), due to the stringent requirement on the precision of photon energy placed by the narrow-band optics, although photon indistinguishability is not required for such applications3. Stark effect, which describes the shift of spectra lines by an external electric field (Fig 1c), can precisely control SPE photon energy and be facilely integrated into quantum systems6, advantageous over other tuning methods such as temperature27, strain28 and magnetic field29. The Stark effect (Fig 1c) has been used to tune the emission energy of quantum dots6, single photon emitters in layered WSe25, atomic emitters such as NV centers and SiV centers in diamond7 -- 9,11 and organic dye molecules10, all at liquid helium temperature, because either the emitters only produce single photons at low temperature, or the effect was too weak to observe at room temperature. In this paper, we report the first room-temperature Stark effect of an SPE, discovered in an hBN color center. Surprisingly, we observe a giant tuning range around 31 meV at 300 K which is one order greater than previously reported in color center SPEs30. Moreover, we develop a characterization method to resolve the angular dependence of the Stark effect to determine the underlying symmetry of the color center. With this method, we directly observe, for the first time, a dipolar pattern of the Stark shift that is well aligned with the optical polarization. This dipolar pattern unambiguously reveals an electric permanent dipole moment which proves the breaking of inversion and rotation symmetries at the hBN SPE. The discovered remarkably giant room-temperature Stark effect and the significant advance in understanding its atomic structure could enable new possibilities of quantum information technologies, such as WDM and indistinguishable single photon sources, at room temperature. To achieve large Stark shift of our hBN SPEs and to fully characterize its dependence on the amplitude and orientation of the local electric fields, we design the nano-scale four-electrode 3 device (Fig. 1). SPEs in multi-layer hBN nano-flakes are chosen due to their much better optical performance compared to those in monolayers4. Multiple electrodes are carefully designed to surround the SPEs such that we can control not only the amplitude but also the direction of the electric field. Taking the advantage of the nanofabrication, we are able to develop a down-scaled four-electrode device with gaps of 200 nm and 400 nm between the adjacent and diagonal electrodes, respectively, to achieve a large electric field of ~ 0.1 V/nm, orders of magnitudes higher than previous reports8,30. Fig. 1a, b shows microscope images of the fabricated four-electrode device. We locate the SPE on the hBN flake by a localization analysis of its photoluminescence (PL) profile with respect to the electrodes (details in Extended Data Figure 1). The high-quality single-photon emission from the hBN color center is verified by PL spectroscopy at room temperature before applying external electric fields (Fig 2a). The majority of its PL emission is attributed to the zero-phonon line (ZPL) at 2.088 eV. The narrow full width at half maximum (~ 7 meV) provides evidence as a high-quality emitter. Two small phonon sidebands (PSBs) are observed at 1.921 eV (PSB1) and 1.753 eV (PSB2), with the frequency difference of ~1370 cm-1 corresponding to the E2g phonon of hBN31. The well resolved doublet at PSB1 is a typical feature for hBN nanoflake15 -- 17. A few tiny peaks are visible that might result from the PL emission of other color centers in the collected region. The emission of the hBN SPE is linearly polarized (blue circles and curve in Fig. 3b), which fits well to a cosine-squared function with a visibility of 0.72. Many SPEs are characterized under the identical pump laser polarization and the detected photons are linearly polarized in various directions, thus the polarization observed here is specific to the SPE and not due to optical excitation. We measure the second-order coherence function (g(2)) after device fabrication using a Hanbury Brown and Twiss (HBT) setup, from which single photon emission is confirmed by a raw anti-bunching dip of g(2)(0) = 0.45 (Fig. 2a inset). By fitting the g(2) data to a single exponential decay function, we 4 estimate the lifetime of our single photon emitter to be 4.2 ns. After characterizing and confirming the SPE optical properties, we apply voltages within ±100 V between the electrodes A and B (Fig. 2b inset) to study the Stark effect. PL emission spectra are collected at each voltage. Fig. 2b plots the ZPL PL intensity map as a function of photon energy and applied voltage. A huge Stark shift of 31 meV is clearly observed, about one order of magnitude greater than the values previously achieved in color center SPEs7,8,30 and 4-fold larger than its room-temperature linewidth (~ 7meV). To further analyze the effect, we extract the ZPL peak position as a function of applied voltage in Fig 2c. The Stark shift is approximately linear to the applied voltage with a tuning efficiency of 137 μeV/V and reverses sign at opposite electric field, which suggests a non-zero electric permanent dipole moment at the color center. The presence of degenerate states is ruled out, because there is neither ZPL splitting nor multi-peaks with applied fields (even at different excitation polarizations)8. The deviation from linearity is possibly due to the light induced ionization in nearby non-emitting defects that can modify the local electric field8. A gradual decrease in its intensity is noted when voltage changes from -100 V to 100 V (Extended Data Fig. 2a), which is possibly due to the change of coupling to dark state as reported previously in diamond NV center7. The repeatability and stability of such room-temperature giant Stark effect is further characterized in multiple emitters as shown in Extended Data Figures 3-7. Stark shift depends not only on the magnitude of the applied field, but also its orientation32. The underlying symmetry of the atomic structure of the color center thus can be probed by the angular dependence of the Stark effect. We further develop a Rotating Field Method that fully control both the magnitude and orientation of the total electric field at the location of SPE by assigning the voltages of multiple electrodes (Methods). With a fixed local electric field magnitude of 0.08 V/nm, the ZPL is 1.13 meV red (1.32 meV blue) shifted when the applied 5 field pointing to 140° (320°), while the shift with the electric field along 230° is negligible. The Stark shifts hΔν as a function of the angle θ of the local field F is well-fitted with electric permanent dipole moment model (Fig 3b) based on perturbation theory to the first order: h n D = -D × = - D μ F μ F cos ( ) f q - (1) where Δμ and ϕ denote the dipole moment responsible for the Stark effect and its orientation angle, respectively. Such result further justifies that the Stark effect is dominated by an electric permanent dipolar term. From the fitting we estimate the magnitude of the dipole is Δμ = 0.65 ± 0.04 D where 1 D = 3.33e-30 Cm, which is on the same order as the NV center in diamond7. The discovered electric permanent dipole moment corresponds to the asymmetric charge distribution at the hBN SPE, which will facilitate the future study of the atomic structure and electronic levels of the color center. In contrast, the linear polarization of emitted photons characterized in earlier reports is determined only by the optical transition dipole and features cosine-squared angular dependence, which omits the information for inversion symmetry of atomic structure (see Methods). We clarify such distinctive difference in Fig. 3b (blue circles). It should be noted that the direction for maximum Stark shift is coincident with that of the emission polarization (Fig. 3b Inset), which corresponds to the intersection of the mirror symmetry plane of the color center and the hBN atomic layer plane. We emphasize that the key to observe the room-temperature giant Stark shift is a combined effort of several crucial factors. First, the large band gap of hBN crystal and low phonon scattering make a superb room-temperate SPE. Second, the layered structure of hBN likely leads to an in-plane dipole moment4, such that an applied in-plane electric field can be well aligned with the dipole orientation. Third, closely spaced electrodes (within a few hundreds of nanometers) enabled by nanofabrication can easily be integrated with the nano-scale hBN flake such that large voltages (even above 100V) can be applied for local electric fields (on the order of 0.1V/nm) around one order of magnitude greater than previous works in diamond color 6 centers30. To conclude, for the first time, we report the room-temperature Stark effect of an SPE, with a giant shift of 31 meV in hBN color center. The demonstrated large and reliable energy tuning and modulation is important for room-temperature quantum applications such as wavelength-division-multiplexing for practical quantum communication. Moreover, for the first time we uncover the intrinsic broken symmetries of color centers in hBN by introducing the Rotating Field Method to enable angular mapping of the Stark effect. Our results provide fundamental knowledge for the understanding and applications of color centers in vdW materials and open a new route towards scalable solid-state quantum information systems at room temperature. 7 Methods Device fabrication. The hBN nanoflakes are purchased from Graphene Supermarket, in the form of a liquid suspension and drop-cast on silicon substrate with ~ 280 nm thermal oxide on top. The samples are then annealed at 1000 C for 30 min in an Ar/H2 environment followed by slow cooling down. Individual single photon emitters are found by fluorescence microscopy and characterized by PL emission spectroscopy, polarization analysis and g(2) measurement. Electron beam lithography is used to define the electrode pattern around the located single photon emitters. The electrodes are made of 5 nm Ti and 100 nm Au deposited by electron beam evaporation. Rotating Field Method. When we apply an electric field via multiple electrodes, the total external field can be considered as linear combination of the fields generated by individual electrodes. In our experiment, two voltage signals are applied to electrodes A and C (Fig. 3b inset), while the other two electrodes and substrate are grounded. From linear combination, we have the equation below: E E x y é ê ë ù ú û = K V é A ê V ë C ù ú û , where Ex and Ey are the x and y components of external electric field at the single photon emitter location, K is a 2-by-2 matrix, and VA and VC are the voltages applied to electrodes A and C respectively. Matrix K is obtained from 3D FEM simulation (COMSOL). The simulated geometry is extracted from the real device. In order to obtain the first column of K, we assign VA = 1 V and ground electrodes B, C, D. The obtained Ex and Ey form first column of K. And the second column can be calculated similarly. In order to generate a local field with specific amplitude F0 and direction θ, we consider the following equation: E E x y é ê ë ù ú û = ( ) 0 cos q é F ê ( ) L sin q ê ë ù ú ú û . 8 Here we follow previous works and use Lorentz approximation to calculate the local field from the external field. L = (εr+2)/3 is the Lorentz factor. The relative permittivity εr is taken from a reference31. Combining the two equations above gives ( ) cos q ) ( sin q V é ê V ë C F 0 L ù ú û K = - é 1 ê ê ë ù ú ú û A . The structure information from the discovered electric permanent dipole moments. The electric permanent dipole moments correspond to the charge distributions of the electronic states of the SPEs. It can be calculated for the ground and excited states of an SPE as μ g,e = ò y y , g e r e * , g e 3 dr where ψ is the wave function of electronic states, and the subscriptions g and e correspond to ground and excited states, respectively. Such dipole moments contribute to Stark shift through Δµ = µe-µg (Eq (1)). A non-vanishing Δµ indicates non-zero µe and/or µg, which must result from non-inversion symmetric probability densities of electrons ψe2 and/or ψg2 as well as atomic structure. μ On the contrary, the optical polarization is determined by optical transition dipole moment ® = ò the directions normal to it. After polarizer, the detected optical intensity has a squared-cosine , which emits an optical wave with electric field parallel to the dipole, along y y e r e * dr 3 g g e dependence I(θ) = I0cos2(θ) on polarization angle θ, which returns itself after θ→ θ+180°. As such measurement is always inversion symmetric, it cannot tell whether inversion symmetry breaks or not at the emitter. Data availability. The data that support the findings of this study are available from the corresponding author upon reasonable request. 9 References 1. Charles Santori, David Fattal & Yoshihisa Yamamoto. Single-photon devices and applications. (Wiley-VCH, 2010). 2. O'Brien, J. L., Furusawa, A. & Vučković, J. Photonic quantum technologies. Nature Photonics 3, 687 -- 695 (2009). 3. Aharonovich, I., Englund, D. & Toth, M. Solid-state single-photon emitters. Nat Photon 10, 631 -- 641 (2016). 4. Tran, T. T., Bray, K., Ford, M. J., Toth, M. & Aharonovich, I. Quantum emission from hexagonal boron nitride monolayers. Nat Nano 11, 37 -- 41 (2016). 5. Chakraborty, C. et al. Quantum-Confined Stark Effect of Individual Defects in a van der Waals Heterostructure. Nano Lett. 17, 2253 -- 2258 (2017). 6. Dots, N. Q. Quantum-Confined Stark Effect in Single CdSe. Science 278, 2114 -- 2114 (1997). 7. Tamarat, P. et al. Stark Shift Control of Single Optical Centers in Diamond. Phys. Rev. Lett. 97, 083002 (2006). 8. Bassett, L. C., Heremans, F. J., Yale, C. G., Buckley, B. B. & Awschalom, D. D. Electrical Tuning of Single Nitrogen-Vacancy Center Optical Transitions Enhanced by Photoinduced Fields. Phys. Rev. Lett. 107, 266403 (2011). 9. Bernien, H. et al. Two-Photon Quantum Interference from Separate Nitrogen Vacancy Centers in Diamond. Physical Review Letters 108, (2012). 10. Lettow, R. et al. Quantum Interference of Tunably Indistinguishable Photons from Remote 10 Organic Molecules. Phys. Rev. Lett. 104, 123605 (2010). 11. Sipahigil, A. et al. Quantum Interference of Single Photons from Remote Nitrogen-Vacancy Centers in Diamond. Physical Review Letters 108, (2012). 12. Geim, A. K. & Novoselov, K. S. The rise of graphene. Nature Materials 6, nmat1849 (2007). 13. Xia, F., Wang, H., Xiao, D., Dubey, M. & Ramasubramaniam, A. Two-dimensional material nanophotonics. Nat Photon 8, 899 -- 907 (2014). 14. Geim, A. K. & Grigorieva, I. V. Van der Waals heterostructures. Nature 499, 419 -- 425 (2013). 15. Sundaram, R. S. et al. Electroluminescence in Single Layer MoS2. Nano Lett. 13, 1416 -- 1421 (2013). 16. Ye, Y. et al. Monolayer excitonic laser. Nature Photonics 9, 733 (2015). 17. Liu, M. et al. A graphene-based broadband optical modulator. Nature 474, 64 (2011). 18. Koperski, M. et al. Single photon emitters in exfoliated WSe2 structures. Nat Nano 10, 503 -- 506 (2015). 19. Tonndorf, P. et al. Single-photon emission from localized excitons in an atomically thin semiconductor. Optica 2, 347 (2015). 20. Srivastava, A. et al. Optically active quantum dots in monolayer WSe2. Nat Nano 10, 491 -- 496 (2015). 21. He, Y.-M. et al. Single quantum emitters in monolayer semiconductors. Nat Nano 10, 497 -- 502 (2015). 22. Chakraborty, C., Kinnischtzke, L., Goodfellow, K. M., Beams, R. & Vamivakas, A. N. 11 Voltage-controlled quantum light from an atomically thin semiconductor. Nat Nano 10, 507 -- 511 (2015). 23. Palacios-Berraquero, C. et al. Atomically thin quantum light-emitting diodes. Nature Communications 7, ncomms12978 (2016). 24. Pospischil, A. et al. CMOS-compatible graphene photodetector covering all optical communication bands. Nature Photonics 7, nphoton.2013.240 (2013). 25. Knill, E., Laflamme, R. & Milburn, G. J. A scheme for efficient quantum computation with linear optics. nature 409, 46 -- 52 (2001). 26. Aaronson, S. & Arkhipov, A. The Computational Complexity of Linear Optics. in Proceedings of the Forty-third Annual ACM Symposium on Theory of Computing 333 -- 342 (ACM, 2011). doi:10.1145/1993636.1993682 27. Reithmaier, J. P. et al. Strong coupling in a single quantum dot -- semiconductor microcavity system. Nature 432, 197 -- 200 (2004). 28. Flagg, E. et al. Interference of Single Photons from Two Separate Semiconductor Quantum Dots. Physical Review Letters 104, (2010). 29. Stevenson, R. M. et al. A semiconductor source of triggered entangled photon pairs. Nature 439, 179 -- 182 (2006). 30. Müller, T. et al. Wide-range electrical tunability of single-photon emission from chromium-based colour centres in diamond. New J. Phys. 13, 075001 (2011). 31. Geick, R., Perry, C. H. & Rupprecht, G. Normal Modes in Hexagonal Boron Nitride. Phys. Rev. 12 146, 543 -- 547 (1966). 32. Brunel, C., Tamarat, P., Lounis, B., Woehl, J. C. & Orrit, M. Stark Effect on Single Molecules of Dibenzanthanthrene in a Naphthalene Crystal and in a n -Hexadecane Shpol'skii Matrix. The Journal of Physical Chemistry A 103, 2429 -- 2434 (1999). 33. Akselrod, G. M. et al. Probing the mechanisms of large Purcell enhancement in plasmonic nanoantennas. Nature Photonics 8, 835 -- 840 (2014). 34. Grange, T. et al. Cavity-Funneled Generation of Indistinguishable Single Photons from Strongly Dissipative Quantum Emitters. Physical Review Letters 114, (2015). 13 Figure 1 Device and physics of the Stark effect in hBN single photon emitter (SPE) at room temperature. a, Two-channel optical image of the fabricated four-electrode device (Yellow: bright-field image of the gold electrodes; Red: photoluminescence (PL) image of the hBN SPE). b, Zoom-in pseudo-color SEM image of the same device. The hBN nano-flake (purple) hosts the SPE (red dot) whose position is found from a localization image analysis of (a) (details in Extended Data Fig. 1). A, B, C and D (yellow) denote the four gold electrodes where voltages are applied to generate external electric fields. c, Illustration of the Stark effect of the SPE (represented by a two-level system) with an optical transition from the excited state 𝑒⟩ to the ground state 𝑔⟩. The emitted photon energy is tuned via shifting the electronic levels by a energy are broadened (characterized by ∆𝐸(𝑇)) due to electron-phonon scattering, which sets local electric field F. At room temperature, the electronic levels and thus the emitted photon the minimum Stark shift needed for practical use. 14 Figure 2 Observation of room-temperature giant Stark effect in hBN SPE. a, Photoluminescence (PL) spectrum of the SPE at 300 K without applying electric field. It shows a dominating zero-phonon line (ZPL) at 2.088 eV with 7 meV full width at half maximum (FWHM) and two phonon sidebands (PSB1 and PSB2). The small peak at 1.988 eV stems from another emitter nearby. Inset: The measured (circles) and fitted (red curve) second order coherence function g(2) of the SPE PL after device fabrication. The g(0)(0) of 0.45 demonstrates the single-photon nature. The measurement data is well fitted by a single exponential decay with a lifetime of 4.2 ns. b, ZPL spectra of the hBN SPE as a function of voltage applied to electrodes A and B (inset) with equal magnitude at opposite signs. The achieved Stark shift is up to 31 meV, one order greater than previous reports in color center SPEs7,8,30 and 4 times greater than its own room-temperature linewidth. c, Voltage controlled ZPL energy extracted from emission spectra 15 fitting from b. The blue and orange dots correspond to experiment data obtained during the forward and backward sweeping of voltages, respectively. Error bar, 95% confidence interval of the fitting. A tuning efficiency of 137 µeV/V is obtained by linear regression (yellow solid line). Insets show g(2) of the device measured at ±10 V, certifying that the single photon emission remains under external electric fields. The spectra and g(2) are measured under the excitations of continuous-wave 473 nm and 532 nm lasers, respectively. The acquisition time for g(2) is 10 s. The excitation intensity is 100 uW/μm2 for all measurements. 16 Figure 3 Angle-resolved Stark effect of hBN SPE and the discovered symmetry breaking. a, ZPL spectra of hBN SPE recorded with electric fields applied in various orientations θ (defined in the inset) with a fixed magnitude (F = 0.08 V/nm). Such electric fields are generated by applying voltages to electrodes A and C (inset see Methods for details). The zero-field spectrum is plotted for comparison. When we apply electric field along 140° (320°) directions, 1.32 meV red (1.13 meV blue) shift of ZPL is observed. On the contrary, the electrical field along 230° does not cause noticeable change in ZPL spectrum. b, Angle-resolved Stark shift (orange color, left y axis) and optical polarization data (blue color, right y axis). The x axis corresponds to the orientation angle of the applied electric field and that of the polarizer in front of the photodetector, respectively. The orange squares (blue circles) are the measured Stark shift (ZPL intensity) of the single photons, and the solid orange line (dashed blue line) is the fitting curve according to the electric permanent dipole model in Eq. (1) in text (to the linearly polarized emission in cosine-squared function). The unveiled electric permanent dipole moment uncovers the broken inversion and rotation symmetries at the atomic color center. The electric permanent dipole moment aligns well with the emission polarization. The inset shows the same data in polar coordinates. Three vertical dashed lines in the main panel (blue, magenta and red) correspond to the three spectra in (a). The photon energies in (b) are obtained by fitting the ZPLs with the Lorentzian line shape. The error bars from fitting are smaller than 0.03 meV. The Stark shift is measured at 80 K to reduce ZPL fitting uncertainty at small shifts, taking advantage of 17 linewidth at the narrow room-temperature. low temperature. The optical polarization is measured at 18 Author Contributions Y.X., Q.L. and X.Z. conceived the idea, initiated the project; Y.X. designed and fabricated the devices; W.B. assisted the fabrication; Q.L. prepared the emitters and performed optical measurements; J.K. did emitter localization microscopy; Y.X. and Q.L. measured the Stark effect and analyzed data; C.G. assisted the data analysis; Y.X., and Q.L. prepared the manuscript; X.Z., Y.W. and S.Y. supervised the project; all authors contributed to discussions and manuscript revision. 19
1008.0297	1	1008	2010-07-29T10:44:57	Manipulating thermal conductivity through substrate coupling	[ "cond-mat.mes-hall", "physics.comp-ph" ]	We report a new approach to the thermal conductivity manipulation -- substrate coupling. Generally, the phonon scattering with substrates can decrease the thermal conductivity, as observed in recent experiments. However, we find that at certain regions, the coupling to substrates can increase the thermal conductivity due to a reduction of anharmonic phonon scattering induced by shift of the phonon band to the low wave vector. In this way, the thermal conductivity can be efficiently manipulated via coupling to different substrates, without changing or destroying the material structures. This idea is demonstrated by calculating the thermal conductivity of modified double-walled carbon nanotubes and also by the ice nanotubes coupled within carbon nanotubes.	cond-mat.mes-hall	cond-mat	Manuscript Manipulating thermal conductivity through substrate coupling Zhixin Guo, Dier Zhang and Xin-Gao Gong∗ MOE Key Laboratory for Computational Physics, and Surface Physics Laboratory (National Key), Fudan University, Shanghai 200433, China∗ (Dated: October 24, 2018) We report a new approach to the thermal conductivity manipulation -- substrate coupling. Generally, the phonon scattering with substrates can decrease the thermal conductivity, as observed in recent experiments. However, we find that at certain regions, the coupling to substrates can increase the thermal conductivity due to a reduction of anharmonic phonon scattering induced by shift of the phonon band to the low wave vector. In this way, the thermal conductivity can be efficiently manipulated via coupling to different substrates, without changing or destroying the material structures. This idea is demonstrated by calculating the thermal conductivity of modi- fied double-walled carbon nanotubes and also by the ice nanotubes coupled within carbon nanotubes. Keywords: Thermal conductivity, manipulate, substrate coupling, molecular dynamics PACS numbers: With the shrinkage of electronic devices to the nanoscale[1 -- 4] and the revival of thermoelectrics[5 -- 7], thermal transport property of nanomaterials has become more and more significant. So far, great efforts have been done on finding/synthesizing new nanomaterials with particularly high/low thermal conductivities.[7 -- 10] Recently, some efforts have taken headway on the sub- ject of manipulating the thermal conductivity via doping, adsorbing, or generating defects.[11 -- 15] These processes can only reduce but hardly enhance the thermal con- ductivity, since the phonon scattering in a crystal lattice usually be less than that with the doping or defects. Fur- thermore, most of these treatments largely destroy the structure and thus the corresponding properties of nano- materials, which make them inapplicable in the nanode- vices. Very recently, the effects of coupling with substrates on In general, thermal conductivity were reported[16 -- 18]. when a conductive material is coupled with substrate, its thermal conductivity is expected to be decreased owning to the additional phonon scattering with the substrate as observed in the recent experiments.[16, 17] In this letter, we find that, at certain regions, the effect of phonon scat- tering can be suppressed and the thermal conductivity of nanomaterials can be surprisingly increased due to the coupling induced shift of phonon band to the low wave vector. Based on this finding, we propose a new approach to thermal conductivity manipulation -- coupling with dif- ferent substrates. This approach naturally extends the capability of conventional treatments on thermal con- ductivity without destroying the structures of materials. Since in the production of nanodevices, the conductive nanomaterials are always placed on certain substrates, our approach has great potential for advancing the per- formance of nanoelectronic and thermoelectric devices. ∗Electronic address: [email protected] TH TL Chain-1 Chain-2 FIG. 1: Schematic configuration of our approach to manipu- late thermal conductivity. Chain-1 and chain-2 represent the conductive material and the substrate, respectively. The two particles on the left and right ends of chain-1 are put into contact with thermostats of temperature TH and TL, respec- tively. In order to demonstrate this approach, we start with a coupled Fermi-Pasta-Ulam(FPU) chain[19, 20] model. Then we use two examples, thermal conductivity of both modified double-walled carbon nanotubes (DWNTs) and coupled ice nanotubes (Ice-NTs)[21], to demonstrate the applicability in real systems. Fig. 1 shows the schematic configuration of our ap- proach, which is illustrated by a coupled atom chain model. The upside chain (chain-1) represents the con- ductive material and the underside chain (chain-2) rep- resents the substrate. The two ends of chain-1 are con- tacted with the thermostats with temperature TH and TL, respectively, while, chain-2 is free of thermostat con- tact. Each node of chain-1, except for the two ends con- tacted with thermostat, is coupled to the corresponding underside node of chain-2. This ladder-like construction corresponds well to the real systems, where the conduc- 0.04 0.03 0.02 J 0.01 0.00 Case (a) Case (b) Case (c) p /4 p /8 p /16 p /32 1 10 (m , k ) 2 2 p /32 p /16 p /8 p /4 FIG. 2: Heat flux of chain-1 and chain-2 with variation of resonance angle Ψ in the coupling system for three cases: (a) k2 =1, m2 is varied; (b) m2=1, k2 is varied; (c) both m2 and k2 are varied but keeping m2=k2. The chain length N is 50, and TH =0.3, TL=0.2. The solid squares, triangles, and circles represent the heat flux of chain-1 for cases (a), (b), and (c), respectively; the corresponding open ones represent heat flux of chain-2 for the three cases. The short dashed line shows the heat flux of isolated chain-1. Chain-1's heat flux monotonously decreases with Ψ increasing, which can be either larger or smaller than that of the isolated one depending on the specific Ψ value. The inset shows (m2, k2) dependence of the resonance angle for the three cases. The resonance angle is very sensitive to the (m2, k2) parameters. tive materials are always fixed on some substrates. Two coupled FPU chains are firstly considered to rep- resent the approach in details, with all the Hamiltonian parameters being in the reduced units.[19, 20] In the model, the atomic mass and spring constant of chain- 1 (m1, k1) are fixed to be 1.0, the anharmonic coefficient of chain-1 and chain-2 (β1, β2) and the coupling strength kc are all set to be 0.5. The atomic mass and spring constant of chain-2 (m2, k2) are variables to manipulate thermal conductivity of chain-1 (see Supporting Informa- tion, SI.I). The nonequilibrium molecular dynamics (NEMD)[19, 20, 22] method is used to calculate the heat flux of chain- 1 and chain-2 in the coupling system. It is found that the phonon resonance between chain-1 and chain-2 plays an important role on the thermal conductivity. Here to describe the phonon resonance strength, where λ1 and λ2 are the phonon amplitudes of chain-1 and chain-2 after coupling (see Supporting Information, SI.II). It is obvious that the resonance strength becomes a maximum (minimum) at Ψ = π we use the resonance angle Ψ=(cid:12)(cid:12)(cid:12) arctan( λ2 λ1 )(cid:12)(cid:12)(cid:12) 4 (Ψ = 0). Fig. 2 shows heat flux J of chain-1 and chain-2 with variation of the resonance angle. Since Ψ is not uniquely determined by m2 or k2, many (m2, k2) sets can be cor- responding to a Ψ value, we mainly consider three cases: (a) k2 =1, m2 is varied; (b) m2=1, k2 is varied; (c) both 2 m2 and k2 are varied but keeping m2=k2. For all the three cases, the heat flux of chain-1 monotonously de- creases with Ψ increasing, while the heat flux of chain-2 has a contrary behavior. Particularly, when the phonon resonance becomes strong enough (with Ψ nearby π 4 ), heat flux of chain-1 gets much smaller than that of the isolated one (J = 0.034), suggesting a substantial reduc- tion of thermal conductivity[23]; when the phonon res- onance becomes small enough, however, the heat flux of chain-1 gets obviously larger than that of the iso- lated one, showing an increment of thermal conductiv- ity, which has not been explored so far. This interesting phenomenon shows that the thermal conductivity of con- ductive material can be efficiently manipulated through the substrate. We note that the resonance angle is very sensitive to the (m2, k2) parameters (Inset of Fig. 2). This suggests that one can easily find a proper substrate to manipulate the thermal conductivity in the application. For example, to decrease the thermal conductivity, the substrate needs to have similar atomic mass and spring constant with the conductive material; to increase the thermal conductiv- ity, the atomic mass and/or spring constant of substrate only needs to be several times larger than that of conduc- tive material. Moreover, the heat-flux curves of cases (a), (b), and (c) are very close to each other, indicating that the thermal conductivity of chain-1 is insensitive to the specific type of (m2, k2) compositions. Thus there would be rich choices of substrate candidates for the thermal conductivity manipulation. Consequently, the present approach can be easily realized in the nanotechnologies, which provides a clear direction for designing advanced nanoelectronic and thermoelectric devices. The reduction of thermal conductivity after coupling can be understood from the phonon-resonance effect, which induces strong phonon scattering and thus reduces the thermal conductivity.[17, 24 -- 27] To understand the increment of thermal conductivity, however, we need to invoke phonon band theory. Here we consider an extreme condition, where the resonance angle is zero and the in- crement of thermal conductivity is maximum. From the coupling-harmonic-oscillator (CHO) model, the phonon dispersion of chain-1 in the coupling system can be writ- 2 q) + kc. Compared with ten as ω(q) = q k1 that of the isolated case ω(q) = 2q k1 2 q), the phonon band of chain-1 has an obvious upshift after cou- pling, the magnitude of which is proportional to √kc (In- set of Fig. 3). m1 · q4 sin2( 1 m1 · sin( 1 Based on the phonon dispersion, the phonon energy density (PED) distributions of chain-1 with wave vector q can be further calculated (see Supporting Information, SI.III). Compared with that of isolated case, the PED peak has an obvious shift to the small-q direction own- ing to the upshift of phonon band after coupling (Fig. 3), suggesting more energy has been carried and trans- ported by the small-q phonons. The phonons' scattering power is proportional to the phase difference between Y Y ) . . ( U A S O D E 2.0 1.6 1.2 0.8 0.4 0.0 0 Coupled case Isolated case 1 2 3 q Coupled case Isolated case 0.0 0.5 1.0 2.0 2.5 3.0 1.5 q FIG. 3: PED distributions with the wave vector q of chain-1 in both the isolated and the coupled (with Ψ = 0) cases at T=0.25. The PED peak has an obvious shift to the small-q direction after coupling. The inset shows the corresponding phonon dispersions. 0.040 0.036 J 0.032 0.028 0.024 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 N J Coupled case Isolated case K~L0.48 K~L0.40 50 100 150 200 250 300 N Isolated case 0.2 0.4 T 0.6 FIG. 4: (a) Temperature dependence of heat flux of isolated chain-1 (N =50). The phonon scattering effect has become dominant at our simulation temperature (T=0.25). The inset shows chain length N dependence of total heat flux J N of chain-1 in both the isolated and the coupled (with Ψ = 0) cases. The larger fitting exponent implies that the increment of thermal conductivity by the coupling gets more and more distinguished with the chain length increasing. different atoms, which is hence proportional to the q value.[28] More small-q phonons being responsible for the heat transport corresponds to smaller scattering power, thus larger phonon mean free paths (PMFPs) in chain- 1, which increases the thermal conductivity (positive ef- fect). On the other hand, the upshift of the phonon band also results in less phonons being excited for the heat transport, which in turn reduces the thermal conductiv- ity (negative effect). The thermal conductivity varia- tion induced by the phonon band upshift is attributed to such two effects competing with each other (phonon- band-upshift effect). 3 We have also calculated the temperature dependence of heat flux of isolated chain-1 (Fig. 4) and found the heat flux monotonically decreases with the temperature increasing from T=0.15 to 0.65. It is known that in- creasing temperature has two competitive effects on the thermal conductivity: (i) It excites more high-frequency phonons that enhance the thermal conductivity. (ii) It increases phonon-phonon scattering that reduces the thermal conductivity.[13, 20] Fig. 4 shows that effect (ii) has become dominant at our simulation temperature (T=0.25), indicating that the upshift of phonon band, which has similar effect with that of decreasing temper- ature, has more positive effect than the negative effect on the thermal conductivity. Hence the thermal conduc- tivity can be increased due to the phonon-band-upshift effect. Moreover, inset of Fig. 4 shows the chain length dependence of total heat flux(J N ) of chain-1 in both the coupled (with Ψ = 0) and isolated cases. As is shown, J N of the isolated chain-1 diverges as J N ∝ N 0.4, consis- tent with the results in previous reports.[19, 20] While, for the coupled case, J N ∝ N 0.48. The larger fitting exponent implies that the increment of thermal conduc- tivity by the coupling gets more and more distinguished with the chain length increasing. Consequently, in the coupling system, the thermal con- ductivity variation is owed to both the phonon-band- upshift effect and the phonon-resonance effect that com- pete with each other. From the results above (Fig. 2), the phonon-resonance effect can be easily manipulated by changing the atomic mass and/or spring constant of the substrate. Thus we can efficiently manipulate the ther- mal conductivity of conductive material through coupling it to different substrates. As a demonstration of this approach in the real sys- tems, we have calculated the thermal conductivity of a (7,0) carbon nanotube (CNT) coupled within (14,0) CNT as a substrate((7,0)@(14,0) DWNT, see Supporting In- formation, SI.IV). We set the atomic mass of (14,0) CNT (MCN T 14) as parameter to see how the thermal conduc- tivity of (7,0) CNT changes with the resonance angle. In the coupling system, the resonance angle Ψ and thus the thermal conductivity can be efficiently manipulated through changing MCN T 14. As shown in Fig. 5(a), ther- mal conductivity of (7,0) CNT obviously increases with MCN T 14 increasing from one to several times that of atomic mass of (7,0) CNT (carbon-atom mass, MCN T 7) owing to the reduction of phonon-resonance effect. Com- pared with that of the isolated case, the thermal conduc- tivity of (7,0) CNT can be either substantially decreased or increased depending on the value of MCN T 14, consis- tent with the results of coupled FPU chain model, which can be well understood form the coupling mechanism dis- cussed above. When MCN T 14 = MCN T 7 (Ψ ≈ π 4 ), the phonon-resonance effect that decreases the thermal con- ductivity is dominant and the thermal conductivity is decreased by coupling. With MCN T 14 increasing, the phonon-resonance effect becomes more minor. When MCN T 14 ≥ 2MCN T 7 (Ψ < π 32 )[29], the phonon-band- w ) / K m W K ( (a) (7,0) CNT 249 246 243 240 237 234 231 228 1 2 3 4 M /M CNT14 CNT7 ) / K m W K ( (b) 8 7 6 5 Isolated case Coupled case 4 6 Tube index(N) FIG. 5: (a) Thermal conductivity of (7,0) CNT coupled within (14,0) CNT, with variation of (14,0) CNT's atomic mass MCN T 14 (MCN T 7 is kept to be carbon-atom mass). The short dashed line shows the thermal conductivity value of isolated (7,0) CNT. The thermal conductivity of (7,0) CNT can be either substantially decreased or increased depending on the value of MCN T 14. The inset shows the structure of (7,0)@(14,0) DWNT. (b) Thermal conductivity of Ice-NTs. Thermal conductivity of Ice-NTs has an obvious increment after coupling. The inset shows optimized structures of ice- NTs coupled within CNTs. upshift effect that increases the thermal conductivity be- comes dominant, and thus the thermal conductivity is increased by coupling. Ice-NT coupled within a CNT can be considered as another realistic illustration for our approach, where the Ice-NT and CNT correspond to the conductive mate- rial and the substrate. Since the CNTs have much larger spring constant than the Ice-NTs (from Supporting Information,Ψ ∼ 0), an increment of thermal conductiv- 4 ity of Ice-NTs is expected after coupling with CNTs. Fig. 5(b) shows the calculated thermal conductivity of various Ice-NTs both with and without CNT coupling (see Sup- porting Information, SI.V). As one can see, the thermal conductivity of Ice-NTs has an obvious increment after coupling, which is independent on the specific tube types. These realistic illustrations further confirm the feasibility of our approach for applications. In summary, we have proposed a new approach to ma- nipulate the thermal conductivity. By coupling with dif- ferent substrates, thermal conductivity of the conduc- tive nanomaterial can be either remarkably decreased or increased, which can be realized in the device applica- tions. To decrease the thermal conductivity, the sub- strate needs to have similar atomic mass and spring con- stant with the conductive material; to increase the ther- mal conductivity, the atomic mass and/or spring con- stant of the substrate needs to be several times larger than that of the conductive material. Through the illus- trations of double-walled carbon nanotubes and coupled ice nanotubes, we have further shown that this approach is applicable in the real systems. Compared with the conventional treatments which only reduce the thermal conductivity, our approach can truly realize the thermal conductivity manipulation in solid nanomaterials with- out destroying their structures. Acknowledgments: We gratefully thank Z. Y. Zhang and S. H. Wei for fruitful discussions. This work was partially supported by the Special Funds for Major State Basic Research, National Science Foundation of China, Ministry of edu- cation and Shanghai municipality. The computation was performed in the Supercomputer Center of Shanghai, the Supercomputer Center of Fudan University. [1] Collins, P. G.; Zettl, A.; Bando, H.; Thess, A.; Smalley, R. E. Science 1997, 278, 100-103. [2] Huang, X. M. H.; Zorman, C. A.; Mehregany, M.; Roukes, M. L. Nature 2003, 421, 496-496. [3] Huang, Y.; Duan, X.; Cui, Y.; Lauhon, L. J.; Kim, K. H.; Lieber, C. M. Science 2001, 294, 1313-1317. [4] Li, M.; Pernice, W. H. P.; Tang, H. X. Nature Nanotech. Nika, D. L.; Balandin, A. A.; Bao, W.; Miao, F.; Lau, C. N. Appl. Phys. Lett. 2008, 92, 151911. [10] Biercuk, M. J.; Llaguno, M. C.; Radosavljevic, M.; Hyun, J. K.; Johnson, A. T. Appl. Phys. Lett. 2002, 80, 2767- 2769. [11] Yang, N.; Zhang, G.; Li, B. Nano Lett. 2008, 8, 276-280. [12] Samvedi, V.; Tomara, V. J. Appl. Phys. 2009, 105, 2009, 4, 377-382. 013541. [5] Chen, G.; Dresselhaus, M. S.; Dresselhaus, G.; Fleurial, J. P.; Caillat, T. Int. Mater. Rev. 2003, 48, 45-66. [13] Zhang, G.; Li, B. J. Chem. Phys. 2005, 123, 114714. [14] Padgett, C. W.; Brenner, D. W. Nano Lett. 2004, 4, [6] Dresselhaus, M. S.; Chen, G.; Ren, Z. F.; Dresselhaus, 1051-1053. G.; Henry, A.; Fleurial, J. P. JOM 2009, 61, 86-90. [7] Dresselhaus, M. S.; Chen, G.; Tang, M. Y.; Yang, R.; Lee, H.; Wang, D.; Ren, Z.; Fleurial, J. P.; Gogna, P. Adv. Mater. 2007, 19, 1043-1053 . [8] Cahill, D. G.; Ford, W. K.; Goodson, K. E.; Mahan, G. D.; Majumdar, A.; Maris, H. J.; Merlin, R. ; Phillpot, S. R. J. Appl. Phys. 2003, 93, 793-818. [9] Ghosh, S.; Calizo, I.; Teweldebrhan, D.; Pokatilov, E. P.; [15] Donadio, D.; Galli, G. Phys. Rev. Lett. 2007, 99, 255502. [16] Seol, J. H.; Jo, I.; Moore, A. L.; Lindsay, L.; Aitken, Z. H.; Pettes, M. T.; Li, X.; Yao, Z.; Huang, R.; Broido, D.; Mingo, N.; Ruoff, R. S.; Shi, L. Science 2010, 328, 213-216. [17] Ghosh, S.; Bao, W.; Nika, D. L.; Subrina, S.; Pokatilov, E. P.; Lau, C. N.; Balandin,A. A. Nature Mater. 2010, 9, 555-558. [18] Savin, A. V.; Kivshar, Y. S.; Hu, B. Europhys. Lett. 2009, 88, 26004. [19] Lepri, S.; Livi, R.; Politi, A. Phys. Rep. 2003, 377, 1-80. [20] Hu, B.; Li, B.; Zhao, H. Phys. Rev. E 1998, 57, 2992- 2995. [21] Koga, K.; Gao, G. T.; Tanaka, H.; Zeng, X. C. Nature 2001, 412, 802-805. [22] Mai, T.; Dhar, A.; Narayan, O. Phys. Rev. Lett. 2007, 98, 184301. [23] Since the temperature difference is fixed to be 0.1, from the Fourier law, the thermal conductivity is proportional to the heat flux value. [24] Yan, X. H.; Xiao, Y.; Li, Z. M. J. Appl. Phys. 2006, 99, 124305. [25] Pohl, R. O. Phys. Rev. Lett. 1962, 8, 481-483. [26] Tse, J. S.; Shpakov, V. P.; Belosludov, V. R.; Trouw, F.; Handa Y. P.; Press, W. Europhys. Lett. 2001, 54, 354-360. [27] Krivchikov, A. I.; Yushchenko, A. N.; Korolyuk, O. A.; Bermejo, F. J.; Fernandez-Perea, R.; Bustinduy, I.; Gonz´alez, M. A. Phys. Rev. B 2008, 77, 024202. [28] Prasher, R. J. Heat Transfer 2006, 128, 627-637. [29] From the Tersoff potential, we evaluated the spring con- stant of C-C bond of CNTs (corresponding to k1 and k2 of the coupled FPU chain model) to be about 1250 N/m. From LJ potential, the CNT-CNT intertube cou- pling spring constant (corresponding to kc of the cou- pled FPU chain model) was evaluated to be about 33 N/m. Then from the Supporting Information, SI.II, we get Ψ < π 32 . I. SUPPORTING INFORMATION "Manipulating thermal conductivity through substrate coupling", Zhixin Guo, Dier Zhang, and Xin-Gao Gong SI.I Hamiltonian of the coupled FPU chains The Hamiltonian of the coupled FPU chain model can be written as: amplitude ratio of chain-2 and chain-1 after coupling can be expressed as: 5 λ2 λ1 = r m1 m2 tan(α) (m2 ≥ m1), (4) c where λ1 and λ2 are the amplitudes of chain- and chain-2, tan(α) = 1 (ω1−ω2)±√(ω1−ω2)2+4ω2 respectively. ∈ [−1, 1], and ωc = kc√m1m2 , ω1 = , respectively. Then the resonance an- Here k1+kc , ω2 = k2+kc m2 2ωc m1 gle is defined as Ψ = (cid:12)(cid:12)(cid:12) . When Ψ = 0, the phonon dispersion of chain-1 in the coupling system can be written as: arctan( λ2 λ1 )(cid:12)(cid:12)(cid:12) ω(q) = r k1 m1 ·r4 sin2( 1 2 q) + kc, (5) Thus the phonon band of chain-1 has an obvious upshift after coupling, the magnitude of which is proportional to √kc. SI.III Calculation of PED distribution with wave vec- tor ) . . ( U A S O D E H = H1 + H2 + Hc, (1) 0.0 0.5 1.0 1.5 2.0 Hn = Xi n)2 (cid:20) (pi mn + 1 2 kn(xi n − xi−1 n )2 + 1 4 βn(xi n − xi−1 n )4(cid:21) , (n = 1, 2), FIG. 6: PED distributions with phonon frequency ω of iso- lated chain-1 at T=0.25. Hc = Xi (cid:20) 1 2 kc(xi 2 − xi 1)2(cid:21) , where H1, H2, and Hc are the Hamiltonian of chain-1, chain-2, and the coupling term, respectively. xi n , pi n represent the displacement from the equilibrium position, the momentum of the ith particle; and mn, kn, βn are the mass, spring constant and the anharmonic coefficient of chain-n (n = 1, 2). SI.II Definition of resonance angle Ψ Using the coupling-harmonic-oscillator (CHO) model, we can calculate the phonon resonance strength. The (2) (3) The q dependence of PED has an expression of u(q) = ω(q) × η[ω(q), T ],with η[ω(q), T ] representing the phonon distribution function at temperature T for a certain frequency ω. Since the phonon distribution in a heat-transport system is mainly determined by the heat source that supplies phonons. We suppose the conduc- tive chain is coupled to an ideal heat source that supplies phonons like the black-body, which can be represented by the Nos´e-Hoover thermostat[1, 2]. Then η(ω, T ) can be written as: η(ω, T ) = 8πω2 c3 1 eω/kbT − 1 , (6) w 6 Using equation (6), we also calculated the PED distri- butions with phonon frequency ω of the isolated chain-1. As is shown in Fig. 1, the result is consistent with that calculated by Li et al. using the numerical method, which confirms the reliability of our consideration.[3] SI.IV NEMD simulation for the double-walled carbon nanotube In the simulation, the Tersoff potential was used to the describe the carbon-carbon interaction[4], and the coupling interaction between two CNTs was described by the LJ potential,[5] which was truncated at 10 A by a switching function. The simulated (7,0)@(14,0) DWNT has a length of 121 A, containing 2352 atoms. The wall- thickness of CNT was chosen to be 1.44 A for the calcula- tion of cross-section area.[6, 7] Fixed boundary condition was applied along the axial direction of CNTs, where the outmost two layers of each head were fixed.[7, 8] Then two layers of each end of (7,0) CNT were put into contact with the Nos´e-Hoover thermostat with temperatures 310 and 290 K, respectively, while the (14,0) CNT was free of thermostat contact. To integrate equations of motion, the velocity Verlet method was employed with a fixed time step of 1 f s. All results were obtained by averaging about 10 ns after a sufficient long transient time (10 ns) to set up a nonequilibrium stationary state. SI.V NEMD simulation for the Ice-NTs Water-water intermolecular interaction was described by the TIP4P[9] potential and carbon-carbon interaction was described by the Tersoff[4] potential. As for the cou- pling term, the CNT-water interaction was described by a carbon-oxygen LJ potential.[10] All the pair interactions were truncated at 10 A by a switching function. The sim- ulation box has a length of 121 A and the total number of water molecules inside is 44×n for the (n,0) Ice-NT, where n = 4, 5, and 6. The Ice-NTs' wall-thickness was chosen to be 2.75 A for the calculation of cross-section area[11]. Fixed boundary condition was applied along the axial direction of Ice-NT/CNT, where the outmost two layers of each head were fixed.[7, 8] Then two lay- ers of each end of Ice-NT were put into contact with the Nos´e-Hoover thermostat with temperatures 110 and 90 K, respectively, while the CNT was free of thermostat contact. To integrate equations of motion, the velocity Verlet method was employed with a fixed time step of 1 f s. All results were obtained by averaging about 5 ns after a sufficient long transient time (5 ns) to set up a nonequilibrium stationary state. [1] Nos´e, S. J. Chem. Phys. 1984, 81, 511-519. [2] Hoover, W. G. Phys. Rev. A 1985, 31, 1695-1697. [3] Li, B., Lan, J.; Wang, L. Phys. Rev. Lett. 2005, 95, 104302. [4] Tersoff, J. Phys. Rev. B 1989, 39, 5566-5568. [5] Girifalco, L. A.; Hodak, M.; Lee, R. S. Phys. Rev. B [8] Wu, G.; Li, B. Phys. Rev. B 2007, 76, 085424. [9] Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.; Klein, M. L. J. Chem. Phys. 1983, 79, 926-935. [10] Meng, L.; Li, Q.; Shuai, Z. J. Chem. Phys. 2008, 128, 134703. 2000, 62, 13104-13110. [11] Guo, Z.; Zhang, D.; Zhai, Y.; Gong, X. G. Nanotechnol- [6] Zhang, G.; Li, B. J. Chem. Phys. 2005, 123, 114714. [7] Guo, Z.; Zhang, D.; Gong, X. G. Appl. Phys. Lett. 2009, ogy 2010, 21, 285706. 95, 163103.
1705.01572	2	1705	2018-01-09T17:58:17	Controlling Chiral Domain Walls in Antiferromagnets Using Spin-Wave Helicity	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	In antiferromagnets, the Dzyaloshinskii-Moriya interaction lifts the degeneracy of left- and right-circularly polarized spin waves. This relativistic coupling increases the efficiency of spin-wave-induced domain wall motion and leads to higher drift velocities. We show that in biaxial antiferromagnets, the spin-wave helicity controls both the direction and magnitude of the magnonic force on chiral domain walls. By contrast, in uniaxial antiferromagnets, the magnonic force is propulsive with a helicity dependent strength.	cond-mat.mes-hall	cond-mat	Controlling Chiral Domain Walls in Antiferromagnets Using Spin-Wave Helicity Alireza Qaiumzadeh, Lars A. Kristiansen, and Arne Brataas Center for Quantum Spintronics, Department of Physics, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway (Dated: July 12, 2018) In antiferromagnets, the Dzyaloshinskii-Moriya interaction lifts the degeneracy of left- and right- circularly polarized spin waves. This relativistic coupling increases the efficiency of spin-wave- induced domain-wall motion and leads to higher drift velocities. We demonstrate that, in biax- ial antiferromagnets, the spin-wave helicity controls both the direction and the magnitude of the magnonic force on chiral domain walls. In this case it is shown that the domain-wall velocity is an order of magnitude faster in the presence of the Dzyaloshinskii-Moriya interaction. In uniaxial an- tiferromagnets, by contrast, the magnonic force is always propulsive but its strength is still helicity dependent. The Dzyaloshinskii-Moriya interaction (DMI) was in- troduced to describe the weak ferromagnetism in antifer- romagnetic (AFM) materials [1, 2]. The DMI arises from the relativistic spin-orbit coupling that occurs when there is a broken inversion symmetry. The DMI is an antisym- metric exchange interaction between two adjacent spins Si−1 and Si, wDM = −Pi(−1)iD0 · (Si−1 × Si), where D0 is a vector with constant direction and amplitude [3]. The direction of D0 is dictated by the point group sym- metry of materials. The DMI can originate in the bulk in non-centrosymmetric crystals. There also can be interfa- cial DMIs in ultrathin films and at interfaces with heavy metals [4 -- 7]. The discovery of magnetic skyrmions and chiral domain walls (CDWs) [8 -- 13] has led to renewed interest in the DMI. In addition to their fundamental in- terest, the control of these intriguing textures may have applications in next-generation information storage and processing devices. AFM materials are ordered magnetic materials with- out any net magnetization. Their lack of stray fields and THz response are promising for novel ultra-dense and ultra-fast magnetic devices. Topics in AFM spintronics include spin angular momentum transport and transfer, spin-orbit coupling, and the manipulation of AFM do- mains and solitons [14 -- 23]. Coherent or thermal spin waves (SWs) [24 -- 27] and currents [21, 28, 29] can in- duce movement of AFM domain walls (AFM-DWs). In the absence of the DMI, circularly polarized SWs push AFM-DWs via the transfer of linear momentum [24 -- 26] but pull ferromagnetic (FM) domain walls (FM-DWs) via the transfer of spin angular momentum [30 -- 32]. The SWs in FM systems are always right-handed. By con- trast, in AFM systems, both left- (L) and right-circularly (R) polarized SWs exist [33 -- 35]. In the absence of the DMI, the dispersions associated with the two helicities are degenerate, but the DMI lifts this degeneracy [36]. The helicity degree of freedom in AFM-SWs, among other degrees of freedom in the nature, such as the spin of electrons and polarization of light, recently has attracted great attention [34 -- 40] and promises many applications in novel low dissipation magnonic data processing. In this Rapid Communication, we demonstrate the possibility of all-magnonic helicity-dependent AFM-DW motion in the presence of DMI. This paves the way for faster and improved control of AFM textures. Model. -- We consider a two-sublattice AFM insula- tor with equal spins, SA = SB = S. The unit vec- tors along the directions of the magnetic moments are mA(r, t) = SA/S and mB(r, t) = SB/S. At equilib- rium, mA(r, t) and mB(r, t) are antiparallel. We in- troduce the magnetization m = (mA + mB)/2 and the staggered order parameter n = (mA − mB)/2, where n · m = 0 and n2 + m2 = 1. We consider an effec- tive one-dimensional (1D) model along the x direction to describe the AFM-DW motion. In the exchange approximation, the total Lagrangian density of an AFM system is L[n, m] = LB[n, m] − U [n, m], (1) where the Berry-phase-induced term LB and the free en- ergy density U are [17, 41 -- 43] LB[n, m] = ρ∂tn · (n × m), + Lm · ∂xn + U [n, m] = m2 2χ + wDM . A 2 (∂xn)2 − Kx 2 (2a) (n · x)2 (2b) Here, ρ = S/a is the spin angular momentum density, a is the lattice constant, χ is the magnetic susceptibility, A is the exchange stiffness, L is the parity-breaking term amplitude [41], and Kx > 0 is the easy axis anisotropy energy density. The DM free energy density in the con- tinuum model, wDM = d·(m×n)+D·(∂xn×n), consists of homogeneous and inhomogeneous DMIs as expressed by the first and second terms, respectively. The magneti- zation is a slave variable and can be found by solving the equation of motion, χ−1m = ρ∂tn × n − L∂xn + d × n. To simplify the expressions, we use natural units of time, length and energy: t0 = ρpχ/Kx, λ0 = pA/Kx, and ǫ0 = √AKx, where the exchange stiffness and easy axis anisotropy are renormalized as follows: A → A−χL2 and Kx → Kx−χd2. In this way, the contributions to the Lagrangian density as functions of the staggered order parameter read simply as [44] LB[n] = 1 U [n] = 2 (∂tn)2, 1 2 (∂xn)2 − 1 2 (n · x)2 + Dn · (x × ∂xn). (3a) (3b) The kinetic part of the AFM Lagrangian, Eq. (3a), is Lorentz invariant [45], and the effective velocity of light is the maximum velocity of magnons in isotropic systems, c = 1. As in ferromagnets [46], Eq. (3b) implies that when D > 1, the ground state is a helical state with a spatial period of 2π/D. By contrast, when D < 1, there are two collinear, degenerate ground states: n0 = σx, where σ = ±. A domain wall (DW) is a transition between these two discrete degenerate ground states [47]. In the following, we assume that D is smaller than the critical value of D < 1. By minimizing the free energy Eq. (3b) with respect to the boundary conditions n0(x → ±∞) = σx, we can find the profile of a CDW. This profile is represented by n0 = (cos θ, sin θ cos φ, sin θ sin φ) with cos θ = tanh[(x − x0)√1 − D2] and φ = (x − x0)D + Φ, where √1 − D2 is the effective DW width, D is the rate of DW twisting, Φ is the DW tilt, and x0 is the position of the DW center [46]. When D → 0, the DW profile is of the N´eel type. In the limit of D → 1, the system gradually approaches a spiral state. The energy of CDW is E = 2√1 − D2. DW motion in a uniaxial AFM system. -- We com- pute the SW dispersions in uniaxial uniform domains and CDWs ωu and ωc, respectively. In doing so, we assume that there is a small transverse deviation of the staggered field on top of a static ground state n = n0(x) + δn. It is convenient to use a global basis to express the SWs [25, 26]. The orthogonal unit vectors are e1 = ∂n0/∂θ, e2 = (∂n0/∂φ)/ sin θ, and e3 = e1 × e2 = n0. A SW is represented by a complex field ψ = δn · (e1 + ie2) [48]. We obtain the effective Lagrangian density for SWs by expanding the total Lagrangian to the second order in ψ. The effective SW equation is Hu(c)ψ = ω2 u(c)ψ. The SW Hamiltonians in a uniform domain and in a CDW are Hu = −∂2 Hc = −∂2 x + i2σD∂x + 1, x + (1 − D2)h1 − 2sech2(xp1 − D2)i . (4a) (4b) The eigenvalues are given by ω2 u = 1 − 2σDk + k2, ω2 c = 1 − D2 + k2, (5a) (5b) and the eigenfunctions are expressed as ψ(t, x) = Ψψ0(x)e−iωu(c) t+ikx, where Ψ = δn is the SW ampli- tude far from the AFM-DW and k is the wave number. The eigenfunction ψ, together with its dispersion rela- (5), describes a circularly polar- tion as given in Eq. ized SW. The dispersion relation of Eq. (5) has both 2 positive- and negative-frequency solutions. The SWs with ω < 0 (ω > 0) are right-circularly (left-circularly) polarized SWs. Since the phase velocity is ω/k, the left- circularly polarized SWs are rightward-moving for k > 0 and leftward-moving for k < 0, whereas the opposite is true for the right-circularly polarized SWs. In AFM materials, the DMI lifts the degeneracy of the circularly polarized SWs, as has recently been ob- served experimentally [36]. Equation (5a) shows that the SW spectrum is non-reciprocal in uniaxial AFM do- mains, such that ωu(k) 6= ωu(−k). However, in chiral AFM-DWs, the dispersion is symmetric and degener- ate, such that ωc(k) = ωc(−k), with a renormalized en- ergy gap. The group velocity in a uniform domain is vu = (k − σD) /ωu, whereas in a CDW, it is vc = k/ωc. The effective potential energy of a static AFM-DW, the Poschl-Teller potential in the SW Hamiltonian Hc of Eq. (4b), is reflectionless. Nevertheless, a circularly polarized SW exerts a torque on an AFM-DW. As a result, the DWs in a uniaxial AFM material rotate and no longer remain reflectionless [24 -- 26]. Consequently, there is a force on the AFM-DWs; see Fig. 1. To capture these effects, we transform from the labo- ratory frame F into a uniformly moving and precessing frame F . The new frame of reference F rotates with an angular frequency of Ω around the x axis and moves with a linear velocity of V along the x direction [25, 26]. In the new frame, the profile of a CDW is described by cos θ = tanh[(x − x0)p1 − D2] and φ = (x − x0) D + Φ, where we mark the variables in F with a tilde. The nat- ural units in F are λr = Γλ0, tr = Γt0, and ǫr = ǫ0/Γ, where Γ = 1/p1 − Ω2 + 2γV DΩ and γ = 1/√1 − V 2. Additionally, we define Ω = Γ(Ω − γV D) and D = ΓγD. Finally, in F , the Lagrangian density is given by LB = 1 U = 2 (∂t 1 2 (∂xn)2 − n)2 + Ω∂t n · (x × n), 1 2 (n · x)2 + Dn · (x × ∂xn). (6a) (6b) The second term in the kinetic part of the Lagrangian, Eq. (6a), is due to the Coriolis force [26]. By expanding the effective Lagrangian in the new reference frame L up to the second order in ψ and calculating the Euler- Lagrange equation, we find an eigenproblem expressed as Hcψ = ω2ψ, with the following effective Hamiltonian for SWs on a precessing CDW: Hc =Hc[x → x, D → D] + 2ω Ω tanh(xp1 − D2). (7) The additional hyperbolic tangent potential energy in the Hamiltonian Hc causes reflections [25, 26]. The reflection r and transmission t amplitudes for the 1D hyperbolic tangent scattering potential are known [49]: 3 2 (k+ − k−)] 2 (k+ + k−)] sinh2[ π sinh2[ π r2 = r2k− + t2k+ = k−. , (8a) (8b) In F , the wave numbers of the transmitted and incoming SWs k+ and k−, respectively, are given by k2 ± = ω2 ∓ ω Ω − 1 + D2. The wave numbers are the same in F and F ; thus, k2 ± − 1 + D2. The frequency shift between an incoming SW of frequency ω− = ω + (Ω − γV D) and an outgoing SW of frequency ω+ = ω− (Ω− γV D) is our first central result: ± = ω2 ∆ω = −2(Ω − γV D). (9) Equation (9) demonstrates that the DMI causes a shift in the SW frequency that is linear in the DW velocity [25, 26]. This frequency shift significantly changes the interaction between SWs and DWs, especially in biaxial systems, where the angular frequency Ω is suppressed. A left-circularly (right-circularly) polarized SW incident on a DW has a positive (negative) frequency and induces a precession of the DW in the positive (negative) direction. In both cases, the effect of the angular frequency term, the first term in Eq. (9), is the same: The precession of the DW reduces the frequency of the transmitted SW by the same amount for incoming SWs of the same am- plitude but opposite helicity. By contrast, the velocity- dependent term, the second term in Eq. (9), reinforces the redshift for left-circularly polarized SWs and reduces it for right-circularly polarized SWs. Let us now derive the reactive magnonic force and torque. We employ the conservation of the energy- momentum tensor T αβ = gαγT β γ , where gαγ = diag (1,−1) is the inverse of the metric tensor in the Minkowski space (x0, x1) = (t, x) and T β α = ∂αn · ∂L/∂(∂β n)− δβ αL. When Lagrangian density is invariant with respect to the space-time, the Noether's theorem implies continuity equations for the energy-momentum tensor ∂βT αβ = 0, and current ∂αjα = 0 [50], respec- tively. The translational and rotational symmetries in uni- axial AFM systems dictate the conservation of total linear-momentum P , and angular-momentum J. Then the force and torque on AFM-CDWs are defined as F = T 11(−∞) − T 11(+∞) = dP/dt, and τ = j1(−∞) − j1(+∞) = dJ/dt, respectively. The total reactive magnonic force F = Freflection + Fredshift + FDM consists of the reflection force, the redshift force, and the DMI force, Freflection = 2Ψ2r2k2 −, Fredshift = Ψ2(1 − r2)k−(k− − k+), FDM = 2γDΨ2(1 − r2)k−. (10a) (10b) (10c) (a) (c) (b) (d) FIG. 1: The angular frequency Ω [(a) and (c)] and linear velocity V [(b) and (d)] of an AFM-DW as functions of the SW amplitude Ψ for k = 0.8. The solid lines are based on Eqs. (8) and (10)-(12), whereas the circles represent solutions of LLG equations. The black curves are the results in the absence of DMI and the red/blue curves show the results for a finite DMI in the presence of left-/right-circularly polarized SWs. Both the redshift and DMI forces are due to the trans- mitted SWs. The total reactive torque exerted by SWs is given by τ = 2Ψ2(1 − r2)k−. (11) Equation (10) shows that when the SWs are hard (k± ≫ 1) and the DMI is large and comparable to the critical value D . 1, Freflection ≪ Fredshift, FDMI. If k+ ≃ k−, then the DMI force dominates (Fredshift ≪ FDMI). When k+ ≪ k−, both the redshift and DMI forces are of the same order. For soft SWs (k± ≪ 1) and a strong redshift (k+ ≪ k−), the reflection force dominates. Thus far, we have not considered the ubiquitous dis- sipation. We include dissipative effects via a Rayleigh's dissipation function density R = α(∂tn)2/2, where α is the Gilbert damping. We define a viscous force F v and a torque τ v such that dP/dt = F + F v and dJ/dt = τ + τ v. In the steady state, F = −F v and τ = −τ v, and we find that 2αγ (cid:0)D(Ω − γV D) − V (1 − D2 − (Ω − γV D)2)(cid:1) , (12a) (12b) F = − p1 − D2 − (Ω − γV D)2 τ = 2αγ(Ω − γV D) p1 − D2 − (Ω − γV D)2 . Equations (8) and (10)-(12) form a closed set that we can solve numerically to find the steady state. We plot the results of the related simulations in Fig. 1. We see that the directions of the angular frequencies of the DWs are opposite for different SW helicities. We check our results based on Eqs. (8) and (10)-(12) against those of another, more direct numerical proce- dure. For the latter calculations, we follow Ref. [24]. 0.4 0.3 V 0.2 0.1 0.0 1.0 1.5 2.0 L R R & L, D=0 0.2 0.1 0.0 -0.1 V -0.2 -0.3 -0.4 3.0 3.5 4.0 1.0 1.5 2.0 2.5 (cid:1)0 4 L R R & L, D=0 3.0 3.5 4.0 2.5 (cid:1)0 FIG. 2: The DW velocity V as a function of the external magnetic field frequency, ω0, for both L- and R-polarized SWs in a uniaxial, κ = 0, AFM system with D = 0.5. The black dots represent the DW velocity in the absence of the DMI, D = 0. We solve the coupled nonlinear Landau-Lifshitz-Gilbert (LLG) equations for the staggered field n and the mag- netization m [24]. A circularly-polarized magnetic field of frequency ω0 and amplitude h0 excites SWs in a region far from the DW center. The amplitude of the external magnetic field is in natural units. Figure 2 shows the DW velocity as a function of ω0 for left- and right-circularly polarized SWs excited by a magnetic field of h0 = 0.001. In the presence of a high frequency left-handed (right-handed) SW, the ve- locity of a chiral AFM-DW is higher (lower) than that of its non-chiral counterpart. The DW velocity exhibits non-monotonic behavior with respect to the frequency ω0. At lower frequencies, CDWs have higher velocities than their non-chiral counterparts for both helicities. Note that these calculations are based on the conser- vation of linear momentum [24 -- 26] and they are valid for the small Gilbert damping regime [51]. DW motion in a biaxial AFM system. -- We model a biaxial AFM material by introducing an additional contribution to the free-energy density expressed in Eq. (3b), U → U + Uani. The transverse-axis anisotropy en- ergy density is Uani = κ(n · z)2, where κ = Kz/2Kx > 0 and Kz is the anisotropy energy density in the z- direction. This fixes the CDW tilt angle such that Φ = 0 or π. For biaxial AFM systems, the spin-current j1 is no longer conserved. In a collinear biaxial AFM system, the SW polarization is elliptical with a dispersion relation: b = 1 + κ + k2 ± p4D2k2 + κ2. ω2 (13) In a biaxial AFM system, the spectrum remains symmet- ric, such that ωb(k) = ωb(−k). Note that, even in the ab- sence of the DMI, a transverse anisotropy lifts the helicity degeneracy of the polarized SWs; however, this degener- FIG. 3: The same as in Fig. 2 but for a biaxial AFM system. The hard-axis anisotropy is κ = 0.25. acy breaking is independent of momentum and leads to a finite gap between the two branches of the SW excita- tions. The group velocity of the SWs in a collinear biaxial AFM system is vg = (cid:0)1 ± 2D2/√4D2k2 + κ2(cid:1) k/ωb. The equations of motion in biaxial AFM systems are com- plicated; here, we present only the results of numerical calculations. In a biaxial AFM system, in the absence of the DMI, a DW approaches the SW source regardless of the SW helicity. There is no rotation of the DW. The potential in the SW Hamiltonian that is induced by the AFM-DW texture causes no reflections. Thus, the motion induced by elliptical SWs is similar to that induced by linearly polarized SWs in a uniaxial AFM system; the DW mo- tion is slow compared with that induced by circularly polarized SWs and is toward the SW source because of the conservation of linear momentum [24]. In a biaxial AFM system, the DMI does not cause the CDWs to rotate, but Eq. (9) shows that the out- going SWs acquire a DMI-dependent frequency shift. The hard-axis anisotropy suppresses the angular fre- quency of the CDWs. The frequency shift simplifies to ∆ω ≃ 2γV D. The DMI and redshift forces dominate the total magnonic force since the reflection force can be ne- glected. Figure 3 shows that DMI dramatically increases the DW velocity and leads to helicity-dependent DW mo- tion. This is our second central result. The helicity of the SWs controls the direction of the induced motion of chiral AFM-DWs. The velocity of CDW is antisymmet- ric respect to the helicity of SWs. The DW velocity is approximately two times higher in the presence of one he- licity (left-circularly polarized SWs, in our configuration) than it is in the presence of the opposite helicity. Discussion and conclusion. -- Using KMnF3 parame- ters [55], we find a DW velocity of 2 km/s which is at least two orders of magnitude faster than the SW-driven CDW motion in biaxial FM systems [32] and one order of magnitude larger than the FM-DW driven by spin-orbit torque [56]. On the other hand, although in SW-driven CDW motion only a change in the sign of the DMI merely reverses the direction of motion [32, 57], via a linear- momentum transfer mechanism [58], in AFMs the SW helicity also controls the direction. Recent developments in atomic-scale resolution mi- croscopy enables us to trace the AFM-DW position via techniques, such as spin-polarized scanning-tunneling mi- croscopy [59], photoelectron emission microscopy [60], and magnetic exchange force microscopy [61]. The po- sition of AFM-DW center also can be detected by using the anisotropic magneto resistance effect, indirectly [21]. The DMI results in a faster and more controlled motion of AFM-DWs. In a biaxial AFM system, the SW helic- ity determines both the direction and the magnitude of the DW velocity. These features enable the magnonic helicity-dependent DW motion. By contrast, in a uniax- ial AFM system, a CDW always is recoiled from the SWs' source via a magnonic force with a helicity-dependent magnitude. Acknowledgments The authors thank E. G. Tveten and Y. Tserkovnyak for fruitful discussions. The research leading to these re- sults received funding from the European Research Coun- cil via Advanced Grant No. 669442 "Insulatronics" and was partially supported by the Research Council of Nor- way through its Centres of Excellence funding scheme, Project No. 262633, "QuSpin". 5 [14] T. Jungwirth, X. Marti, P. Wadley and J. Wunderlich, Nat. Nanotechnol. 11, 231 (2016). [15] V. Baltz, A. Manchon, M. Tsoi, T. Moriyama, T. Ono, and Y. Tserkovnyak, arXiv:1606.04284. [16] A. V. Kimel, B. A. Ivanov, R. V. Pisarev, P. A. Usachev, A. Kirilyuk, and T. Rasing, Nature Phys. 5, 727 (2009). [17] B. A. Ivanov, Low Temp. Phys. 31, 635 (2005); A. M. Kosevich, B. A. Ivanov, and A. S. Kovalev, Phys. Rep. 194, 117 (1990). [18] E. V. Gomonai, B. A. Ivanov, V. A. L'vov, and G. K. Oksyuk, Sov. Phys. JETP 70, 174 (1990). [19] V. G. Baryakhtar, B. A. Ivanov, and M. V. Chetkin, Sov. Phys. Usp. 28, 563 (1985). [20] A. Qaiumzadeh, H.Skarsvag, C. Holmqvist, and A. Brataas, Phys. Rev. Lett. 118, 137201 (2017). [21] E. G. Tveten, A. Qaiumzadeh, O. A. Tretiakov, and A. Brataas, Phys. Rev. Lett. 110, 127208 (2013). [22] J. Barker and O. A. Tretiakov, Phys. Rev. Lett. 116, 147203 (2016). [23] E. G. Galkina, A. Yu. Galkin, B. A. Ivanov, and F. Nori, Phys. Rev. B 81, 184413 (2010). [24] E. G. Tveten, A. Qaiumzadeh, and A. Brataas, Phys. Rev. Lett. 112, 147204 (2014). [25] M. M. Bogdan and Z. V. Slyozova, Frontiers in Mag- netism of Reduced Dimension Systems, High Technol- ogy NATO ASI Series 3, Vol. 49, P. 323 (Kluwer Aca- demic Publishers, Dordrecht/Boston/London, 1998); M. M. Bogdana and O. V. Charkina, Low Temp. Phys. 40, 84 (2014). [26] S. K. Kim, Y. Tserkovnyak, and O. Tchernyshyov, Phys. Rev. B 90, 104406 (2014). [27] S. Selzer, U. Atxitia, U. Ritzmann, D. Hinzke, and U. Nowak, Phys. Rev. Lett. 117, 107201 (2016). [28] O. Gomonay, T. Jungwirth, and J. Sinova, Phys. Rev. Lett. 117, 017202 (2016). [29] T. Shiino, S.-H. Oh, P. M. Haney, S.-W. Lee, G. Go, B-G. Park, and K-J. Lee, Phys. Rev. Lett. 117, 087203 (2016). [30] A. Mikhailov and A.I. Yaremchuk, JETP Lett. 39, 296 (1984). [1] I. Dzyaloshinskii, J. Phys. Chem. Solids 4, 241 (1958). [2] T. Moriya, Phys. Rev. 120, 91 (1960). [3] N. Papanicolaou, Phys. Rev. B 55, 12290 (1997). [4] A. N. Bogdanov, U. K. Rossler, M. Wolf, and K-H. Muller, Phys. Rev. B 66, 214410 (2002). [5] A. Bogdanov, and A. Shestakovr, Phys. Solid State 40, [31] P. Yan, X.S. Wang, and X.R. Wang, Phys. Rev. Let. 107, 177207 (2011). [32] W. Wang, M. Albert, M. Beg, M.-A. Bisotti, D. Chernyshenko, D. Cort´es-Ortuno, I. Hawke, and H. Fan- gohr, Phys. Rev. Lett. 114, 087203 (2015). [33] F. Keffer, H. Kaplan, and Y. Yafet, Am. J. Phys. 21, 250 1350 (1998). (1953). [6] A. N. Bogdanov and D. A. Yablonskii, Sov. Phys. JETP 69, 142 (1989). [7] A. Fert, V. Cros, and J. Sampaio, Nat. Nanotech. 8, 152 (2013). [8] S. Muhlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R. Georgii, and P. Boni, Science 323, 915 (2009). [9] N. Nagaosa and Y. Tokura, Nat. Nanotechnol. 8, 899 (2013). [10] A. Soumyanarayanan, N. Reyren, A. Fert, and C. [34] T. Satoh, S.-J. Cho, R. Iida, T. Shimura, K. Kuroda, H. Ueda, Y. Ueda, B. A. Ivanov, F. Nori, and M. Fiebig, Phys. Rev. Lett. 105, 077402 (2010). [35] N. Kanda, T. Higuchi, H. Shimizu, K. Konishi, K. Yosh- ioka, and M. Kuwata-Gonokami, Nat. Commun. 2, 362 (2011). [36] G. Gitgeatpong, Y. Zhao, P. Piyawongwatthana, Y. Qiu, L. W. Harriger, N. P. Butch, T. J. Sato, and K. Matan, aXiv:1702.04889. [37] R. Cheng, M. W. Daniels, J-G. Zhu, and D. Xiao, Sci. Panagopoulos, Nature 539, 509 (2016). Rep. 6, 24223 (2016). [11] V. Flovik, A. Qaiumzadeh, A. K. Nandy, C. Heo, T. Ras- [38] R. Cheng, S. Okamoto, and D. Xiao, Phys. Rev. Lett. ing, Phys. Rev. B, 96, 140411(R) (2017). 117, 217202 (2016). [12] M. Heide, G. Bihlmayer, and S. Blugel, Phys. Rev. B 78, [39] V. A. Zyuzin and A. A. Kovalev, Phys. Rev. Lett. 117, 140403 (2008). 217203 (2016). [13] F. J. Buijnsters, Y. Ferreiros, A. Fasolino, and M.I. Kat- [40] Y. Shiomi, R. Takashima, E. Saitoh, Phys. Rev. B 96, snelson, Phys. Rev. Lett. 116, 147204 (2016). 134425 (2017). 6 [41] E. G. Tveten, T. Muller, J. Linder, A. Brataas, Phys. Rev. B 93, 104408 (2016). [42] E.M. Lifshitz and L.P. Pitaevskii, Statistical Physics, Course of Theoretical Physics Vol. 9 (Pergamon, Oxford, 1980), Pt. 2. [43] The parameters of the 1D micromagnetic model are de- fined in terms of the atomistic spin-lattice parameters as follow: χ−1 = 4J0S 2/a, A = 2aJ0S 2, Kx = 2K0S 2/a, d = 2D0S 2/a, and D = D0S 2, where J0 and K0 are the Heisenberg exchange energy and anisotropy energy, respectively, and a is the lattice constant. [44] In our non linear σ model, we disregarded the topological term [χL n · (∇n × n)] and another term arising from the homogeneous DMI [χ n · d], because these terms are total derivatives and do not affect the equations of motion. [45] I. V. Baryakhtar and B. A. Ivanov, Zh. Eksp. Teor. Fiz. 85, 328 (1983) [Sov. Phys. JETP 58, 190 (1983)]; B. A. Ivanov and A. K. Kolezhuk, Phys. Rev. Lett. 74, 1859 (1995); B. A. Ivanov and A. K. Kolezhuk, Fiz. Nizk. Temp. 21, 355 (1995) [Low Temp. Phys. 21, 275 (1995)]; I. Affleck, Phys. Rev. Lett. 57, 1048 (1986). [46] O. A. Tretiakov and Ar. Abanov, Phys. Rev. Lett. 105, 157201 (2010). [47] M. Kardar, Statistical Physics of Fields (Cambridge Uni. Press, Cambridge, 2007). [48] The staggered field can be expressed in terms of ψ: n = Re[ψ]e1 + Im[ψ]e2 + (cid:2)1 − (Re[ψ])2/2 − (Im[ψ])2/2(cid:3) e3. [49] L. D. Landau and E. M. Lifshitz, Quantum Mechanics, 3rd ed. (Pergamon, Oxford, 1991). presence of circularly polarized SWs in systems with uni- axial anisotropy. This is similar to the contribution of so- called entropic torque (or force) on the DW motion in the presence of a temperature gradient. The entropic torque arises from the temperature dependence of the effective exchange field as discussed in Ref [52]. [52] S. K. Kim and Y. Tserkovnyak, Phys. Rev. B 92, 020410(R) (2015). [53] S. J. Pickart, M. F. Collins, and C. G. Windsor, J. Appl. Phys. 37, 1054 (1966). [54] I. V. Solovyev, V. V. Mazurenko, A. A. Katanin, Phys. Rev. B 92, 235109 (2015). [55] We use the material parameters for KMnF3 in 1D: S = 5/2, a = 0.418nm, Ms = 5 × 10−14Am, Kx = 2.03 × 10−14J/m, D = 1 × 10−22J, and A = 1.15 × 10−30 Jm, adapted from Ref. [53]. Also we consider α = 10−4. Other candidate materials for the observation of SW- induced AFM-DW motion include metallic AFM materi- als such as the uniaxial KMnF3 [53] and insulating AFM materials such as the uniaxial Cu2V2O7 [36] as well as the biaxial Sr2IrO4 and Ba2IrO4 [54]. [56] I. M. Miron, T. Moore, H. Szambolics, L. D. Buda- Prejbeanu, S. Auffret, B. Rodmacq, S. Pizzini, J. Vogel, M. Bonfim, A. Schuhl, and G. Gaudin, Nature Mater. 10, 419 (2011). [57] X.-G. Wang, L. Chotorlishvili, G.-H. Guo, A. Sukhov, V. Dugaev, J. Barna´s, and J. Berakdar, Phys. Rev. B 94, 104410 (2016). [58] P. Yan, Y. Cao, and J. Sinova, Phys. Rev. B 92, 100408 [50] The conserved spin-currents are expressed as jα = (2015). ny∂L/∂(∂αnz) − nz∂L/∂(∂αny). [51] In the presence of larger damping constants another mechanism becomes dominant in uniaxial AFMs. In this case SWs decay quickly thus there are much more SWs on one side of AFM-DW than on the other side, con- sequently this leads to a large gradient of the effective stiffness field amplitude. This effective stiffness field gra- dient leads to a backward motion of AFM-DWs in the [59] M. Bode, E. Y. Vedmedenko, K. von Bergmann, A. Ku- betzka, P. Ferriani, S. Heinze and R. Wiesendanger, Na- ture Mater. 5, 477 (2006). [60] N. B. Weber, H. Ohldag, H. Gomonaj, and F. U. Hille- brecht, Phys. Rev. Lett. 91, 237205 (2003). [61] R. Wiesendanger, Rew. Mod. Phys. 81, 1495 (2009).
1211.4919	1	1211	2012-11-21T02:44:14	Broadband Absorbers and Selective Emitters based on Plasmonic Brewster Metasurfaces	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci", "physics.optics" ]	We discuss the possibility of realizing utlrabroadband omnidirectional absorbers and angularly selective coherent thermal emitters based on properly patterned plasmonic metastructures. Instead of relying on resonant concentration effects that inherently limit the bandwidth, we base our design on the combination of two inherently nonresonant effects: plasmonic Brewster funneling and adiabatic plasmonic focusing. With this approach, we demonstrate compact, broadband absorption and emission spanning terahertz, infrared and optical frequencies, ideal for various energy and defense applications.	cond-mat.mes-hall	cond-mat	Broadband Absorbers and Selective Emitters based on Plasmonic Brewster Metasurfaces Christos Argyropoulos1, Khai Q. Le1, Nadia Mattiucci2, Giuseppe D’Aguanno2, Andrea Alù1,* 1Dept. of Electrical & Computer Engineering, The University of Texas at Austin 2AEgis Tech., Nanogenesis Division, Huntsville, AL 35806, USA *[email protected] We discuss the possibility of realizing utlrabroadband omnidirectional absorbers and angularly selective coherent thermal emitters based on properly patterned plasmonic metastructures. Instead of relying on resonant concentration effects that inherently limit the bandwidth, we base our design on the combination of two inherently nonresonant effects: plasmonic Brewster funneling and adiabatic plasmonic focusing. With this approach, we demonstrate compact, broadband absorption and emission spanning terahertz, infrared and optical frequencies, ideal for various energy and defense applications. PACS: 71.45.Gm, 41.20.Jb, 78.67.Pt, 79.60.Dp Infrared (IR) and optical emitters are currently based on semiconductor devices, such as light- emitting diodes (LEDs) and quantum cascade lasers [1]-[2]. In general, their operation is inherently narrowband, fabrication is quite challenging and their overall efficiency and emissivity are limited. In the last few years, thermal sources were suggested as an alternative way to overcome many of these limitations. An ideal thermal emitter follows Planck’s law of 1 blackbody radiation [3], but conventional thermal sources in general exhibit smaller and more narrowband emission than a blackbody, depending on material and design constraints. In addition, thermal sources emit incoherently over a relatively broad angular range, whereas in many applications spatial coherence and narrow emission angles are very desirable. Substantial research efforts have been recently dedicated to overcome these limitations and tame blackbody, angularly coherent emission features. One of the most promising venues is based on plasmonic and metamaterial effects [4]: resonant plasmonic gratings and metamaterials can significantly boost emission at the desired wavelength, up to the level of an ideal black-body, but usually this comes at the price of stringent bandwidth constraints [5]-[6]. Resonant nanoantenna emitters [7]- [8], periodic grooves [9], metallic gratings [10], microstrip patches [11] and wire medium [12] have also been successfully employed for similar purposes, with analogous limitations based on their resonant mechanisms. Since the emission properties of a device in thermodynamic equilibrium are directly related to its absorption features [13], many of these solutions have also been explored to realize narrowband absorbers and filters [14]-[18]. Different from emitters, in the case of absorbers a large acceptance angle may be desirable, where instead resonant solutions based on plasmon polaritons often lead to a sensitive response to the incident angle [19]-[20]. Realizing an ideally broadband blackbody response in a compact device with the added capability of tailoring its angular emission would open groundbreaking venues in sources, emitters and absorbers technology for a variety of energy and defense applications. In this regard, a few recent efforts have been devoted to realizing broadband operation based on tapered plasmonic geometries or combined multiple resonances [21]-[25]. These solutions still show bandwidth limitations, as they fundamentally rely on resonant processes. In this Letter, on the contrary, we rely on strictly nonresonant phenomena to achieve ultrabroadband emission and 2 absorption with controllable angular selectivity, spanning with a single device THz, IR and visible frequencies. Our concept is based on the combination of two nonresonant effects: plasmonic Brewster light funneling at a single interface [26]-[27] and adiabatic plasmonic focusing [28]. Consider a one-dimensional (1D) grating with period d , formed by an array of slits carved in gold and infinitely extended along y with unit cell shown in Fig. 1(a). The slits have width w and length l , terminated by a taper with length tapl designed to adiabatically dissipate the energy transmitted through the slits [28]-[29]. The taper is then terminated by a gold back plate much thicker than the skin depth, ensuring that an external impinging wave can only be reflected or absorbed by the structure. We assume a relative Drude permittivity    Au  pf 2 / f    f   i    for gold with parameters pf  2069 THz,  17.65 THz,   1.53 [30], under an e  t i time dependence. In order to model the varying taper width, which becomes rapidly comparable to the electron mean free path in the metal [31], and of the temperature dependence of the gold resistivity [16], we assume an increased collision frequency in the taper portion tap  176.5 THz. This assumption is not necessary for the proposed concept, as discussed in [32], but is considered here to make our model more realistic. As originally shown in [26], the grating interface may be tailored to be perfectly impedance matched to a transverse-magnetic (TM) impinging plane wave, provided that the incident angle satisfies the condition: cos  B  w  s k d  S 0 , 3 (1) where S is the material permittivity filling the slits, k 0 c / is the free-space wave number, c is the velocity of light in free-space and s is the guided wave number in a parallel-plate plasmonic slit with thickness d . At this angle, similar to the Brewster condition for a homogeneous interface, zero reflection and total transmission through the interface are expected [26]. Interestingly, this phenomenon weakly depends on frequency, as long as the plasmonic mode in the slit is weakly dispersive. This simple analytical model is a very accurate description of the anomalous funneling mechanism through the slits as long as the impinging wavelength is longer than d [26], ensuring that the impinging energy can funnel into the slits from DC to very high frequencies, up to the wavelength d  . Independent confirmation of these findings was 0 reported in [33]-[34]. This funneling phenomenon is purely based on impedance matching, without resorting on any resonance, and therefore the transmitted wave may be interestingly absorbed inside the slits without affecting at all the reflection coefficient or the bandwidth of operation. This functionality is very different from any other tunneling mechanism through narrow slits relying on resonant mechanisms, which would be severely affected by absorption. Absorption is achieved in our design by introducing a proper taper behind the Brewster interface, which can adiabatically absorb the transmitted plasmonic mode without reflections. The tapering angle and the corresponding length tapl determine, following [28]-[29],[35]-[36], the largest wavelength over which the transmitted energy gets fully absorbed in the metallic walls by the time it reaches the taper termination. Since the efficiency of adiabatic absorption depends on the taper length compared to the excitation wavelength, in fact, a given choice of tapering length fixes the limit on the minimum frequency of operation to achieve perfect absorption. 4 Figure 2 shows the absorption coefficient 1A R   , where R is the calculated reflectance computed with the finite-integration method [37], for a grating as in Fig. 1(a) with d  96 nm , w  24 nm , l  200 nm , S  and 1 tapl  980 nm , consistent with the grating design proposed in [26]-[27] to support Brewster funneling at B  70o , as predicted by Eq. (1). As expected, around the Brewster angle total absorption may be achieved over a very broad range of wavelengths, effectively spanning from   min 200nm to  max  10 m . Consistent with the previous discussion, this range may be further broadened, as the upper cut-off (shorter wavelength) is determined by the transverse period, whereas the lower limit is fixed by the taper length. The angle of total absorption is rather flat until 5 6 m , and the angular range of absorption can be controlled by the ratio /d w [27]. In this design, large absorption A  70 % is achieved for all incident angles in the frequency range of interest, even at normal incidence, except for angles very close to grazing incidence beyond the Brewster angle. The proposed structure forms an omnidirectional, ultrabroadband absorber based on inherently nonresonant mechanisms. Compared to recent designs based on resonant effects [24],[38], we achieve broader absorption bandwidths and larger acceptance angles. Interestingly, by simply tailoring the design parameters as discussed above, we can further broaden both frequency and angular bandwidth as desired. In [32], we consider more lossy metals, such as platinum, which provide even better absorption features for similar total thickness. The proposed broadband omnidirectional absorber may mimic the absorption performance of an ideal blackbody with exciting applications in energy harvesting, absorbers and bolometers. 5 Seen these interesting absorption features, the same concept may be applied to realize black- body like thermal emission. The overall emissivity can be easily computed by multiplying its absorption coefficient by the Planck’s law distribution of blackbody radiation [3]: B T ( )   2 hc 2 5  1 k T  B hc e ,  1 (2) where h is the Planck’s constant, Bk the Boltzmann constant and T the absolute temperature. In this scenario, angular selectivity may be desirable, in order to channel the emitted energy towards specific directions in free-space and not waste it into other directions. Again, by simply tailoring the ratio /w d , we may be able to realize ultrabroadband emission with much sharper angular directivity. We consider an emitter with dimension d m 1.44 , w  90 nm , l m 2 , tapl m 9.1 , 1 S  . First, we assume an operating temperature T  700 K , for which B is centered in the IR range, spanning an emission bandwidth shown in the bottom part of Fig. 3(a). The panel shows the thermal emission as a function of frequency and angle, normalized to the maximum value of blackbody radiation at this temperature. We essentially achieve an emission bandwidth equal to the one of an ideal black-body, with much large angular selectivity confined in a narrow beamwidth around the Brewster angle B  84o , with large, and tailorable by design, spatial coherence, ideal for a variety of applications, including thermo-photovoltaic and energy. It is very remarkable that the emission angle is very flat over the whole emission range, of great practical interest. The emitter does not ‘waste’ energy in the entire angular spectrum, as an omnidirectional ideal blackbody would, but instead routes it only towards a specific direction of interest. Its large spatial coherence is another important feature for IR sources [5]. 6 It is important to remark that the operational bandwidth of the grating is broader than the blackbody spectrum at the temperature of interest, which fundamentally determines the emission bandwidth in Fig. 3a. This implies that the same structure can emit different IR frequencies by simply changing the temperature of operation! For example, in Fig. 3(b) we demonstrate thermal emission for the same grating at room temperature T  300 K . The structure emits with the same selective properties, at the same angle, but a different, redshifted IR spectrum, due to the lower temperature. By increasing the operating temperature to, e.g., T  3000 K it may be possible to extend these concepts to optical emission. In this case, gold would melt at these high temperatures and other metals, such as tungsten [16] may be alternatively considered, as we discuss in [39]. So far, the proposed geometries are 1D configurations, operating only for TM excitation in the plane of the grating. We have recently extended the Brewster funneling concepts to 2D [40], showing that a mesh of orthogonal slits may provide light funneling independent of the plane of polarization. We show the corresponding design for 2D absorbers and emitters in Fig. 1(b). The structure is formed by crossed slits, tapered in 2D to allow adiabatic focusing and absorption (and reciprocally emission) on all planes of TM polarization. In order to test the performance of this device in both absorption and emission, we analyze their functionality in the worst-case scenario of an azimuthal angle 45o between the two orthogonal sets of slits. The structure is expected to perform equal or better than its functionality at this angle. Figure 4(a) compares the performance (Fig. 2) of the 1D absorber in Fig. 1(a) at normal incidence and at the Brewster angle B  70o with the equivalent 2D case monitored on the 45o plane (which is the worst-case scenario), obtained by simply introducing another set of 7 orthogonal slits with same period and width, as shown in Fig. 1(b). We notice that the 2D device has remarkable very similar performance with the 1D case, extending its functionality to all polarization planes. As expected, both designs show very large, broadband absorption, especially large at the Brewster angle (red lines), but consistently large for any angle, even at normal incidence (black lines). The design of Fig. 1(b) effectively provides an omnidirectional ultrabroadband absorber extending from far-IR to the optical range, and on all planes of polarization. Figure 4(b) shows a similar comparison for a 2D broadband selective emitter with the same slit and grating dimensions as in Fig. 3 and for T  700 K . Again, very good agreement is obtained between the two structures, with a good contrast between emission at the Brewster angle compared to normal emission. This ensures that the 2D grating can operate as a broadband, angularly selective coherent thermal emitter, producing in this configuration a 2D conical directive radiation focused at the Brewster angle. Very exciting applications are envisioned based on the proposed devices, such as directive thermo-photovoltaic radiation. To conclude, we have proposed here a novel concept to realize ultrabroadband omnidirectional absorbers and emitters based on 1D and 2D plasmonic gratings. Plasmonic Brewster funneling of energy and adiabatic focusing and absorption in tapered plasmonic slits have been combined to achieve large nonresonant absorption features over a controllable angular range, offering the possibility to realize omnidirectional absorbers, but also spatially coherent and angularly selective emitters. Design parameters can easily control the performance of our device: the grating period controls the upper frequency cut-off and the taper length, or equivalently the depth of our structure, controls the lower frequency of operation, whereas the angular spectrum is 8 determined by the ratio /w d . For emission, the spectrum is fundamentally controlled by the temperature, as in the case of an ideal blackbody. We notice that the limitation on the minimum frequency of operation on the device thickness is consistent with the fundamental bounds on thickness to bandwidth ratio [41] for radar absorbers. The proposed structures may increase the efficiency of optical and IR energy harvesting devices and lead to novel bolometer designs. Our recent experimental setup [42] for Brewster microwave funneling, may be used to realize similar effects also for high-gigahertz emission and absorption. Finally, the proposed thermal emitters with small angular beamwidth may lead to novel directional IR thermal sources with broadband coherent emission, as well as novel thermo- photovoltaic screens for efficient heat conversion and radiation. These thin covers may be realized with nanoskiving [43] or related nanofabrication techniques. More efficient solar cells may be built based on the proposed concepts, generally leading to new robust photonic devices. This work has been supported by the ARO STTR project “Dynamically Tunable Metamaterials”, the AFOSR YIP award No. FA9550-11-1-0009, the ONR MURI grant No. N00014-10-1-0942 and the DARPA SBIR project “Nonlinear Plasmonic Devices”. References [1] R. N. Hall, G. E. Fenner, J. D. Kingsley, T. J. Soltys, R. O. Carlson, Phys. Rev. Lett. 9, 366 (1962). [2] J. Faist, F. Capasso, D. L. Sivco, C. Sirtori, A. L. Hutchinson, A. Y. Cho, Science 264, 553 (1994). [3] M. Planck, Ann. Phys. 309, 553 (1901). 9 [4] J. A. Schuller, E. S. Barnard, W. Cai, Y. C. Jun, J. S. White, M. L. Brongersma, Nature Mater. 9, 193 (2010). [5] J. J. Greffet et al., Nature (London) 416, 61 (2002). [6] X. Liu, T. Tyler, T. Starr, A. F. Starr, N. M. Jokerst, W. J. Padilla, Phys. Rev. Lett. 107, 045901 (2011). [7] J. A. Schuller, T. Taubner, M. L. Brongersma, Nature Photon. 3, 658 (2009). [8] A. G. Curto, G. Volpe, T. H. Taminiau, M. P. Kreuzer, R. Quidant, N. F. van Hulst, Science 329, 930 (2010). [9] H. T. Miyazaki, et al., Appl. Phys. Lett. 92, 141114 (2008). [10] N. Mattiucci, G. D’Aguanno, A. Alù, C. Argyropoulos, J. V. Foreman, M. J. Bloemer, Appl. Phys. Lett. 100, 201109 (2012). [11] I. Puscasu, W. L. Schaich, Appl. Phys. Lett. 92, 233102 (2008). [12] G. D’Aguanno, N. Mattiucci, A. Alù, C. Argyropoulos, J. V. Foreman, M. J. Bloemer, Opt. Express 20, 9784 (2012). [13] R. Siegel, J. R. Howell, Thermal Radiation Heat Transfer (Taylor & Francis, New York, 2002). [14] D. L. C. Chan, M. Soljačić, J. D. Joannopoulos, Phys. Rev. E 74, 036615 (2006). [15] N. Liu, M. Mesch, T. Weiss, M. Hentschel, H. Giessen, Nano Lett. 10, 2342 (2010). [16] M. Diem, T. Koschny, C. M. Soukoulis, Phys. Rev. B 79, 033101 (2009). [17] J. Hao, L. Zhou, M. Qiu, Phys. Rev. B 83, 165107 (2011). [18] J. Hendrickson, J. Guo, B. Zhang, W. Buchwald, R. Soref, Opt. Lett. 37, 371 (2012). [19] C. Hu, Z. Zhao, X. Chen, X. Luo, Opt. Express 17, 11039 (2009). [20] A. Tittl, et al., Nano Lett. 11, 4366 (2011). 10 [21] T. V. Teperik, F. J. García de Abajo, A. G. Borisov, M. Abdelsalam, P. N. Bartlett, Y. Sugawara, J. J. Baumberg, Nat. Photon. 2, 299 (2008). [22] K. Aydin, V. E. Ferry, R. M. Briggs, H. A. Atwater, Nat. Commun. 2, 517 (2011). [23] M. K. Hedayati, et al., Adv. Mater. 23, 5410 (2011). [24] Y. Cui, K. H. Fung, J. Xu, H. Ma, Y. Jin, S. He, N. X. Fang, Nano Lett. 12, 1443 (2012). [25] T. Søndergaard, S. M. Novikov, T. Holmgaard, R. L. Eriksen, J. Beermann, Z. Han, K. Pedersen, S. I. Bozhevolnyi, Nat. Commun. 3, 969 (2012). [26] A. Alù, G. D’Aguanno, N. Mattiucci, M. Bloemer, Phys. Rev. Lett. 106, 123902 (2011). [27] C. Argyropoulos, G. D’Aguanno, N. Mattiucci, N. Akozbek, M. J. Bloemer, A. Alù, Phys. Rev. B 85, 024304 (2012). [28] M. I. Stockman, Phys. Rev. Lett. 93, 137404 (2004). [29] A. R. Davoyan, I. V. Shadrivov, A. A. Zharov, D. K. Gramotnev, Y. S. Kivshar, Phys. Rev. Lett. 105, 116804 (2010). [30] P. B. Johnson, R. W. Christy, Phys. Rev. B 6, 4370 (1972). [31] J. J. Lermé, Phys. Chem. C 115, 14098 (2011). [32] See supplementary material for different grating designs, which can achieve ultra- broadband omnidirectional absorption. [33] X. R. Huang, R. W. Peng, R. H. Fan, Phys. Rev. Lett. 105, 243901 (2010). [34] R.H. Fan, R.W. Peng, X.R. Huang, J. Li, Y. Liu, Q. Hu, M. Wang, X. Zhang, Adv. Mater. 24, 1980 (2012). [35] E. Verhagen, A. Polman, L. Kuipers, Opt. Express 16, 45 (2008). [36] E. J. R. Vesseur, R. de Waele, M. Kuttge, A. Polman, Nano Lett. 7, 2843 (2007). 11 [37] CST Design Studio 2011, www.cst.com. [38] E. Popov, S. Enoch, N. Bonod, Opt. Express 17, 6770 (2009). [39] See supplementary material for similar thermal emission results to Fig. 3 but for higher temperatures T  1500 K . [40] K. Q. Le, C. Argyropoulos, N. Mattiucci, G. D’Aguanno, M. J. Bloemer, A. Alù, Journ. Appl. Phys. 112, 094317 (2012). [41] K. N. Rozanov, IEEE Trans. Antennas Propag. 48, 1230 (2000). [42] N. Akozbek, N. Mattiucci, D. de Ceglia, R. Trimm, A. Alù, G. D’Aguanno, M. A. Vincenti, M. Scalora, M. J. Bloemer, Phys. Rev. B 85, 205430 (2012). [43] Q. Xu, R. M. Rioux, M. D. Dickey, G. M. Whitesides, Acc. Chem. Res. 41, 1566 (2008). 12 Figures Figure 1 – a) Geometry of a 1D periodic grating combined with a plasmonic taper. b) Geometry of a 2D grating obtained combining two crossed plasmonic tapers. Both devices are illuminated at oblique incidence by a TM polarized wave. Notice that these model sketches are not in scale. 13 Figure 2 – Angular absorption spectra for the structure in Fig. 1(a) with period d  96 nm and nm 980 , slit width w  24 nm with 1 S  . The grating and taper lengths are l  200 nm and tapl  respectively. Broadband omnidirectional absorption is obtained at optical and IR frequencies. 14 Figure 3 – Thermal emission of a directional plasmonic emitter with d m 1.44 , w  90 nm , l m 2 , tapl m 9.1 , s  at a) T = 700 K and b) T = 300 K. The insets in the bottom of the 1 two panels show the emission spectrum of an ideal blackbody at the same temperature. Both plots are normalized to the maximum blackbody radiation. 15 Figure 4 – a) Comparison of absorption for the 1D [Fig. 1(a)] and 2D [Fig. 1(b)] structures with same dimensions as in Fig. 2, both at the Brewster angle and at normal incidence. b) Similar comparison for emission at T  700 K for the design in Fig. 3. In the 2D geometries, the azimuthal angle is fixed to 45o , which represents the worst-case scenario. 16 Supplementary material for the paper ‘Broadband Omnidirectional Absorbers and Selective Thermal Emitters based on Plasmonic Brewster Transmission’ In this supplemental material, we further analyze the dependence and robustness of our presented results on several assumptions and design parameters considered in the main text. First of all, as discussed in the main manuscript, we have considered increased losses in the gold plasmonic taper, assuming a collision frequency tap , where  is the collision frequency of bulk gold 10  [19]. This was motivated by two fundamental issues: the effective collision frequency of a metal is modified (a) when the dimensions of a plasmonic waveguide become comparable to the electron mean free-path and (b) when the temperature of operation increases, leading to an effectively larger absorption coefficient. As we show in the following, this assumption does not qualitatively affect the overall concepts presented in the paper, other than the fact that a reduced absorption leads to a longer tapering length required to absorb the impinging energy. In Fig. S1, we consider a plasmonic grating analogous to Fig. 1(a) of the main paper, but in which we have removed the entrance layer ( l  ) and assumed a collision frequency 0    tap 17.65 THz tapl m 1.76 [30]. Its absorption features for , calculated for simplicity using our analytical TL model, validated with full-wave 1 S  and 6w nm  , 24 nm , d simulations for several examples, are shown in Fig. S2(a). The results are consistent with the ones obtained in the paper, after slightly increasing the length of the taper. In order to decrease the required length tapl of the plasmonic taper we can use a more lossy plasmonic material, such as platinum (Pt), with experimental values of permittivity extracted from [1]. The absorption for an optimized design with w  24 nm , d  72 nm , 1 S  and tapl  980 nm is shown in Fig. S2(b), providing similarly good performance. Figure S1 – Geometry of the 1D plasmonic taper. The device is illuminated by an oblique incidence TM polarized wave. Finally, it may be desirable for some applications to extend the proposed plasmonic directional thermal emission to higher frequencies, such as near-infrared and optical. In order to achieve these selective emission properties, we need to increase the temperature to higher values than T  1000 K , for which gold will melt. Different metals needs to be considered for these cases with higher melting points, such as tungsten [1]. A directional thermal emitter based on tungsten is shown in Fig. S3(a) with dimensions 6w nm , d  192 nm , 1 S  , l  200 nm and tapl  980 nm . The tungsten grating follows a relative Drude permittivity dispersion with parameters  W 1   pf 2 / f    f   i   ,  pf  1448 THz, 13  THz [1]. The taper has length tapl and is also made of tungsten, which follows a similar Drude model but with increased collision frequency tap  130 THz to take into account the temperature effects and the reduced slit width, similar to the previous discussion. Figure S2 – a) Angular absorption spectra for the structure shown in Fig. S1 made of gold with period d  24 nm , slit width 6w nm and length tapl m 1.76 . b) Angular absorption spectra for the structure shown in Fig. S1 made of platinum with period d  72 nm , slit width w  24 nm and length tapl  980 nm . Both structures’ slits are loaded with free-space permittivity 1 S  . Broadband omnidirectional absorption is obtained at optical and IR frequencies for both cases. The selective thermal emission distribution at temperature T  1500 K is shown in Figs. S3(b), centered at near-infrared frequencies. This may be further extended to optical frequencies in case we further increase the temperature, since the melting point of tungsten is T  3695 K . Broadband thermal emission with essentially the same bandwidth of an ideal blackbody (see inset in the bottom of Fig. S3(b)), is clearly obtained, confined to a narrow angular range with large spatial coherence. This directive near-infrared emission is confined around the plasmonic Brewster angle, as discussed in the manuscript. Figure S3 – a) Geometry of the 1D selective thermal emitter based on tungsten. b) Emission of the directional plasmonic emitter at T  1500 K normalized to the maximum of the blackbody radiation at the same temperature, shown in the bottom part of (b). References [1] E. D. Palik, Handbook of Optical Constants of Solids (Academic Press, New York, 1985).
1809.04722	3	1809	2019-02-14T01:24:04	Observation of topological edge modes in a quasi-periodic acoustic waveguide	[ "cond-mat.mes-hall" ]	Topological boundary and interface modes are generated in an acoustic waveguide by simple quasi-periodic patternings of the walls. The procedure opens many topological gaps in the resonant spectrum and qualitative as well as quantitative assessments of their topological character are supplied. In particular, computations of the bulk invariant for the continuum wave equation are performed. The experimental measurements reproduce the theoretical predictions with high fidelity. In particular, acoustic modes with high Q-factors localized in the middle of a breathable waveguide are engineered by a simple patterning of the walls.	cond-mat.mes-hall	cond-mat	Observation of topological edge modes in a quasi-periodic acoustic waveguide David J. Apigo,1 Wenting Cheng,1 Kyle F. Dobiszewski,2 Emil Prodan,3 and Camelia Prodan1 1Department of Physics, New Jersey Institute of Technology, Newark, NJ, USA 2Albert Dormans Honors College, New Jesrey Institute of Technology, Newark, NJ, USA 3Department of Physics, Yeshiva University, New York, NY, USA Topological boundary and interface modes are generated in an acoustic waveguide by simple quasi- periodic patterning of the walls. The procedure opens many topological gaps in the resonant spectrum and qualitative as well as quantitative assessments of their topological character are supplied. In particular, computations of the bulk invariant for the continuum wave equation are performed. The experimental measurements reproduce the theoretical predictions with high fidelity. In particular, acoustic modes with high Q-factors localized in the middle of a breathable waveguide are engineered by a simple patterning of the walls. The ideas based on topological concepts [1, 2] have revolutionized the field of condensed matter physics and led to the discovery of topological insulators and super- conductors. The latter have been classified at the end of the previous decade [3 -- 6] and a table of strong topolog- ical phases has been conjectured. One of their common characteristics is the emergence of disorder-immune boundary modes whenever a sample is halved. Physics akin to that of topological condensed matter systems has been also predicted in classical wave-supporting mate- rials [7, 8] and many examples of topological metamate- rials have been reported in the literature [9 -- 22]. At the same time, it has been pointed out that the pe- riodic table of topological systems is highly enhanced if more complex systems are considered, such as the quasi-periodic or quasi-crystalline ones [23 -- 26]. In [27], K-theoretic arguments [28, 29] were applied for quasi- periodically coupled discrete mechanical resonators. The finding was that, if these are single-mode resonators, then every gap in the bulk resonant spectrum is topo- logical, in the sense that it will be completely filled by boundary spectrum under any boundary condition. The practical value of the finding is that the quasi-periodic Hamiltonians display a large number of topological gaps, hence one can generate localized wave-modes in both space and energy by simply halving the system. In this work, we put these general principles to the test in a completely different regime and we implement them for the first time using sound waves. Acoustic setups have been successfully used in the past to generate topo- logical edge modes [30 -- 32] and even to map the Hofs- tadter butterfly [33]. In particular, [32, 33] introduced re-configurable acoustic resonant structures where the building blocks are sealed acoustic chambers connected via thin bridges. They have isolated resonant modes, hence these structures fall under the umbrella of pat- terned resonators introduced in [27] and they can be analyzed by similar methods. However, these types of acoustic structures are not breathable, which is a key requirement for many practical applications. As such, here we ask the question: Can one generate topological edge and interface modes by patterning the walls of an acoustic waveguide without impeding the air flow? As we shall see, the answer is yes, but the methods of analysis are very different from those introduced in [27]. Indeed, the picture of coupled discrete resonators is no longer applicable and a full continuum medium treatment must be employed for the theoretical analysis. Furthermore, the topological character of the spectral gaps cannot be taken for granted because the waveguide supports many overlapping modes. As such, a new as- sessment of the topological character is introduced based on the continuum version of the lattice non-commutative Chern number proposed in [25], achieved in [34]. This invariant is here evaluated numerically using the meth- ods developed in [35, 36]. Let us recall from [27] that the role of aperiodicity in this type of applications is to generate virtual dimensions and, as we shall see [37], the Chern number mentioned above is defined on a 3- dimensional non-commutative manifold, while for dis- crete patterns is on a 2-dimensional manifold. At the experimental level, challenges exist because some of the spectral bands are very narrow and this, to- gether with the aperiodicity, can lead to irregular mode profiles, although the bulk states are extended. As such, the only way to accurately map the bulk spectrum is to collect data from a large number of points along the waveguide. Following this protocol, we map not only the frequecy but also the spatial profile of the bulk modes. Furthermore, inside the topological bulk gaps, we were able to detect sharp edge modes, which flow with the phason degree of freedom in a manner consis- tent with the computed Chern numbers. The quasi-periodic acoustic waveguide consists of a uniform cylindrical tube decorated with walls. The parts were 3D-printed out of polylactic acid (PLA) using an Ul- timaker 3 and then assembled as in Fig. 1.The walls have identical thickness but the spacings between adjacent walls are modulated according to the algorithm: Ln = Lavg + ∆L sin(nθ + φ), (1) The geometric parameters used in the experiments are supplied in Fig. 1. To make the above labels meaning- n ∈ Z. 9 1 0 2 b e F 4 1 ] l l a h - s e m . t a m - d n o c [ 3 v 2 2 7 4 0 . 9 0 8 1 : v i X r a 2 FIG. 1. Top: photograph of the waveguide configuration used to measure topological interface modes. Bottom: Cross-section and geometrical parameters. The waveguide consists of interlocking 3D printed PLA parts as shown in the insert and it is mirrored relative to the domain wall indicated by the dashed line. For experimentation, a speaker is placed at portholes accessible in each chamber and a piezoelectric microphone is inserted into an opposite porthole. The portholes that are not in use are sealed. The lengths Ln were generated with Eq. (1) and their average was fixed at Lavg = 40 mm. The parameters in Eq. (1) were fixed at −3), which ∆L = 0.2Lavg and θ = 2π√ was used in some of the numerical calculations. The system was also run without a domain wall, for bulk and edge measurements. 117 . This particular irrational fraction of 2π accepts a good rational approximation θ = 9π + O(10 48 ful, we assume that the waveguide is centered at a point inside L0. In Equation (1), θ is an angle incommensurate with 2π, which will be kept fixed during the measure- ments, and φ is the phason, which should be let to vary. For example, a simple relabeling n → n + m, which cor- responds to re-centering the waveguide, will change φ into (φ + mθ)mod 2π. Since θ is incommensurate, these relabelings alone will sample the phason densely in the [0, 2π] interval. Lavg in Equation (1) is the average dis- tance between the walls and ∆L sets the magnitude of the fluctuations in Ln. In the inset of Fig. 1, we show a front view of the waveguide, confirming that air can flow freely through the structure. It is then somewhat striking that, with the proposed patterning, we can stop sound propagation over several intervals of frequencies and, furthermore, we can generate, very much on demand, topological sound modes localized at any desired location along the tube. As opposed to an ordinary resonant mode pro- duced in a fully sealed acoustic chamber, the interface modes produced in the present work have less contact with the boundary, hence they are expected to have very high Q-factors, a much desired characteristic for practi- cal applications. To understand the effect of the patterning, we report in Fig. 2 the dispersion of the acoustic modes for clean and periodically (Ln = Lavg) patterned waveguides, as well as the resonant spectrum of the aperiodically pat- terned waveguide (Ln set by (1)). As expected for quasi 1-dimensional wave propagation, the periodic pattern opens spectral gaps in the gapless spectrum of the clean tube. These gaps, however, are not topological. The role of aperiodicity is to open additional gaps in the spectrum that, as one can see, resemble quite closely the Hofstadter butterfly [38], when mapped as function of θ. As we shall see, these are the gaps that carry non-trivial bulk topological invariants prompting the topological edge and interface modes. Let us mention that the spectra in Fig. 2 were produced with an in-house Fortran code, which diagonalizes the Laplace operator expressed in the cylindrical coordinates (ρ, z) and resolved over the azimuthal symmetry sectors. In appropriate units, the operator reads: ∆m = − 1 , m = 0,±1, . . . , (2) ρ ∂ ∂ρ + m2 ρ2 ∂ ∂ρ ρ − ∂2 ∂z2 and von Neumann condition is considered at the bound- ary. Recall that the latter is set by θ and φ, hence ∆m de- pends in a fundamental way on these parameters. The Laplace operator was discretized using finite differences. The protocol for acoustic data acquisition was as fol- lows. Sinusoidal signals of duration 1 s and amplitude of 0.5 V were produced by a Rigol DG1022 function genera- tor, amplified by a Crown XLS 2502 power amplifier with the gain set to 6, and then applied on a CUI Inc. GF0501 speaker, placed at one of the portholes. A PCB Piezotron- ics Model-378C10 microphone and a PCB Piezotronics Model-485B12 power conditioner acquired the acoustic signals at a porthole opposite the speaker (see Fig. 1). To account for the frequency-dependent response of the components, a separate measurement is performed with the waveguide removed but speaker and microphone kept in the same positions. All readings are normalized by the output of these measurements. The outputs were read by a custom LabVIEW code via a National Instru- ments USB-6112 data acquisition box and the ratio of the two measurements is stored on a computer for graphic renderings. 3 FIG. 3. Bulk resonant spectrum for the geometry described in Fig. 1. Left: Theoretical resonant spectrum reproduced from Fig. 2(c), with arrows indicating the topological gaps. The ver- tical marking identifies θ = 2π√ 117 , used in experiments. Center: Normalized microphone readings from the center of 48 cham- bers, recorded over a wide frequency interval. Right: Collapse on the frequency axis of the intensity plot reported in the mid panel. Three spectral gaps can be clearly identified in the experimental data and seen to be well aligned with the theo- retical calculations. The values of the Chern numbers for the two topological gaps are also indicated. boundary modes experimentally, the acquisition proto- col was applied on the second chamber from the left physical edge, which was plugged. The frequency was swept from 2.0 to 2.6 kHz in steps of 25 Hz and the value of the phason was modified by moving the phys- ical edge sequentially to the right, hence from L0 to Ln, n = 1, 2, . . .. The results are presented in Fig. 4b and they indeed confirm the existence of one chiral band in the upper topological gap and two such bands in the lower topological gap. For reference, we reproduced in panel c) the experimental data from Fig. 3, from where the ex- act position of the bulk edges can be inferred. As one can see, the boundary resonances occur inside the bulk gaps and the dispersion with φ is consistent with the theoretical prediction. We now demonstrate that a localized topological edge mode can be created without the assistance of any plug. For this, we consider a domain wall configuration: . . .L31L30L29L29L30L31 . . . where the waveguide is mirror-reflected relative to left edge of L29 chamber. This particular index was chosen because moving the origin to that chamber generates a phason φ = (29θ)mod 2π, which coincides with the FIG. 2. a) Dispersion of the acoustic modes for the un-patterned waveguide, for m = 0 sector. b) The band structure of a peri- odically patterned waveguide (i.e. θ = 0), for m = 0 sector. c,d) Resonant spectrum of a patterned waveguide as function of θ, for m = 0, 1 sectors, respectively. For the bulk measurements, the protocol was repeated for all 48 chambers of a patterned waveguide, with fre- quency scans from 500 to 6000 Hz in 25 Hz steps. The re- sults are reported in Fig. 3. When the data is rendered as function of frequency and chamber index, clear extended acoustic modes can be identified. Furthermore, when the data is collapsed on the frequency axis, clear spectral gaps can be identified, two of which are predicted to be topological. Unfortunately, the m = 0, 1 spectra over- lap above the non-topological gap (see Fig. 2) and the higher frequency topological gaps could not experimen- tally resolved. Let us note that the agreement between experiment and theory in Fig. 3 is less than 5%. To assess the topological character of the gaps, we em- ploy the bulk-boundary correspondence for continuum models established in [34]. The bulk-topological invari- ant is supplied by the non-commutative Chern number of the gap projection PG = χ(−∞,G] (cid:16) (cid:17) (cid:16) ∆m(φ) − G (cid:17), PG[∂φPG, [Z, PG]] : Ch(PG) = TrL (3) where Z is the position operator parallel to the tube and TrL is the trace per length. The invariant can be com- puted at any arbitrary but fixed phason value, which is a consequence of Birkhoff ergodic theorem [39]. With the Laplacian discretized on a lattice via finite differ- ences, Eq. (3) was evaluated using methods which are by now standard [35, 36]. The results are reported in Fig 3, confirming that the smaller gaps are topological. Furthermore, [34] established the existence of a bound- ary topological invariant which counts the number of chiral boundary bands, as well as the equality between the bulk and boundary invariants. The presence of chiral modes, in accordance to the above bulk-boundary principle, is confirmed by our nu- merical simulations reported in Fig. 4a. To map the 4 FIG. 4. Topological edge spectrum. a) Theoretical prediction of the spectral flow against the phason parameter φ, demonstrat- ing the existence of chiral bands. The red/blue marks relate to the left/right edge of the waveguide, respectively. b) Experi- mental mapping of the spectral flow, confirming the existence of chiral bands. c) The measurements for bulk spectrum, re- produced from Fig. 3, indicating the position of the bulk gap edges. FIG. 5. a) Topological interface mode, measured for a waveg- uide configuration similar to that in Fig. 1. The spatial localiza- tion of the interface mode was mapped by moving the speaker and microphone incrementally away from the domain wall. b) The topological interface mode is also observed in COMSOL simulations. Red, blue and green colors represent high, low and zero pressure variations, respectively. value where strong mid-gap edge modes were observed in the first topological gap. Since Eq. 3 is odd under re- flection, with this patterning, an interface between topo- logical systems with opposite Chern numbers is created As such, the bulk-boundary principle predicts the emer- gence of 2× Ch acoustic modes localized at the interface. The experimental measurements are reported in Fig. 5(a). The frequencies were swept as in Fig. 4 and, in order to probe the localization of the acoustic modes, the speaker and microphone were placed at several port- holes at and away from the interface. A strong and sharp resonance was detected in the first topological gap (Ch = 1), when the measurements were performed one and two chambers away from the interface. The reso- nance was not detectable further away from the interface or at the interface itself. A similar resonance can be de- tected at the other side of the interface, leading to a full confirmation of the topological bulk-boundary predic- tion. The interface mode is also observed in a standard COMSOL simulation, as shown in Fig. 5(b). In conclusion, we have demonstrated that topologi- cal edge and interface modes can be created by a simple quasi-periodic patterning of an acoustic waveguide. The topological gaps can be easily identified when the res- onant spectrum is mapped as function of modulation parameter θ. Furthermore, a topological invariant was computed and shown to be in agreement with the num- ber of observed topological chiral edge modes. As we have seen, quasi-periodicity opens topological gaps inside the bands of the periodic structure, which resemble the Hofstadter butterfly when mapped as function of θ. Optimization over ∆L in Eq. (1) and the geometric parameters of the tube, as well as improve- ments in materials (e.g. by replacing the polymer with metal), can highly enhance these topological gaps and the Q-factors of the topological boundary and interface modes. Other than that, the procedure requires no further fine tuning and, due to its simplicity, we believe it can be easily incorporated in practical applications. The present analysis can also serve as a model for acoustic implementations of many other promising aperiodic structures [40]. All authors acknowledge support from the W. M. Keck Foundation. to determine the operator algebra which supplies the as- sociated physical observables. As it is now well known [28], for continuous 1-dimensional models, this algebra is the crossed product A = C(Ω) (cid:111)τ R. The elements of this algebra belong to a certain class of complex valued functions over R × Ω and the multiplication rule is: dξ f1(ξ, τξ−zω) f2(z − ξ, ω). (cid:90) (5) The algebra accepts a canonical representation on L2(R): ( f1 ∗ f2)(z, ω) = (cid:104) (πω f )ψ(cid:105) (z) = R (cid:90) dξ f (z − ξ, τ−ξω)ψ(ξ). (6) R Here, z is the coordinate along the axis of the waveguide. The dispersion equation for our waveguide is defined over L2([0, R]) ⊗ L2(R), where [0, R] is the interval where the radial coordinate ρ takes values. The transversal modes, however, from topology point of view, brings nothing significant because all spectral projectors of the dispersion operator can be generated from the algebra K⊗A (via the above representation), where K is the alge- bra of compact operators over L2([0, R]). This is the case because the resolvent of the radial part of the Laplace operator is compact when ρ is restricted to a finite inter- val. the machinery developed in [34]: Topological Invariant. We now can specify the input for • If (ω1, ω2) are the coordinates of ω ∈ T2 = S × S, then we have the derivations ∂ω1 and ∂ω2 (= ∂φ), as well as: 5 (7) (8) (9) SUPPLEMENTAL The aperiodic continuum systems are quite different from the discrete aperiodic ones for the following rea- sons: • They can be halved at any point of their axis. Hence, the phason φ alone does not specify com- pletely the configuration of the waveguide, for we also need to know where the origin of the Eu- clidean space is located relative to the walls of the patterned waveguide. This is so because, by con- vention, it is at this origin where the cut is made and the edge modes emerge. • The bulk topological invariant, while still a formal Chern number, is defined on a different algebra of observables. • The proof [34] of the quantization and stability of the topological invariants for continuum models also proceeds quite differently from the one for discrete models [29]. The aim of this note is to walk the reader through [34], as adopted to the acoustic waveguide analyzed in the main text. The continuous hull. The continuous hull Ω of the pat- terned waveguide is the topological space traced by the pattern when one continuously translates the waveguide along the axis [28]. Here we show that Ω is a 2-torus. With the elements introduced in diagram 6, let ω ∈ T2 be an arbitrary point and imagine the blue line as being a physical rope wound around the torus. Let the red line, whose equation is supplied in the diagram, be soaked with ink so that, every time when the rope crosses the red line, a mark is imprinted. Let us label these marks as shown in the diagram. Then, after we unwind the rope and lay it flat and parallel to axis of the tube, one will find that: (cid:16) xn+1 − xn = Lavg + y (cid:17) − y(nθ + φω) (n + 1)θ + φω = Lavg + ∆L sin(nθ + φω). The marks xn will overlay perfectly over the centers of the walls if the origin of the Euclidean space is fixed at xω (the position of ω on the rope)! The conclusion is that every rigidly translated waveg- uide configuration can be uniquely characterized by a point ω ∈ T2, hence the continuous hull is the 2- torus. Furthermore, the group of translations parallel to the waveguide's axis induces an action τ of R on T2, which amounts to shifting ω along the winded rope. As such, the hull becomes a topological dynamical system (Ω, τ, R). Algebra of physical observables. When defining a topo- logical invariant for aperiodic systems, the first task is f ∈ A. (∂z f )(z, ω) = z f (z, ω), • The trace Tr ⊗ T on K ⊗ A with: dω (2π)2 f (0, ω) T ( f ) = T2 (cid:90) Then, for a projection p ∈ K ⊗ A, [34] showed that: (4) Ch(p) = 2π Tr ⊗ T(cid:16) (cid:17) p[∂ω2p, ∂zp] equals the index of a certain Fredholm operator, which ensures the quantization and stability of this Chern num- ber. We can use the physical representation to write this invariant. Indeed, if Pω = πω(p), then: Ch(P) = dρ dω(cid:104)0, ρPω[∂φPω, [Z, Pω]]0, ρ(cid:105), (10) T2 where Z is the position operator parallel to the tube. Lastly, since τ acts ergodically on T2, we have from Birkhoff theorem [39]: dρ dω(cid:104)0, ρ . . .0, ρ(cid:105) = TrL(. . .), (11) T2 (cid:90) (cid:90) (cid:90) (cid:90) 6 FIG. 6. The virtual manifold associated to the patterned waveguide is a 2-torus. the latter being the trace per length introduced in the main text. With this simplification, Eq. (10) becomes identical with the one supplied in the main text. Bulk-boundary correspondence. According to [34], the topological class of Pω is mapped into the K1-class of the torus T2 generated by the function eıφ. Since it involves only the vertical coordinate of the torus 6, the horizon- tal coordinate plays no role in the bulk-boundary corre- spondence treated in our work. [1] D. Thouless, M. Kohmoto, M. Nightingale, M. den Nijs, Quantized Hall Conductance in a Two-Dimensional Peri- odic Potential, Phys. Rev. Lett. 49, 405-409 (1982). [2] F. D. M. Haldane, Model for a Quantum Hall-Effect with- out Landau levels: Condensed-matter realization of the parity anomaly, Phys. Rev. Lett. 61, 2015-2018 (1988). [3] A. P. Schnyder, S. Ryu, A. Furusaki, A. W. W. Ludwig, Classification of topological insulators and superconduc- tors in three spatial dimensions, Phys. Rev. B 78, 195125 (2008). [4] X.-L. Qi, T. L. Hughes, Shou-Cheng Zhang, Topological field theory of time-reversal invariant insulators, Phys. Rev. B 78, 195424 (2008). [5] A. Kitaev, Periodic table for topological insulators and su- perconductors, (Advances in Theoretical Physics: Landau Memorial Conference) AIP Conference Proceedings 1134, 22-30 (2009). [6] S. Ryu, A. P. Schnyder, A. Furusaki, A. W. W. Ludwig, Topological insulators and superconductors: tenfold way and dimensional hierarchy, New J. Phys. 12, 065010 (2010). [7] F. D. M. Haldane, S. Raghu, Possible realization of direc- tional optical waveguides in photonic crystals with broken time-reversal symmetry, Phys. Rev. Lett. Lett. 100, 013904 (2008). [8] E. Prodan, C. Prodan, Topological phonon modes and their role in dynamic instability of microtubules, Phys. Rev. Lett. 103, 248101 (2009). [9] Z. Wang, Y. Chong, J. D. Joannopoulos, M. Soljacic, Obser- vation of unidirectional backscattering-immune topolog- ical electromagnetic states, Nature 461, 772 -- 775 (2009). [10] L. M. Nash, D. Kleckner, A. Read, V. Vitelli, A. M. Turner, W. T. M. Irvine, Topological mechanics of gyroscopic meta- materials, Proc. Nat. Acad. Sci. 112, 14495-14500 (2015). [11] M. Hafezi, S. Mittal, J. Fan, A. Migdall, J. M. Taylor, Imag- ing topological edge states in silicon photonics, Nature Photonics 7, 1001-1005 (2013). [12] L.-H. Wu, X. Hu, Scheme for Achieving a Topological Photonic Crystal by Using Dielectric Material, Phys. Rev. Lett. 114, 223901 (2015). [13] R. S usstrunk, S. Huber, Observation of phononic helical edge states in a mechanical topological insulator, Science 349, 47-50 (2015). [14] C. Kane, T. Lubensky, Topological boundary modes in isostatic lattices, Nature Physics 10, 39-45 (2013). [15] J. Paulose, B. G. Chen, V. Vitelli, Topological modes bound to dislocations in mechanical metamaterials, Na- ture Physics 11, 153-156 (2015). [16] E. Prodan, K. Dobiszewski, A. Kanwal, J. Palmieri, C. Pro- dan, Dynamical Majorana edge modes in a broad class of topological mechanical systems, Nature Communications 8, 14587 (2017). [17] A. Slobozhanyuk, S. H. Mousavi, X. Ni, D. Smirnova, Y. S. Kivshar, A. B. Khanikaev, Three-dimensional all-dielectric photonic topological insulator, Nature Photonics 11, 130- 136 (2017). [18] S. H. Mousavi, A. B. Khanikaev, Z. Wang, Topologically protected elastic waves in phononic metamaterials, Na- ture Communications 6, 8682 (2015). [19] M. Miniaci, R. K. Pal, B. Morvan, M. Ruzzene, Observation of topologically protected helical edge modes in Kagome elastic plates, arXiv:1710.11556 (2017). qy𝜙x0x1x2x-1𝜔Origin𝑦𝜙=−Δ𝐿cos𝜙−𝜃/22sin𝜃/2x3𝜙𝜔[20] R. Chaunsali, C.-W. Chen, J. Yang, Subwavelength and directional control of flexural waves in zone-folding in- duced topological plates, Phys. Rev. B 97, 054307 (2018). [21] H. Chen, H. Nassar, G. Huang, Topological mechanics of edge waves in Kagome lattices, arXiv:1802.04404 (2018). [22] R. K. Pal, M. Schaeffer, M. Ruzzene, Helical edge states and topological phase transitions in phononic systems using bilayered lattices, J. Appl. Phys. 119, 084305 (2016). [23] Y. E. Kraus, Y. Lahini, Z. Ringel, M. Verbin, O. Zilberberg, Topological states and adiabatic pumping in quasicrystals, Phys. Rev. Lett. 109, 106402 (2012). [24] M. Verbin, O. Zilberberg, Y. E. Kraus, Y. Lahini, Y. Sil- berberg, Observation of topological phase transitions in photonic quasicrystals, Phys. Rev. Lett. 110, 076403 (2013). [25] E. Prodan, Virtual topological insulators with real quan- tized physics, Phys. Rev. B 91, 245104 (2015). [26] F. Baboux, E. Levy, A. Lemaitre, C. Gomez, E. Galopin, L. Le Gratiet, I. Sagnes, A. Amo, J. Bloch, E. Akkermans, Measuring topological invariants from generalized edge states in polaritonic quasicrystals, Phys. Rev. B 95, 161114 (2017). [27] D.J. Apigo, K. Qian, C. Prodan, and E. Prodan, Topological edge modes by smart patterning, Phys. Rev. Materials 2, 124203 (2018). ∗ [28] J. Bellissard, K-theory of C -Algebras in solid state physics, Statistical mechanics and field theory: mathematical as- pects, (Springer, Berlin, Heidelberg, 1986, 99-156). [29] E. Prodan, H. Schulz-Baldes, Bulk and boundary invari- ants for complex topological insulators: From K-theory to physics, (Springer, Berlin, 2016). [30] M. Xiao, G. Ma, Z. Yang, P. Sheng, Z.Q. Zhang, C.T. Chan, 7 Geometric phase and band inversion in periodic acoustic systems, Nature Physics, 11:240, (2015). [31] C. He, X. Ni, H. Ge, X. Sun, Y. Chen, M. Lu, X. Liu, Y. Chen, Acoustic topological insulator and robust one-way sound transport, Nature Physics, 12:1124, (2016). [32] X. Ni, M. Weiner, A. Al, A. B. Khanikaev, Observation of bulk polarization transitions and higher-order embedded topological eigenstates for sound, arXiv:1807.00896 (2018). [33] O. Richoux, V. Pagneux, Acoustic characterization of the Hofstadter butterfly with resonant scatterers, EPL (Euro- physics Letters), Vol 59, 34 (2002). [34] C. Bourne, A. Rennie, Chern numbers, localisation and the bulk-edge correspondence for continuous models of topological phases, Math. Phys. Anal. Geom. 21: 16. https://doi.org/10.1007/s11040-018-9274-4 (2018). [35] E. Prodan, A computational non-commutative geometry program for disordered topological insulators, Springer Briefs in Mathematical Physics, Springer, 2017. [36] C. Bourne, E. Prodan, Non-commutative Chern numbers for generic aperiodic discrete systems, J. Phys. A: Math. Theor. 51, 235202 (2018). [37] See Supplemental Material at [URL] for a computation of the continuous hull of the pattern as well as for the definition of the bulk topological invariant. [38] D. R. Hofstadter, Energy levels and wave functions of Bloch electrons in rational and irrational magnetic fields, Phys. Rev. B 14, 2239-2249 (1976). [39] G. D. Birkhoff, Proof of the ergodic theorem, Proc. Natl. Acad. Sci. USA 17, 656-660 (1931). [40] E. Prodan, Y. Shmalo, The K-Theoretic Bulk-Boundary Principle for Dynamically Patterned Resonators, Journal of Geometry and Physics 135, 135-171 (2019).
1304.2871	2	1304	2013-04-11T06:49:47	Single Hole Transport in a Silicon Metal-Oxide-Semiconductor Quantum Dot	[ "cond-mat.mes-hall" ]	We describe a planar silicon metal-oxide-semiconductor (MOS) based single hole transistor, which is compatible with conventional Si CMOS fabrication. A multi-layer gate design gives independent control of the carrier density in the dot and reservoirs. Clear Coulomb blockade oscillations are observed, and source-drain biasing measurements show that it is possible to deplete the dot down to the few hole regime, with excited states clearly visible. The architecture is sufficiently versatile that a second hole dot could be induced adjacent to the first one.	cond-mat.mes-hall	cond-mat	Single Hole Transport in a Silicon Metal-Oxide-Semiconductor Quantum Dot R. Li,1 F. E. Hudson,2, 3 A. S. Dzurak,2, 3 and A. R. Hamilton1, a) 1)School of Physics, University of New South Wales, Sydney NSW 2052, Australia 2)Australian National Fabrication Facility, University of New South Wales, Sydney NSW 2052, Australia 3)Centre of Excellence for Quantum Computation and Communication Technology, School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney NSW 2052, Australia (Dated: 4 August 2021) We describe a planar silicon metal-oxide-semiconductor (MOS) based single hole transistor, which is compati- ble with conventional Si CMOS fabrication. A multi-layer gate design gives independent control of the carrier density in the dot and reservoirs. Clear Coulomb blockade oscillations are observed, and source-drain biasing measurements show that it is possible to deplete the dot down to the few hole regime, with excited states clearly visible. The architecture is sufficiently versatile that a second hole dot could be induced adjacent to the first one. Over the past 15 years much effort has gone into the development and study of electron quantum dots as arti- ficial atoms1,2, ultra-sensitive electrometers3, and quan- tum bits4 for quantum information applications. To use an electron in a quantum dot as a spin qubit requires 5,6. Signif- long spin life-time T1 and coherence-time T2 icant progress has been made with III-V semiconductor based devices, although T2 is limited by the hyperfine interaction between the electron spin and nuclei in the host crystal7. Spin qubits based on group-IV semicon- ductors have recently shown long T1 and T2 times8,9. However even in silicon based devices challenges remain due to the presence of nonzero nuclear spin in isotopes of Si, the valley degree of freedom in conduction band10, and disorder at the Si/SiO2 interface. Recently holes in quantum dots have attracted signifi- cant interest11,12 since the strong spin-orbit coupling en- ables all electrical spin manipulation13,14, while the hy- perfine interaction between holes and nuclear spins is strongly suppressed15, promising longer T2. Besides, for holes in silicon there is no valley degeneracy. However, to date there have been only a few studies of holes in gate defined quantum dots16 -- 19. In this letter, we describe a planar silicon metal-oxide-semiconductor (MOS) based single hole transistor, which is compatible with conven- tional Si CMOS fabrication. The MOS structure studied in this work was fabri- cated from a high-resistivity (ρ > 10 kΩ·cm) (100) sili- con substrate. Field-oxide, boron-diffused ohmic regions, and thin gate-oxide (thickness ∼ 5.9 nm) were defined by standard micro-fabrication techniques. Subsequently, multi-level aluminum gates were patterned by electron- beam lithography and lift-off. The gates were insulated from each other by a thin native AlOx layer20. The fi- nal stage was a forming gas anneal to reduce the Si/SiO2 interface trap density and enhance low-temperature per- formance20. a)Electronic mail: [email protected] (a) L1 S UG P2 P1 L2 D Hole Reservoir Dot (b) ) S µ ( G intrinsic Si substrate aluminum gates SiO2 gate oxide p+ ohmic regions p+ oxidized aluminum eg 0.8 0.6 0.4 0.2 0.0 -1.76 -1.72 VP1 (V) -1.68 FIG. 1. (a) Schematic cross section of the device. The hole reservoirs are induced by lead gates L1 and L2. For the data shown below the dot was induced below plunger gate P1, while UG and P2 were biased to extend the hole reser- voir. (b) Conductance vs plunger gate bias, showing peri- odic Coulomb blockade oscillations in the many hole regime. VL1=VL2=VP2=-4 V, and VUG=0 V. Inset shows the SEM image of a typical device. The white scale bar is 200 nm. Figure 1(a) shows a schematic cross section of the de- vice. There are three layers of gates: the first is the two plunger gates (P1 and P2), each 30 nm wide with a sepa- ration of 30 nm between them. The middle layer consists of the lead gates (L1 and L2) which were kept at -4 V to induce the source and drain hole reservoirs. The upper gate (UG) has a width of 50 nm and extends over P1 and P2. The multiple gates allow considerable flexibil- ity over device operation. Gates L1 and L2 were always negatively biased to induced holes into the leads, but the remaining gates could either be biased negative to in- duced holes underneath them, or positive to form tunnel barriers between regions of holes. In the following exper- iments we used a single gate, P1, to localize holes into a quantum dot and control the dot occupancy. In this mode of operation the entrance and exit tunnel barriers were formed due to the oxidized aluminum layer between different gates, and the upper gate was grounded as it had little effect on the dot. Gate P2 was kept at a large negative bias, to ensure that it was transparent. In this biasing arrangement the lithographic dimensions of the dot were defined by the width of P1 (30 nm) and the fringing field from L1 (150 nm in width), so we estimate the dot area ∼ 3 × 103 nm2. Several devices were tested at 4 K, with a yield of ∼ 50%. Further measurements were performed on one device in a dilution fridge with base temperature of 30 mK, using standard two-terminal lock-in techniques with a 100 µV ac excitation voltage. Fig. 1(b) shows the Coulomb blockade oscillations obtained when sweeping gate P1, demonstrating that the device functions as a single hole transistor. The number of holes in the dot was estimated to be ∼ 25. (a) . 6 3 - ) V ( 2 P V . 8 3 - . 0 4 - (b) G (µS) 0 1.0 2.0 . 4 0 ) V ( G U V . 2 0 . 0 0 -1.9 -1.8 -1.7 -1.6 -1.9 -1.8 -1.7 -1.6 VP1 (V) VP1 (V) (a) P2 with FIG. 2. Charge stability diagram of P1 vs. VUG=0 V, and (b) UG with VP2=-4 V. In both plots, VL1=VL2=-4 V. The dense parallel lines intercepting VP1 axis indicates single dot operation and that the dot was strongly coupled to P1. TABLE I. Gate capacitance to the dot. The capacitance val- ues are estimated from the average line-spacing in charge sta- bility diagrams. Gate P1 P2 L1 L2 UG C (aF) 16.8 0.70 1.48 0.28 0.83 Figure 2 shows the conductance of the quantum dot as a function of the biases on gates P1, P2 and UG. These charge stability diagrams show almost vertical lines, in- dicating that the dot was most strongly coupled to the P1 gate, and the P2 and UG gates had a much weaker capacitive coupling to the dot. Similar results were ob- tained for the L1 and L2 gates. The large number of periodic oscillations shows that the dot cannot be due to unintentional dopants or defects. The capacitance of the various gates to the dot was determined from the period- icity of the oscillations in the charge stability diagrams, as shown in Table I. These data confirm that the dot was 2 located under P1, since the capacitance to all other gates was much smaller. We also estimated the size of the dot using a simple parallel plate capacitor model with a sili- con oxide thickness d = 5.9 nm, and obtained an area of ∼ 2900 nm2, in good agreement with the area estimated from the lithographic dimensions of the dot. (a) 1 10 100 G (nS) 20 10 V s d ( m V ) 0 0 -10 -20 G (nS) 25 20 15 10 5 0 -1.60 -1.55 VP1 (V) -1.50 -1.45 -1.70 -1.65 (b) 10 ) V m ( d s V 5 0 -5 -10 -1.56 -1.54 -1.52 -1.50 VP1 (V) FIG. 3. (a) Source-drain biasing of the hole quantum dot, showing Coulomb diamonds down to last few holes. VL1=VL2=VP2=-4 V, and VUG=0V. (b) A close-up of the Coulomb diamonds, showing that excited states can be re- solved. Figure 3(a) shows source drain bias spectroscopy mea- surements, with well resolved Coulomb diamonds. At high gate bias (VP1 > 1.6 V), the number of holes is greater than ten and the charging energy of the dot was approximately constant at Ec ∼ 5 meV. As the dot was pinched off, by making VP1 more positive, the charging energy increased suggesting that the dot was shrinking in size and approaching the few hole limit. Fi- nally for VP1 > −1.5 V the charging energy increased rapidly and the Coulomb diamonds no longer close. It is tempting to ascribe this opening of the last Coulomb diamond as signaling the last occupied hole state in the dot. However, the observation of excited states at VP1 = −1.47 V shows that there must be at least one hole in the dot for VP1 > −1.47 V, suggesting the last hole charge state could not be reached in these mea- surements. No Coulomb diamonds could be resolved for VP1 >∼ −1.47 V, as the device became so pinched off that the conductance dropped below the background noise level of 1 nS (I=0.1pA). However, the ∼ 10 meV charging energy of the last diamond is a strong indication that we were approaching the last few holes in the dot. The well defined confining potential of the dot is fur- ther highlighted by the slope of the edges of the Coulomb diamonds. The slopes are the same for all diamonds in Fig. 3(a), giving a lever-arm α=CP1/CΣ=0.36. This sug- gests that the dot was defined underneath the central region of P1, and was not affected by disorder even in the few hole limit (since α would change and additional features would be observed in the bias spectroscopy if disorder induced parasitic dots were forming). Further- more the slope of the diamonds allowed the capacitive coupling to the source and drain reservoirs to be esti- mated, giving CS/CP1=1.1 and CD/CP1=0.77. The dot was more strongly coupled to the source reservoir than the drain reservoir, consistent with the geometry of the device. Figure 3(b) is a close up of the Coulomb diamonds, showing the excited states of the hole quantum dot. The excited states manifest as thin lines of high-conductance running parallel to the edge of the diamonds outside the Coulomb blockade region. The spacing of the excited states was ∆E ∼ 800 µeV at VP1 = −1.51 V, although even larger energy spacings ∼ 2 meV could also be re- solved. For comparison, measurements of a silicon elec- tron quantum dot fabricated using the same approach and with similar lithographic dimensions showed ∆E up to 600 µeV20. Since the hole mass is significantly larger than the electron mass, this would suggest the excited state spacing ∆E measured for the hole device should be smaller than in Ref. 20. However, the hole band structure is more complex than the electron bands, and is further complicated by the lateral confinement in the quantum dot. The thickness of the 2D hole system is ∼ 10 nm, comparable to the length-scale of the in-plane confine- ment geometry, indicating that the quantization of the hole states should be treated in 3D. The precise nature of the hole states, including the spin properties, shape of the orbital states and the degree of light and heavy hole mixing, is highly sensitive to the confining potential and will be a fruitful area for future research. Finally we show that this device can also be operated as a hole double quantum dot, by changing the bias on gate P2 so that a second dot formed as sketched in Fig- ure 4(a). The resulting charge stability diagram is pre- sented in Fig. 4(b), where the bias on gate P2 has been reduced from -4 V to a bias similar to VP 1. Dark regions indicate Coulomb blockade where the double dot main- tains the same charge configuration, and bright lines in- dicate current transport through the double dot, demark- ing regions where the hole occupation changes. In the top right of Fig. 4(b) the vertical lines show that gate P1 con- trols the hole number in dot 1, while the horizontal lines (a) S D Hole Reservoir Dot 1 Dot 2 Si AlOx Al p+ Ohmic SiO2 SiO2 (b) 4 7 1 - . ) V ( 2 P V 8 7 1 - . G ( µ S ) 3 2 1 0 -2.30 VP1 (V) -2.28 FIG. 4. (a) Schematic cross section of the device when the gates are biased to form a double hole quantum dot system. The difference to Fig. 1(a) is that P1 and P2 were in the same bias range and that a dot was induced under each of them. (b) Charge stability diagram obtained by sweeping gates P1 and P2 for VL1=VL2=-4 V and VUG=0 V. show that P2 controls the occupancy of the second dot. The horizontal line spacing is approximately twice that of the vertical spacing, suggesting that the capacitance between gate P2 and dot 2 is twice that between gate P1 and dot 1. This is consistent with the lithographic gate dimenions, since the width of gate L1 is twice that of L2 (see SEM image in Fig. 1(b). At more negative VP 1 and VP 2 the number of holes in the dots increased and there is evidence of coupling between the dots, as the lines become more diagonal. In summary, we have fabricated single hole transis- tors based on a planar silicon MOS structure. A well- defined hole quantum dot could be induced, and operated in both the many-hole and few-hole regimes. Bias spec- troscopy measurements show that the device can be oper- ated down to the few hole regime, showing large charging and excited state energies. The flexibility of the multi- gate structure also made it possible to form a second hole dot, with the charge stability diagram displaying weak coupling between the two dots. These devices will allow future studies of individual hole spins in standard silicon MOS structures. The authors thank N. S. Lai and A. Morello for help- ful discussions, A. Rossi for help with fabrication, L. A. Yeoh and A. Srinivasan for help with the dilution refrig- erator, and J. Cochrane for technical support. A.R.H. ac- knowledge support from the Australian Research Council (DP120102888 and DP120101859). F.E.H and A.S.D. ac- knowledge support from the Australian Research Council (CE11E0096) and the U.S. Army Research Office (con- tract W911NF-13-1-0024). Experimental devices for this study were fabricated with support from the Australian National Fabrication Facility, UNSW. Note added: After completing these measurements we became aware of similar experiments underway elsewhere21. 1L. P. Kouwenhoven, D. G. Austing, and S. Tarucha, Rep. Prog. Phys. 64, 701 (2001). 2W. H. Lim, C. H. Yang, F. A. Zwanenburg, and A. S. Dzurak, Nanotechnology 22, 335704 (2011). 3C. Barthel, M. Kjaergaard, J. Medford, M. Stopa, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. B 81, 161308 (2010). 4 4J. 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). 5D. Loss and D. P. DiVincenzo, Phys. Rev. A 57, 120 (1998). 6I. Zuti´c, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004). 7R. Hanson, L. P. Kouwenhoven, J. R. Petta, S. Tarucha, and L. M. K. Vandersypen, Rev. Mod. Phys. 79, 1217 (2007). 8A. Morello, J. J. Pla, F. A. Zwanenburg, K. W. Chan, K. Y. Tan, H. Huebl, M. Mottonen, C. D. Nugroho, C. Yang, J. A. van Donkelaar, A. D. C. Alves, D. N. Jamieson, C. C. Escott, L. C. L. Hollenberg, R. G. Clark, and A. S. Dzurak, Nature 467, 687 (2010). 9J. J. Pla, K. Y. Tan, J. P. Dehollain, W. H. Lim, J. J. L. Morton, D. N. Jamieson, A. S. Dzurak, and A. Morello, Nature 489, 541 (2012). Rev. Lett. 107, 176811 (2011). 14P. Szumniak, S. Bednarek, B. Partoens, and F. M. Peeters, Phys. Rev. Lett. 109, 107201 (2012). 15Z. K. Keane, M. C. Godfrey, J. C. H. Chen, S. Fricke, O. Klochan, A. M. Burke, A. P. Micolich, H. E. Beere, D. A. Ritchie, K. V. Trunov, D. Reuter, A. D. Wieck, and A. R. Hamilton, Nano Lett. 11, 3147 (2011). 16U. Dotsch, U. Gennser, C. David, G. Dehlinger, D. Grutzmacher, T. Heinzel, S. Luscher, and K. Ensslin, Appl Phys. Lett. 78, 341 (2001). 17B. Grbic, R. Leturcq, K. Ensslin, D. Reuter, and A. D. Wieck, Appl Phys. Lett. 87, 232108 (2005). 18O. Klochan, A. P. Micolich, A. R. Hamilton, K. Trunov, D. Reuter, and A. D. Wieck, Phys. Rev. Lett. 107, 076805 (2011). 19Y. Hu, F. Kuemmeth, C. M. Lieber, and C. M. Marcus, Nat. Nanotechnol. 7, 47 (2012). 10B. Koiller, X. Hu, and S. Das Sarma, Phys. Rev. Lett. 88, 027903 20S. J. Angus, A. J. Ferguson, A. S. Dzurak, and R. G. Clark, Nano (2001). 11D. V. Bulaev and D. Loss, Phys. Rev. Lett. 95, 076805 (2005). 12B. D. Gerardot, D. Brunner, P. A. Dalgarno, P. Ohberg, S. Seidl, M. Kroner, K. Karrai, N. G. Stoltz, P. M. Petroff, and R. J. Warburton, Nature 451, 441 (2008). 13M. D. Schroer, K. D. Petersson, M. Jung, and J. R. Petta, Phys. Lett. 7, 2051 (2007). 21P. C. Spruijtenburg, J. Ridderbos, F. Mulller, A. W. Leenstra, M. Brauns, A. A. I. Aarnink, W. G. van der Wiel, and F. A. Zwanenburg, arXiv:1304.2870 (unpublished).
1711.01459	2	1711	2019-12-01T15:44:20	Effects of noise on fidelity in spin-orbit qubit transformations	[ "cond-mat.mes-hall" ]	We analyse non-adiabatic non-Abelian holonomic transformations of spin-qubits confined to a linear time dependent harmonic trap with time dependent Rashba interaction. For this system exact results can be derived for spin rotation angle which also enables exact treatment of white gate-noise effects. We concentrate in particular on the reliability of cyclic transformations quantified by fidelity defined by the probability that the qubit after one full cycle remains in the ground-state energy manifold. The formalism allows exact analysis of spin transformations that optimise final fidelity. Various examples of time dependent fidelity probability distributions are presented and discussed.	cond-mat.mes-hall	cond-mat	December 3, 2019 3:15 WSPC/INSTRUCTION FILE UlcakarRamsak 9 1 0 2 c e D 1 ] l l a h - s e m . t a m - d n o c [ 2 v 9 5 4 1 0 . 1 1 7 1 : v i X r a International Journal of Modern Physics B c(cid:13) World Scientific Publishing Company Effects of noise on fidelity in spin-orbit qubit transformations Lara Ulcakar J. Stefan Institute, Ljubljana, Slovenia [email protected] Anton Ramsak Faculty of Mathematics and Physics, University of Ljubljana, Ljubljana, Slovenia J. Stefan Institute, Ljubljana, Slovenia [email protected] We analyse non-adiabatic non-Abelian holonomic transformations of spin-qubits confined to a linear time dependent harmonic trap with time dependent Rashba interaction. For this system exact results can be derived for spin rotation angle which also enables exact treatment of white gate-noise effects. We concentrate in particular on the reliability of cyclic transformations quantified by fidelity defined by the probability that the qubit after one full cycle remains in the ground-state energy manifold. The formalism allows exact analysis of spin transformations that optimise final fidelity. Various examples of time dependent fidelity probability distributions are presented and discussed. Keywords: Rashba interaction; holonomic qubit transformation; white noise; fidelity 1. Introduction Spintronics, as a new branch of electronics, is a quantum information technol- ogy promising better performance with smaller power consumption.1,2,3 The spin of electrons plays the central role4 and the main challenge is to manipulate the spin of a single electron precisely and locally. Employing magnetic fields, a natural way of spin rotation, usually cannot be applied locally in a small region so other mechanisms should be applied. A possible such solution is to use semiconductor heterostructures5,6 with spin-orbit interaction (SOI) and particularly strong Rashba interaction7,8 that can be tuned externally using voltage gates.9,10,11,12,13,14,15,16,17 Recently a simple scheme for the spin-qubit manipulation was proposed in which an electron is driven along a linear quantum wire with time dependent spin-orbit interaction, tuned by external time-dependent potential.18,19 One limitation of such linear systems is posed by fixed axis of spin rotation, but it can be eliminated in quantum ring structures, exhibiting a rich range of phenomena.20,21,22,23,24,25,26,27 For quantum ring structures consisting of a narrow ring with superimposed time dependent harmonic trap and controllable time dependent Rashba interaction exact solutions were presented most recently.28,29 1 December 3, 2019 3:15 WSPC/INSTRUCTION FILE UlcakarRamsak 2 Lara Ulcakar and Anton Ramsak In linear as well as in ring systems controlled by external gates there are several possible sources of noise which can not be avoided. In particular, noise can be in- duced due to fluctuating electric fields, caused by the piezoelectric phonons30,31,32,33 or due to phonon-mediated instabilities in molecular systems with phonon assisted potential barriers, which introduce noise in the confining potentials.34,35 For qubits realised as spin of electrons carried by surface acoustic waves the noise can be caused by the electron-electron interaction.36,37,38 Since exact solutions for qubit manipu- lation scheme considered here are possible, the analysis of environment effects can for some sources of noise be performed analytically.39 The paper is organised as follows. After the introduction is in Section 2 presented the model where also a brief overview of the exact solution together with the analysis of effects due to white noise is revealed. Section 3 is devoted to the fidelity of qubit transformations. The derivation of influences of noise on fidelity is presented in detail and explicit examples are given. Results are summarised in Section 4. 2. Model, exact solution and white noise We consider an electron in a quantum wire confined in a harmonic trap.18,19 The centre of such one-dimensional quantum dot, ξ(t), can be arbitrarily translated along the wire by means of time dependent external electric fields. Spin-orbit Rashba in- teraction couples the electron spin with orbital motion, resulting in the Hamiltonian m∗ω2 p2 2m∗ I + [x − ξ(t)]2I + α(t)p n·σ, 2 H(t) = (1) where m∗ is the electron effective mass, ω is the frequency of the harmonic trap, α(t) is the strength of spin-orbit interaction, possibly time dependent due to appropriate time dependent external electric fields. The spin rotation axis n is fixed and depends on the crystal structure of the quasi-one-dimensional material used and the direction of the applied electric field.42 σ and I are Pauli spin matrices and unity operator in spin space, respectively, and p is the momentum operator. Exact solution of the time dependent Schrodinger equation corresponding to the Hamiltonian equation (1) is given by19 Ψms(t)(cid:105) = e−i[θ(t)I+φ(t)n·σ/2)]AαXξψm(x)(cid:105)χs(cid:105), ac(t)ac(t)/ω2, e−im∗ac(t)xn·σ, θ(t) = ωmt + φα(t) + φξ(t) + m∗ Aα = e−i ac(t)pn·σ/ω2 Xξ = eim∗ (cid:82) t Here ψm(x) represents the m-th eigenstate of a harmonic oscillator with eigenenergy ωm = (m + 1/2)ω and χs(cid:105) is spinor of the electron in the eigenbasis of operator σz. 0 Lξ(t(cid:48))dt(cid:48) is the coordinate action integral, where Lξ(t) = The phase φξ(t) = − m∗ x2 c(t)/2 − m∗ω2[xc(t) − ξ(t)]2/2 is the Lagrange function of a driven harmonic oscillator and xc(t) is the solution to the equation of motion of a classical driven [x−xc(t)] xc(t)e−ixc(t)pI. (3) (4) (5) (2) December 3, 2019 3:15 WSPC/INSTRUCTION FILE UlcakarRamsak Effects of noise on fidelity in spin-orbit qubit transformations 3 oscillator (cid:82) t (6) 0 Lα(t(cid:48))dt(cid:48), with c(t)/(2ω2)− m∗[ac(t)− α(t)]2/2 + m∗α2(t)/2 being the Lagrange func- Another phase factor is the SOI action integral phase φα(t) = − Lα(t) = m∗ a2 tion of another driven oscillator, satisfying ac(t) + ω2ac(t) = ω2α(t). xc(t) + ω2xc(t) = ω2ξ(t). In this paper we consider particularly interesting cyclic transformations with periodic drivings ξ(T ) = ξ(0) and α(T ) = α(0) with zero values and time derivatives of responses xc and ac at times t = 0 and t = T . The spin-qubit is for such drivings rotated around n by the angle φ = −2m∗(cid:82) T We assume noise in the driving function ξ(t) = ξ0(t) + δξ(t) consisting of ideal driving part without noise ξ0(t) with superimposed stochastic part with vanishing mean (cid:104)δξ(t)(cid:105) = 0. We consider the Ornstein-Uhlenbeck coloured noise40,41 char- et(cid:48)−t(cid:48)(cid:48)/τξ , with noise acterized by the autocorrelation function (cid:104)δξ(t(cid:48))δξ(t(cid:48)(cid:48))(cid:105) = ξ and correlation time τξ. A general solution of equation (6) xc(t) with 0 ac(t(cid:48))ξ(t(cid:48))dt(cid:48).19 σ2 ξ 2τξ intensity σ2 xc(0) = ξ0(0) and xc(0) = 0 is given by xc(t) = ξ0(0) + ω )]ξ(t(cid:48) )dt(cid:48), (7) (cid:90) t sin[ω(t − t(cid:48) )dt(cid:48)(cid:90) t+∆t 0 0 which due to the noise term δξ is normally distributed with the variance evaluated as equal-times autocorrelation function, )]δξ(t(cid:48) )]δξ(t(cid:48)(cid:48) )dt(cid:48)(cid:48) (cid:90) t (8) x(t) = ω2 lim σ2 ∆t→0(cid:104) 0 sin[ω(t − t(cid:48) sin[ω(t − t(cid:48)(cid:48) (cid:105). For the Ornstein-Uhlenbeck noise considered here the integrals can be evaluated exactly. Nevertheless, here we consider only the white noise limit where τξ → 0 and (cid:104)δξ(t(cid:48))δξ(t(cid:48)(cid:48))(cid:105) = σ2 ξ δ(t(cid:48) − t(cid:48)(cid:48)) leading to the variances x(t) = and σ2 ξ (2ωt − sin 2ωt) 1 4 σ2 x(t) = ωσ2 1 4 ω3σ2 ξ [2ωt + sin(2ωt)] , (9) corresponding to xc(t) and xc(t), respectively. Additionally to the coordinate noise is also normally distributed noise in SOI driving function α(t) = α0(t)+δα(t), where α0(t) is ideal noiseless driving. SOI noise δα(t) is similar to the previous case of spatial driving and is again of the Ornstein- Uhlenbeck type of autocorrelation function (cid:104)δα(t(cid:48))δα(t(cid:48)(cid:48))(cid:105) with noise intensity σ2 and correlation time in the white noise limit τα → 0, leading to the time-dependent variances σ2 x(t) for ac(t) and ac(t), respectively. a(t) = (σα/σξ)2σ2 a(t) = (σα/σξ)2σ2 x(t) and σ2 α 3. Fidelity of noisy qubit transformations As an example of effects of noise to spin-qubit transformations we consider driving corresponding to the class of circular paths in two dimensional coordinate-SOI space Cad ∼ α0[ξ0], ξ0(t) = ξ0 cos(ωt/n) and α0(t) = α0 sin(ωt/n), (10) December 3, 2019 3:15 WSPC/INSTRUCTION FILE UlcakarRamsak 4 Lara Ulcakar and Anton Ramsak φad = −2m∗(cid:90) T φ0 = −2m∗(cid:73) 0 C n2 − 1 = −2m∗(cid:73) where n ≥ 2 is integer, and the period of the transformation is T = 2πn/ω. Periodic responses represent contours C ∼ a0 a0 c(t) = α0 c[ξ0], where n [n sin (ωt/n) − sin(ωt)] (11) , with the phases given by the area in the coordinate-SOI plane, α0(t(cid:48) )ξ(t(cid:48) )dt(cid:48) α0[ξ0]dξ = −2πm∗ξ0α0, Cad a0 c[ξ0]dξ = n2 n2 − 1 φad. (12) (13) (cid:18) ωσξσα (cid:19)2 The adiabatic angle φad corresponds to the one when circular driving is of type n → ∞. Transformation angle φ = φ0 + δφ is due to the noise distributed normally around the mean φ0, with the variance after one cycle given by39 σ2 φ,n φ2 ad = n(1 + n2) π(n2 − 1)2 ωσ2 ξ ξ2 0 + 2n3 π(n2 − 1)2 ωσ2 α α2 0 + n2 . (14) ξ0α0 c(0), ξ0(0)] and final noiseless values [a0 In figure 1(a) are shown spin orbit responses as a function of time and in figure 1(b) is shown the contour C for the case of circular driving equation (10) with n = 6. In both panels the dashed black lines denote noiseless spin orbit driving α0(t) and the red line noiseless spin orbit response a0 c(t). The focus is on the set of 10 spin orbit responses ac(t) to 10 different realisations of white noise in α(t). σ2 a(t) manifests as a spread of these curves around the ideal noiseless red line. Bullets correspond to c(T ), ξ0(T )] of noiseless response and initial [a0 show that final values of ac(T ) deviate from the desired ones. The noisy response is not periodic, resulting in open loop in parameter space unlike the case of noiseless C and noiseless adiabatic driving Cad. Consequently the angle of spin rotation cannot be expressed as an area enclosed by the contour as in equation (13) and in figure 1(b) pink shaded. It should be noted that in general the total angle of spin rotation φ is less prone to noise because the noisy curves oscillate around the ideal value and so contributions to final error partially cancel out.39 This analysis of spin-rotation angle demonstrated that due to gate noise in the driving functions, spin transformations are not completely faithful. For non-adiabatic qubit manipulations the electron state is determined by the time- dependent Hamiltonian during the evolution and is in general a superposition of excited states, ultimately becoming the ground state when the transformation is complete. Therefore in addition to correct transformation of the spin direction, one has also to take care that the electron state has not left the starting energy man- ifold at the final time. As shown in Refs.18,19,28 such motions in parametric space can easily be performed if the driving functions are appropriately chosen. Here an important question is relevant: how well does the final state of the electron relax to the desired final state energy manifold after the transformation if the driving function is not ideal as in the presence of noise? December 3, 2019 3:15 WSPC/INSTRUCTION FILE UlcakarRamsak Effects of noise on fidelity in spin-orbit qubit transformations 5 Fig. 1. Responses to circular driving with n = 6 are shown. In (a) are as functions of time shown noiseless driving α0(t) (dashed line), noiseless response a0 c (t) (red) and 10 responses ac(t) ω). Bullets (black lines) to different realisations of white noise δα(t) with intensity σα = α0/(20 √ c (T ), ξ0(T )]. In (b) are shown the denote noiseless starting [a0 ω). The same quantities as in (a) but as a function of coordinate driving ξ(t) with σξ = ξ0/(20 c [ξ0] form closed loops, Cad and C, respectively. Note that φ is noiseless contours α0[ξ0] and a0 proportional to pink shaded area enclosed by C. In all calculations α0 = ξ0 = 1 (in dimensionless units) was used. c (0), ξ0(0)] and ending positions [a0 √ In order to demonstrate how to answer this question in general we consider the (t)(cid:105), equation (2), which is at t = 0 in the ground state of qubit wave function Ψ0 1 the harmonic quantum dot (with m = 0) and spin 1 2 . We observe its relaxation to the ground state manifold that is spanned by two basis states18 of time dependent Hamiltonian equation (1) at time t, 2 [x−ξ(t)]α(t)n·σψ0[x − ξ(t)](cid:105)χs(cid:105). s (cid:101)Ψ0s(cid:105)(cid:104)(cid:101)Ψ0s is the projector onto the ground As the appropriate measure of the relaxation accuracy we define fidelity F = (cid:104)Ψ0 1 state manifold. We choose n perpendicular to the z-axis and a straightforward (t)(cid:105) with the basis states at derivation leads to the expression for overlaps of Ψ0 1 time t, (t)P0Ψ0 1 (15) 2 2 2 (cid:101)Ψ0s(cid:105) = e−im∗ (t)(cid:105), where P0 = (cid:80) (cid:104)(cid:101)Ψ0± 1 (t)Ψ0 1 2 2 (t)(cid:105) = 1 2 [e− 1 2 E+(t) ± e− 1 2 E−(t)], where E±(t) = m∗ 2ω {[ω(xc(t) − ξ(t)) ± ac(t)/ω]2 + [ xc(t) ∓ (ac(t) − α(t))]2} mations with spin-fidelities Fs = (cid:104)(cid:101)Ψ0sΨ0 1 resembles classical energy with additional terms for spin-orbit coupling and is equal to the classical energy if the spin-orbit driving is constant.18 Ideal qubit transfor- are achieved by applying ideal drivings, where the energies E± vanish at final time t = T , i.e., when xc = ξ, ac = α, xc = 0, and ac = 0. 2(cid:105)2 = δs 1 2 (16) (17) (a)(b)CCad•••December 3, 2019 3:15 WSPC/INSTRUCTION FILE UlcakarRamsak 6 Lara Ulcakar and Anton Ramsak The fidelity at arbitrary time t is obtained by summation over final spin states, [e−E+(t) + e−E−(t)]. 1 2 (18) (cid:88) s Fs(t) = F (t) = The presence of noise in spin-orbit and spatial driving terms makes fidelity a random quantity, F (t) = F 0(t)+δF (t), where F 0(t) represents the result of noiseless driving and δF (t) is the deviation from this value. Fidelity is therefore characterized by some probability density function dP (F ) dF . It can be calculated from the probability density for variables E± which are functions of independent random variables and normally distributed. The probability density functions for E± can at time t be calculated using the formula dP±(E) (cid:90)(cid:90)(cid:90)(cid:90) (19) (cid:12)(cid:12)(cid:12)t dE δ[E − E±(xc, xc, ac, ac)] dP x( xc) dPa(ac) dP a( ac) dPx(xc) dxc d xc dac d ac dxcd xcdacd ac. The result is obtained by first calculating the characteristic functions, (cid:113) 2σ−1 1 σ−1 1 − ik)(2σ−2 2 (2σ−2 2 − ik) dP±(E) −∞ dE eikEdE = (cid:18) 2m∗ (cid:18) 2m∗ ω (cid:19) (cid:19) ω σ2 1(t) = σ2 2(t) = [ω2σ2 x(t) + σ2 a(t)/ω2], [σ2 x(t) + σ2 a(t)]. , (20) (21) (22) p±(k) = with = × (cid:90) ∞ (cid:12)(cid:12)(cid:12)t Note the equality p+(k) = p−(k) which after the inverse Fourier transform yields equal functional forms for E+ and E−, dP±(E±) dE± = 2σ−1 1 σ−1 2 I0[(σ−2 1 − σ−2 2 )E±]e−(σ −2 1 +σ −2 2 )E±, (23) where I0(z) is the modified Bessel function of the first kind. x(t) = σ2 x(t) = σ2 a(t)/ω4 and σ2 dF exactly when σ2 Since the fidelity is a sum of two dependent random variables, its probability distribution is calculated from the joint probability distribution function for those two variables, which in general cannot be evaluated analytically. However, one can examine dP a(t), which is satisfied for t = T if the coordinate and the SOI driving noise intensities are equal, i.e., σα = ωσξ. In this case E+ and E− become independent random variables and dP dF can be calculated as the convolution of probability distributions for e−E+ and e−E− . At t = T the exact result for F ≥ 1 , σ−2 F , σ−2 , σ−2 where B(x, a, b) is the incomplete beta function and σ−2 F = σ−2 the probability distribution is dP F )](2F )2σ 1 + σ−2 2 . For F < 1 2 −2 F −1, where B(a, b) is F ) − B(1 − (cid:12)(cid:12)(cid:12)t=T dF = 2σ−4 F B(σ−2 F , σ−2 2 is given by F , σ−2 F )(2F )2σ = 2σ−4 −2 F −1, F [B( dP (F ) 1 2F 1 2F (24) dF December 3, 2019 3:15 WSPC/INSTRUCTION FILE UlcakarRamsak Effects of noise on fidelity in spin-orbit qubit transformations 7 √ In panel (a) the noiseless fidelity F 0(t) (red line) and 10 fidelities F (t) (black lines) for Fig. 2. different realizations of white noise with intensities σξ/ξ0 = σα/α0 = 1/(20 ω) are shown as functions of time when driving the system circularly with n = 2 and α0 = ξ0 = 1 (in dimensionless units). Orange, grey and green shaded regions at times t/T = 0.5, 0.75, 1, correspondingly, show the spread of noisy fidelities around the exact value. Noiseless contours in parameter space [ac, ξ] are for n = 2 (red), 8 (blue, dashed) and n → ∞ (black, dashed) presented in panel (b). Panel (c) shows the same as panel (a) for circular driving with n = 8 and the noiseless fidelity is denoted with dashed blue line. Bullets mark initial and final values of noiseless fidelity in (a) and (c) and contour in parameter space in (b). In panel (d) probability density distributions of fidelity for n = 2 at times t/T = 0.5 (orange), t/T = 0.75 (green) and t = T (grey) are shown. Colour codes coincide with the area of fidelity spreading shown in (a). Distributions were generated from N = 107 samples. the beta function. In practice where noise intensities are small the most relevant regime is σF → 0 for which the probability distribution equation (24) simplifies to dP F . Due to similar dependence of Fs and F on E± it is easy to derive analytical results also for spin-fidelity probability distributions dPs (not dFs shown here). dF ∝ (1 − F )F 2σ −2 In figure 2(a) different realizations of noisy fidelity (black lines) are compared to the noiseless one (red) for n = 2. One can observe that noisy fidelity starts to devi- ate from noiseless one for t/T (cid:38) 0.1, reaches maximum deviation at t/T ∼ 0.5 and then deviations are again lowered when approaching t → T . The same quantities are presented in figure 2(c) for circular driving with n = 8 where noiseless curve is n=2∞8•(b)(d)n=2December 3, 2019 3:15 WSPC/INSTRUCTION FILE UlcakarRamsak 8 Lara Ulcakar and Anton Ramsak denoted with blue colour. Figure 2(b) shows noiseless curves ac[ξ0] in parametric space during the transformation with n = 2 (red), n = 8 (blue, dashed) and n → ∞ (black, dashed), the latter corresponding to the adiabatic limit. Bullets denote ini- tial and final values of ac(t) and ξ(t). Note that the motion is periodic with period T and that ac(0) = ac(T ) = α(0) = α(T ) as is manifested also in noiseless fidelity being equal to 1 at t = 0 and t = T , as a demonstration that the system returns to the ground state manifold with probability 1. This can be seen from positions of bullets in figures 2(a) and 2(c). Figure 2(d) shows the probability density distribu- tion of fidelity at times t/T = 0.5 (orange), t/T = 0.75 (green) and t = T (black). It should be mentioned that the distribution for t = T is given also by exact formula, equation (24). The colour code of distributions corresponds to the code of the shad- ing of fidelity spreading around the noiseless value in figure 2(a). Distributions are centred around noiseless values and their variances are proportional to spreadings observed in figure 2(a), the distribution at t/T = 0.5 having the largest variance which is lower at t/T = 0.75 and even lower at t = T . 4. Summary We presented an analysis of spin-qubit non-adiabatic manipulation of an electron traped in a moving linear harmonic trap and in the presence of time dependent Rashba interaction. One of the main challenges here is a precise tuning of driving fields since the electron starting from the ground state should after performing one cycle with time-dependent Hamiltonian return to the ground state, although during the cycle the state of the electron is a superposition of excited eigenstates of the moving trap. The problem is even more subtle because there will always be present some noise in driving functions, which means that spin-qubit transformation will always deviate from the ideal one. Since for the model considered here exact solutions are available for a broad class of drivings, we concentrated also to the exact analysis of the influence of small deviations from ideal qubit manipulation. In particular, we focused to an explicit example and demonstrated how one can analyse the effects of a general noise to the transformation angle and we showed the results for the Ornstein-Uhlenbeck type of noise. An example, considered in detail, is the case of circular driving in the space of parameters for which exact analytical formulae are given and analysed for white noise. In view of the fact that for non-adiabatic regimes a non-trivial point is the ability of the system to return to the ground state after an arbitrary time-dependent driving, our analysis was focused to the fidelity -- the overlap of the actual wave function with the desired ideal. For white noise explicit formulae are derived for symmetric noise intensities in position and spin-orbit driving functions. A detailed derivation and analysis of fidelity is presented. Additionally, analytical results are illustrated by special cases of driving together with numerically generated noisy drivings and the corresponding responses. December 3, 2019 3:15 WSPC/INSTRUCTION FILE UlcakarRamsak Effects of noise on fidelity in spin-orbit qubit transformations 9 Acknowledgements The authors acknowledge support from the Slovenian Research Agency under con- tract no. P1-0044. References 1. S. A. Wolf, D. D. Awschalom, R. A. Buhrman, J. M. Daughton, S. von Moln´ar, M. L. Roukes, A. Y. Chtchelkanova and D. M. Treger, Science 294, 1488 (2001). 2. I. Zuti´c and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004). 3. E. I. Rashba, Future Trends in Microelectronics (John Wiley & Sons, Inc., Hoboken, NJ, USA, 2007). 4. D. D. Awschalom, L. C. Bassett, A. S. Dzurak, E. L. Hu and J. R. Petta, Science 339, 1174 (2013). 5. R. Winkler, Spin -- Orbit Coupling Effects in Two-Dimensional Electron and Hole Sys- tems, Springer Tracts in Modern Physics, Vol. 191 (Springer Berlin Heidelberg, Berlin, Heidelberg, 2003). 6. H.-A. Engel, E. I. Rashba and B. I. Halperin, Handbook of Magnetism and Advanced Magnetic Materials (John Wiley & Sons, Ltd, Chichester, UK,2007). 7. G. Dresselhaus, Phys. Rev. 100, 580 (1955). 8. E. I. Rashba, Sov. Phys. Solid State 2, 1109 (1960). 9. J. Nitta, T. Akazaki, H. Takayanagi and T. Enoki, Phys. Rev. Lett. 78, 1335 (1997). 10. T. Scapers, G. Engels, J. Lange, T. Klocke, M. Hollfelder and H. uth, J. Appl. Phys. 83, 4324 (1998). 11. J. Nitta, F. E. Meijer and H. Takayanagi, Appl. Phys. Lett. 75, 695 (1999). 12. J. Schliemann, J. C. Egues and D. Loss, Phys. Rev. Lett. 90, 146801 (2003). 13. J. Wunderlich, B.-G. Park, A. C. Irvine, L. P. Zarbo, E. Rozkotov´a, P. Nemec, V. Nov´ak, J. Sinova and T. Jungwirth, Science 330, 1801 (2010). 14. A. G´omez-Le´on and G. Platero, Phys. Rev. B 86, 115318 (2012). 15. J. Pawlowski, P. Szumniak and S. Bednarek, Phys. Rev. B 93, 045309 (2016). 16. J. Pawlowski, P. Szumniak and S. Bednarek, Phys. Rev. B 94, 155407 (2016). 17. Jingtao Fan, Yuansen Chen, Gang Chen et al., Scientific Reports 6, 38851 (2016). 18. T. Cadez, J. H. Jefferson and A. Ramsak, New J. Phys. 15, 013029 (2013). 19. T. Cadez, J. H. Jefferson and A. Ramsak, Phys. Rev. Lett. 112, 150402 (2014). 20. M. Buttiker, Y. Imry and R. Landauer, Phys. Lett. A 96, 365 (1983). 21. A. Fuhrer, S. Luscher, T. Ihn, T. Heinzel, K. Ensslin, W. Wegscheider and M. Bichler, Nature 413, 822 (2001). 22. A. Aronov and Y. Lyanda-Geller, Phys. Rev. Lett. 70, 343 (1993). 23. T.-Z. Qian and Z.-B. Su, Phys. Rev. Lett. 72, 2311 (1994). 24. A. Mal'shukov, V. Shlyapin and K. Chao, Phys. Rev. B 60, R2161 (1999). 25. K. Richter, Physics 5, 22 (2012). 26. F. Nagasawa, D. Frustaglia, H. Saarikoski, K. Richter, and J. Nitta, Nature Commu- nications 4, 2526 (2013). 27. H. Saarikoski, J. E. V´azquez-Lozano, J. P. Baltan´as, F. Nagasawa, J. Nitta, and D. Frustaglia, Phys. Rev. B 91, 241406(R) (2015). 28. A. Kregar, J. H. Jefferson and A. Ramsak, Phys. Rev. B 93, 075432 (2016). 29. A. Kregar and A. Ramsak, Int. J. Mod. Phys. B 30, 1642016 (2016). 30. P. San-Jose, G. Zarand, A. Shnirman and G. Schon, Phys. Rev. Lett. 97, 076803 (2006). 31. P. San-Jose, B. Scharfenberger, G. Schon, A. Shnirman and G. Zarand, Phys. Rev. B 77, 045305 (2008). December 3, 2019 3:15 WSPC/INSTRUCTION FILE UlcakarRamsak 10 Lara Ulcakar and Anton Ramsak 32. P. Huang and X. Hu, Phys. Rev. B 88, 075301 (2013). 33. C. Echeverr´ıa-Arrondo and E. Y. Sherman, Phys. Rev. B 87, 081410(R) (2013). 34. J. Mravlje, A. Ramsak and T. Rejec, Phys. Rev. B 74, 205320 (2006). 35. J. Mravlje and A. Ramsak, Phys. Rev. B 78, 235416 (2008). 36. T. Rejec, A. Ramsak and J. H. Jefferson, J. Phys., Condens. Matter 12, L233 (2000). 37. J. H. Jefferson, A. Ramsak and T. Rejec, Europhys. Lett. 74, 764 (2006). 38. G. Giavaras, J. H. Jefferson, A. Ramsak, T. P. Spiller and C. Lambert, Phys. Rev. B 74, 195341 (2006). 39. L. Ulcakar and A. Ramsak, New J. Phys. 19, 093015 (2017). 40. M. C. Wang and G. E. Uhlenbeck, Rev. Mod. Phys. 17, 323 (1945). 41. J. Masoliver, Phys. Rev. A 45, 706 (1992). 42. S. Nadj-Perge, et al., Phys. Rev. Lett. 108, 166801 (2012).
1809.02985	1	1809	2018-09-09T14:54:06	First principles investigation of nanopore sequencing using variable voltage bias on graphene-based nanoribbons	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci", "cond-mat.str-el" ]	In this study, we examine the mechanism of nanopore-based DNA sequencing using a voltage bias across a graphene nanoribbon. Using density functional theory and a non-equilibrium Green's function approach, we determine the transmission spectra and current profile for adenine, guanine, cytosine, thymine, and uracil as a function of bias voltage in an energy minimized configuration. Utilizing the transmission current, we provide a general methodology for the development of a three nanopore graphene-based device that can be used to distinguish between the various nucleobases for DNA/RNA sequencing. From our analysis, we deduce that it is possible to use different transverse currents across a multi-nanopore device to differentiate between nucleobases using various voltages of 0.5, 1.3, and 1.6 V. Overall, our goal is to improve nanopore design to further DNA/RNA nucleobase sequencing and biomolecule identification techniques.	cond-mat.mes-hall	cond-mat	First principles investigation of nanopore sequencing using variable voltage bias on graphene-based nanoribbons Hannah L. McFarland,1 Towfiq Ahmed,2 Jian-Xin Zhu,2, 3 Alexander V. Balatsky,4, 5 and Jason T. Haraldsen6 1Department of Biology, James Madison University, Harrisonburg, VA 22802, USA 2Theoretical Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545 3Center for Integrated Nanotechnologies, Los Alamos National Laboratory, Los Alamos, New Mexico 87545 4Institute of Material Science, Los Alamos National Laboratory, Los Alamos, New Mexico 87545 5Nordic Institute for Theoretical Physics, KTH Royal Institute of Technology and Stockholm University, 6Department of Physics and Astronomy, James Madison University, Harrisonburg, VA 22802, USA Roslagstullsbacken 23, 106 91 Stockholm, Sweden (Dated: September 11, 2018) In this study, we examine the mechanism of nanopore-based DNA sequencing using a voltage bias across a graphene nanoribbon. Using density function theory and a non-equilibrium Green's function approach, we determine the transmission spectra and current profile for adenine, guanine, cytosine, thymine, and uracil as a function of bias voltage in an energy minimized configuration. Utilizing the transmission current, we provide a general methodology for the development of a three nanopore graphene-based device that can be used to distinguish between the various nucleobases for DNA/RNA sequencing. From our analysis, we deduce that it is possible to use different transverse currents across a multi-nanopore device to differentiate between nucleobases using various voltages of 0.5, 1.3, and 1.6 V. Overall, our goal is to improve nanopore design to further DNA/RNA nucleobase sequencing and biomolecule identification techniques. I. INTRODUCTION the individual bases. Advances in medicine and the understanding of biomolecule interactions is on the forefront of DNA (de- oxyribonucleic acid) sequencing research1,2. Errors or mutations in the nucleobase coding sequence have been shown to lead to the deregulation of gene products in- cluding proteins and enzymes3. These coding errors can be caused by various environmental factors like radiation or chemical exposure and even viruses, which can lead to many diseases like cancer4. Therefore, there is a unique need to identify variations in genes with the resolution of individual nucleobase sequences, which can help lead to advances in medicines and therapies for a wide vari- ety of genetic-related diseases as well as cancer5. With the human genome project working towards the ability to sequence hundreds of genomes more efficiently and cost effectively6,7, the ability to sequence DNA or RNA (ri- bonucleic acid) down to the individual nucleobase is in great demand. Technologies that can help sequence and compare multiple genomes more accurately could lead to the possibility of diagnosing and developing treatments for patients within a personalized medicine approach. There are numerous approaches for nucleobase sequencing8 -- 10, where the most common technique is the Sanger sequencing method that uses a chain termination approach11. More recently, there have been a number of next generation sequencing methods that have gained attention12,13, but many of these methods are either in- efficient and/or very expensive. However, one major av- enue for individual base sequencing is the use of nanopore technology, where a single DNA/RNA strand (ssDNA or ssRNA) is drawn through a nanoscopic pore in a con- ductive material and measurements of variations in the optical or electronic properties can be used to identify Nanopore-based technology has the potential to be an efficient method for nucleobase sequencing14 -- 19, as well as an identifier of other biomolecules for various biologi- cal sensors. There have a number of realizations of this technology using gold and other material substrates20. However, the thickness of the nanopore is critical for the identification of individual nucleobases due to resulting noise and resolution problems21 -- 23, which has led to a number of publications suggesting the use of 2D materi- als for nanopores because the atomically-thin materials provide the proper resolution needed for individual base identification27,28. However, while nanopores can yield a superior resolution, there are many challenges regarding strength, durability, and overall systematic noise. There- fore, the design and material makeup of the device must ∆V = V 1 ∆V = V 2 ∆V = V 3 FIG. 1: Schematic representation of nanopore sequencing de- vice consisting of three graphene nanoribbons. address these challenges. Recently, graphene has been suggested as a possible 2D nanopore material due to its increased electron mobility and its high tensile strength27,28. Graphene consists of an atomically thin layer of carbon atoms arranged in a honeycomb lattice (Fig. 1)24, where the lattice struc- ture and sp2 hybridized 2D graphene sheet with dan- gling p-orbitals contributes to its large tensile strength and allows for conductivity of the valence electrons25,26. Therefore, the use of graphene nanopores will provide the electronic conduction and material strength needed while also having the necessary atomic resolution. While this addresses the durability and strength issues, it has been shown that the device setup can greatly reduce the general noise to signal ratio through the use of multiple sequential nanopore ribbons (shown in Fig. 1), where the use of three separate nanopores help to reduce and distinguish random and systematic noise parameters27. In this study, we examine the viability of a multi- nanoribbon graphene nanopore device as a nucleobase sequencer. Using density functional theory and a non- equilibrium Green's function (NEGF) method, we simu- late transmission spectra and calculate the ballistic cur- rents and tunneling conductance as functions of voltage bias for adenine (A), thymine (T), cytosine (C), guanine (G), and uracil (U) translocating through the graphene nanopore. Through the analysis of I-V curves, we show that by applying various bias voltages across a graphene ribbon for the general, energy-minimized position of the translocated nucleobase, it is possible to distinguish in- dividual bases using the resulting current. Furthermore, we provide a general mechanism that can be used for identification of the nucleobases with specific voltage bi- ases. II. COMPUTATIONAL METHODOLOGY For our simulation, we construct a graphene nanorib- bon structure with a single 0.5 nm nanopore in the cen- ter. Hydrogen atoms were used to passivate the dangling bonds of the dangling orbitals of the carbon atoms at the edge of the nanoribbon and nanopore, which pre- vents the nucleobases from bonding to the nanoribbon and provides stability translating molecule. In addition, we place electrodes at each end of the ribbon that will simulate with a finite voltage bias (shown in Fig. 1). To help isolate the individual base responses, the sugar-phosphate backbone, typically found in ssDNA, is ignored, and only the individual nucleobases are placed in the nanopore. The backbone produces noise that can interrupt the current response. However, since the back- bone will produce a systematic noise, previous studies have shown that analysis of cross-correlations between multiple nanoribbons can be used to reduce and possibly eliminate the systematic noise, as well as thermal and fluid fluctuations due to general flow of the DNA through the nanopore27. Calculations were performed on a single 2 nanopore graphene nanoribbon and one individual nu- cleobase in a translocating position of 60◦ determined through molecular dynamics simulations32. Using Atomistix Toolkit (ATK) by Quantumwise R(cid:13), we performed density functional calculations using a gen- eralized gradient approach (GGA)29 -- 31. From a NEGF method, we determine the transmission spectra for each base and various voltage biases. The transmission spec- tra is then determined from the transmission coefficient, T (ω, V ) = Xk Xnl tnl(ω, k, V )t† ln(ω, k, V ), (1) where tnℓ(k) represents the transmission amplitude from ψn(k) in the left (L) electrode to ψℓ(k) in the right (R) electrode. This is determined through a calculation of the Kohn-Sham Hamiltonian and the density matrix, which is given as ¯D = 1 π Z µL −∞ ¯G(ω)Im[ ¯ΣL] ¯G(ω)†dω ¯G(ω)Im[ ¯ΣR] ¯G(ω)†dω. (2) + 1 π Z µR −∞ Here, ¯G(ω) is the retarded Green's function and ¯Σ is the self energy for the left and right electrodes33. The chem- ical potentials of the left electrode, µL = EL F − eVL and the right electrode, µR = EL F − eVR are defined relative to the Fermi level of the left electrode EL F . and related to the applied bias through µR −µL = eV and V = VL −VR. Using various simulated voltages, we determined the current versus voltage for each nucleobase in the nanopore. The current is independent of how the left and right voltages are applied and only depends on volt- age bias or difference. III. RESULTS AND DISCUSSION Figure 2 shows the calculated transmission coefficients as a function of energy for the graphene nanoribbon with a single nanopore for V = 0.0, 1.0, and 2.0 V. The dif- ferent panels show the various base configurations in the nanopore (empty, adenine, cytosine, guanine, thymine, and uracil). In this calculation of the transmission spec- trum, the background contribution from the large phos- phate backbone is ignored, since the background noise from the heavy and rigid backbone structure can be iden- tified and subtracted from the general spectra. Through an integration of transmission coefficient, the ballistic current for nucleobase can be calculated as a function of voltage I(V ) = e h Z ∞ −∞ T (ω, V )(cid:2)nF (ω−µL)−nF (ω−µR)(cid:3)dω (3) 3 FIG. 2: Transmission coefficient as a function of energy for the graphene nanoribbon with an empty nanopore (a), adenine (b), cytosine (c), guanine (d), thymine (e), and uracil (f) in the nanopore. where nF is the Fermi function and voltage is defined as the difference between the left and right electrode chemical potentials, µR − µL = eV . In Fig. 3(a), we show the calculated current as a function of voltage for the various nucleobases (adenine, thymine, guanine, cytosine, and uracil), where the empty graphene nanopore is presented in the inset. From this data, there are distinct voltage pathways for the differen- tiation of all bases. The sudden rise in current at 1.3 V is due to the presence of the nanopore in the simulation and can be traced back to the graphene itself. Since the simulation assumes a nanopore on the size order of the nanoribbon itself, there is a critical voltage for which the empty graphene nanopore will not produce a current due to a structurally induce energy gap. Therefore, once 1.0 V is achieved, electrons can overcome this energy barrier and produce a sizable current. If no nanopore existed, then the graphene nanoribbon would have its normal con- ductivity. Therefore, the presence of a nucleobase allows for current to be drawn through the nanoribbon at volt- age differences lower than 1.0 V. From the current, the tunneling conductance (dI/dV) 4 FIG. 3: Current as a function of voltage for all nucle- obases (adenine, guanine, cytosine, thymine, and uracil) in a graphene nanopore. The inset shows the current as a func- tion of voltage for a lone graphene nanopore ribbon. The critical voltage in the graphene is due to the presence of the nanopore and not an intrinsic property of graphene. TABLE I: General response for each DNA and RNA nucle- obase in a three nanopore setup. Nanobase Adenine Guanine Cytosine Thymine (DNA) Uracil (RNA) V1 (0.5 V) V2 (1.3 V) V3 (1.6 V) low low high high high high low high high high low low low high high (shown in Fig. 4) can be determined through a differenti- ation of the I-V curve. Therefore, there is the possibility that topological probes can be used to examine the in- dividual bases as well. This may be useful for scanning tunneling microscopy. From Fig. 3(a), there are specific voltages that produce distinct current variations. Therefore, in order to pro- duce a nanopore device that can distinguish individual nucleobases, the use of multiple nanoribbons is needed. FIG. 4: The tunneling conductance as a function of the volt- age for the nanopore and various bases translocating through the nanopore. The transmission spectrum is shown for volt- age biases of 0.0 V, 1.0 V, and 2.0 V. This is not a problem since multiple nanoribbons are needed for the appropriate noise reduction. Figures 3(b)- (d) zoom in around the characteristic voltages (V1, V2, V3 as illustrated in Fig. 1) for nucleobase differentiation are 0.5, 1.3, and 1.6 V, respectively. Through a com- parison to background, the utilization of these voltages allows us to distinguish each base by evaluating a high or low signal or current response. This is shown in Table I for all nucleobases. Figure 5 illustrates the methodology needed to charac- terize the bases. The first nanoribbon will provide a small voltage bias (V1 = 0.5 V), and will be able to determine pairs of bases by have a low (A or G) or high (C or T) current compared to the normalized baseline. The nucle- obase will then pass through a second nanoribbon with a moderate voltage bias (V2 = 1.3 V), which can be used +#,-.# !"#!$ (#)#+#,-.# !3#$#"&4!## %&'$ (# !"#!$ /1#(1#01#)#+#,-.# !"#$#%&'!# %&'$ /# !"#!$ /#)*#0# !2#$#"&3!# %&'$ 0# FIG. 5: Illustrates the differentiation pathway for the bases translocating through three nanopores for DNA bases (a) and RNA bases (b). This reveals the distinct possibility for the use of graphene nanopores as a sequencing device. to identify either G (low) or A (high) assuming a low first voltage. The third nanoribbon at a higher voltage bias (V3 = 1.6 V) will be used to determine C (low) and T (or U) (high). Once the nucleobase has translocated through all three nanopores, the characteristic current sequence will allow for the identification of the individual base. IV. CONCLUSIONS Graphene has been shown to be a potential material for nanopore-based sequencing, due to its atomic thick- 5 ness and relative strength. Using density functional cal- culations, we find that the five nucleobases, including uracil, can be distinguished through the use of multiple nanoribbons using variable voltage biases. From the sim- ulated transmission spectra, we calculate the I-V curves for these nucleobases. By examining specific differences in the calculated current, the precise nucleobase that is translocating through the nanopore can be determined. We focus on voltages of 0.5, 1.3, and 1.6 V as a proof of principle for a specific nanopore sequencing device. Future work includes performing conductance calcu- lations for specific voltages for a better microscopic un- derstanding as well as looking at surface plasmon reso- nances. In addition, increasing the size of the calculation, such as a large nanopore or a full DNA strands calcu- lation. Further investigations include a time-dependent translocation through the nanopore that includes ther- mal fluctuations. Here, we focused on graphene as possi- ble 2D materials. However, there should also be a push forward with other 2D materials for comparison. Acknowledgement H.L.M. and J.T.H. thank the support of James Madi- son University and useful discussions with J.-H. Kim. The work at Los Alamos National Laboratory was carried out under the auspice of the U.S. DOE and NNSA un- der Contract No. DEAC52-06NA25396 and supported by U.S. DOE Basic Energy Sciences Office (T.A. and A.V.B.). This work was also, in part, supported by the Center for Integrated Nanotechnologies, a U.S. DOE Of- fice of Basic Energy Sciences user facility (J.-X.Z). 1 Jones, P. A.; Baylin, S. B. Nat. Rev. Genet 2002, 3, 415- 428. 2 Ehrlich, M. Oncogene 2002, 21, 5400-5413. 3 Griffiths, A.J.F.; Gelbart, W.M.; Miller, J.H.; Lewontin, R.C. The Molecular Basis of Mutation. In: Modern Ge- netic Analysis; W.H. Freeman and Company, New York 1990. 4 Grivennikov, S. I.; Greten, F. R.; Karin, M. Cell 2010,140, 883899. 5 Smithies, O.; Maeda, N. Proc Natl Acad Sci 1995, 92, 52665272. 6 Collins, F. S.; Morgan, M.; Patrinos, A. Science 2003, 300, 286290. 7 Adams, M. D. et al. Science 1991, 252, 16511656. 8 Maxam, A. M.; Gilbert, W. Proc Natl Acad Sci 1977, 74, 14 Kilina, S.; Tretiak, S.; Yarotski, D. A.; Zhu, J.-X.; Modine, N.; Taylor, A.; Balatsky, A. V. J. Phys. Chem. C 2007, 111, 14541-14551. 15 Kilina, S.; Yarotski, D. A.; Talin, A. A.; Tretiak, S.; Taylor, A. J.; Balatsky, A. V. J. Drug Delivery 2011, 2011, 415621. 16 Tanaka, H.; Kawai, T. Nat. Nanotechnol. 2009, 4, 518-522. 17 Wanunu, M.; Cohen-Karni, D.; Johnson, R. R.; Fields, L.; Benner, J.; Peterman, N.; Zheng, Y.; Klein, M. L.; Drndic, M. J. Am. Chem. Soc. 2011, 133, 486-492. 18 Garaj, S.; Hubbard, W.; Reina, A.; Kong, J.; Branton, D.; Golovchenko, J. A. Nature 2010, 467, 190-193. 19 Shim, J.; Humphreys, G. I.; Venkatesan, B. M.; Munz, J. M.; Zou, X.; Sathe, C.; Schulten, K.; Kosari, F.; Nardulli, A. M.; Vasmatzis, G.; Bashir, R. Sci. Rep. 2013, 3, 1389. 20 Pathak, B. et al. Applied Physics Letters 2012, 100, 560564. 023701. 9 Staden, R. A Nucleic Acids Res 1979, 6, 26012610. 10 Braslavsky, I.; Hebert, B.; Kartalov, E.; Quake, S. R. Proc Natl Acad Sci 2003, 100, 39603964. 11 Sanger, F.; Nicklen, S.; Coulson, A. R. Proc Natl Acad Sci 1977, 74, 54635467. 12 Brenner, S. et al. Nat Biotech 2000, 18, 630634. 13 Shendure, J. et al. Science 2005 309, 17281732. 21 Tsutsui, M.; Taniguchi, M.; Yokota, K.; Kawai, T. Nature Nanotechnol. 2010, 5, 286290. 22 Chang, S.; Huang, S.; He, J.; Liang, F.; Zhang, P.; Li, S.; Chen, X.; Sankey, O.; Lindsay, S. Nano Lett. 2010, 10, 10701075. 23 Ohshiro, T.; Matsubara, K.; Tsutsui, M.; Furuhashi, M.; Taniguchi, M.; Kawai, T. Sci. Rep. 2012, 2, 10701075. 24 Geim, A. K.; Novoselov, K. S. Nat Mater 2007, 6, 183191. 25 Castro Neto, A. H.; Guinea, F.; Peres, N. M. R.; Novoselov, K. S.; Geim, A. K. Rev. Mod. Phys. 2009 81, 109162. 26 Lee, C.; Wei, X.; Kysar, J. W.; Hone, J. Science 2008, 321, 385388. 27 Ahmed, T.; Haraldsen, J. T.; Rehr, J. J.; Ventra, M. D.; Schuller, I.; Balatsky, A. V. Nanotechnology 2014, 25, 125705. 28 Ahmed, T.; Haraldsen, J. T.; Zhu, J.; Balatsky, A. V. J. Phys. Chem. Lett. 2014, 5, 2601-2607. 29 Atomistix ToolKit version 13.8, QuantumWise A/S (www.quantumwise.com) 30 Brandbyge, M.; Mozos, J.-L.; Ordej´on, P.; Taylor, J.; Stok- bro K. Phys. Rev. B 2002, 65, 165401. 31 Soler, J.M.; Artacho, E.; Gale, J.D.; Garc´ıa, A.; Junquera, J.; Ordej´on, P.; S´anchez-Portal D. J. Phys. Condens. Mat- ter 2002, 14, 2745. 32 Wells, D. B.; Belkin, M.; Comer, J.; Aksimentiev, A. Nano Lett. 2012, 12, 4117-4123. 33 Stokbro, K, J. Phys.: Condens. Matter 2008, 20, 064216. 6
1301.3481	2	1301	2013-08-19T15:33:29	Manipulation of Ferromagnets via the Spin-Selective Optical Stark Effect	[ "cond-mat.mes-hall" ]	We investigate the non-resonant all-optical switching of magnetization. We treat the inverse Faraday effect (IFE) theoretically in terms of the spin-selective optical Stark effect for linearly or circularly polarized light. In the dilute magnetic semiconductors (Ga,Mn)As, strong laser pulses below the band gap induce effective magnetic fields of several teslas in a direction which depends on the magnetization and light wave vectors. Our theory demonstrates that the polarized light catalyzes the angular momentum transfer between lattice and the magnetization.	cond-mat.mes-hall	cond-mat	Manipulation of Ferromagnets by the Spin-Selective Optical Stark Effect Alireza Qaiumzadeh,1 Gerrit E. W. Bauer,2, 3, 4 and Arne Brataas1 1Department of Physics, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway 2Institute for Materials Research, Tohoku University, Sendai 980-8577, Japan 3WPI Advanced Institute for Materials Research, Tohoku University, Sendai 980-8577, Japan 4Delft University of Technology, Kavli Institute of NanoScience, 2628 CJ Delft, The Netherlands (Dated: August 17, 2018) We investigate the non-resonant all-optical switching of magnetization. We treat the inverse Faraday effect (IFE) theoretically in terms of the spin-selective optical Stark effect for linearly or circularly polarized light. In the dilute magnetic semiconductors (Ga,Mn)As, strong laser pulses below the band gap induce effective magnetic fields of several teslas in a direction which depends on the magnetization direction as well as the light polarization and direction. Our theory demon- strates that the polarized light catalyzes the angular momentum transfer between the lattice and the magnetization. PACS numbers: 75.78.-n, 78.20.Bh, 71.70.Ej, 75.40.Gb I. INTRODUCTION An essential challenge in magnetoelectronics is finding new methods to manipulate the magnetization and in- crease the switching speed for realizing faster and higher- density data storage and information processing. Tra- ditionally, the magnetization is switched by applying nearly collinear external magnetic fields for longer than 100 ps1. Current-induced spin transfer torque switch- ing is not faster2. So-called precessional switching can be achieved by a magnetic field pulse perpendicular to the magnetization1,3, leading to magnetization reversal as fast as, but not faster than, a few ps3. In this pa- per, we explain how off-resonant optical pulses generate strong effective magnetic fields that can lead to ultrafast magnetic reversal. Recently, it has been demonstrated that a 40 fs circu- larly polarized pulse can reverse the magnetization of the metallic ferrimagnet GdFeCo with perpendicular mag- netic anisotropy4. A magnetic write event as short as 30 ps by using 100 fs circularly polarized light pulses has been reported5, i.e. a potential data storage rate of about 10 Tbit/s. Simulations suggest that the magnetization reversal is not realized via precession, but is caused by a linear process in which the magnitude of the magneti- zation passes through zero, in the presence of magnetic fields of ∼ 20 T4,5. The magnetization dynamics can also be triggered by light that is linearly polarized in a direc- tion non-collinear to the crystal axis6,7. For a comprehen- sive review on ultrafast all-optical magnetization dynam- ics, see Ref.8. Phenomenologically, these experiments are attributed to two reactive effects: (i) the inverse Faraday effect (IFE)9,10 - the ability of the electric field compo- nent of circularly polarized light E(ω) to induce a static magnetization MIFE(0) ∝ E(ω) × E∗(ω) - and (ii) the inverse Cotton-Mouton effect (ICME)10 - the magneti- zation induced by polarized light in the presence of an external magnetic field Bext, MICME(0) ∝ E(ω)2Bext. The microscopic origin of the large magnetic field in- duced by light and the induced magnetization dynam- ics are still a subject of debate8,11,12. Several theoretical mechanisms have been proposed, such as the optical Bar- nett effect or the inverse Einstein-de Haas effect13, light- induced circular currents in the collisionless limit14,15, the impulsive stimulated Raman-like process8, and pho- tonic angular momentum transfer via deflection of the scattered photons16. A dissipative IFE under THz irra- diation has been computed for dirty metals with extrinsic spin-orbit interaction17,18. Also experimentally the situ- ation is not clear. Recent experiments on the ferrimag- netic metallic alloy GdFeCo found optical magnetization switching without light polarization in a certain range of light intensities and sample temperatures, casting doubt on the ubiquity of the IFE19. Also Such a behavior was not reported for TbCo films, however20. In this paper, we predict huge effective magnetic fields induced by the below band gap polarized light through the spin-selective ac (optical) Stark effect21, i.e. the shift of electronic energy levels connected through finite op- tical matrix elements10. In perturbation theory, this process is closely related to Raman scattering. Since the electronic structure of amorphous alloys is compli- cated and experiments concerning the role of the IFE are inconclusive, we focus here on GaMnAs as a generic model system, since it can be grown with perpendicular anisotropy and its electronic structure is well known. Di- lute magnetic semiconductors are interesting spintronic materials by themselves22. Although their Curie temper- ature at present is below the room temperature, study- ing these materials can improve our understanding of novel physical phenomena that are also present in other magnets23 -- 25. Photo-injected carriers induced by linearly polarized light with frequency slightly above the Γ or L band edges have been shown to induce magnetization dy- namics in GaMnAs23,26. In contrast, we focus here on excitation with frequencies below the fundamental band gap, which is dissipationless, since no free carrier are ex- cited. Our approach is quite general and can be applied to arbitrary electronic structures and computed from first principles. s µ∗ for GaAs parameters BIFE = ℏΩ/ (g∗ In one scenario4,5, extrinsic processes due to the high- intensity laser pulse heat up the ferromagnet so that it becomes paramagnetic while the circularly polarized light generates the spin-selective optical Stark effect or effec- tive magnetic field BIFE that triggers linear reversal, i.e. the modulus of the magnetization passes through zero during switching. The maximum achievable field can be B) ∼ huge, e.g. 100 T, where Ω is the Rabi frequency for a light in- tensity of 5 GWcm−2, and g∗ s µ∗ B/ℏ ∼ 500 GHz/T is the effective gyromagnetic ratio. In practice, the light frequency should be sufficiently below the band gap to reduce heating that destroys the sample at high inten- sities. We therefore formulate the IFE here in second order perturbation theory and compute the resulting ex- pressions for the GaAs band structure and wave func- tions. We find that for light frequencies safely below the energy gap, the effective field amounts to several teslas in GaMnAs, which suffices to nucleate a ferro- magnetic state during the cooling phase5,27. In a sec- ond scenario, we assume that the material remains fer- romagnetic under the laser excitation, possibly with a reduced magnetization. In this case, both linearly and circularly polarized light can trigger both precessional and linear switching mechanisms. In both scenarios, the required angular momentum is not supplied by the photons, but by the lattice via the spin-orbit interac- tion. We show that, in general, the light-induced ef- fective field has three components, depending on the light polarization and initial magnetization direction, viz. Blight = BIFE q + BICME M0 + B⊥q × M0. The sign of the IFE effective field BIFE depends on the helicity and points along the light propagation direction q. BICME, the magnetic field associated with the ICME, is directed along the magnetization vector M0, while a field with strength B⊥ is perpendicular to both. In Sec. This paper is organized as follows. II we present the Kane band model we have used to de- scribe light-matter interaction and introduce the effec- tive hamiltonian within the second order time dependent perturbation theory. In Sec. III we present and discuss our main analytical results for the light-induced magnetic field in dilute magnetic semiconductors. Finally, in Sec. IV we present a summary of our main conclusions. II. MODEL HAMILTONIAN AND SECOND-ORDER TIME-DEPENDENT PERTURBATION THEORY In the Coulomb gauge, the 8-band Kane model Hamil- tonian for a zinkblende semiconductor at the Γ point reads H = H0 + Hp−d + Hint, where H0 = p2 2m + 4m2c2 p · (s × ▽▽▽Vp) + Vp, Hsp−d = −J M0 · s, A · p + Hint ≃ e mc e2 2mc2 A2. 2 (1) Here p, e, and m are the momentum operator, elec- tron charge, and electron mass, respectively, is the reduced Planck constant, c is the light velocity, s is vector of 2 × 2 Pauli matrices, A is the vector poten- tial of the monochromatic light field, Vp is the periodic lattice potential, and Hsp−d is the sp-d mean-field ex- change interaction between the magnetization direction of the localized d-spins M0, and the itinerant s- or p- spins, controlled by the exchange potential J. The A · p interaction term describes the annihilation of a photon and the creation of an electron-hole pair and vice versa, while A2 represents a photon scattering processes. In perturbation theory, two-photon transitions can be in- duced by either A· p to second order or A2 to first order in the interaction Hamiltonian. To leading order in the light-matter interaction, A2 does not induce spin rever- sal and will therefore be disregarded in the following. A · p induces only two-photon virtual interband transi- tions, since the light frequency is below the band gap. Intra-band transitions are disregarded because they are impurity-mediated and weak. Direct-band-gap semicon- ductors can be treated in the effective mass approxima- tion and projected on the well-established 8-band Kane model for H including the conduction (cb±i), the heavy- hole (hh±i), the light-hole (lh±i), and the spin-orbit split-off (so±i) bands28. In the following, we will dis- regard the band dispersion, a common approximation in theories of Raman scattering29 that is allowed for low doping levels and/or large detuning. The electric field component of monochromatic light with frequency ω0 and wave-vector qkz is E(t) = eE0e−i(ω0t−q.r) + c.c., for light propagating along the z- direction with polarization e = ex x + ey y. For circular polarization ex = 1/√2, ey = λi/√2, where λ = ±1, and for linear polarization with angle α relative to the x-axis, ex = cos α and ey = sin α. When the pulse duration Tp is sufficiently longer than (Eg/ − ω0)−1, where Eg is the energy gap, transient effects can be disregarded30. For a laser pulse width of 40-100 fs with a frequency that is not too close to the resonance, the above criterion is satisfied by /Eg ≈ 0.5 fs for our material. der perturbation for the A · p interaction term read9 The matrix elements of the Hamiltonian in second or- hmHki = e2E2 0 m2ω2 0c2 Xl −hmpγeγlihlpβe∗ ω0 − (ǫm − ǫl) βlihlpγeγki (cid:20)hmpβe∗ ω0 + (ǫk − ǫl) βki (cid:21) , (2) where β(γ) = x, y, z. mi, ki, and li are the initial, final, and intermediate states including the spin and mo- mentum quantum numbers, with the energies ǫm, ǫk, and ǫl respectively. tors are C(n) σ = C(n) ρ = 2P 2 3m (cid:18) −ω0 g − 2ω2 E2 3m (cid:18) 2Eg g − 2ω2 E2 2P 2 0 0 + + ω0 (Eg + ∆)2 − 2ω2 Eg + ∆ (Eg + ∆)2 − 2ω2 0(cid:19) , 0(cid:19) , 3 (4) (5) III. LIGHT-INDUCED EFFECTIVE MAGNETIC FIELD IN GaAs AND GaMnAs Unlike an external magnetic field, conduction and va- lence bands experience different light-induced effective fields31. For conduction band (valence bands) Blight = s Tr [s2]/3(cid:1), where s is the vector of 2 × 2 Bg∗ 2Tr [sH] /(cid:0)µ∗ Pauli spin matrices for 1/2-spins in the conduction band (the 4×4 spin matrices for 3/2-spins in the valence band), Tr is the trace over electron states (hole states), µ∗ B is ef- fective Bohr magneton of electron (hole), and g∗ s is the electron (hole) effective Land´e g-factor. Note that we lump heavy and light holes together by the trace and adopt an average value of µ∗ s for the valence band. Bg∗ Bg∗ As illustrated by Fig. 1, the below-band gap light field induces a Zeeman-like splitting, called the spin- selective optical Stark shift10, which can be interpreted as an effective magnetic field experienced by each band, δEStark = −µ∗ s BIFE/2. The effective field, BIFE, is a reactive response and therefore essentially instantaneous as long as Tp > /Eg. The effective field eventually gives rise to a non-equilibrium spin-polarization in the con- ne ∝ N (n)δEStark, where N (n) is the density of states per unit volume at the Fermi level of the conduction band (with analogous relations for the holes) and h. . .i denotes the expectation value on intermediate time scales. This spin polarization is generated by a re- population of the states (see Fig. 1), on the scale of the spin-flip scattering time Tp ∼ τsf , which is expected to be in the ps range under high excitation conditions. duction band, hσzi(n) A. Paramagnetic case In second-order time-dependent perturbation theory, the spin susceptibility of a paramagnet is defined as KIFE = hσine/(ξE2 reads 0) with ξ = e2/(cid:0)mc2(cid:1), which 0 /ω2 KIFE = − 1 L3 Xkm f (ǫk) − f (ǫm) (ǫk − ǫm) + iηhksmi× (3) hCρe · I · e∗hmki + iCσ(e × e∗) · hmskii , where ∆ is the spin-orbit splitting energy and P the inter- band momentum matrix element. The coefficients of Cσ and Cρ are identical to the spin- and charge-density exci- tation coefficients in the theory of Raman scattering29. In our formulation, the incoming and outgoing photons have identical polarization, Eq. (3), which means that there is no direct angular momentum transfer from the light to the medium. Angular momentum of spin-flip processes is hence supplied from the lattice via spin-orbit coupling during the spin-flip relaxation process. In a paramagnetic n-doped semiconductor, the Stark effective field, or IFE field, is oriented along the light propagation direction q, as IFE = −2ξE2 B(n) 0 s ω2 µ∗ Bg∗ 0 KIFE N (n) = − 2λC(n) σ ξE2 0 µ∗ Bg∗ s ω2 0 q. (6) For Tp & τsf spin accumulation hσi(n) −2λN (n)C(n) 0 q/ω2 0. σ ξE2 this magnetic field leads ne = −Kz,IFEξE2 to the 0 = 0 q/ω2 In paramagnetic p-doped systems, the effective field experienced by the hole bands are B(p) IFE = − 3λC(p) σ ξE2 s ω2 µ∗ Bg∗ 0 0 q , (7) σ ξE2 0 q/ω2 σ,ρ = C(n) According to Eqs. ne = λN (p)C(p) scales is hσi(p) where C(p) σ,ρ (∆ → ∞) since the matrix elements with the spin-orbit split-off bands vanish. In the p-doped case, the non-equilibrium spin polarization on longer time 0, where N (p) is the average density of states at the Fermi level of hole bands. (6) and (7), polarized light with frequency ω0 < Eg/ induces a magnetic field along q. Its sign is governed by the light helicity λ, while its mag- nitude is proportional to the light intensity E2 0 and van- ishes with the spin-orbit coupling since C(n) σ (∆ = 0) = 0. In n-doped systems and in the large detuning limit ω0 ≪ Eg/, to leading order, it yields hσzine ∝ ∆/ω0 for ∆ < Eg. This optical Stark shift-induced non- equilibrium spin polarization can be compared with the magnetization induced by the circular currents in re- sponse to the rotating electric field of the circularly polar- ized light14. The latter scales with frequency like ∝ ω−3 0 , thus should be small at optical frequencies. The spin- transfer torques induced by the circular currents might be significant, however15. where L is the system size, η is a positive infinitesimal, f (ǫ) is the Fermi-Dirac distribution function in equilib- rium, and I is the unit dyadic in Pauli spin space. The interband couplings Cσ and Cρ for n-doped semiconduc- This perturbation theory is valid in the limit δEStark, εF ≪ Eg, and ω0 < Eg. For n-GaAs with s ≃ −0.44, m∗ ≃ 0.067m Eg = 1.52 eV, ∆ = 341 meV, g∗ and 2P 2/m ≃ 20 eV32, a light intensity of 10 GW / cm2 4 FIG. 2: Both linearly and circularly polarized light exert a torque T, on the equilibrium magnetization M, and may trig- ger magnetization dynamics. in contrast to the IFE, which is even in the exchange en- ergy and odd in the helicity. In the large detuning limit, the ICME scales like ω−2 0 , while the helicity dependent 0 . In this case qk M0, and then the effective IFE is ∝ ω−1 field does not trigger magnetization precession dynamics. Linear reversal through zero magnetization can occur if the light-induced effective field is sufficiently larger than coercive field, which is dramatically reduced down to a few teslas near the Curie temperature27. Magnetization reversal by precession might be possible in principle, but would require much longer light pulses for qk M0. Second, we consider a magnetization direction per- pendicular to the light propagation direction as M0 = x cos φ + y sin φ, where φ is the azimuthal angle. The average light-induced effective field induced by circularly polarized light on the holes in the valence bands is B(p) light = B(p) ICME IFE + B(p) ≃ −(cid:18)3λC(p) σ q − J 5∆ C(p) ρ M0(cid:19) ξE2 0 s ω2 µ∗ Bg∗ 0 . (9) ne ∝ M0 × B(p) a spin accumulation hσi(p) T = J M0 × hσi(p) The first term, the IFE field, is along the light wave vector and changes sign with light helicity. The IFE field acts on the holes that relax very fast to generate ne , which by the exchange inter- actions exerts a strong torque on the local magnetization 34. The other term corresponds to the ICME and is strictly longitudinal, which enhances or suppresses magnetization, but does not trigger magnetization precession33. With J/∆ ∼ 0.1 we estimate B(p) s ) for a light in- tensity of 10 GW / cm2, which is large considering that Bg∗ µ∗ Also the effective magnetic field induced by linearly po- larized light, in perpendicular configuration q⊥ M0, is s ∼ 10−4 − 10−5 eV/T. ICME ∼ 10−4 eV / (µ∗ Bg∗ light FIG. 1: Illustration of the changes in the majority and minor- ity population due to the Stark shift (δE n Stark), in the presence of non-resonant and intense circularly polarized laser field. Stark < δE p at frequency ω0 = 1.24 eV (λ0 = 1 µm) then generates an effective magnetic field of 9 T. This estimate is more than three orders of magnitude larger than what has been predicted in disordered metals involves intraband tran- sitions with the same laser intensities17,18. The high- intensity laser power used in ultrafast opto-magnetic ex- periments leads to heating and demagnetization of sam- ples even at below band gap frequencies due to multiple photon absorption, band tails, disorder etc. The opti- cally induced spin accumulation then can nucleate a per- sistent magnetization when the samples cools after the pump pulse4,5. B. Ferromagnetic case Consider now hole-doped ferromagnetic semiconduc- tors. As before, we assume small hole densities εF ≪ Eg, ∆ and thus limit the discussion to the optical transi- tions at Γ. We investigate the weak ferromagnetic regime in which J ≪ ∆ < Eg, therefore it is sufficient to calcu- late effective fields to the lowest order of J/∆. First, we assume an equilibrium magnetization direc- tion M0 along the light propagation direction q. The average effective field experienced by the valence bands is B (p) light = B (p) ICME (p) IFE + B ≃ −(cid:18)3λC(p) σ + 2J 5∆ C(p) ρ (cid:19) ξE2 0 q s ω2 0 µ∗ Bg∗ . (8) The first term, the helicity (λ) dependent term, is the IFE field. The helicity independent term, the second term, corresponds to the ICME field and enhances or suppresses the magnetization even for a linearly polar- ized beam. The ICME is an odd function of the ex- change coupling and, in the small magnetization limit, is linearly proportional to the exchange coupling. This is given by B(p) ⊥ light = B(p) ICME + B(p) 2J 5∆ 3 2 C(p) ≃ + ρ h M0(cid:0)3 cos2(φ − α) − 1(cid:1) q × M0 sin 2(φ − α)(cid:21) ξE2 0 s ω2 µ∗ Bg∗ 0 . (10) This field has two components, the conventional ICME parallel to M0, and a term along q × M0, which exerts a torque on the local magnetization in the z-direction. Eqs. (9) and (10) show that in the perpendicular con- figuration both linearly and circularly polarized light in- duce effective fields that exert torques on the equilibrium magnetization and induce precessional dynamics. Note, however, that in our model, unpolarized light or just a heat pulse does not generate effective magnetic fields. Linearly polarized light does not carry net angular mo- mentum, but nevertheless induces spin precession by in- ducing angular momentum transfer between lattice and exchange field, thereby rotating its plane of polarization (Faraday effect), see Fig. 2. Circularly polarized pho- tons can directly transfer angular momentum from the light to the spin of electrons only when absorbed. At typical laser intensities, the amount of available angular momentum is by far not enough to reverse the magne- tization. We thus demonstrate that in optomagnetism the lattice and exchange fields act as sources and sinks of angular momentum via the spin-orbit and exchange couplings35. We present here a microscopic theory of 5 light-induced magnetic fields. In order to compare with experiments, the magnetization dynamics under effective magnetic field and heat pulses will have to be computed. A realistic micromagnetic simulation in the presence of such an effective field has been carried out in Refs. [5] and [27]. A repetition of these calculations for III-V mag- netic semiconductors is far beyond the scope of this pa- per, however. IV. SUMMARY AND CONCLUSIONS In summary, we studied the magnetic response to in- tense and non-absorptive, linearly and circularly polar- ized lights in para- and ferromagnetic III-V semiconduc- tors. The strong spin-orbit coupling plays a vital role to supply the required angular momentum. As a result, the light-induced field strength in GaMnAs is huge, up to sev- eral teslas, which is sufficient to reverse magnetization by either linearly or precessional paths. We found that the spin-selective optical Stark effect in ferromagnets induces effective magnetic fields in different directions depending on the light orientation and the magnetization direction. V. ACKNOWLEDGMENTS This work was supported by EU-ICT-7 contract No. 257159 "MACALO", the FOM Foundation, the ICC- IMR and DFG Priority Programme 1538 "Spin-Caloric Transport". A. Q. would like to thank A. G. Moghaddam for useful discussions. 1 J. Stohr and H. C. Siegmann, Magnetism: From Funda- mentals to Nanoscale Dynamics (Springer-Verlag, Berlin, 2006). 2 S. Serrano-Guisan, K. Rott, G. Reiss, J. Langer, B. Ocker, and H. W. Schumacher, Phys. Rev. Lett. 101, 087201 (2008). 3 I. Tudosa, C. Stamm, A. B. Kashuba, F. King, H. C. Sieg- mann, J. Stohr, G. Ju, B. Lu and D. Weller, Nature 428, 831 (2004). 4 C. D. Stanciu, F. Hansteen, A. V. Kimel, A. Kirilyuk, A. Tsukamoto, A. Itoh, and Th. Rasing, Phys. Rev. Lett. 99, 047601 (2007). 5 K. Vahaplar, A. M. Kalashnikova, A. V. Kimel, D. Hinzke, U. Nowak, R. Chantrell, A. Tsukamoto, A. Itoh, A. Kiri- lyuk, and Th. Rasing, Phys. Rev. Lett. 103, 117201 (2009). 6 A. M. Kalashnikova, A. V. Kimel, R. V. Pisarev, V. N. Gridnev, A. Kirilyuk, and Th. Rasing, Phys. Rev. Lett. 99, 167205 (2007); A. M. Kalashnikova, A. V. Kimel, R. V. Pisarev, V. N. Gridnev, P. A. Usachev, A. Kirilyuk, and Th. Rasing, Phys. Rev. B 78, 104301 (2008). 7 R. Iida, T. Satoh, T. Shimura, K. Kuroda, B. A. Ivanov, Y. Tokunaga, and Y. Tokura, Phys. Rev. B 84, 064402 (2011). 8 A. Kirilyuk, A. V. Kimel, and Th. Rasing, Rev. Mod. Phys. 82, 2731 (2010). 9 P. S. Pershan, J. P. van der Ziel, and L. D. Malmstrom, Phys. Rev. 143, 574 (1966). 10 Y. R. Shen, The Principles of Nonlinear Optics (Wiley, New York, 1984). 11 D. Steil, S. Alebrand, A. Hassdenteufel, M. Cinchetti, and M. Aeschlimann, Phys. Rev. B 84, 224408 (2011). 12 K. Vahaplar, A. M. Kalashnikova, A. V. Kimel, S. Gerlach, D. Hinzke, U. Nowak, R. Chantrell, A. Tsukamoto, A. Itoh, A. Kirilyuk, and Th. Rasing, Phys. Rev. B 85, 104402 (2012). 13 J. Hohlfeld, C. D. Stanciu, and A. Rebei, Appl. Phys. Lett. 94, 152504 (2009). 14 B. Talin, V. P. Kaftandjian, and L. Klein, Phys. Rev. A 11, 648 (1975); R. Hertel, J. Magn. Magn. Mater. 303, L1 (2006). 15 F. K. Joibari and G. E. W. Bauer, unpublished. 16 S. R. Woodford, Phys. Rev. B 79, 212412 (2009). 17 V. M. Edelstein, Phys. Rev. Lett. 80, 5766 (1998). 18 K. Taguchi and G. Tatara, Phys. Rev. B 84, 174433 (2011). 19 J. A. de Jong, I. Razdolski, A. M. Kalashnikova, R. V. Pisarev, A. M. Balbashov, A. Kirilyuk, Th. Rasing, and A. V. Kimel. Phys. Rev. Lett. 108, 157601 (2012); A. R. Khorsand, M. Savoini, A. Kirilyuk, A. V. Kimel, A. Tsukamoto, A. Itoh, and Th. Rasing Phys. Rev. Lett. 108, 127205 (2012); T. A. Ostler, J. Barker, R. F. L. Evans, R. W. Chantrell, U. Atxitia, O. Chubykalo-Fesenko, S. El Moussaoui, L. Le Guyader, E. Mengotti, L. J. Heyderman, F. Nolting, A. Tsukamoto, A. Itoh, D. Afanasiev, B. A. Ivanov, A. M. Kalashnikova, K. Vahaplar, J. Mentink, A. Kirilyuk, Th. Rasing, and A. V. Kimel, Nat. Commun. 3, 666 (2012). 20 S. Alebrand, M. Gottwald, M. Hehn, D. Steil, M. Cinchetti, D. Lacour, E. E. Fullerton, M. Aeschlimann, and S. Man- gin, Appl. Phys. Lett. 101, 162408 (2012) 21 M. Y. A. Raja, D. Allen, and W. Sisk, Appl. Phys. Lett. 67 2123 (1995). 22 K. S. Burch, D. D. Awschalom, D. N. Basov, J. Magn. Magn. Mater. 320, 3207 (2008); T. Jungwirth, Jairo Sinova, J. Masek, J. Kucera and A. H. MacDonald, Rev. Mod. Phys. 78, 809 (2006). 23 J. Wang, C. Sun, Y. Hashimoto, J. Kono, G. A. Khoda- parast, L. Cywi´nski, L. J. Sham, G. D. Sanders, C. J. Stanton, and H. Munekata, J. Phys.: Condens. Matter. 18, R501 (2006). 24 N. Tesarova, P. Nemec, E. Rozkotova, J. Subrt, H. Re- ichlova, D. Butkovicova, F. Trojanek, P. Maly, V. Novak, T. Jungwirth, Appl. Phys. Lett. 100 102403 (2012). 25 C. Phelps, T. Sweeney, R. T. Cox, and H. Wang, Phys. Rev. Lett. 102, 237402 (2009); S. G. Carter, Z. Chen, and S. T. Cundiff, Phys. Rev. B 76, 201308(R) (2007). 26 M. D. Kapetanakis, I. E. Perakis, K. J. Wickey, C. Pier- marocchi, and J. Wang, Phys. Rev. Lett. 103, 047404 6 (2009); J. Wang, I. Cotoros, D. S. Chemla, X. Liu, J. K. Furdyna, J. Chovan, and I. E. Perakis, Appl. Phys. Lett. 94, 021101 (2009); Y. Hashimoto, S. Kobayashi, and H. Munekata, Phys. Rev. Lett. 100, 067202 (2008). 27 N. Kazantseva, D. Hinzke, R. W. Chantrell and U. Nowak, Europhys. Lett. 86, 27006 (2009). 28 E. O. Kane, J. Phys. Chem. Solids 1, 249 (1957). 29 G. Abstreiter, M. Cardona and A. Pinczuk, in Light Scattering in Solids IV, edited by M. Cardona and G. Guntherodt (Springer-Verlag, New York, 1984). 30 D. Popova, A. Bringer, and S. Blugel, Phys. Rev. B 85, 094419 (2012). 31 A. H. M. Reid, A. V. Kimel, A. Kirilyuk, J. F. Gregg, and Th. Rasing, Phys. Rev. Lett. 105, 107402 (2010); Phys. Rev. B 81, 104404 (2010). 32 C. Hermann and C. Weisbuch, Phys. Rev. B 15, 823 (1977). 33 T. Makino, F. liu, T. Yamasaki, Y. Kozuka, K. Ueno, A. Tsukazaki, T. Fukumura, Y. Kong, and M. Kawasaki, Phys. Rev. B 86, 064403 (2012). 34 A. Matos-Abiague, and R. L. Rodr´ıguez-Su´arez, Phys. Rev. B 80, 094424 (2009). 35 B. Koopmans, M. van Kampen, J. T. Kohlhepp, and W. J. M. de Jonge, Phys. Rev. Lett. 85, 844 (2000).
1512.00534	1	1512	2015-12-02T00:45:04	Dynamical Axion Field in a Magnetic Topological Insulator Superlattice	[ "cond-mat.mes-hall", "hep-ph" ]	We propose that the dynamical axion field can be realized in a magnetic topological insulator superlattice or a topological paramagnetic insulator. The magnetic fluctuations of these systems produce a pseudoscalar field which has an axionic coupling to the electromagnetic field, and thus it gives a condensed-matter realization of the axion electrodynamics. Compared to the previously proposed dynamical axion materials where a long range antiferromagnetic order is required, the systems proposed here have the advantage that only a uniform magnetization or a paramagnetic state is needed for the dynamic axion. We further propose several experiments to detect such a dynamical axion field.	cond-mat.mes-hall	cond-mat	Dynamical Axion Field in a Magnetic Topological Insulator Superlattice 1Department of Physics, McCullough Building, Stanford University, Stanford, California 94305-4045, USA Jing Wang,1, 2 Biao Lian,1 and Shou-Cheng Zhang1, 2 2Stanford Institute for Materials and Energy Sciences, SLAC National Accelerator Laboratory, Menlo Park, California 94025, USA (Dated: September 17, 2018) We propose that the dynamical axion field can be realized in a magnetic topological insulator superlattice or a topological paramagnetic insulator. The magnetic fluctuations of these systems produce a pseudoscalar field which has an axionic coupling to the electromagnetic field, and thus it gives a condensed-matter realization of the axion electrodynamics. Compared to the previously proposed dynamical axion materials where a long range antiferromagnetic order is required, the systems proposed here have the advantage that only a uniform magnetization or a paramagnetic state is needed for the dynamic axion. We further propose several experiments to detect such a dynamical axion field. PACS numbers: 73.20.-r 75.70.-i 14.80.Va The search for topological quantum phenomena has attracted considerable interest in condensed matter physics. Topological phenomena are determined by some topological structures in physical systems and are thus usually universal and robust against perturbations [1]. The recent discovery of the time-reversal (T ) invari- ant (TRI) topological insulator (TI) brings the oppor- tunity to realize a large family of new topological phe- nomena [2, 3]. The electromagnetic response of three- dimensional (3D) insulators is described by the Maxwell action SM = (1/8π)(cid:82) d3xdt(E2 − B2/µ), together with a topological θ term Sθ = (θ/2π)(α/2π)(cid:82) d3xdtE· B [4]. Here, E and B are the conventional electromagnetic fields inside the insulator, and µ are material-dependent di- electric constant and magnetic permeability, α = e2/c is the fine structure constant, e is the charge of an elec- tron, and θ is the dimensionless pseudoscalar parameter describing the insulator, which refers to the axion field in high energy physics [5, 6]. Physically θ depends on the band structure of the insulator and has an explicit micro- scopic expression of the momentum space Chern-Simons form [4]: (cid:90) (cid:20) (cid:21) θ = 1 4π d3kijkTr Ai∂jAk + i AiAjAk 2 3 , (1) k∂iu i (k) = −i(cid:104)uµ k(cid:105) is the momen- where ∂j = ∂/∂kj, Aµ k(cid:105) and u k(cid:105) tum space non-abelian gauge field, with uµ referring to the periodic part of the Bloch function of the occupied bands. All physical quantities in the bulk de- pend on θ only modulo 2π. Sθ generally breaks the parity P and T symmetry except for two TRI points θ = 0 and θ = π, which describe trivial insulator and TI, respec- tively [4]. The term Sθ with a universal value of θ = π in TIs gives rise to new physical effects such as image magnetic monopole [7], quantized Kerr effect [8, 9], and quantized topological magnetoelectric effect [4, 10 -- 12]. The axion field θ is static in a TRI TI. However, as is first suggested in Ref. [13], when a long range anti- ferromagnetic (AFM) order is established in a TI, θ can deviate from π due to T symmetry breaking and becomes a dynamical field associated with the magnetic fluctua- tions. The resulting system is a new state of quantum matter which realizes the axion electrodynamics in con- densed matter physics. The axionic excitations in such an unconventional AFM insulator can lead to novel ef- fects such as the axionic polariton [13]. There have been great efforts devoted to searching for such dynamical ax- ion state of matter [14 -- 19]. However, such AFM materi- als are still lacking. In this paper, we propose a much simpler route to re- alize the dynamic axion in a magnetic TI superlattice. In particular, we clarify that the realization of a dynam- ical axion field does not necessarily require an AFM or- der and a TI parent material, but what is important is a proper coupling between the electrons and magnetic fluc- tuations [13 -- 15]. The magnetic TI superlattice we adopt consists of alternating layers of a parent magnetic TI and a spacer normal insulator (NI), as shown in Fig. 1(a). The magnetic TI layer is doped with Cr and Mn on top and bottom halves of TI film, respectively. We show that the phase diagram of this system contains a dynamic ax- ion phase when uniform magnetization is achieved, where T symmetry is broken and the static value θ0 (cid:54)= 0, π. We also propose to realize the dynamic axion in a topologi- cal paramagnetic (PM) insulator, where T symmetry is present and θ0 = 0 or π. The PM fluctuations can cou- ple to electrons, which induces a dynamical axion field. Such a PM insulator can be achieved by doping magnetic elements into TI materials close to the topological quan- tum critical point (QCP) [shown in Fig. 1(b)]. Finally, we propose several experiments to detect this dynamical axion field. The first system we propose is a magnetic TI superlat- tice as described above. Recent experiments have shown that the thickness and magnetic doping concentration of thin film TIs can be well controlled through layer-by- 2 proximation, the exchange field of Mn and Cr are given by ∆ν = yνJν(cid:104)Sz ν(cid:105)/2. Here y is the doping concentra- tion of magnetic ion, (cid:104)Sz(cid:105) is the MF expectation value of the ion spin in the z direction, ν = c, m. The thickness dependent parameters ts and tn describe the tunneling between the top and bottom surface states within the same (ts) or neighboring (tn) TI layer. For simplicity, we assume Jc = −Jm = J > 0. First, we examine the phase diagram of the system. The momentum space Hamiltonian now is H(k) = da(k)Γa + mbΓ12, (3) 5(cid:88) a=1 (cid:104) mb ±(cid:112)m2 (cid:105)2 a + t2(kz) , (4) where d1,2,3,4,5(k) = (vF ky,−vF kx,−tn sin(kz(cid:96)), ts + tn cos(kz(cid:96)), ma), and the Dirac Γ matrices Γ1,2,3,4,5 = (τ zσx, τ zσy, τ y, τ x, τ zσz), Γ12 = [Γ1, Γ2]/2i = σz. The T and P transformation are defined as T = iσyK (with K being the complex conjugation operator) and P = τ x, respectively. The band dispersion is given by x + k2 F (k2 ε2 k± = v2 where t(kz) = (cid:112)t2 y) + s + t2 b −m2 b − m2 c1 ≡ (ts − tn)2 < m2 n + 2tstn cos(kz(cid:96)) and (cid:96) is the superlattice period along the growth z direction with (cid:96) = (cid:96)m + (cid:96)c + (cid:96)n. In the absence of exchange field, i.e., ma = mb = 0, the system is fully gapped when ts (cid:54)= tn, while it has a gapless Dirac node when ts/tn = ±1 [25]. For convenience we assume here ts/tn ≥ 0. When ts/tn = 1, the Dirac point is located at kx = ky = 0, kz = π/(cid:96). Such a critical Dirac point opens a gap when ts/tn deviates from unity, resulting in a 3D TI (tn > ts) or NI (tn < ts). Since both P and T symmetries are re- spected, the axion field θ given by Eq. (1) is either 0 or π in this case, as shown in Fig. 2(b). In the case of ma,b (cid:54)= 0, the band structure has two nondegenerate Weyl nodes a < (ts + tn)2 ≡ t2 when t2 c2, located on the kz axis at kz = π/(cid:96) ± k0 where k0(cid:96) = arccos[(m2 n)/2tstn]. Such a Weyl semimetal phase occurs in a finite region in the phase diagram as shown in Fig. 2(a). When m2 c2, the sys- tem is a 3D quantum anomalous Hall (QAH) insulator characterized by a quantized Hall conductivity e2/h per magnetic TI layer. Interesting physics happens when b − m2 m2 c1. The system is fully gapped, however, as we will show below, it is not a simple NI but an ax- ionic insulator (AI) with θ (cid:54)= 0, π. Furthermore, the FM fluctuations in the AI lead to a dynamical axion field. Since θ is odd under T and P, only T - and P-breaking perturbations can induce a change of θ. The term maΓ5 breaks both T and P, which varies the value of θ to the linear order. mbΓ12 breaks T but respects P, therefore it does not affect θ to the leading order. To compute the axion field θ, a lattice regularization is necessary. Ex- plicitly, the value of θ in this model can be calculated b − m2 a−t2 a > t2 s−t2 a < t2 FIG. 1. (color online). Schematic drawing of the proposed systems to realize the dynamical axion field. (a) The mag- netic TI superlattice structure. The upper and lower halves of TI films are doped with Cr and Mn, respectively. The ar- rows in TI layers indicate the magnetization direction. In each crystalline unit cell, the z-direction thickness of Mn-doped TI, Cr-doped TI, and NI layers are (cid:96)m, (cid:96)c, and (cid:96)n. (b) Topolog- ical PM insulator: 3D TI materials close to topological QCP doped with Cr, where a PM state is realized. xi (cid:80) layer growth via molecular beam epitaxy [20], therefore, such a superlattice is quite realistic to be fabricated. The magnetic ions will have a local exchange coupling with Sν(xi) · s, where the band electrons described by Jν Sν(xi) denotes the magnetic impurity spin at the position xi, ν = c, m denotes the Cr and Mn, respectively, and s = σ/2 is the local electron spin. The main advantage of such a superlattice is the two types of magnetic ions have opposite signs of exchange coupling parameters in TI materials, namely, Jm < 0 and Jc > 0. This is experi- mentally verified by opposite signs of the anomalous Hall conductance in the insulating regime of Mn-doped [21] and Cr-doped [22] Bi2Te3 family materials. Therefore a uniform magnetization in the superlattice will induce op- posite exchange fields in the upper and lower halves of a TI layer. The Hamiltonian describing the superlattice can be written as H = vF τ z (z × σ) · k(cid:107)δi,j + maτ zσzδi,j + mbσzδi,j (cid:88) (cid:104) k(cid:107),i,j +tsτ xδi,j + tn 2 τ +δi+1,j + τ−δi−1,j tn 2 † c k(cid:107)i ck(cid:107)j . (2) (cid:105) where i and j label distinct magnetic TI layers, σβ and τ β (β = x, y, z) are Pauli matrices acting on the spin and the top/bottom surface of the parent TI layer, respectively. The first term in the Hamiltonian describes the top and bottom surface states of a parent TI layer, where a single 2D Dirac node is considered for Bi2Te3 family materi- als [23]. vF is the Fermi velocity. k(cid:107) = (kx, ky) is the in- plane momentum. The second and third terms describe the Zeeman-type spin splitting for surface states induced by the ferromagnetic (FM) exchange couplings ∆c of Cr and ∆m of Mn along z axis, where ma = (∆c − ∆m)/2 is the staggered Zeeman field and mb = (∆c + ∆m)/2 is the uniform Zeeman field [24]. In the mean field (MF) ap- Cr-doped TINIMn-doped TI(a)(b)CrCr-doped TI3 TABLE I. The parameters of ts, sign of mean J mc eff and pos- sible magnetic order along z direction in Cr-doped and Mn- doped (Bi0.1Sb0.9)2Te3 superlattice with different thickness. Here we set (cid:96)m = (cid:96)c, and assume tn (cid:54)= 0. (cid:96)m + (cid:96)c 2 QL 4 QL 6 QL ts (eV) 0.116 0.029 0.004 sgn (J mc eff ) − + − possible order FM AFM FM AI Yes ? Yes (cid:82) π/a3−π/a3 eff ((cid:96)z) ∝ −JmJc eff < 0 for Mn-doped TI film [21], J cc been shown to be FM with an easy axis z, which in- dicates J mm eff < 0 for Cr-doped TI film [22] and γ < 1. The sign of J mc eff is determined by the Ruderman-Kittel-Kasuya- Yoshida (RKKY) type interaction [29] along the z direc- tion. J mc dqzχzz(qz)eiqz(cid:96)z , where a3 = 1 nm is the z direction lattice constant of parent TI material, i.e., thickness of a quintuple layer (QL). χzz(qz) is the z direction magnetic susceptibility of TI with mo- mentum q = (0, 0, qz) obtained by Kubo formula [29]. (cid:96)z is the vertical distance between Mn and Cr, which we set to their mean distance in a magnetic TI layer as (cid:96)z = ((cid:96)m + (cid:96)c)/2. The calculated sign of J mc eff ((cid:96)z) oscil- lates as a function of (cid:96)z [30], is listed in Table I. The sign of J mc eff is opposite to that in Ref. [30] since JmJc < 0. We note that the exact sign of the interlayer coupling has not been settled yet by experiments. According to Table I, J mc eff < 0 for (cid:96)m = (cid:96)c = 1, 3 QL, possibly leading to a FM ground state, and the system becomes an AI. For (cid:96)m = (cid:96)c = 2 QL, the system may develop an AFM order, yet one can still reach an AI state by tuning tn. Next we show the axion θ becomes a dynamical field θ = θ0 + δθ(x, t) in the presence of the FM fluctua- tions. The magnetic fluctuations in the TI originate from the quantum nature of spin interactions (for γ (cid:54)= 0) and the thermal fluctuations. For convenience we de- fine the magnetization per unit volume of Cr and Mn as Mν = gLµByν(cid:104)Sν(cid:105)/a3 with ν = c, m. Here gL is the Land´e factor, µB is the Bohr magneton, a is the average lattice constant of TI. They can be regrouped into the FM and AFM magnetization as M± = (Mc ± Mm)/2. In the below, we assume yc = ym = y. The fluctua- tion of M± can be generally written as M± = (M± 0 + δM± y (x, t)y. To the linear order, it can be deduced from Eq. (5) that the axion field θ is only coupled to d5 = ma = (∆c − ∆m)/2 ∝ M + z . Therefore, only the FM fluctuations along z axis δM + z are relevant. The corresponding effective Lagrangian is LM = KM [(∂tδM + z )2], where KM , vi and ms are the stiffness, velocity and mass of the spin-wave mode δM + z . The fluctuation of θ is now given by δθ(x, t) = δma(x, t)/g = (Ja3/4gLµBg)δM + z (x, t), where the coefficient g can be determined from Eq. (5). z (x, t))z + δM± z )2 − (vi∂iδM + x (x, t)x + δM± z )2 − m2 s(δM + FIG. 2. (color online). (a) Phase diagram of the proposed magnetic TI superlattice with two variables: ts and tn. An AI phase emerges with nonzero θ. When ma > mb, the phase diagram will be AI phase only (not shown). (b) Typical value of θ as a function of tn − ts in the AI phase, where the pa- rameter is setted as tn + ts = 1. as [13, 14], (cid:90) where d = ((cid:80)5 1 4π θ = d3k 2d + d4 (d + d4)2d3 ijkldi∂xdj∂ydk∂zdl, (5) a=1 d2 a)1/2, and the repeated index i, j, k, l take values from 1, 2, 3, 5 and indicates summation. Typ- ical θ values in AI phase is calculated in Fig. 2(b). As expected, θ deviates gradually from 0 or π as ma in- creases away from 0. When tn/ts (cid:29) 1, the θ value tends to π; for tn/ts → 0, θ converges quickly towards zero. Therefore, θ can be tuned by the layer thickness. Note that θ is only well defined in the insulating regime when b − m2 m2 c1. For ma > mb, this condition is always satisfied and the whole phase diagram in Fig. 2(a) will be occupied by the AI phase only. Physically, ma > mb means the magnetic moments in Mn and Cr are polarized in the same direction. Different from the previous pro- posals [13, 14], such nonquantized θ is coupled to the FM order parameter instead of AFM order, due to opposite signs of Jc and Jm. a < t2 a < t2 To realize the AI phase, the system should have an appropriate magnetic ordering. If ma > mb, the sys- tem will have a FM long range order along z axis and become an AI. Conversely if ma < mb, the system will have an AFM long range order, where the spins of Mn and Cr in each parent magnetic TI layer will point along the +z and −z direction, respectively. In this case, b − m2 m2 c1 must be satisfied to realize the AI, which may be fulfilled by adjusting the doping concentration yc,m and tuning the layer thicknesses (cid:96)c,m,n. The mag- netic properties of this system are determined by the ef- fective interaction between neighbouring magnetic impu- (cid:107) (cid:107) rity spins J ρν ρ(xi) · S ν (xj) + γS ν(xj)], where (cid:107) S ρ denotes the in-plane impurity spin, and ρ, ν = c, m labels the ion type. Such effective spin interactions are mediated by the band electrons of TIs [26 -- 28]. The in- teractions between the same types of magnetic ions have ρ (xi)Sz eff [Sz 0022bamm−22bamm−3DQAHWeylSemimetalAxionInsulator(a)(b)0.50.40.10.20.30.00.51.00.0-0.5-1.00abmm==0.10.05abmm==nstt−2θπ0.40.05abmm==stntAxionInsulator(cid:18) L = LMaxwell + Lθ + Laxion α + B2 = (cid:19) The effective Lagrangian density describing the axion coupled electromagnetic response is then given by 1 8π +g2KA E2 − 1 µ (cid:2)(∂tδθ)2 − (vi∂iδθ)2 − m2 sδθ2(cid:3) , 4π2 (θ0 + δθ) E · B (6) eff − 2J mc where the three terms describe the conventional Maxwell action, the topological coupling between the axion and the electromagnetic field, and the dynamics of the mas- sive axion. KA = KM (4µBgL/Ja3)2. The axion mass at temperature T is ms ∼ kBT − JF /, where JF = eff /2 is of the same order as the Curie J cc eff +J mm temperature and decays exponentially with the mean dis- tance between magnetic ions ξ = y−1/3a. The coeffi- cient KM ∼ ξ3/g2 Bms, while the velocity vi ∼ ξms. in a typical magnetic TI system, For an estimation, ms ∼ kBT ∼ JF ∼ 1 meV, the bulk gap is ts − tn ∼ 0.1 eV and the Zeeman field is ma ∼ 0.06 eV. Therefore, ms (cid:28) ma < ts − tn, justifying the above low-energy description of the system. Lµ2 The coupling between the dynamic axion field θ and the electromagnetic field gives rise to a number of novel topological phenomena, which can be used in experi- ments as a unique signature of dynamic θ. For instance, it leads to the formation of axion polariton, which be- comes gapped in the presence of a background mag- netic field. It also leads to the double frequency re- sponse on the cantilever torque magnetometry [13]. Here we mention another interesting phenomena proposed in Ref. [15], that the massive axion in L exhibits an in- stability in the presence of an external electric field E0. lead to a complete screening Such an instability will of electric field above a critical value Ecrit. In other words, when 0E0 < Ecrit, the field inside the sys- tem is E = 0E0/ and B = 0; when 0E0 > Ecrit, one gets E = Ecrit and B = ±(cid:112)µEcrit(0E0 − Ecrit). Ecrit = (ms/α)(cid:112)8π3g2KA/µ. For θ0 = 0, a second- Here 0 is the dielectric constant outside the system, and order phase transition happens at 0E0 = Ecrit, while for θ0 (cid:54)= 0, the phase transition becomes a crossover. More details on these experimental proposals are presented in the Supplemental Material [31]. The relative permit- tivity, axion mass, and axion coupling of the magnetic TI system are estimated to be ∼ 100, ms ∼ 1 meV, g ∼ 0.08 eV, J ∼ 1.5 eV, y ∼ 0.1, and a = 0.5 nm. This gives Ecrit ∼ (g/J)(cid:112)ms/ya3 = 2 × 106 V/m, which ments. The critical field Ecrit ∝(cid:112)ms/y could be reduced is much smaller than the breakdown field of the typi- cal semiconductors and in the range accessible by experi- by adjusting the doping concentration y of the system. For extremely low temperatures, ms ∼ JF ∝ e−λy−1/3 , and Ecrit becomes smaller as y decreases. For relatively high temperatures when ms ∼ kBT is independent of y, Ecrit will be reduced as y increases. 4 In the above discussion, we show that to realize the dynamic axion, it is not essential to start from a nontriv- ial TI or a magnetic order. In fact, in such a TI system, the effect of the dynamical axion may be suppressed in the bulk since the electromagnetic field mainly couples to the surface states [13]. Instead, a topologically triv- ial insulator with magnetic fluctuations properly coupled to the electrons is also able to produce dynamic axions, and the low-energy physics is dominated by the bulk. This motivates us to propose the second dynamic axion system which is a PM insulator. Such a system can be realized by doping magnetic elements such as Cr into 3D TI materials to the vicinity of the topological QCP, for example, Bi2(SexTe1−x)3 with x ≥ 0.66 [32]. The sys- tem is topologically trivial and exhibits a PM response at low temperature, which is caused by the reduced ef- fective spin-orbit coupling strength of CryBi2−y result- ing from the Cr substitution of Bi. The Hamiltonian of the system is the Dirac model Hb = (cid:80)5 a=1(cid:101)da(k)(cid:101)Γa as in Ref. [13], (cid:101)da(k) = (sin kx, sin ky, sin kz, m4(k), m5), (cid:101)Γa = (τxσx, τxσy, τy, τz, τxσz), τ i refers to orbit index. m4 is topologically trivial mass, while m5 = 0 on aver- age, leading to a mean value θ0 = 0. The AFM fluc- tuation δM− z of Cr spins inside a unit cell will induce a fluctuation δm5, leading to a dynamical axion field δθ = δm5/g [31]. The advantage of such a system is that it is close to the PM to FM transition [32], therefore the magnetic fluctuation is strong and the axion mass ms is small. To distinguish with the previous proposals, this material may be called topological PM insulator which is a TRI AI with a dynamic axion field. In summary, we show that the dynamical axion field can be realized in a magnetic TI superlattice. We empha- size that each magnetic TI layer does not need to exhibit QAH effect, but only a uniform magnetization is neces- sary. We hope the theoretical work here could aid the search for the axionic state of matter in real materials. This work is supported by the US Department of En- ergy, Office of Basic Energy Sciences, Division of Mate- rials Sciences and Engineering, under Contract No. DE- AC02-76SF00515 and in part by the NSF under grant No. DMR-1305677. [1] D. J. Thouless, Topological Quantum Numbers in Non- relativistic Physics (World Scientific, Singapore, 1998). [2] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [3] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [4] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, 195424 (2008). [5] R. D. Peccei and H. R. Quinn, Phys. Rev. Lett. 38, 1440 (1977). [6] F. Wilczek, Phys. Rev. Lett. 58, 1799 (1987). 5 [7] X.-L. Qi, R. Li, J. Zang, and S.-C. Zhang, Science 323, [21] J. G. Checkelsky, J. Ye, Y. Onose, Y. Iwasa, and 1184 (2009). [8] J. Maciejko, X.-L. Qi, H. D. Drew, and S.-C. Zhang, Phys. Rev. Lett. 105, 166803 (2010). [9] W.-K. Tse and A. H. MacDonald, Phys. Rev. Lett. 105, 057401 (2010). [10] K. Nomura and N. Nagaosa, Phys. Rev. Lett. 106, 166802 (2011). Y. Tokura, Nature Phys. 8, 729 (2012). [22] 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). [23] H. Zhang, C.-X. Liu, X.-L. Qi, X. Dai, Z. Fang, and [11] J. Wang, B. Lian, X.-L. Qi, and S.-C. Zhang, Phys. Rev. S.-C. Zhang, Nature Phys. 5, 438 (2009). B 92, 081107 (2015). [24] J. Wang, B. Lian, and S.-C. Zhang, Phys. Rev. B 89, [12] T. Morimoto, A. Furusaki, and N. Nagaosa, Phys. Rev. 085106 (2014). B 92, 085113 (2015). [25] A. A. Burkov and L. Balents, Phys. Rev. Lett. 107, [13] R. Li, J. Wang, X. L. Qi, and S. C. Zhang, Nat. Phys. 127205 (2011). 6, 284 (2010). [26] Q. Liu, C.-X. Liu, C. Xu, X.-L. Qi, and S.-C. Zhang, [14] J. Wang, R. Li, S.-C. Zhang, and X.-L. Qi, Phys. Rev. Phys. Rev. Lett. 102, 156603 (2009). Lett. 106, 126403 (2011). [27] R. Yu, W. Zhang, H.-J. Zhang, S.-C. Zhang, X. Dai, and [15] H. Ooguri and M. Oshikawa, Phys. Rev. Lett. 108, Z. Fang, Science 329, 61 (2010). 161803 (2012). [28] J. Wang, B. Lian, and S.-C. Zhang, Phys. Rev. Lett. [16] X. Wan, A. Vishwanath, and S. Y. Savrasov, Phys. Rev. 115, 036805 (2015). Lett. 108, 146601 (2012). [17] C. H. Kim, H. S. Kim, H. Jeong, H. Jin, and J. Yu, Phys. Rev. Lett. 108, 106401 (2012). [18] K. Shiozaki and S. Fujimoto, Phys. Rev. B 89, 054506 (2014). [19] P. Goswami and B. Roy, Phys. Rev. B 90, 041301 (2014). [20] Y. Zhang, K. He, C.-Z. Chang, C.-L. Song, L.-L. Wang, X. Chen, J.-F. Jia, Z. Fang, X. Dai, W.-Y. Shan, S.-Q. Shen, Q. Niu, X.-L. Qi, S.-C. Zhang, X.-C. Ma, and Q.-K. Xue, Nature Phys. 6, 584 (2010). [29] T. Dietl and H. Ohno, Rev. Mod. Phys. 86, 187 (2014). [30] M. Li, W. Cui, J. Yu, Z. Dai, Z. Wang, F. Katmis, and J. Moodera, Phys. Rev. B 91, 014427 W. Guo, (2015). [31] See Supplemental Material for details on (i) the electric field screening, and (ii) derivation of the Lagrangian for dynamic axion field. [32] J. Zhang, C.-Z. Chang, P. Tang, Z. Zhang, X. Feng, K. Li, L.-l. Wang, X. Chen, C. Liu, W. Duan, K. He, Q.-K. Xue, X. Ma, and Y. Wang, Science 339, 1582 (2013).
1606.03698	2	1606	2016-10-11T22:42:03	Nonsymmorphic symmetry-required band crossings in topological semimetals	[ "cond-mat.mes-hall", "cond-mat.str-el", "quant-ph" ]	We show that for two-band systems nonsymmorphic symmetries may enforce the existence of band crossings in the bulk, which realize Fermi surfaces of reduced dimensionality. We find that these unavoidable crossings originate from the momentum dependence of the nonsymmorphic symmetry, which puts strong restrictions on the global structure of the band configurations. Three different types of nonsymmorphic symmetries are considered: (i) a unitary nonsymmorphic symmetry, (ii) a nonsymmorphic magnetic symmetry, and (iii) a nonsymmorphic symmetry combined with inversion. For nonsymmorphic symmetries of the latter two types, the band crossings are located at high-symmetry points of the Brillouin zone, with their exact positions being determined by the algebra of the symmetry operators. To characterize these band degeneracies we introduce a \emph{global} topological charge and show that it is of $\mathbb{Z}_2$ type, which is in contrast to the \emph{local} topological charge of Fermi points in, say, Weyl semimetals. To illustrate these concepts, we discuss the $\pi$-flux state as well as the SSH model at its critical point and show that these two models fit nicely into our general framework of nonsymmorphic two-band systems.	cond-mat.mes-hall	cond-mat	Nonsymmorphic symmetry-required band crossings in topological semimetals Y. X. Zhao1, ∗ and Andreas P. Schnyder1, † 1Max-Planck-Institute for Solid State Research, D-70569 Stuttgart, Germany (Dated: October 13, 2016) We show that for two-band systems nonsymmorphic symmetries may enforce the existence of band crossings in the bulk, which realize Fermi surfaces of reduced dimensionality. We find that these unavoidable crossings originate from the momentum dependence of the nonsymmorphic symmetry, which puts strong restrictions on the global structure of the band configurations. Three different types of nonsymmorphic symmetries are considered: (i) a unitary nonsymmorphic symmetry, (ii) a nonsymmorphic magnetic symmetry, and (iii) a nonsymmorphic symmetry combined with inversion. For nonsymmorphic symmetries of the latter two types, the band crossings are located at high- symmetry points of the Brillouin zone, with their exact positions being determined by the algebra of the symmetry operators. To characterize these band degeneracies we introduce a global topological charge and show that it is of Z2 type, which is in contrast to the local topological charge of Fermi points in, say, Weyl semimetals. To illustrate these concepts, we discuss the π-flux state as well as the SSH model at its critical point and show that these two models fit nicely into our general framework of nonsymmorphic two-band systems. PACS numbers: 03.65.Vf, 71.20.-b, 73.20.-r,71.90.+q I. INTRODUCTION Since the experimental discovery of topological insula- tors1,2, symmetry protected topological phases have be- come a major research subject3–8. Recent studies have been concerned with topological phases that are pro- tected by spatial symmetries, such as topological crys- talline insulators9–12 and topological semimetals stabi- lized by reflection, inversion, or other crystal symme- tries13–15. Until recently, the study of these topologi- cal crystalline materials has focused on the role of point group symmetries. However, besides point group symme- tries the space group of a crystal can also contain non- symmorphic symmetries, which are combinations of point group operations with nonprimitive lattice translations. It has been shown that the presence of nonsymmorphic symmetries leads to new topological phases, which can be insulating16–21, or semimetallic with Dirac points pro- tected by nonsymmorphic symmetries22,23. In the latter case, the Dirac points possess local topological charges, which guarantees their local stability. However, as we show in this paper, nonsymmorphic symmetries restrict the form of the band structure not only locally but also globally, which may lead to unavoid- able band crossings in the bulk24–28. Indeed, the nonsym- morphic symmetries can put so strong constraints on the global properties of the band structure that the system is required by symmetry to be in a topological semimetal phase, with Fermi surfaces of reduced dimensionality. These symmetry-enforced semimetals possess low-energy excitations with unconventional dispersions and may ex- hibit novel topological response phenomena and unusual magneto-transport properties. In the following we con- sider three different types of nonsymmorphic symmetries: (i) Unitary nonsymmorphic symmetries, (ii) nonsymmor- phic symmetries combined with inversion, and (iii) non- symmorphic magnetic symmetries. We first rigorously prove that for any one-dimensional (1D) two-band system unitary nonsymmorphic symmetries enforce the existence of band crossings, due to global topological constraints on the band structure. In the presence of an additional inversion symmetry, the symmetry enforced band degen- eracies are located either at the origin or at the boundary of the Brillouin zone (BZ), depending on the algebra of the symmetry operators. The same holds true for non- symmorphic magnetic symmetries, which are composed of a unitary nonsymmorphic symmetry followed by an anti-unitary time-reversal symmetry. We present gener- alizations of these results to higher dimensions, for which nonsymmorphic symmetries may enforce the existence of zero- or higher-dimensional band crossings. In all of the above cases we find that the nonsymmorphic symmetries restrict the momentum space structure in the BZ both lo- cally and globally. To characterize the global topological features we introduce a novel global topological charge, which as we show, is always of Z2 type. Hence, the global topological features exhibit a Z2 classification, which is in contrast to the local topological characteristics, which possess a Z classification. Finally, we illustrate these findings by considering two prototypical examples: (i) the π-flux square lattice model and (ii) the SSH model at its critical point. Within our unified framework, we show that the former model can be viewed as the higher- dimensional generalization of the latter. II. UNITARY NONSYMMORPHIC SYMMETRY We start by considering a general 1D two-band Hamil- tonian H(k) with the two-fold unitary nonsymmorphic symmetry (cid:18)0 e−ik (cid:19) , (1) G(k) = 1 0 6 1 0 2 t c O 1 1 ] l l a h - s e m . t a m - d n o c [ 2 v 8 9 6 3 0 . 6 0 6 1 : v i X r a III. NONSYMMORPHIC SYMMETRY COMBINED WITH INVERSION SYMMETRY 2 We note that while a unitary nonsymmorphic symme- try guarantees the existence of a band crossing point, it does not fix the position of this degeneracy point in momentum space. However, in the presence of an additional inversion symmetry, the band crossings are pinned to either the origin or the boundary of the BZ. To demonstrate this, let us consider the inversion symmetry P = σ2ı, where ı inverses the momentum. We find that P 2 = −1. (6) [H, P ] = 0, P G(k) P −1 = −GT (−k), n qneink. Since q(k) is a periodic function, we expand it as q(k) = It follows from Eq. (4) that q−(n+1) = q∗ n, which, as a recursion relation, allows us to express q(k) as (cid:80) ∞(cid:88) n=0 ∞(cid:88) n=0 q(k) = (qneink + q∗ ne−i(n+1)k). (7) From Eq. (6) it follows that σ2H(−k)σ2 = H(k), which implies q(k) = −q∗(−k) or equivalently qn = −q∗ n. Since qn are all purely imaginary, we find that q(k) = (eink − e−i(n+1)k), λn i (8) with λn being real numbers. We observe that indepen- dent of λn there always exists a band crossing point at k = 0. For example, by keeping only the zeroth term in Eq. (8), one finds as a simple concrete model, H0(k) = λ sin kσx + λ(1 − cos k)σy. (9) We note that the nonsymmorphic symmetry G(k) re- lates seemingly independent terms to each other in the Hamiltonian. This is exemplified by Eq. (9), where all three terms (which are usually independent) have the same coefficients. Obviously, higher order terms in Eq. (8), which constitute symmetry-preserving pertur- bations, cannot split the band crossing point of H0 at k = 0. That is, the gapless mode at k = 0 described by the low-energy effective Hamiltonian Heff(k) = λkσx is stable against symmetry-preserving perturbations. The fact that the Hamiltonian given by Eq.(8) exhibits a band crossing at k = 0 can directly be seen by comput- ing the eigenstate of G(k) and H(k). Because G(k) and H(k) commute [see Eq.(2)], they can be simultaneously diagonalized by the same set of eigenstates H(k)±, k(cid:105) = E±(k)±, k(cid:105), G(k)±, k(cid:105) = g±(k)±, k(cid:105), (10) where the eigenfunctions ±, k(cid:105) are given by (cid:18) 1 (cid:19) ei k 2 +, k(cid:105) = 1√ 2 −, k(cid:105) = , 1√ 2 (cid:19) (cid:18) 1 −ei k 2 (11) Illustration of nonsymmorphic symmetries in one- FIG. 1: (a) The nonsymmorphic symmetry is dimensional lattices. composed of a π rotation followed by a half translation a/2, where a is the lattice constant. (b) The nonsymmorphic mag- netic symmetry is composed of the two operations of (a) fol- lowed by the exchange of black and white balls, which repre- sents time-reversal symmetry. which acts on H(k) as G(k)H(k)G−1(k) = H(k). (2) Since G2(k) = e−ikσ0, the eigenvalues of G(k) are ±e−ik/2. Therefore, the nonsymmorphic symmetry G(k) can be viewed as an operation on internal degrees of free- dom (e.g., pseudospin) followed by a half translation, as illustrated in Fig. 1(a). Observing that G(k) anti- commutes with σ3, the Hamiltonian can be written as (cid:18) 0 (cid:19) H(k) = q∗(k) q(k) 0 . (3) Without loss of generality, we have dropped the term proportional to the identity, which only shifts the energy of eigenstates. Inserting Eqs. (3) and (1) into Eq. (1), we find that due to the non-symmorphic symmetry G(k), q(k) must satisfy q(k)eik = q∗(k). (4) We claim that any periodic function q(k) satisfying Eq. (4) has zeros, and thus any two-band model with the nonsymmorphic symmetry (1) is required to be gap- less. To see this, we introduce f (z) = q(k) with z = eik, from which it follows that zf (z) = f∗(z). If q(k) or f (z) is nonzero on the unit circle S1, then z = f∗(z)/f (z), (5) which however is impossible. This is because for z ∈ S1, the two sides of Eq. (5) both define functions from S1 to S1, but the left-hand side has winding number 1, while the winding of the right-hand side is even, since f∗(z)/f (z) = e2iArc[f (z)]. Thus, q(k) must vanish at some momentum by contradiction. For the topological argu- ment to work for multi-band theories, we may replace q(k) in Eq.(3) by the determinant of the off-diagonal en- try, which is discussed in Sec.VII. 3 FIG. 2: (a) Energy spectrum E±(k) of Hamiltonian (9). Blue and orange correspond to the eigenstates E+ and E−, respectively. The two eigenstates are connected smoothly at the boundary of the BZ and cross each other at the center of the BZ. (b) E±(k) as a function of the phases φ of the nonsymmorphic symmetry eigenvalue g±(k). In the space of the eigenvalues g±(k) of the nonsymmorphic symmetry, the two bands are smoothly connected with each other, without any crossing point. (c) k as a function of the phase φ of the eigenvalues g±(k). The two eigenvalue branches are connected at φ = π 2 (= − π 2 ), leading to a winding number 2. (d) Trajectory of the two bands in the (k, φ,E) space. As a problem of essential three parameters, the two bands are connected as a circle in the (k, φ,E) space, corresponding to (2, 1) ∈ H1(S1 × S1 × R) ∼= Z ⊕ Z. 2 and φ = 3π and the eigenvalues are E±(k) = ±2 λn sin ∞(cid:88) n=0 (cid:18) nk + (cid:19) k 2 g±(k) = ±e−i k 2 . , (12) We find that the energy and nonsymmorphic symme- try eigenvalues of +, k(cid:105) at k = ±π are continuously connected to the corresponding eigenvalues of −, k(cid:105) at k = ∓π, see Fig.2. That is, we have (cid:40)E+(−π) = E−(π) E−(−π) = E+(π) (cid:40) , g+(−π) = g−(π) g−(−π) = g+(π) . (13) We note that the eigenfunctions ±, k(cid:105) become degen- erate in energy at k = 0 [i.e., E+(0) = E−(0)], while their nonsymmorpic symmetry eigenvalue remains non- degenerate at k = 0 [i.e., g+(0) (cid:54)= g−(0)]. Therefore the two bands ±, k(cid:105) must cross each other. To see the topological features of the band structure, we first note that the eigenvalues g±(k) of the nonsym- morphic symmetries G form a manifold as a function of momentum k. That is, the eigenvalues g±(k) are mul- tivalued functions of k, with different branches being smoothly connected. Inversely, k is a single-valued con- tinuous function of the eigenvalues of the symmetry G. For the two-fold nonsymmorphic symmetry (1), the mo- mentum k ∈ S1 has winding number 2 as a function of the eigenvalue g±(k) ∈ U (1), which indicates a non- trivial topology [see Fig.2(c)]. To better understand this nontrivial topology, it is instructive to draw the mutual dependence of the energy eigenvalues E±, the nonsym- morphic eigenvalues g±, and the momentum k in terms of a trajectory in the three-dimensional space (k, φ,E). For the two-band model (9) this is shown in Fig. 2(d). The projections of this trajectory onto the three orthogonal planes (E, k), (E, φ), and (k, φ) are shown in Figs.2(a), 2(b) and 2(c), respectively. We can see that the two bands E± are connected as a circle in (k, φ,E) space, cor- responding to the element (2, 1) in the homology group H1(S1 × S1 × R, Z) ∼= Z ⊕ Z. Instead of P = σ2ı, another possible choice for P is P = σ1ı with the symmetry relations [H, P ] = 0, P G(k) P −1 = GT (−k), P 2 = 1. (14) With this choice, we find the following relations for q(k) and qn, q(k) = q∗(−k), qn = q∗ n. (15) (16) Using Eq. (7), it follows that ∞(cid:88) q(k) = λn(eink + e−i(n+1)k). n=0 Hence, there always exists a band crossing point at k = π. Let us now show that the algebra obeyed by the sym- metry operators determines whether the band crossing point is at k = 0 or k = π. To that end, we recall that for the choice P = σ2ı the operators at the inversion in- variant point k = π, P = σ2ı, G(π) = −iσ2, and H(π) are mutually commuting, see Eq. (6). At the other inversion invariant point k = 0, however, P = σ2ı and G(0) = σ1 are anti-commuting, while H(0) commutes with P and G(0), i.e., [H(0), P ] = 0 and [H(0), G(0)] = 0. It follows that the two degenerate eigenstates of H at k = 0 can be written as eigenstates of P with different eigenvalues. 2 −, 0(cid:105) is an eigen- 2 +, 0(cid:105) + 1−i Explicitly, we find that 1+i 2 +, 0(cid:105)+ 1+i 2 −, 0(cid:105) state of P with eigenvalue +1, while 1−i is an eigenstate of P with eigenvalue −1. Therefore, the band crossing, which is protected by P , occurs at k = 0. A similar analysis can be performed for the choice P = σ1ı, i.e., the Hamiltonian given by Eq. (16). In that case, we find that at k = 0 the operators H(0), G(0), and P are mutually commuting, while P and G(k) anti-commute at k = π, where the band degeneracy is located. We conclude that the algebraic relations obeyed by the symmetry operators determine the location of the symmetry-enforced band crossing, see Table I. IV. NONSYMMORPHIC MAGNETIC SYMMETRY From the discussion in the previous section it follows that not all the symmetry constraints are necessary to en- force the existence of the band crossing. As we shall see, a single nonsymmorphic antiunitary symmetry, namely a magnetic nonsymmorphic symmetry, is sufficient to en- sure the existence of a band crossing at k = 0 or k = π. As illustrated in Fig. 1(b), a magnetic nonsymmorphic symmetry can be viewed as the combination of a nonsym- morphic symmetry G(k) with a time-reversal symmetry T . We only require that the combined symmetry G T is satisfied. That is, both G and T may be broken individ- ually, but the combination must be preserved. In what follows we assume that T 2 = +1 and consider two possi- ble choices for T , namely, (i) T = Kı and (ii) T = σ3 Kı, where K denotes the complex conjugation operator. By use of Eq. (1), we find that in case (i) [ T , G(k)] = 0, while in case (ii) { T , G(k)} = 0. Let us start with the discussion of case (i), where the combined symmetry G = G T = G(k) Kı acts on H(k) as [ G(k),H(k)] = 0. q(−k), and qn = q−(n+1) by use of q(k) = (cid:80) Inserting Eq.(3) into Eq. (17), one obtains eikq(k) = n qneink, (17) which implies ∞(cid:88) ∞(cid:88) q(k) = qn(eink + e−i(n+1)k). (18) n=0 We observe that due to the symmetry constraint Eq. (17), q(π) = 0, i.e., there is a band crossing point at k = π. Note that the term fz(k)σ3 is not forbidden by the combined symmetry G T . However, Eq. (17) requires that fz(k) is an odd function of k. Hence fz(k)σ3 must vanish at the high-symmetry points k = 0 and k = π, due to the periodicity of the BZ, i.e., fz(k) = fz(k + 2π). To summarize, the magnetic nonsymmorphic symmetry G T is sufficient to enforce the existence of a band crossing at k = π. Next, we discuss choice (ii) for T , in which case the combined symmetry G(cid:48) ≡ G T = Gσ3 Kı acts on H(k) as [ G(cid:48)(k),H(k)] = 0. (19) Combining Eq. (3) with Eq. (19) we find that eikq(k) = −q(−k) . Hence, the Fourier components qn must satisfy qn = −q−(n+1), which yields q(k) = qn(eink − e−i(n+1)k). (20) n=0 Since q(0) = 0 in Eq. (20), there is an unavoidable band crossing at k = 0. As before, the term fz(k)σ3 is symme- try allowed, with fz(k) an odd function. Hence, fz(k)σ3 4 Position k = 0 k = π G, P G T P G = −G† P {G, T} = 0 P G = G† P [G, T ] = 0 TABLE I: The positions of the band crossings in the BZ are determined by the algebra of the symmetry operators. vanishes at the high-symmetry points k = 0 and k = π, and therefore cannot gap out the band crossing point. From the above discussion, one infers that the com- mutation relation between T and G(k) determines the position of the band-degeneracy points. Namely, for [G, T ] = 0 [case (i)] and {G, T} = 0 [case (ii)], we have (G T )2 = +e−ikσ0 and (G T )2 = −e−ikσ0, respectively. Hence, we find that for case (i) (G T )2 = −1 at k = π, while for case (ii) (G T )2 = −1 at k = 0. Since G T is an anti-unitary operator, (G T )2 = −1 leads to a band degeneracy, in analogy to Kramers theorem. Thus, for [G, T ] = 0 the band-crossing point is at k = π, while for {G, T} = 0 it is at k = 0, see Table I. We note that a number of recent works16,17,21,29 have discussed edge band structures of two-dimensional (2D) nonsymmorphic insulators that are similar to the 1D bulk band structures studied here. In contrast to conven- tional topological insulators, the edge bands of these 2D nonsymmorphic insulators do not connect valence and conduction bands. Hence, our results for the bulk band structures of 1D systems, can be applied directly to the edge spectrum of these 2D nonsymmorphic insulators. This suggests, in particular, that the crossing of the edge bands of these 2D systems is, at least in some cases, en- forced by the nonsymmorphic symmetry of the edge the- ory. In closing this section, we note that the existence of a band crossing cannot be enforced by a nonsymmorphic particle-hole symmetry, which is discussed in detail in Appendix A. V. TOPOLOGICAL CLASSIFICATION OF BAND-CROSSING POINTS Let us now derive the classification of the global topo- logical properties of the considered Hamiltonians. By global topology, we mean a band structure is allowed to be deformed smoothly in the whole momentum space with the symmetries being preserved, in contrast to the ordinary local one, where deformations are restricted only in a open neighborhood around the band-crossing point. The group structure is given by the direct sum of the Hamiltonians. To that end, we study whether the band crossings of the doubled Hamiltonians H ⊗ τ0 and H⊗τ3 can be gapped out by symmetry-preserving terms. Here, τµ's represent an additional set of Pauli matrices and τ0 is the 2 × 2 identity matrix. The symmetry oper- ators for the doubled Hamiltonians are G(k)⊗ τ0, P ⊗ τ0, and T ⊗ τ0. We observe that diag(H, λH) can be contin- uously deformed to diag(H,−λH) without breaking the symmetries and without opening a gap at k = 0 or π. In addition, we find that mσ0⊗ τ1 is a symmetry preserving mass term that gaps out the spectrum of H ⊗ τ3 in the entire BZ. It follows that the global topological features of H possess a Z2 classification, namely even number of copies can be gapped with symmetries being preserved, while odd number cannot. It is noted that although in the above simple deformation the gap is fully closed at λ = 0, a more carefully chosen deformation can be made such that at any intermediate stage band crossing hap- pens only at finite number of momenta. To infer the classification of the local topological prop- erties of the band crossing points, we enclose the degen- eracy point by an S0 sphere (consisting of two points on the left and right of the degeneracy point) and con- sider adiabatic deformations that do not close the gap on the chosen S0. The only possible gap opening term is fz(k)σ3, which however vanishes at the high-symmetry points k = 0, π due to the nonsymmorphic symmetry. This also holds for multiple copies of H. From this we conclude that the local topological features of the band crossing points exhibit a Z classification. VI. HIGHER-DIMENSIONAL GENERALIZATIONS Our results for symmetry-enforced band crossings in 1D can be readily generalized to higher dimensions. We assume that the fractional translation is along the kx direction for d-dimensional systems. The d-dimensional Hamiltonian H(k) can then be decomposed into a fam- ily of 1D Hamiltonians hk⊥ (kx) = H(kx, k⊥), which are parametrized by the (d − 1) momenta k⊥ perpendicu- lar to kx. Let us briefly discuss how the three different types of nonsymmorphic symmetries that we considered above constrain this d-dimensional Hamiltonian. (i) If H(k) is invariant under a unitary nonsymmorphic sym- metry G(kx), then there are in general several branches of Fermi surfaces (possibly of dimension d > 0) that are parametrized by k⊥. (ii) If there exists in addition a reflection symmetry reversing kx, then the Fermi sur- faces are pinned at kx = 0 or kx = π, depending on the algebraic relations obeyed by the symmetry opera- tors, as specified in Table I. (iii) If we consider symme- tries that relate k to −k, such as a nonsymmorphic mag- netic symmetry (or an additional inversion), then there exist 2d−1 1D inversion invariant subsystems hka⊥ (kx) (a = 1,··· , 2d−1) of hk⊥ (kx) , which are labeled by the perpendicular momenta ka⊥ that are invariant under k⊥ → −k⊥. These subsystems have band degeneracies at kx = 0 or kx = π, as determined by the algebraic rela- tions in Table I. The other 1D subsystems hk⊥ , where k⊥ is not invariant under k⊥ → −k⊥, are generally gapped. 5 FIG. 3: Illustration of the SSH model. (a) For α (cid:54)= 0 the SSH model breaks the nonsymmorphic symmetry G(k), Eq. (1). (b) At the critical point α = 0 the translation symmetry is promoted from 2Z to Z, such that the nonsymmorphic sym- metry G(k) is satisfied. (c) Energy spectrum at α = 0 in (I) the original BZ and (II) the unfolded BZ. A. Examples We illustrate our theoretical results by two examples: (i) The Su-Schrieffer-Heeger (SSH) model30 and (ii) the π-flux square lattice model. The Hamiltonian of the SSH model is given by (cid:32) (cid:33) HSSH(k) = 0 ∆(k) ∆†(k) 0 , (21) where ∆(k) = (t + α) + (t − α)e−ik, see Fig. 3(a). At the critical point α = 0 the system is invariant under a half translation followed by an exchange of A and B sub- lattices, which corresponds to the nonsymmorphic sym- metry G(k), Eq. (1), see Fig. 3. Since the SSH model also has the inversion symmetry P = σ1ı, we find in ac- cordance with Table I that there is a gapless point at k = π. Observe that at α = 0 the translation sym- metry is promoted from 2Z (with translator 2a) to Z (with the translator a) and the SSH model becomes a 1D tight-binding model of free fermions with the dispersion E(k) = 2t cos(k) [see Fig. 3(c)]. It is noted that in order to create a bandcrossing point protected by nonsymmor- phic symmetry in a tight-binding model through dimer- ization, one must ensure that the original non-dimerized model has two chiral gapless modes. A nonvanishing α, on the other hand, reduces the translation symmetry from Z to 2Z and breaks the nonsymmorphic symmetry G(k), leading to a topological (α < 0) or trivial (α > 0) insulator, depending on the sign of α. The Hamiltonian of the π-flux square lattice model reads in momentum space H(k) = 2t sin kx t + te−iky t + teiky −2t sin kx , (22) (cid:32) (cid:33) where t denotes the nearest neighbor hopping amplitude, see Fig. 4(a). The model is invariant under the non- symmorphic magnetic symmetry G(ky) Ki, which corre- sponds to a time-reversal symmetry T = Kı followed by (I)(II)6 VII. DISCUSSIONS ABOUT MULTI-BAND THEORIES It is noted that the topological arguments in this work are not limited to two-band models, which is exemplified by two cases in follows. First, if a multi-band theory has a chiral symmetry, then the Hamiltonian can be anti- diagonalized with the upper-right entry being a matrix ∆(k), and equation (4) still holds for q(k) = Det(∆(k)). The topological argument around Eq.(5) implies that Det(∆(k)) has to vanish somewhere in momentum space, namely there exists band-crossing points enforced by the chiral and nonsymmorphic symmetry. Secondly, let us extend the spinless time-reversal sym- metry discussed in Sec.IV to the spinful one T = −iσ2 Kı, which acts in spinful four-band theories. If both T and G are preserved (G acts on the space of τ ), equations (4) and (18) hold for q(k) = Det(∆(k)), which implies Det(∆(k)) vanishes at k = π. But diagonal terms have to vanish at k = 0 and π as required by the symme- tries, except the chemical potential term. Thus bands are enforced to cross at k = π. Note that the vanishing of Det(∆(k)) at k = π needs only the combined symme- try G T , but both T and G are required for that of the diagonal terms. Acknowledgments The authors thank C.-K. Chiu for useful discussions. FIG. 4: (a) Illustration of the π-flux square lattice model. This model is invariant under the nonsymmorphic magnetic symmetry G(ky) Ki, namely a time-reversal symmetry fol- lowed by the nonsymmorphic symmetry G(ky). (b) Energy spectrum of the π-flux square lattice model. a half translation along y and an exchange of A and B sublattices. The high-symmetry 1D subsystems kx = 0 and kx = π have gapless Dirac points at ky = π, as shown in Fig. 4(b). This is in agreement with Table I, since [G(ky), Ki] = 0. Similar to the SSH model, the π-flux state is driven into a topological or trivial insu- lating phase by a dimerization α along y, that breaks the nonsymmorphic symmetry G(ky). In closing, we ob- serve that the π-flux model can be viewed as a higher- dimensional generalization of the SSH model. ∗ [email protected] † [email protected] 1 M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buhmann, L. W. Molenkamp, X.-L. Qi, and S.-C. Zhang, Science 318, 766 (2007). 2 D. Hsieh, D. Qian, L. Wray, Y. Xia, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nature (London) 452, 970 (2008). 3 M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). 4 X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). 14 Y.-H. Chan, C.-K. Chiu, M. Y. Chou, and A. P. Schnyder, Phys. Rev. B 93, 205132 (2016). 15 Y. X. Zhao, A. P. Schnyder, and Z. D. Wang, Phys. Rev. Lett. 116, 156402 (2016). 16 K. Shiozaki, M. Sato, and K. Gomi, Phys. Rev. B 91, 155120 (2015). 17 C. Fang and L. Fu, Phys. Rev. B 91, 161105 (2015). 18 C.-X. Liu, R.-X. Zhang, and B. K. VanLeeuwen, Phys. Rev. B 90, 085304 (2014). 19 X.-Y. Dong and C.-X. Liu, Phys. Rev. B 93, 045429 (2016). 20 L. Lu, C. Fang, L. Fu, S. G. Johnson, J. D. Joannopoulos, 5 M. Z. Hasan and J. E. Moore, Annual Review of Condensed and M. Soljacic, Nat Phys 12, 337 (2016). Matter Physics 2, 55 (2011). 21 K. Shiozaki, M. Sato, and K. Gomi, Phys. Rev. B 93, 6 A. P. Schnyder and P. M. R. Brydon, Journal of Physics: 195413 (2016). Condensed Matter 27, 243201 (2015). 22 S. M. Young and C. L. Kane, Phys. Rev. Lett. 115, 126803 7 T. Senthil, Annual Review of Condensed Matter Physics (2015). 6, 299 (2015). 8 C.-K. Chiu, J. C. Y. Teo, A. P. Schnyder, and S. Ryu, ArXiv e-prints (2015), 1505.03535. 9 Y. Ando and L. Fu, Annual Review of Condensed Matter 23 T. Bzdusek, Q. Wu, A. Ruegg, M. Sigrist, and A. A. Soluyanov, ArXiv e-prints (2016), 1604.03112. 24 L. Michel and J. Zak, Physics Reports 341, 377 (2001). 25 A. Konig and N. D. Mermin, Phys. Rev. B 56, 13607 Physics 6, 361 (2014). (1997). 10 C.-K. Chiu, H. Yao, and S. Ryu, Phys. Rev. B 88, 075142 26 H. Watanabe, H. C. Po, M. P. Zaletel, and A. Vishwanath, (2013). ArXiv e-prints (2016), 1603.05646. 11 K. Shiozaki and M. Sato, Phys. Rev. B 90, 165114 (2014). 12 T. Morimoto and A. Furusaki, Phys. Rev. B 88, 125129 27 S. A. Parameswaran, ArXiv e-prints (2015), 1508.01546. 28 S. A. Parameswaran, A. M. Turner, D. P. Arovas, and (2013). A. Vishwanath, Nat Phys 9, 299 (2013). 13 C.-K. Chiu and A. P. Schnyder, Phys. Rev. B 90, 205136 29 Z. Wang, A. Alexandradinata, R. J. Cava, and B. A. (2014). Bernevig, Nature 532, 189 (2016). 30 A. J. Heeger, S. Kivelson, J. R. Schrieffer, and W. P. Su, Rev. Mod. Phys. 60, 781 (1988), URL http://link.aps. org/doi/10.1103/RevModPhys.60.781. 7 Hence, the off-diagonal component of H(k) must satisfy eikq(k) = −q(−k), and its Fourier components obey qn = −q−(n+1). Therefore, q(k) can be written as ∞(cid:88) n=0 Appendix A: Nonsymmorphic Particle-hole symmetry q(k) = qn(eink − e−i(n+1)k), (A2) In this section, we discuss whether the existence of a band crossing can also be enforced by a nonsymmorphic particle-hole symmetry. As it turns out this is not possi- ble. To see this, we consider for instance an anti-unitary non-symmorphic particle-hole symmetry G(cid:48)(cid:48) = G(k) K, which acts on H as G(k)H(k)G−1(k) = −H∗(−k). (A1) which vanishes at k = 0. However, there does not exist a symmetry protected band crossing at k = 0, since the gap opening term z(k)σ3, with z(k) = z(−k) an even func- tion, preserves the non-symmorphic particle-hole symme- try G(cid:48)(cid:48). To protect the band crossing point at k = 0, an additional symmetry is needed which forbids the z(k)σ3 term. For example, the chiral symmetry {H, S} = 0, with S = σ3 and { S, G(cid:48)(cid:48)} = 0, prevents the z(k)σ3 term.
1808.02809	1	1808	2018-08-08T14:54:27	Electrically controlled crossover between $2\pi$- and $4\pi$-Josephson effects through topologically-confined channels in silicene	[ "cond-mat.mes-hall" ]	We propose a tunable topological Josephson junction in silicene where electrostatic gates could switch between a trivial and a topological junction. These aspects are a consequence of a tunable phase transition of the topologically confined valley-chiral states from a spin-degenerate to a spin-helical regime. We calculate the Andreev bound states in such a junction analytically using a low-energy approximation to the tight-binding model of silicene in proximity to s-wave superconductors as well as numerically in the short- and long-junction regime and in the presence of intervalley scattering. Combining topologically trivial and non-trivial regions, we show how intervalley scattering can be effectively switched on and off within the Josephson junction. This constitutes a topological Josephson junction with an electrically tunable quasiparticle poisoning source.	cond-mat.mes-hall	cond-mat	Electrically controlled crossover between 2π- and 4π-Josephson effects through topologically-confined channels in silicene Daniel Frombach,1 Sunghun Park,2 Alexander Schroer,1 and Patrik Recher1, 3 1Institut fur Mathematische Physik, Technische Universitat Braunschweig, D-38106 Braunschweig, Germany 2Departamento de F´ısica Te´orica de la Materia Condensada, Condensed Matter Physics Center (IFIMAC) and Instituto Nicol´as Cabrera, Universidad Aut´onoma de Madrid, Spain 3Laboratory for Emerging Nanometrology Braunschweig, D-38106 Braunschweig, Germany (Dated: August 9, 2018) We propose a tunable topological Josephson junction in silicene where electrostatic gates could switch between a trivial and a topological junction. These aspects are a consequence of a tun- able phase transition of the topologically confined valley-chiral states from a spin-degenerate to a spin-helical regime. We calculate the Andreev bound states in such a junction analytically using a low-energy approximation to the tight-binding model of silicene in proximity to s-wave superconduc- tors as well as numerically in the short- and long-junction regime and in the presence of intervalley scattering. Combining topologically trivial and non-trivial regions, we show how intervalley scat- tering can be effectively switched on and off within the Josephson junction. This constitutes a topological Josephson junction with an electrically tunable quasiparticle poisoning source. I. INTRODUCTION The fractional (4π) Josephson effect is one of the key signatures in junctions between topological super- conductors hosting Majorana bound states (MBS)1 -- 4. MBS have been proposed as the fundamental building blocks for topological quantum computation5,6. The 4π-Josephson effect is inherently a non-equilibrium ef- fect which sensitively depends on a protected crossing of many-particle states of opposite fermion parity. In the absence of quasiparticle poisoning2,7, the parity is con- served and the current exhibits a 4π periodicity which is known as a fermion parity anomaly (for a review see, e.g. Ref. 8). However, the 4π-Josephson effect could in prin- ciple also originate from a ballistic spin degenerate chan- nel where the appearance of a crossing of Andreev bound states (ABS) at a phase difference of φ = π is universal9. In the absence of scattering and equilibration effects, the current should also exhibit a 4π-Josephson effect carried by each Kramer's pair. Depending on the global fermion parity of both Kramers' pairs, the current is either 4π periodic (even global fermion parity) or 2π periodic (odd global fermion parity). This result was actually derived for the quantum spin Hall (QSH) effect samples where the superconductors cover both helical edge states, and the scattering between different Kramers' pairs is prohib- ited by the insulating bulk, separating the two edges of the sample4. It means that a ballistic Josephson junction could exhibit both, a 2π- or 4π-Josephson effect depend- ing on the global fermion parity. Experiments to detect the fractional Josephson ef- fect so far have concentrated on ac-properties10 -- 12 and it could be interesting to think about probably simpler dc-measurements. It was theoretically shown3 that the critical current of a Josephson junction should behave differently in the short and long junction limit regard- ing the difference between a 4π- and 2π-junction. This probe only needs a static dc measurement which is not FIG. 1. Topological edge states can form at the interface (gray dashed line) between two gapped regions of silicene with different applied perpendicular electric fields E1 and E2. Two superconductors (S 1 and S 2) are placed on the edge states and induce superconductivity in them via the proximity effect. The superconductors exhibit a phase difference φ and form a Josephson junction of length l with Andreev bound states. sensitive to the dynamics of φ. Therefore, to pinpoint the existence of a 4π-Josephson junction, a spectacular test would be the creation of a Josephson junction that could be switched (with a knob) between a 4π- and 2π- situation. This is what we would like to propose and investigate in this work using the peculiarities of silicene. We investigate a novel Josephson junction that can be tuned between a 2π- and a 4π-Josephson junction us- ing the valley chiral states at a mass domain in silicene, where the sign of the mass can be tuned electrically13 -- 18 due to the buckled structure of its honeycomb lattice. In addition, the presence of a sizeable spin-orbit cou- pling allows one to switch between valley-chiral but spin- degenerate states and a valley-chiral and spin helical regime. We analytically calculate the ABS in the short junction limit using the Bogoliubov de Gennes (BdG) equation of silicene19 including the generic case of inter- valley scattering. We supplement the silicene Josephson junction with numerical simulations on a square lattice by discretizing the low energy model as well as by using the full tight-binding model. The intervalley scattering affects the topologically trivial regime and the ABS de- velop an anticrossing around φ = π, whereas the crossing is protected in the valley-chiral and spin-helical regime by the same symmetries as for the single edge of a QSH sys- tem. This insight can be exploited to create a controllable poisoning source by constructing regions of different mass terms such that a dissipative helical channel -- attached to an additional normal reservoir -- is coupled to another helical channel along a small part of the Josephson junc- tion by intervalley scattering (see Fig. 7). The presence or absence of the dissipative channel can be tuned by electric gates. In the absence of intervalley scattering the transition between the topologically different junctions is reminis- cent of the transition between a single QSH edge and a pair of QSH edges between two superconducting con- tacts. II. MODEL (cid:90) We consider a sheet of silicene with two regions sep- arated by the x-axis that can be distinguished by the applied perpendicular electric field E(y) (Fig. 1). Two s- wave superconductors on top of the sheet induce a super- conducting pairing potential in silicene via the proximity effect19. The superconductors exhibit a phase difference φ and form a Josephson junction. At low energies, the system can be described by the Hamiltonian (setting the chemical potential to zero) H = 1 2 d2x Ψ†(x)H(x)Ψ(x), (1) with H(x) = H0 + HS + HI to be described below. We use the Nambu basis † † Ψ(x) = (ψ↑(x), ψ↓(x), ¯ψ ↓(x),− ¯ψ ↑(x))T , (2) where ψs(x) = (cAKs(x), cBKs(x), cAK(cid:48)s(x), cBK(cid:48)s(x))T and ¯ψs(x) is obtained from ψs(x) by the substitution K ↔ K(cid:48). The operator cστ s(x) annihilates an electron at position x on sublattice σ, in valley τ and with spin- polarization s. The Hamiltonian describing the system without superconductivity and intervalley scattering is H0 = −ivF (∂xρzτzσx+∂yρzσy)+∆SOρzszτzσz+mρzσz. (3) It consists of the kinetic part (first term) with the Fermi velocity vF = 5.42× 105 m/s13, intrinsic spin orbit inter- action with ∆SO = 3.9 meV13 and a staggered potential with mass m = 0.23 A×Ez resulting from the perpendic- ular electric field due to the buckled lattice structure of silicene17,20. The matrices σi, τi, si and ρi are four sets of Pauli matrices corresponding to the sublattice subspace, 2 FIG. 2. Schematics of the energy gaps confining the topolog- ical channel: the energy gap sizes in both half spaces of the silicene sheet are marked on an arbitrary energy scale (hor- izontal arrow), forming a box. The left and right sides cor- respond to two different possible configurations of the mass parameters while the top and bottom halves correspond to vanishing and finite spin orbit interaction, respectively. If zero is contained in a box, the sign of the energy gap which is proportional to the Chern number changes across the inter- face of the two half spaces and a topological edge state exists at the interface. Since the spin in the z-direction is conserved, the Chern numbers can be considered independently for each spin (blue, red). The intrinsic spin-orbit interaction strength is equal in both half spaces, so it only results in a shift of the boxes. Since the boxes for both spins get shifted in opposite directions (depicted by dashed arrows), a spin helical regime (bottom right) results in the regime mi < ∆SO < mj, i, j = 1, 2 (i (cid:54)= j). the valley subspace, the electron spin and the particle- hole subspace, respectively21. The proximity-induced s- wave superconductivity takes the form HS = ∆(cos(φ)ρx + sin(φ)ρy), (4) with the superconducting pairing potential ∆eiφ. In clean silicene samples, the two valleys are independent of each other, however, in disordered samples intervalley scattering HI = δρzτx (5) (cid:80) n εnγ† of strength δ will be present due to atomic scale impurities22. The Hamiltonian H can be diagonalized H = 1 n = 2 (cid:82) d2x Ψ†(x)Λn(x), where Λn(x) is a 16-component eigen- nγn with the creation operator γ† spinor of the matrix H(x) with eigenvalue εn. Due to the electron-hole symmetry of the BdG-equation, {Ξ,H(x)} = 0 where Ξ = ρysyτxC with C denoting the operator of complex conjugation. The solutions come in pairs (Λn(x), εn) and (ΞΛn(x),−εn) with the corre- sponding operators γεn = γ † −εn . III. SPIN-HELICITY OF VALLEY-CHIRAL BOUNDARY STATES We start by investigating H0 in the absence of su- perconductivity and intervalley scattering. Topologically confined edge states can be found at the interface between the two half spaces (1 with y < 0; 2 with y > 0) that can be distinguished by the two different perpendicular elec- tric fields14,16 (assumed to be homogenous within each half space). The bulk dispersion of silicene in the pres- ence of an electric field (cid:112)(vF k)2 + (m + ηξ∆SO)2, (6) εk = ± where η and ξ are the eigenvalues of τz and sz, features a spin and valley dependent energy gap. Since the spin in the z-direction and the valley quantum number are conserved by H0, the Chern number (cid:88) (cid:90) 1 2π n = α ∈ filled bands with the Berry curvature Fα = [∇k × i(cid:104)φαk∇kφαk(cid:105)]· ez and where the spinor φαk is connected to the solution 23, can be calculated for each Λαk(x) = exp(ik · x)φαk spin and each valley separately15, nηξ = η 2 sgn(m + ξη∆SO). (8) With these Chern numbers a topological Z2 invariant (cid:26) 0 1 (cid:88) η ν = nη↑ − nη↓ 2 mod 2 = m > ∆SO m < ∆SO (9) distinguishing the topologically trivial (ν = 0) from the non trivial (ν = 1) phase can be defined. If the Chern numbers of the two half spaces differ, topological edge states exist at the interface due to the bulk-boundary correspondence23. These edge states are valley chiral because the Chern number is proportional to η, i.e., if a left-moving spin-up channel exists at the K valley, a right-moving spin-down channel will exist at the K(cid:48) val- ley, which is a manifestation of H0 being time reversal invariant (see also Fig. (2)). Spin helicity (one Kramers pair of helical edge states) is achieved by tuning the elec- tric fields to fulfill the condition mi < ∆SO < mj , (10) i, j = 1, 2 (i (cid:54)= j). This is because in this regime the argument of the sgn-function m1,2 ± ∆SO changes (does not change) its sign when replacing m1 by m2 for one of the two signs so that the channels do (do not) exist. Since the sign is given by ξη one spin polarization is suppressed per valley and the resulting edge states are spin helical. This regime can for instance be accomplished by tuning the second electric field so that ∆SO < m2 while the first one vanishes. Conversely, when tuning E1 such that m1 < −∆SO, edge states of both spin polarisations exist at the interface of the two half spaces and the edge states are spin degenerate. This crossover between spin helical and spin degenerate edge states can therefore be achieved by tuning the external electric fields. d2k Fα, (7) IV. TUNABLE JOSEPHSON EFFECT 3 By solving the Schrodinger equation H0(x)Λ(x) = εΛ(x) in both half spaces independently and matching their wave functions at the interface the dispersion rela- tion of the edge states ε(kx) = ±ηvF kx (11) can be calculated without further approximations. It is independent of both the mass parameters and the spin orbit interaction strength. The prefactor ±, however, de- pends on the concrete realisation of the mass parameters in both half spaces. If the topological edge states in silicene are spin helical they effectively mimic those at the sample edge of a QSH insulator with the main difference being that the energy dispersions at zero energy do not cross at the Γ point in momentum space but are located at the K and K(cid:48) Dirac points in the first Brillouin zone. When building a Josephson junction mediated by these edge states, i.e. upon including the proximity induced superconductivity described by HS, where we choose the phase of the super- conducting pairing potential of one of the superconduc- tors to vanish while the other is equal to φ (see Fig. (1)), ABS localized inside the junction emerge2. We note in passing that electrically tunable Josephson junctions in silicene have been considered theoretically in the con- text of 0−π-24 -- 27 and/or ϕ0-junctions25,27,28 where time- reversal symmetry is broken explicitly in these works. We first develop an effective model by projecting the full Hamiltonian in the absence of HI onto the subspace spanned by the boundary states obtained above. Without loss of generality we assume that the electric fields are arranged such that the channel on the K (K(cid:48)) valley is spin up (down) polarized (i.e. m1 < ∆SO and m2 < −∆SO). The field operators c↑(x) = c↓(x) = dyf (y) dyf (y) 1 √2 1 √2 [cAK↑(x) − cBK↑(x)] , [cAK(cid:48)↓(x) − cBK(cid:48)↓(x)] (12) with f (y) = 1 √N −m+∆SO vF y e , m = (cid:26) m1 m2 y < 0 y > 0 (13) annihilate an electron in these edge states at position x along the channel direction and are connected to the an- nihilation operators of the edge eigenstates via a Fourier transformation in the x direction. The normalization fac- tor N is chosen such that(cid:82) dy f (y)2 = 1. The Hamil- tonian projected (with projection operator P) onto the subspace spanned by the valley chiral states can be writ- (cid:90) (cid:90) Γ1 = Γ2 = where (cid:90) (cid:90) dx ϕ(x) dx ϕ(x) ϕ(x) = θ = arg (cid:105) 1 √2 i √2 (cid:104) † (cid:104) eiθc↑(x) + c ↓(x) e−iθc↓(x) − c (cid:115)√∆2 − ε2 (cid:16) (cid:112) ∆2−ε2 vF (cid:17) vF ε + i √ − e ∆2 − ε2 x (cid:18) φ (cid:19) ε(φ) = ±∆ cos 2 (cid:105) , (15) † ↑(x) . (16) (17) † † ten in the basis Φ(x) = (c↑(x), c↓(x), c ↑(x))T as ↓(x),−c ∆e−iφ 0 −ivF ∂x ivF ∂x  . (14) −ivF ∂x PHP = ∆e−iφ ∆eiφ 0 0 0 0 0 ∆eiφ ivF ∂x 0 0 A. Short-junction limit Here, we assume that the two superconductors are lo- cated next to each other with their edges being perpen- dicular to the mass boundary, forming a short Josephson junction where l (cid:28) ξ0 with the superconducting coher- ence length ξ0 = vF /∆. Solving for the decaying wave- functions on both superconducting sides for ε < ∆ and µ = 0 and matching them at the junction boundary lo- cated at x = 0 leads to two ABS They obey the energy-phase relation (cid:12)(cid:12)(cid:12)(cid:12)ε=0 (cid:12)(cid:12)(cid:12)(cid:12)ε=0 that is 4π-periodic (Fig. 3, blue), so that the Josephson current through the junction at zero temperature being proportional to ∂φε is also 4π-periodic. For φ = π, the ABS lie at zero energy and θ → π/2, so that the linear combinations γ1 = (Γ1 + Γ2) √2 , γ2 = i(Γ1 − Γ2) √2 (18) † 1ε=0 = are Majorana excitations due to the relation Γ Γ2ε=0. The corresponding Majorana wave functions are shown in Fig. (4) and are obtained from Eqs. (15) and (12). Tuning the electric fields such that the topological edge states are not spin helical but spin degenerate can be interpreted as introducing a second set of Kramers pair edge states with the opposite spin polarization. In the absence of intervalley scattering HI , these two sets of Kramers pair edge states are independent of each other and both result in ABS with the 4π-periodic energy-phase relation (17) which again translates into an overall 4π-periodic Josephson current. Since the inter- valley scattering couples electrons with the same spin po- larization of different Dirac cones, it induces scattering 4 FIG. 3. Energy-phase relation of the Andreev bound states. Without intervalley scattering (blue) the curves cross at zero energy and are 4π periodic both in the helical and non-helical case. In the generic case, intervalley scattering δρzτx opens a gap at zero energy (red) in the non-helical regime and the energy-phase relation becomes 2π periodic, whereas the cross- ing of Andreev bound states is protected by parity conserva- tion in the topologically non-trivial regime with underlying helical edge states. The result obtained by implementing the low energy model on a square lattice of 50 × 50 lattice sites (dotted lines) coincides well with the analytical prediction (solid lines). FIG. 4. Probability density of the Majorana wave functions appearing for φ = π. The degenerate wave functions are localized in x-direction within the superconducting coher- ence length while the localization in y-direction is dependent on the electric fields. The fields are chosen such that the mass parameters of both regions are symmetric around −∆SO (m1,2 = −∆SO±δm) so that the wave functions decay equally rapidly in the y-direction on the length scale of vF /δm. The projections on the vertical planes are for specific values of x, y. between the two sets of Kramers pair edge states in the absence of the superconductors or couples the two sets of ABS in the presence of the Josephson junction. Due to this coupling a gap of size 2δ opens at zero energy in the energy-phase relations of the ABS. The resulting 0π2πφ−101ε/∆2δ/∆x/(¯hvF∆)−11y/(¯hvFδm)−11Λ2energy-phase relation (cid:115) ε(φ) = ± ∆2 cos2 (cid:18) φ (cid:19) 2 + δ2 (19) is 2π-periodic (Fig. 3, red) which results in the Josephson current now also being 2π-periodic. We compare the low-energy model with a numerical treatment of the full low-energy Hamiltonian Eq. (1) on a square lattice (see Appendix A for further details). We note that the opening of an energy gap is a consequence of the ABS not being protected in the topologically trivial phase, so that these results should be valid even for more general forms of intervalley scattering. We note that the effect of intervalley scattering on the ABS in the spin-degenerate case depends on the chemical potential µ, which, so far, we have set to zero. By re- peating the calculations for the ABS without intervalley scattering and calculating the matrix elements between ABS of different valleys, the minimal energy of the ABS as a function of φ becomes δ/(1 + (µ/∆)2). For small µ this corresponds to a correction −δ(µ/∆)2 that reduces the gap in the spectrum around φ = π, however, it is parametrically small by the factor (µ/∆)2. In the op- posite limit µ (cid:29) ∆, the effect of intervalley scattering is suppressed by the factor (∆/µ)2. This analysis shows that we can tune the influence of intervalley scattering on the ABS by the chemical potential µ. Intravalley scattering can also be present, but has no effect on the ABS since it corresponds to forward scat- tering due to the valley chirality of all boundary states. In addition, we have neglected the much smaller Rashba effect compared to the intrinsic spin-orbit effect13. The Rashba spin orbit coupling could split the spin degener- acy in the spin degenerate regime, but would not lead to a change of the periodicity of the Josephson effect. 5 FIG. 5. Energy-phase relation of the Andreev bound states in the long-junction regime l (cid:29) ξ0 obtained numerically from the low energy model Eq. (1) implemented on a square lattice of 100 × 100 lattice sites in the helical regime. The length of the junction (50 lattice sites) is large compared to the co- herence length of the system (∼ 1.6 lattice sites) so that the junction is well in the long-junction regime. Top panel: No intervalley scattering, multiple ABS exist inside the junction and feature an energy-phase relation linear in φ. Bottom panel: Andreev bound states with intervalley scattering in- duced by HI where δ = 0.25∆ in the spin-degenerate regime. B. Long-junction limit V. EXPERIMENTAL REALIZATION Our calculation of the ABS is valid in the short junc- tion regime l (cid:28) ξ0. In a real experiment, however, the Josephson junction may be in the long-junction regime 3. With a Fermi velocity vF = 5.42 × 105 m/s l (cid:29) ξ0 in silicene and a superconducting gap of the order of ∆ ∼ 1 meV, ξ0 ∼ 357 nm so that the junctions fabri- cated in Ref. 29 and 12 for HgTe/CdTe quantum wells be- ing 400 nm and 600 nm long, respectively would fall into the intermediate or long-junction regime. In the latter regime, the energy-phase relation has been theoretically shown to depend linearly on the phase difference φ across the junction3,30 -- 32 while its periodicity stays unchanged. In Fig. 5, we present results for the ABS energies using a numerical simulation of the low-energy model Eq. (1) implemented on a square lattice (see Appendix A). This and all other numerical tight-binding simulations were performed with the Kwant code33. The proposed setup could be implemented by using electric top and bottom gates to define the topologi- cally confined channel. For instance, choosing m1 = 0 and m2 = −2∆SO (for which electric fields E1 = 0 and E2 ≈ −2 × 17 meV/A = −34 meV/A are needed17,20) the spin helical edge states will lie inside an energy gap of ∆SO ≈ 3.9 meV. Choosing m1 = 2∆SO and m2 = −2∆SO (for which electric fields E1,2 = ±2∆SO ≈ ±34 meV/A are needed) spin degenerate edge states will again lie in- side an energy gap of ∆SO ≈ 3.9 meV. If the gates do not reach into the superconducting region the channel would not be present below the superconductors and we would assume that m1 and m2 are zero there. We can treat this scenario numerically by describing the silicene sheet by a tight binding model with nearest neighbor hop- ping, inversion symmetry breaking staggered potentials, a Kane-Mele-type intrinsic spin orbit interaction34 and s- 0π2πφ−0.50.00.5ε/∆0π2πφ−0.50.00.5ε/∆wave superconductivity via the BdG-formalism (see Ap- pendix B). We find that, in addition to the energy-phase relation of in-gap states already present with electric fields reaching into the superconducting regions (Fig. 5, top panel), new in-gap states emerge (Fig. 6, top panel) at energies ε ≈ ±∆/2 which appear to be independent of the phase difference across the junction and are localized at the edges of the superconductors (Fig. 6, middle panel) in contrast to the ABS spread homogenously between the two superconductors along the topological channel (Fig. 6, bottom panel). These states are remnants of the topological edge states, which would propagate along the edges of the superconducting regions (but with vanishing ∆) instead of straight into these regions, as would be the case for a finite electric field inside the superconduct- ing regions (see Appendix C). However, as long as the electric fields merely vanish inside the superconducting regions, these states only occur at ε ≈ ± ∆ 2 and leave the low energy excitations, and therefore the main as- pects of our proposal, unchanged. In Appendix D, we calculate analytically these bound states located at the boundary between a superconductor and a region with a mass gap for silicene and show that for the parameters used in Fig. 6, an energy gap on the scale of ∆ appears, similar to the numerics. The spectrum of these states is flat as a function of the superconducting phase difference φ since they are localized near one of the two supercon- ductors with only an negligible overlap with the other superconductor35. VI. TUNABLE POISONING IN A 4π-JUNCTION So far we have shown how one can electrically switch from a spin-helical regime to a spin-degenerate regime for the valley chiral channels, leading to generically different types of Josephson effects. Another promising direction is offered by utilizing the tunable mass-regions for an ad- ditional probe for the spin-helical channel, see Fig. (7). The mass-terms, i.e. the gate voltages, can be tuned such that the helical channel is coupled to another he- lical channel -- forming a small region of weakly coupled (by intervalley scattering) spin-degenerate valley-chiral states -- that is further coupled dissipatively to an elec- tron reservoir. Assuming a phase-biased Josephson junc- tion formed via the spin-helical channel, the coupling to the dissipative channel, if present, can be used to re- lax the Josephson junction formed by the helical chan- nel to its instantaneous ground state 0(φ)(cid:105), leading to a 2π-periodic Josephson current, despite the fact that the channel is helical. Removing the dissipative channel, by switching the electric gates, establishes the return to the 4π-periodic Josephson current. To show this, we use the following effective model for the short junction regime (similar conclusions should also hold for the long-junction regime). The helical edge states in contact with the two superconductors having a phase difference of φ are represented by its 6 FIG. 6. Energy-phase relation (top panel) of in-gap states ob- tained by implementing a tight binding model of silicene on a hexagonal lattice of 100× 100 unit cells. The probability den- sity (middle and bottom) are shown on a linear scale, where red colors correspond to high probability densities. The x− and y−axes are given in units of the length of a single unit cell. The electric fields are only present inside the junction (50 unit cells wide). The superconducting regions (red dashed boxes, middle and bottom panels) and regions with a finite electric field (yellow dashed boxes, middle and bottom pan- els) actually touch and are only drawn with a spatial gap to better visualize the probability density localized at the bor- der between these regions. In addition to the energy-phase relations already present in the case for a finite electric field inside the superconducting regions (Fig. 5) there exist in-gap states whose energies appear to be independent of the phase difference across the junction at ε ≈ ±∆/2. The probability density corresponding to such states (middle panel) is local- ized at the sides of the superconductors and not inside the junction like the ABS shown at the bottom for zero energy. Since the field E1 tunes the region to be topologically trivial no edge channels are present at the top and bottom edges of the sample. 0π2πφ−0.50.00.5ε/∆0 20 40 60 80 100 120 140 0 10 20 30 40 50 60 70 80 0 20 40 60 80 100 120 140 0 10 20 30 40 50 60 70 80 0 20 40 60 80 100 120 140 0 10 20 30 40 50 60 70 80 0 20 40 60 80 100 120 140 0 10 20 30 40 50 60 70 80 7 that change the many-particle states of the Josephson junction from state α to state β are given by Fermi's Golden rule rates (see Appendix E). Here, α, β denote the two possible states of the junction 0(φ)(cid:105) and 1(φ)(cid:105). We derive the following results W10 = γtϕ2(0)f (ε(φ)) and W01 = γtϕ2(0)[1 − f (ε(φ))] where γt is the normal state tunneling rate 2πνt2/ with ν the density of states per spin and length in the dissipative helical channel and f (ε) = [1 + exp(βε)]−1 is the Fermi function in the dissi- pative helical channel with β = 1/kBT the inverse ther- mal energy. We assume that there is no voltage bias between the superconducting reservoirs and the normal conducting reservoir (lead) to which the dissipative chan- nels are coupled. The dynamical state of the Josephson junction is then described by the reduced density matrix with elements ραβ. The off-diagonal elements ρ12 = ρ∗ 21 decay exponen- tially with a rate given by (W10 + W01)/2, whereas the probabilities ρii have a finite stationary value given by (ρ00/ρ11) = (W01/W10) with the constraint ρ00 + ρ11 = 1 which gives the solution for the occupation probability of the excited state 1(φ)(cid:105), ρ11 = W10/(W10 + W01) = f (ε(φ)). For low temperatures T → 0, we obtain the de- sired result ρ11 ∼ exp[−βε(φ)] so that the junction stays in the ground state 0(φ)(cid:105). A phase-biased experiment would therefore exhibit a 2π Josephson effect when slowly sweeping φ in the pres- ence of the dissipative channel. For such a switching experiment to work, the intrinsic poisoning rates of the Josephson junction (i.e. the poisoning rate in the ab- sence of the dissipative channel), need to be slower than γt. We note that Ref. 3 suggested a doubling of the criti- cal current of a long current biased topological Josephson junction in the 4π-regime (weak poisoning) compared to the 2π-regime (strong poisoning). According to Ref. 3, the possibility to switch between the two regimes would require for our setup that γ−1 t (cid:28) τJ (cid:28) τqp, where τJ is the internal phase relaxation time of the Josephson junction and τqp is the intrinsic quasiparticle poisoning time. VII. CONCLUSIONS We have investigated in detail a novel type of tun- able Josephson junction through a channel defined at a mass domain in silicene. Using the electrically tunable mass term in silicene in combination with the intrinsic spin-orbit coupling and s-wave superconductivity via the proximity effect, we have demonstrated the transition between a valley chiral and spin helical boundary state and a valley chiral and spin degenerate boundary state. In both regimes we have calculated the Andreev bound states in the short junction limit analytically and nu- merically in the long-junction regime using a discretiza- tion of the low-energy Dirac model as well as using the full tight-binding model on the honeycomb lattice us- ing the KWANT code33. In the valley chiral and spin FIG. 7. Helical and valley chiral Josephson junction with tunable quasiparticle poisoning that can be used to switch between a 4π- and 2π-junction electrically. Blue (red) lines represent the spin up (down) polarized edge channels. Sec- tions of spin helical and spin-degenerate channels can be con- trolled via tunable mass terms. The spin-degenerate regime can act -- in the presence of intervalley scattering δ -- as a tun- nel junction between the Andreev bound states formed in the helical and valley chiral Josephson junction and a second he- lical and valley chiral channel (of opposite spin) that is cou- pled to a lead allowing for dissipation. This lead has the same chemical potential as the two superconductors. instantaneous eigenstates via the Hamiltonian H0 = † † (εa(φ)Γ† bΓb)/2. Using Γa = Γ aΓa + εb(φ)Γ b and εa(φ) = −εb(φ) due to electron-hole symmetry, we rewrite H0 = εa(φ)(Γ† aΓa − 1/2). The many-particle states are con- sequently 1(φ)(cid:105) = Γ† a0(φ)(cid:105) with energy εa(φ)/2, and 0(φ)(cid:105) with energy ε0(φ) = −εa(φ)/2. Also, it holds that Γa0(φ)(cid:105) = 0. The operator Γa is directly related to Γ1,2 defined in Eq. (15) such that εa(φ) = ∆ cos(φ/2). We model the coupling of the spin-helical Josephson junc- tion to the dissipative channel by a tunneling Hamilto- † σ cLσ(0)c Rσ(0) + h.c., where we assume a pointlike tunneling region in the low-energy model. The tunneling matrix element t originates from the micro- scopic form of intervalley scattering in the sample. nian HT = t(cid:80) (1/√¯l)(cid:80) The field operators in the low-energy model are given for both sides (L, R) by Eq. (12). We expand them on the side hosting the ABS (L) in the operators Γ1 and Γ2 for a fixed phase difference φ and on the R-side in plane-wave states for the spin-helical channel cRσ(x) = k exp(ikx)cRkσ, see Appendix E. Here, cRkσ an- nihilates an electron with spin σ and wave number k in the dissipative helical channel and ¯l is the quantization length for these channels. Without loss of generality we † assume that Γ 1 creates a particle in the ABS with pos- itive energy ε(φ) > 0. Then the tunneling rates Wαβ helical regime, the Josephson effect resembles the one through a helical edge state of a two-dimensional topo- logical insulator2,3. The Josephson current exhibits a fractional (4π) current phase relation in the absence of quasiparticle poisoning. In the valley chiral but spin- degenerate regime, the Josephson effect is generically of the usual 2π type as intervalley scattering will open a gap in the Andreev bound state spectrum around φ = π. In the absence of intervalley scattering and in the pres- ence of fermion parity conservation, the periodicity of the Josephson effect would depend on the global fermion par- ity, similar to the case of two separate spin helical edge states coupled by the same superconductors4. Intraval- ley scattering events and/or Rashba spin-orbit coupling will not change the periodicity of the Josephson current. We also investigated numerically experimentally feasible ways for implementing the tunable electric fields in the setup. We further proposed a scenario where a valley chiral and spin helical channel is subjected to tunable quasi- particle poisoning by coupling a part of the channel to another valley chiral and spin helical channel that is con- nected to a normal reservoir leading to dissipation. In that part, intervalley scattering, generically present, will at small temperatures cool the Josephson junction to its ground state. This quasiparticle poisoning is tunable electrically since the additional channel is only present if certain mass terms are present. This tunable Josephson effect would allow us to detect a topological Josephson junction exhibiting a 4π periodic current phase relation by comparing the critical current of the junction in the cases with and without the (tunable) quasiparticle poi- soning. According to Ref. 3, the critical current of a long Josephson junction should differ by a factor of two in these two cases. This constitutes a new way to search for topological effects in Josephson junctions in silicene. Finally, we would like to emphasize that silicene is not the only material system with such desired tun- able properties. In the recently proposed HgTe double layer quantum wells36 as well as in the topological insu- lator InAs/GaSb type II quantum wells37,38 such mass domains could be implemented with electric fields. To some extend also in single HgTe-based quantum wells, an electrically tunable coupling between helical edge states in constrictions has been predicted39. In both material systems signs of topological superconductivity in edge states have been seen experimentally11,12,40. Another promising material system is bilayer graphene where an electrically tunable gap can be implemented by a volt- age between the layers and valley chiral states appear at voltage domain walls41 with the difference that there are two spin-degenerate states per valley. Such voltage induced mass-gaps in bilayer graphene have been ex- perimentally realized in graphene transistors42 and for the creation of gate-tunable quantum dots43,44. The coupling of such a valley-chiral channel to a supercon- ductor has been recently investigated in the context of Cooper pair splitting45. The problem of a compara- 8 bly small spin-orbit coupling in graphene was recently taken up by considering curved bilayer graphene46 or by combining bilayer graphene with transition metal dichalcogenides47 -- 49. ACKNOWLEDGMENTS We thank Bjorn Trauzettel for useful feedback on the manuscript and the Research Training Group GrK1952/1 "Metrology for Complex Nanosystems", and the Braun- schweig International Graduate School of Metrology B- IGSM for financial support. Appendix A: Low energy tight binding model on a square lattice The tight binding approach of the low energy model on a square lattice can be obtained by first Fourier trans- forming (1) to momentum (k)-space and then discretize real space on a square lattice with lattice constant b. The spinor in momentum space is then represented as ψk = 1 √N e−ik·RψR, (A1) where ψk is the Fourier transform of Ψ(x) defined in Eq. (2), N is the number of lattice sites and R is the position vector of a lattice site of a square lattice. The wave vectors in the Hamiltonian ki = 1 b sin(kib) can be approximated by a sine function. With the or- thonormality relation b kib ≈ 1 (cid:88) R (cid:88) k δR,0 = 1 N eik·R (A2) (cid:88) k ≈ vF b = −i where (cid:88) (cid:88) ψ k (cid:16) vF 2b R the kinetic part of the Hamiltonian transforms to † k[vF (kxρzτzσx + kyρzσy)]ψk ψ † k (sin(kxb)ρzτzσx + sin(kyb)ρzσy) ψk † R+δ1 ψ ρzτzσxψR + ψ † R+δ2 ρzσyψR (cid:17) + H.c. (A3) (cid:18)b (cid:19) (cid:18)0 (cid:19) b . , 0 (A4) δ2 = δ1 = (cid:112)sin(kxb)2 + sin(kyb)2 of the The spectrum ε = ± particle and K valley subspace approximates the corre- sponding spectrum ε = ±vF k of the kinetic part of the low energy model (3) for small wave vectors k but devi- ates significantly from it for large wave vectors. Most notably, in contrast to the original low energy model (1) vF b the spectrum of the discretized model introduces three additional inequivalent Dirac cones at the corners of the first Brillouin zone. Since the topology of a system is linked intimately to the number of Dirac cones present in the spectrum, these additional Dirac cones would sig- nificantly alter results directly linked to topology, i.e. the number of topological edge states. We therefore need to include another term, which gaps the three spurious ad- ditional Dirac cones but leaves the Dirac cone centered at the Γ point unaltered. Furthermore this additional term must respect the symmetry of the system, that is time reversal symmetry. A possible term reads vF b (2 − cos(kxb) − cos(kyb))ρzσz (A5) which we can identify as the "lattice generalization of the mass term" introduced in Chap. 8.3 of Ref. 50. For a more detailed discussion see also Ref. 51. If the inversion symmetry breaking mass parameter m becomes finite, the now gapped additional Dirac cones can be closed vF b . Furthermore, if again if m = −2 vF b , then the resulting bands are non dispersive, m = − so that numerical implementations should satisfy the con- straint or m = −4 vF b This additional term transforms as (cid:20)vF (cid:88) † ψ k k = 2 vF b . m > − (A6) (cid:21) (2 − cos(kxb) − cos(kyb) ρzσzψk (cid:17) (cid:16) (cid:88) (cid:88) R b vF b vF 2b − † RρzσzψR ψ 2(cid:88) (cid:16) R i=1 † R+δi ψ ρzσzψR + H.c. (A7) (cid:17) . Since all other terms in the low energy model Eq. (1) are momentum independent, their Fourier transform is trivial, so that the final tight binding model on a square lattice of the low energy model, that is the discretized low energy model, reads (cid:88) (cid:88) i (cid:104)i,j(cid:105)x H = − it (cid:88) (cid:104)i,j(cid:105) † ψ i εψi − t † i ρzσzψj ψ νijψ † i ρzτzσxψj − it (cid:88) (cid:104)i,j(cid:105)y † i ρzσyψj νijψ (A8) with ε = (m + 4t) ρzσz + ∆SOρzszτzσz + HS + HI , t = vF 2b . (A9) Here (cid:104)i, j(cid:105) denotes all nearest neighbors and (cid:104)i, j(cid:105)i nearest neighbors in i-direction. Furthermore νij = ±1 are a set 9 of signs where νij = 1, if the site i is to the right or top of site j. In Figs. 3 and 5 we report numerical tight binding sim- ulations based on this discretized low energy model. In vF these simulations we set b = 1 as the characteristic energy scale of the system and measure all other energies vF b . The parameters were chosen as follows: in units of spin helical regime spin degenerate regime vF b m1 m2 t 1 0.5 1 0.5 ∆SO 0.75 ∆ 0.2 δ 0 vF 1 b m1 −0.5 1 m2 t 0.5 ∆SO 0 ∆ 0.2 δ 0.05 so that in both cases the constraint Eq. (A6) is satis- fied. While the hopping parameter t is always fixed in the chosen units and the two mass parameters mi can be externally tuned by the applied electric fields, the last three parameters are material specific. Parameter values clearly illustrating their effects have been chosen. The spin orbit coupling strength for silicene is known and non- vanishing, unlike our assumption in the spin degenerate regime. However, the spin orbit coupling strength ∆SO only enters the Chern number Eq. (8) in the argument of the sign-function and determines the localization length of the topological edge states, so only the spin orbit inter- action strength relative to the induced mass parameters is of physical relevance here. Appendix B: Tight binding model of the hexagonal lattice The microscopic tight binding model, to which the Hamiltonian (1) is the low energy approximation, is given by a Kane-Mele-type Hamiltonian34 with s-wave super- conductivity formulated in the Bogoliubov de Gennes for- malism on a honeycomb lattice (cid:88) (cid:88) H = c † i εici− t i (cid:104)i,j(cid:105) † where ci = (ci↑, ci↓, c i↓,−c of undoped silicene (cid:88) (cid:104)(cid:104)i,j(cid:105)(cid:105) † i ρzcj + it2 c † νijc i ρzszcj (B1) † i↑)T . Here the on-site energies εi = ±miρz + ∆i(cos(φ)ρx + sin(φ)ρy) (B2) consist of a staggered potential and an s-wave supercon- ducting pairing term and the plus (minus) sign applies to the A (B) sublattice. Again, (cid:104)i, j(cid:105) denotes all near- est neighbors and (cid:104)(cid:104)i, j(cid:105)(cid:105) all next nearest neighbors. The Haldane phases νij = ±1 are a set of signs depending on the two nearest neighbor bonds connecting the next near- est neighbors i and j. If we have to take a left (right) 10 turn when moving from site j to i the Haldane phase equals +1 (−1). If we take the hopping energies to be , t2 = ∆SO 3√3 (B3) t = 2 3 vF a where a is the distance between two nearest neighbors, this tight binding Hamiltonian results in the low energy model Eq. (1) at low energies. The characteristic energy scale of this model is the nearest neighbor hopping energy t. In the numerical results presented in Fig. 6 we set this characteristic energy to unity and measured all energies in units of t. The parameters were chosen as follows 1 0.5 1 t m2 m1 ∆SO 0.75 ∆ 0.2 where again the values for the mass parameters m1,2 can be varied externally and the absolute value of the superconducting paring potential is induced externally by the proximity effect. The physical strength of the spin orbit interaction may differ from the assumed value ∆SO = 0.75t, however, the number of existing edge chan- nels again only depends on the spin orbit strength com- pared to the mass parameters. Appendix C: Electric fields inside superconducting regions To better understand the existence of the additional in gap states at energies ε ≈ ±∆/2 existing in the case for vanishing electric fields inside the superconducting regions we also calculate the energy-phase relation for the same parameters as in Fig. 6 with the only differ- ence that the electric fields can now reach 5 lattice sites into the superconducting regions (Fig. 8, top). In this case the additional dispersionless states from Fig. 6 are no longer present. Furthermore, when comparing the probability densities of a zero energy state ((Fig. 6, bot- tom) and (Fig. 8, bottom)) we note, that the bound state is well localized in the y direction even at the bound- aries to the superconducting regions if the electric fields reach into the superconducting regions. Both observa- tions support the premise that the additional states at energies ε ≈ ±∆/2 are remnants of the topological edge states without the superconducting pairing, because if the electric fields reach into the superconducting regions, the topological edge states are localized inside the super- conducting regions and are thus suppressed by the super- conducting pairing potential. If, on the other hand, the electric fields like in Fig. 6 do not reach into the supercon- ducting regions, the topological edge states are localized at the boundary of the superconducting regions and thus states localized at the boundary between the supercon- ductor and the regions with a finite electric field can form. FIG. 8. Energy-phase relation (top) for the same setup as in Fig. 6 with the only difference being that the electric fields now reach 5 lattice sites into the superconducting regions. Unlike the corresponding probability density of Fig. 6 (bot- tom) the wave function at zero energy (bottom) now is well localized in the y direction also at the boundary of the super- conductors. Furthermore, we note that the additional localization in the y direction at the boundary of the superconducting regions does not alter the energy-phase relation. Appendix D: Boundary states perpendicular to the channel We provide analytical support to our above argument for the existence of the in-gap states, which are indepen- dent of the superconducting phase difference, shown in Fig. 6. Specifically, we show that the states are subgap bound states at the interface between the normal region with a finite electric field and the superconducting region with a vanishing electric field, and hence do not carry su- percurrent between two superconductors. To this end, we calculate the dispersion relation for in-gap states at the 0π2πφ−0.50.00.5ε/∆0 20 40 60 80 100 120 140 0 10 20 30 40 50 60 70 80 0 20 40 60 80 100 120 140 0 10 20 30 40 50 60 70 80 interface by solving the following BdG equation: (cid:32) H(x)Ψ(x) = εΨ(x), H0(x) ∆(x) ∆∗(x) −H0(x) (cid:33) H = (D1) (D2) , where Ψ(x) is the wave function in the Nambu basis, as defined in Eq. (2), and H0(x) is the Hamiltonian for silicene including a staggered potential, where M is the 2 × 2 matrix given by M = −iηvF κSσx + vF kyσy + ηξ∆SOσz − ε. H0(x) = vF pxσxτz + vF pyσy + ∆SOσzτzsz + m(x)σz. (D3) Eq. (D11) is used below in Eq. (D17) to obtain a disper- sion relation at the interface. Next we calculate wave functions in the normal region The matrices σi, τi, si are Pauli matrices acting on the spaces of sublattice, valley, and spin, respectively. The position-dependent variables m(x) and ∆(x) are given by m(x) = ∆(x) = m for x > 0 for x ≤ 0 0 0 for x ≥ 0 ∆ for x < 0 , , (D4) (D5) (cid:113)2v2 ε = ± (cid:40) (cid:40) (cid:32) 11 Note that we only consider kx = −iκS with κS > 0, as an imaginary momentum kx = iκS corresponds to a wave function which diverges as x → −∞. From Eqs. (D8) and (D10), we have MΨe S(x) + ∆Ψh S(x) = 0, (D11) (D12) (cid:32) by solving the following equation, (cid:33)(cid:32) (cid:33) (cid:32) (cid:33) (cid:48)N 0 Ψe Ψh N (x) N (x) (cid:48)N 0 H 0 0 −H (cid:48)N 0 = ηvF pxσx + vF pyσy + ηξ∆SOσz + mσz. The where H bulk dispersion relation for both an electron and a hole in the normal region is given by N (x) N (x) Ψe Ψh (D13) = ε , x + 2v2 F k2 F k2 y + (ηξ∆SO + m)2, (D14) and a complex wave vector in the x direction correspond- ing to a wave function decaying as x → ∞ is kx = iκN = i(vF )−1(cid:113)2v2 F k2 y + (ηξ∆SO + m)2 − ε2. (D15) The electron and hole components of the wave function are given by N (x) = cee−κN x+ikyy Ψe N (x) = che−κN x+ikyy Ψh (cid:33) (cid:32) vF (iηκN − iky) (cid:32) vF (iηκN − iky) ε − (ηξ∆SO + m) −ε − (ηξ∆SO + m) , (cid:33) , (D16) where ce and ch are coefficients. By matching the wave functions, Ψe/h S (x, y) given in Eq. (D11) and Ψe/h N (x) = Ψe/h S (x) = Ψe/h N (x, y) in Eq. (D16), at x = 0, (cid:32) (cid:33) (cid:32) Ψe Ψh S(0, y) S(0, y) = Ψe Ψh N (0, y) N (0, y) we have (cid:32) (D9) ceM ivF (ηκN − ky) ε − (ηξ∆SO + m) = ch∆ , (D17) (cid:32) (cid:33) −ivF (ηκN − ky) ε(ηξ∆SO + m) , (cid:33) (cid:33) from which we derive the dispersion relation at the in- terface. After some algebra, we find the simplified form as and they are uniform along the y-direction. Here we as- sume that chemical potential is zero. As the spin and valley degrees of freedom are conserved quantities in this model, we can reduce H(x) to a 4 × 4 matrix as (cid:33) (cid:48)(x) = H (cid:48) 0(x) ∆(x) H ∆∗(x) −H (cid:48) 0(x) , (D6) where (cid:48) 0(x) = ηvF pxσx + vF pyσy + ηξ∆SOσz + m(x)σz. H (D7) Here, η = ±1 and ξ = ±1 are eigenvalues of τz and sz, (cid:48)(x) in each of regions, respectively. Below, we solve H x < 0 and x > 0. In the superconducting region of x < 0, the reduced BdG equation is given by (cid:32) (cid:33)(cid:32) (cid:33) (cid:32) (cid:33) ∆ −H (cid:48)S H 0 ∆∗ (cid:48)S 0 = ηvF pxσx + vF pyσy + ηξ∆SOσz. The bulk S(x) S(x) S(x) S(x) Ψe Ψh Ψe Ψh (D8) = ε (cid:48)S 0 , with H dispersion relation in this region is given by ε = ± x + 2v2 F k2 F k2 y + ∆2 SO + ∆2. Then a complex wave vector within the energy gap in the x direction is (cid:113)2v2 kx = −iκS = −i(vF )−1(cid:113)2v2 F k2 y + ∆2 SO + ∆2 − ε2. (D10) ε2 − ∆2 SO − ηξ∆SO − 2v2 F k2 y = 2v2 F κSκN . (D18) This equation has a real solution ε given by (cid:115) 2v2 ε = ± provided that (cid:18) Z = 1 + y + ∆2 (ηξ∆SO + m)2 F k2 m2 + ∆2 (cid:19) (cid:19)(cid:18) ηξ∆SOm ηξ∆SO m 1 − ∆2 , (D19) > 0. (D20) We examine the result with two sets of parameters used in Fig. 6. Case 1. m = 1, ∆SO = 0.75, and ∆ = 0.2. In this case, Z > 0 is satisfied when ηξ = −1, and the solution is given by (cid:115) ε(ky) = ± 2v2 y + ∆2 (∆SO − m)2 F k2 m2 + ∆2 , (D21) 12 superconductors with ABS Γ1 and Γ2, whereas R labels the channels that are coupled to the dissipa- tive reservoir (denoted as "Lead" in Fig. (7)). We expand ΦL(x) in the low-energy ABSs as ΦL(x) ≈ (ϕ(x)/√2)[exp[−iθ(ε(φ))]Γ1, Γ † † 1]T . 1, Γ1,− exp[iθ(ε(φ))]Γ spin-helical channels cRσ(x) = (1/√¯l)(cid:80) The field operators of the R-side we expand in the plane-wave states (wave-number k, spin σ) of the k exp(ikx)cRkσ, where ¯l is the quantization length for these channels. In the remainder of this appendix, we present the de- tails of the calculation of the rates Wαβ that switch the Josephson junction from a many-particle state α to a state β by resorting to a Fermi's Golden Rule approach. We first rewrite HT in the appropriate excitations as (cid:112) 2¯l) (cid:16) (cid:88) HT = (t∗ϕ(0)/ cRk↑Γ × k † 1(φ)eiθ(ε(φ)) + cRk↓Γ1(φ) (cid:17) + h.c. (E1) which has a gap at ky = 0 of size ε(0) ≈ ±0.245∆. Case 2. m = 0.5, ∆SO = 0.75, and ∆ = 0.2. In this case, both cases of ηξ = 1 and −1 give Z < 0, and hence no solution exists within the gap. Appendix E: Poisoning rates W10 and W01 Here, we present the details of the calculation for the poisoning rates and the stationary solution for the ABS occupation presented in subsection VI. We start with the solution for the ABSs given in Eqs. (15). Fixing the superconducting phase difference φ, we have two solutions of opposite energy Γ1(ε(φ)) ≡ † Γ1(φ) and Γ2(−ε(φ)) ≡ Γ2(φ) = Γ 1(φ), due to particle- hole symmetry. Without loss of generality, we assume ε(φ) > 0. (cid:82) dxϕ(x)(1/√2) [exp(iθ(ε)), 0, 1, 0]T Φ(x) A complete set of states are the eigenstates of the Hamiltonian H (cf. Eq. (1)) with operators {γn}. Projecting H to the low energy space spanned by the valley-chiral edge states, the field operator for these edge † † states is given by Φ(x) = (c↑(x), c↓(x), c ↑(x))T ↓(x),−c defined above Eq. (14). Near zero energy, and close to φ = π, the low energy subspace is spanned by Eqn. (15)), the ABSs with energies ε < ∆ (cf. which we write here in the basis of Φ(x) as Γ1(φ) † 1(φ)). We model the coupling to the (and Γ2(φ) = Γ other two channels of opposite spin-valley chirality which are present at the same spatial position, but only along a finite region (see Fig. (7), by a tunneling † Rσ(0) + h.c.. Here, we consider a pointlike tunneling region in the low-energy model. The tunneling matrix element t represents the microscopic form of intervalley scattering in the sample. Let L stand for the channels coupled to the Hamiltonian HT = t(cid:80) σ cLσ(0)c = The rate from the ground state 0(φ)(cid:105)ABS of the Joseph- son junction (Γ10(φ)(cid:105)ABS = 0) to the first excited state † 1(φ)(cid:105)ABS = Γ 10(φ)(cid:105)ABS is given by the rate 2π ρi(cid:104)fHTi(cid:105)2 δ(Ei − Ef ). (cid:88) W10 = (E2) if Here, the possible initial states are i(cid:105) = 0(φ)(cid:105)ABSφi(cid:105)lead where φi(cid:105)lead is the initial state of the dissipative chan- nel coupled to a reservoir at temperature T having the same chemical potential µ as the superconductors, and ρi is the probability with which this initial state oc- curs. The possible final states are determined by HT † and have the form f(cid:105) = c Rk↓1(φ)(cid:105)ABSφi(cid:105)lead and f(cid:105) = cRk↑1(φ)(cid:105)ABSφi(cid:105)lead. Note that the fact that the pro- cess can create or annihilate an electron in the dissipative channel is a consequence of the ABSs being superposi- tions of electrons and holes. Both processes create a spin up excitation in the Fermi sea of the dissipative channel. The initial and final energies of the total (L, R)-system are Ei = εφi − ε(φ)/2 and Ef = εφi ± εk↓/↑ + ε(φ)/2. † † RkσcRkσ] = f (εkσ) RkσcRkσφi(cid:105) ≡ Tr[ρc i ρi(cid:104)φic with f (x) = [1 + exp(βx)]−1 the Fermi function in the dissipative channel, and 1− f (x) = f (−x), we obtain for the total rate Using (cid:80) (E3) W10 = γtϕ2(0)f (ε(φ)), ν = (cid:80) with the tunneling rate per length γt = 2πνt2/ where k δ(εkσ − ε)/¯l is the (constant) density of states A similar calculation holds for the opposite rate W01 where a spin down excitation tunnels into the dissipative channel with the result per spin and length at energy ε counted from µ. W01 = γtϕ2(0)(1 − f (ε(φ)). (E4) 1 A. Y. Kitaev, Phys. Usp. 44, 131 (2001). 2 L. Fu and C. L. Kane, Phys. Rev. B 79, 161408 (2009). 3 C. W. J. Beenakker, D. I. Pikulin, T. Hyart, H. Schomerus, and J. P. Dahlhaus, Phys. Rev. Lett. 110, 017003 (2013). 4 F. Cr´epin and B. Trauzettel, Phys. Rev. Lett. 112, 077002 (2014). 5 C. Nayak, S. H. Simon, A. Stern, M. Freedman, and S. Das Sarma, Rev. Mod. Phys. 80, 1083 (2008). 6 J. K. Pachos, "Introduction to topological quantum com- putation," (Cambridge University Press, 2012). 7 D. Rainis and D. Loss, Phys. Rev. B 85, 174533 (2012). 8 C. L. Kane and F. Zhang, Physica Scripta 2015, 014011 (2015). 9 C. W. J. Beenakker, Phys. Rev. Lett. 67, 3836 (1991). 10 L. P. Rokhinson, X. Liu, and J. K. Furdyna, Nat. Phys. 8, 795799 (2012). 11 J. Wiedenmann, E. Bocquillon, R. S. Deacon, S. Hartinger, O. Herrmann, T. M. Klapwijk, L. Maier, C. Ames, C. Brune, C. Gould, A. Oiwa, K. Ishibashi, S. Tarucha, H. Buhmann, and L. W. Molenkamp, Nat. Commun. 7, 10303 (2016). 12 R. S. Deacon, J. Wiedenmann, E. Bocquillon, F. Dom´ınguez, T. M. Klapwijk, P. Leubner, C. Brune, E. M. Hankiewicz, S. Tarucha, K. Ishibashi, H. Buhmann, and L. W. Molenkamp, Phys. Rev. X 7, 021011 (2017). 13 C.-C. Liu, H. Jiang, and Y. Yao, Phys. Rev. B 84, 195430 (2011). 14 M. Ezawa, New Journal of Physics 14, 033003 (2012). 15 S. K. Wang, J. Wang, and K. S. Chan, New Journal of Physics 16, 045015 (2014). 16 S. Rachel and M. Ezawa, Phys. Rev. B 89, 195303 (2014). 17 M. Ezawa, Journal of the Physical Society of Japan 84, 121003 (2015). 13 (2016). 27 X. Zhou and G. Jin, Phys. Rev. B 95, 195419 (2017). 28 J. Wang, L. Hao, and J.-F. Liu, Phys. Rev. B 93, 155405 (2016). 29 E. Bocquillon, R. S. Deacon, J. Wiedenmann, P. Leubner, T. M. Klapwijk, C. Brne, K. Ishibashi, H. Buhmann, and L. W. Molenkamp, Nature Nanotechnology (2016). 30 C. Ishii, Progress of Theoretical Physics 44, 1525 (1970). 31 J. Bardeen and J. L. Johnson, Phys. Rev. B 5, 72 (1972). 32 A. V. Svidzinsky, T. N. Antsygina, and E. N. Bratus', Journal of Low Temperature Physics 10, 131 (1973). 33 C. W. Groth, M. Wimmer, A. R. Akhmerov, and X. Wain- tal, New Journal of Physics 16, 063065 (2014). 34 C. L. Kane and E. J. Mele, Physical Review Letters 95, 226801 (2005). 35 In the case, where the superconductors dope the silicene sheet below them, potential steps between the valley-chiral channels and superconducting regions develop. We numer- ically checked that the current phase relation is not influ- enced by such doping effects. 36 P. Michetti, J. C. Budich, E. G. Novik, and P. Recher, Phys. Rev. B 85, 125309 (2012). 37 C. Liu, T. L. Hughes, X.-L. Qi, K. Wang, and S.-C. Zhang, Phys. Rev. Lett. 100, 236601 (2008). 38 F. Qu, A. J. A. Beukman, S. Nadj-Perge, M. Wimmer, B.- M. Nguyen, W. Yi, J. Thorp, M. Sokolich, A. A. Kiselev, M. J. Manfra, C. M. Marcus, and L. P. Kouwenhoven, Phys. Rev. Lett. 115, 036803 (2015). 39 V. Krueckl and K. Richter, Phys. Rev. Lett. 107, 086803 (2011). 40 I. Knez, R.-R. Du, and G. Sullivan, Phys. Rev. Lett. 109, 186603 (2012). 41 I. Martin, Y. M. Blanter, and A. F. Morpurgo, Phys. Rev. 18 F. Geissler, J. C. Budich, and B. Trauzettel, New Journal Lett. 100, 036804 (2008). of Physics 15, 085030 (2013). 19 J. Linder and T. Yokoyama, Phys. Rev. B 89, 020504 (2014). 42 J. B. Oostinga, H. B. Heersche, X. Liu, A. F. Morpurgo, and L. M. K. Vandersypen, Nature Materials 7, 151 (2008). 43 M. T. Allen, J. Martin, and A. Yacoby, Nature Commu- 20 N. D. Drummond, V. Z´olyomi, and V. I. Fal'ko, Phys. nications 3, 934 (2012). Rev. B 85, 075423 (2012). 21 For definiteness, we assume ∆SO > 0 in the following. If ∆SO < 0 all results remain the same if we change the basis in Eq. (2): Ψ(x) → U Ψ(x) with U = sx. 22 Eq. (5) is not the only possible form of intervalley scat- tering, but one of six time reversal symmetric and spin independent possibilities52. The term chosen here opens a gap in the spectrum of the spin degenerate valley chiral edge states. Important for our purposes, however, is only some kind of backscattering in the channel that would be present for any type of intervalley scattering. This would open a gap in the ABS spectrum9. 23 M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). 24 J. Wang, Y. H. Yang, and K. S. Chan, Phys. Rev. B 89, 064501 (2014). 44 M. Eich, F. Herman, R. Pisoni, H. Overweg, Y. Lee, P. Rickhaus, K. Watanabe, T. Taniguchi, M. Sigrist, T. Ihn, and K. Ensslin, arXiv:1803.02923. 45 A. Schroer, P. G. Silvestrov, and P. Recher, Phys. Rev. B 92, 241404 (2015). 46 J. Klinovaja, G. J. Ferreira, and D. Loss, Phys. Rev. B 86, 235416 (2012). 47 M. Gmitra and J. Fabian, Phys. Rev. B 92, 155403 (2015). 48 J. Y. Khoo, A. F. Morpurgo, and L. Levitov, Nano Letters 17, 7003 (2017). 49 A. M. Alsharari, M. M. Asmar, and S. E. Ulloa, Phys. Rev. B 97, 241104 (2018). 50 B. A. Bernevig and T. L. Hughes, Topological Insulators and Superconductors (University Press Group Ltd, 2013). 51 Y.-F. Zhou, H. Jiang, X. C. Xie, and Q.-F. Sun, Phys. Rev. B 95, 245137 (2017). 25 D. Kuzmanovski, J. Linder, and A. Black-Schaffer, Phys. Rev. B 94, 180505 (2016). 26 H. Li, R. Wang, and C. S. Ting, Phys. Rev. B 94, 085422 52 E. McCann, K. Kechedzhi, V. I. Fal'ko, H. Suzuura, T. Ando, and B. L. Altshuler, Physical Review Letters 97 (2006).
1408.3681	4	1408	2015-01-23T19:09:36	Effects of quantum interference on the electron transport in the semiconductor$/$benzene$/$semiconductor junction	[ "cond-mat.mes-hall" ]	Using the tight-binding model and the generalized Green's function formalism, the effect of quantum interference on the electron transport through the benzene molecule in a semiconductor/benzene/semiconductor junction is numerically investigated. We show how the quantum interference sources, different contact positions and local gate, can control the transmission characteristics of the electrode/molecule/electrode junction. We also study the occurrence of anti-resonant states in the transmission probability function using a simple graphical scheme (introduced in Ref.[Phys. Chem. Chem. Phys, 2011, 13, 1431]) for different geometries of the contacts between the benzene molecule and semiconductor(silicon and titanium dioxide) electrodes.	cond-mat.mes-hall	cond-mat	Effects of quantum interference on the electron transport in the semiconductor/benzene/semiconductor junction Javad Vahedi∗, Zahra Sartipi Department of Physics, Sari Branch, Islamic Azad University, Sari, Iran. (Dated: July 9, 2018) Using the tight-binding model and the generalized Green's function formalism, the effect of quantum interference on the electron transport through the benzene molecule in a semiconduc- tor/benzene/semiconductor junction is numerically investigated. We show how the quantum inter- ference sources, different contact positions and local gate, can control the transmission characteristics of the electrode/molecule/electrode junction. We also study the occurrence of anti-resonant states in the transmission probability function using a simple graphical scheme (introduced in Ref.[Phys. Chem. Chem. Phys, 2011, 13, 1431]) for different geometries of the contacts between the benzene molecule and semiconductor(silicon and titanium dioxide) electrodes. PACS numbers: 85.35.Ds, 85.65.+h, 81.07.Nb, 72.10.Di I. INTRODUCTION Recently, harnessing quantum interference (QI) effect as a versatile tool to design of single molecule devices as well as controlling the current through it has attracted many attention2–8. QI hapens when electron waves go through a meso/nano junctions phase-coherently. One of the important factors to study electron transport through the mesoscopic junctions is to control the quantum in- terference effect of electron wave related to the symetry that device adopts within the junctions. Regarding the employment of QI as an enabling tool for the execution of molecular switches, logic gates, data strong elements and thermoelectronic devices in single molecule devices, two different proposals have been suggested. One way to induce QI in molecular junction is to use a local gate po- tential to tune the position of induced transmission nodes relative to bias window, and the other is to control the electron transmission through chemical/conformational modification of side groups to aromatic molecules3–5. for semiconductor Studying wave-guides nano- structures was the first ground for signaturing QI, though it has been reported early on theoretically13 and experimentally14,15 that electron transport through a benzene molecule connected in a meta-configuration is strongly diminished in comparing with two ortho and para configuration between the molecule and the leads. This connection of the conductance in the benzene configuration has been recognized as QI and has been described in terms of phase shifts of transmission channels, and it has also been observed through another generic aromatic molecules. On the other hand, some theoretical and experi- mental efforts, have mainly focused on systems with two metal leads15–24 often gold, intriguing physics, oc- ∗email: [email protected] Tel: (+98) 911-1554504 Fax: (+98) 11-33251506 FIG. 1: Schematic representation of three meta, ortho and para configurations. Local gate is shown in light green curs when one or both electrodes are replaced with semiconductors24–28,30? –32,including negative differen- tial resistance and rectification33. Indeed, semiconduc- tor/molecule/semiconductor hybrid junctions with self- assembled mono-layers (SAM) are considered as promis- ing candidates to create functional devices for molecular electronics. Semiconductor substrates are widely used in many applications. Multiple practical uses involving these materials require the ability to tune their phys- ical (bandgap, electron mobility) and chemical (func- tionalization, passivation) properties to adjust those to a specific application. In the work reported in the present article we investigated the effect of quantum interference on the electronic conductance of semicon- ductor/benzene/semiconductor junction which no stud- ies have been reported to date. Using tight-binding model and a generalized Green's function method in the Landauer-Buttiker formalism, we have tackled this prob- lem and the variation of interference condition deter- mined by replacement three different configurations of the benzene molecule (ortho, meta and para) connected to the semiconductor/metal electrodes. Semiconductor modeling is acquirable via some per- turbations of the Newns-Anderson (NA) model as seen in Fig.(2-b). One such model, introduced by Koutecky and Davison (KD) is the generalization of the simple tight binding model, which alternates both the site en- ergies α1, α2 and intersite couplings β1, β2 with nearest- neighbor in tight-binding picture Fig.(2-a). This model may consider each unit cell as an atom, with two sites 5 1 0 2 n a J 3 2 ] l l a h - s e m . t a m - d n o c [ 4 v 1 8 6 3 . 8 0 4 1 : v i X r a 2 face states are created in both AS and AB models. For AB model 0, 2 and 1 surface states are defined as β1 > β2 in both leads, β1 < β2 in both leads and β1 > β2(β1 < β2) in source(drain), respectively. For AS model 0, 2 and 1 surface states are defined as α1 < 0 in both leads, α1 > 0 in both leads and α1 > 0 (α1 < 0) in source(drain), respectively. The paper is organized as follows. In section (II), we summarize the model, Hamiltonian and formalism. More details on the (KD) model are given in the appendix A. In section (III), we present our results for QI in the benzene molecule by numerical calculations and graphical rules. Finally, conclusion is given in section IV. II. COMPUTATIONAL SCHEME We start with Fig.(1),where a benzene molecule which subject to a local gate potential is attached to semi- infinite semiconductor electrode. We have also consid- ered metal electrodes, for comparison. Leads are char- acterized by electrochemical potentials and respectively, under the non-equilibrium condition when an external bias voltage is applied. Both leads have almost the same cross section as the sample to reduce the effect of the scattering induced by a wide-to-narrow geometry of the sample-lead interface. The whole system is de- scribed within a single electron picture by a tight-binding Hamiltonian with nearest-neighbor hopping approxima- tion. The Hamiltonian representing the entire system can be written as a sum of four terms H = HL + HR + HM + HC (1) where HL,(R) represents the Hamiltonian of the left (right) electrode which is described within the tight- binding approximation, HM is the Hamiltonian of the molecule and the last term in Eq.(1) is the Hamiltonian of coupling of electrodes to the molecule. The Hamiltonian of KD model as a semiconductor lead in the tight-binding picture is expressed as follows FIG. 2: Schematic representation of (a) Koutecky-Davison (KD), (b) Newns-Anderson (NA), (c) Alternative site(AS) and (d) Alternative bond (AB) models. states corresponding to s and p orbitals. The coupling be- tween s and p orbitals in the same unit cell describes their overlap, and the coupling across unit cells describes the interatomic bonding. Neighboring s − s and p − p inter- actions are neglected in the simple tight-binding model. The KD model has three important limits. The first one can achieve in the combined limit α → 0, β1 → β2 ≡ β which produces the Newns-Anderson model as seen in Fig.(2-b) that α and β are the on-site energy and intersite hopping energies, respectively, in tight-binding model33. The second one is the limit α1 = −α2, β1 → β2 ≡ β. This alternating site (AS) model has been used to describe ti- tanium dioxide where the site energies (α and −α) cor- respond to the different atoms Fig.(2-c)33. The third one is the limit where the model alternates bonds(AB)Fig.(2- d). The AB model has been used to model silicon and germanium, where the bond disparities(β1 and β2) are related to the orbital hybridization33. Intercepting crystal into two noninteracting segments causes symmetry breaking and surface states arising. The surfaces can exhibit dangling bonds, potentially leading to reconstructions surface states, with densities localized near the surface33–35. Three 0, 1 and 2 sur- (cid:0)χ N(cid:88) j=1 HKD = α † † 2j−1χ2j−1 − χ 2jχ2j (cid:1) + (cid:104) β1 (cid:0)χ N(cid:88) j=1 † 2j−1χ2j (cid:1) + β2 N−1(cid:88) j=1 (cid:0)χ † 2j+1χ2j (cid:105) (cid:1) + h.c (2) † Here χ j (χj) denotes the creation (annihilation) operator of an electron at site j. Besides, its three Hamiltonian limits NA, AS and AB models can be expressed in the following forms, HN A = β HAS = α + β j=1 N(cid:88) N(cid:88) (cid:0)χ 2N−1(cid:88) N(cid:88) j=1 j=1 (cid:1) † † 2j−1χ2j−1 − χ 2jχ2j † χ jχj + h.c † χ jχj + h.c HAB = β1 j=1 † 2j−1χ2j − β2 χ N−1(cid:88) j=1 † 2j+1χ2j + h.c (3) χ the Hamiltonian of benzene can be described as fol- lows: HM = (cid:88) (cid:88) (cid:0)C (cid:1), † i Ci + t iC † i Cj + C † j Ci (4) i <ij> † i (Ci) denotes the creation (annihilation) operator HereC of an electron at site i in the benzene molecule. i and t are the on-site energy and nearest-neighbor hopping integral in the benzene molecule, respectively. The last term in Eq.(1) is the coupling part and given by (cid:88) (cid:0)χ i Ci + h.c)(cid:1), † HC = γc(j,i) (5) i where the matrix elementsγc(j,i) represents the coupling strength between the molecule and electrodes. The Green's function of the system can be written as G(E)C = 1 E1 − HM − ΣL(E) − ΣR(E) , (6) 3 1 stands is the and where ΣL(E)(ΣR(E)) self-energy matrix resulting from the coupling of the benzene molecule to the left (right) electrode and given by identity matrix for the ΣL(R)(E) = γ∗ c gL(R)γc = ΛL(R)(E) − i 2 ΓL(R)(E), (7) where γc is the coupling matrix and will be non-zero only for the adjacent points in the benzene molecule and elec- trode. In Eq.(7), ΛL(R) is the real part of self-energy which correspond to the shift of energy levels of the ben- zene molecule and the imaginary part ΓL(R)(E) of the self-energy represents the broadening of these energy lev- els. In the follow shifts and broadening terms of energy levels are given for KD model33. (cid:115)(cid:2)α2 + (β1 + β2)2 − E2(cid:3)(cid:2)E2 − α2 − (β1 − β2)2(cid:3) γ2 β2 2 E2 − α2 − β2 (E − α)2 (cid:113)(cid:2)E2 − α2 − (β1 − β2)2(cid:3)(cid:2)E2 − α2 − (β1 + β2)2(cid:3) 1 + β2 2 + ΘKD(E) ΘKD(E) = Θ(α2 + (β1 − β2)2 − E2) − Θ(E2 − α2 − (β1 + β2)2), 2 (E − α) 2β2 ΓKD(E) = ΛKD(E) γ2 = (8) (9) where Θ is the sign function. Having ΓKD, ΛKD, one can easily find the other models relations. By applying α → 0,β1 → β2 ≡ β we have ΓKD, ΛKD → ΓN A, ΛN A. Also taking the limit β1 → β2 ≡ β and α → 0 gives relations for the AS and AB models,respectively. In Eq.(7), gL(R) correspond to the Green's functions for the left(right) electrode. The Green's functions for the KD model can be written as follows33: E2 − α2 + β2 1 + β2 gKD = 2 ±(cid:113)(cid:2)E2 − α2 − (β1 − β2)2(cid:3)(cid:2)E2 − α2 − (β1 + β2)2(cid:3) 2 (E − α) 2β2 Some of the details on the derivation are further ex- plained in Appendix A. The transmission probability between reservoirs is given as T (E) = T r(cid:0)ΓLGrΓRGa(cid:1), which electrons transmit from the left to the right elec- trode by propagating through the molecule. Relying on Landauer-Buttiker formula, we evaluate the current as a function of the applied bias voltage dET (E, V )(cid:2)fL(E, V ) − fR(E, V )(cid:3) (11) (cid:90) ∞ (10) I(V ) = 2e h −∞ where Gr and Ga are the retarded and advanced Green's function, respectively. The broadening matrix is de- fined as the imaginary part of the self-energy ΓL(R) = −2Im(ΣL(R)). Transmission function tells us the rate at where fL(R) is the Fermi distribution function at the left (right) electrode with chemical potential µL(R) = EF ± V and Fermi energy EF . 2 III. NUMERICAL RESULTS Here, we present results of the numerical calcula- tion based on the formalism described in sectionII. We have calculated the transmission through the benzene molecule for ortho, meta and para configurations. For simplicity, we chose the on-site energy equal to zero, i.e., ε = 0 without loss of generality. Furthermore, the value of the hopping energy is set as t = −2eV in the tight- binding picture. Local gate potential is applied on site 5 with(εg = 0.5eV ) (see Fig.1, local gate is shown in light green spot). The NA, AB and AS model parameters have listed in Table-I33. As a reference energy, the Fermi en- ergy of electrodes is set EF = 0. The temperature is also set as T = 4K. A. Metal electrode We have plotted the logarithmic scale of transmission function versus energy of the gold/benzene/gold junc- tion for meta, ortho and para configurations (see the top panel of Fig.(3). The dotted lines are transmission func- tion with a gate (G) potential applied on site 5 and the solid lines are the transmission through ungated (WG) benzene for comparison. For an electron with energy E, that comes from the left connection, the probabil- ity of transmission function reaches its saturated value (resonance peaks) for the specific energy values. These resonance peaks are related to the eigenenergies of the in- dividual benzene. When the electron travel from the left electrode to the right one through the benzene molecule, the electron waves propagating along the two branches of benzene may suffer a relative phase shift. Consequently, there might be constructive or destructive interference due to the superposition of the electronic wave func- tions along the various pathways. Therefor, the trans- mission probability will change and some anti-resonance behaviour may arise. In the top panel of Fig.(3), for meta and ortho configurations one can easily see such anti- resonance behaviours (solid lines). These anti-resonance states are related to the quantum destructive interference effect. In contrast, for para configuration in the absence of the gate potential there is not such an anti-resonance behaviour which is referred to the constructive superpo- sition of the electron waves. In what follows, we will apply a local gate potential and check its effects on the transport properties of three con- figurations. To this end, we exploit a simple but versa- TABLE I: Model Parameters for Au, Si, and TiO2 material model α(eV ) Au11 Si TiO2 NA AB AS 1.6 β1(eV ) −8.95 −1.60 −2 γ(eV ) β2(eV ) −0.45 −2.185 −1.0 −1.0 4 (Color online) The top panel FIG. 3: is the logarith- mic scale of transmission function versus energy of the gold/benzene/gold junction for meta, ortho and para con- figurations. The bottom panel is the current-voltage (IV) for meta, ortho and para configurations. The dotted(solid)line corresponding to the gated (ungated) potential. tile graphical method proposed recently36. The authors showed that using the graphical method one can easily detect the presence/absence of QI induced transmission nodes only from the topological structure of molecules. Here, we point out some essential features and rules of graphical method. A detailed description can be found in Ref. [36]. With some mathematical task one can recast the Eq.(10) as follows T (E) = γ(E)2G1N (E)2 (12) −5−4−3−2−1012345−20−15−10−50Meta−5−4−3−2−1012345−20−15−10−50Log(T(E))Ortho−5−4−3−2−1012345−10−50E(ev)Para WGG−6−4−20246−10−50510Meta−6−4−20246−4−2024I(µA)Ortho−6−4−20246−10010VPara WGG5 where, G1N (E) is the (1N ) matrix element of the Green function, and γ(E) = [ΓL(E)]11 = [ΓR(E)]N N . Often we can neglect the energy dependence on the lead coupling strength, both for metal and semiconductor electrodes then QI effects are included in G1N (E). Using Cramer's rule the matrix element G1N (E) is defined C1N (EI − HM ) G1N (E) = det(EI − HM − Σl − ΣR) (13) where C1N (EI−HM ) is the (1N ) co-factor of (EI−HM ) determined as the determinant of the matrix obtained by removing the first row and the N-th column from (EI − HM − Σl − ΣR) and multiplying it by (−1)1+N . Since we consider solely sites 1 and N pair to the left and right electrodes, the elimination of the first row and Nth column entirely removes ΣL,R in the co-factor. There- fore we will concentrate on C1N (EI − HM ) and represent the determinant graphically. The transmission zeros can be defined from the zeros of the co-factor (condition for complete destructive interference). (−1)N +1C1N (EI − HM ) = 0 (14) to understand how the graphical rules follow from Eq.(14), we use the Laplace's formula for evaluating of co-factor. det(A) =(cid:80) Aij(−1)i+jMij, where Mij is the FIG. 4: (Color online) All graphical diagrams determining the transmission zeros. Six sites connected to leads at site 1 and 4 in para configuration. determinant obtain from removing first row and N-th col- umn of the matrix (A). So the elements of co-factor gives an equation for the transmission nodes in certain ener- gies. We start with Hamiltonian of benzene molecule.  HM =  0 0 ε1 t12 t21 ε2 t23 0 0 t32 ε3 t34 0 0 t61 0 t16 0 0 0 0 0 t43 ε4 t45 0 0 t54 ε5 t56 0 0 0 0 t65 ε6 FIG. 5: (Color online)All graphical diagrams determining the transmission zeros. Benzene molecule connected to leads at site 1 and 2 in ortho configuration. (15) Note that for different configuration the Hamiltonian will be unchanged and we have just different co-factor for different meta, ortho and para configurations. On-site energies are set zero except site 5 which we put a local gate ε5 = εg, hopping integrals are also set equal as t. So it is simple to adjust co-factor for different meta, ortho and para configurations. Co-factor in para configuration is achievable solely by omitting 1st row and 4th column of matrix C14(EI − HM ) C14(E − HM ) = (16) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) −t E −t 0 0 −t E 0 0 −t −t 0 0 0 −t −t 0 0 0 0 0 E − εg −t E 0 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) when we write down the elements of co-factor, we reach an equation which determines the transmission nodes en- ergy (transmission zeroes). In general its a tedious work and graphical method circumvents this complexity. FIG. 6: (Color online) All graphical diagrams determining the transmission zeros. Benzene molecule connected to leads at site 1 and 3 in meta configuration. In Fig.(4) we have depicted all the diagrams based on graphical rules that display how the graphical method is applied to predict the possible existence of QI in an en- ergy range benzene connected to leads in para formation. The four terms in the co-factor can be arisen graphically according to the following rules: (i) in each diagram be- tween the external sites i and j, there is only one con- tinuous path connecting them. (ii) at all remaining sites, internal sites (sites other than i and j) there is one out- going and one incoming path or either have an on-site loop. ( iii) each diagram has a sign that is (−1)p where p is the total number of closed hopping loops and on-site loops. (iv) between two sites i and j, only one path can be drawn if they have none-zero hopping elements, tij. By converting these diagrams into a polynomial, the co- factor C14(E−HM ) is in hand. Now we write down poly- nomial function for para configuration based on above conditions as follows {+t3(εg − E)(ε6 − E) − t5 − t5 +t3(ε3 − E)(ε2 − E)} = 0 ⇒ E1 = 1.87, E2 = −2.12 (17) where E1 and E2 are two points in which transmis- sion spectrum vanishes. The derivation of the co-factor C1N (E − HM ) for the benzene molecule in ortho and meta configurations will follow the same methodology of the co-factor by the graphical scheme in para configura- tion. We have drawn all the generalized diagrams deter- mining the transmission zeroes, for ortho and meta con- figurations of benzene molecule in Fig.(5))and Fig.(6)), respectively. so for the ortho configuration {+t(ε3 − E)(ε4 − E)(ε6 − E)(εg − E) + t5 + t5 −t3(εg − E)(ε6 − E) − t3(ε4 − E)(ε3 − E) −t3(ε3 − E)(ε6 − E)} = 0 ⇒ E1 = 2, E2 = −2, E3 (cid:39) −3.09, E4 (cid:39) 2.59 (18) and for the meta configuration {−t2(ε4 − E)(ε6 − E)(εg − E) + t4(ε4 − E) +t4(ε6 − E) − t4(ε2 − E)} = 0 ⇒ E1 = 1.76, E2 = −2.26, E3 = 0 (19) Comparing the above result shows a good agreement with our numerical results based tight-binding method in Fig.(3). More precisely, the transmission zeros points are exactly corresponding to roots of above equations for all three possible configurations of the benzene molecule. In order to provide a deep understanding of the elec- tron transport, we have depicted the current-voltage for three ortho, meta and para configurations in the bottom panel of Fig.(3). An applied voltage shifts the chem- ical potentials of two electrodes relative to each other by eV , with e the electronic charge. When the ben- zene level is positioned within such bias window, current will flow. With consideration to the current profile, it 6 shows staircase-like structure with fine steps versus of the applied bias voltage. This is due to the existence of the sharp resonance peaks in the transmission spec- trum, since the current is computed by the integration of the transmission function Eq.(11). With the increase of the bias voltage the electrochemical potentials on the electrodes are shifted gradually, and finally cross one of the quantized energy levels of the benzene. Therefore, a current channel is opened up which provides a jump in the characteristic. For the same voltage the current mag- nitude of the three ortho, meta and para configurations exhibit different responses to the applied bias voltage. This is due to the quantum interference effects of the electron waves traversing through the different branches of the benzene molecule. By comparing the current plots in the presence and absence of the gate potential, two features can be readily found. First, three configurations ortho, meta and para show different current magnitudes and in this regard meta configuration shows profound re- sponses to the gate potential. The second one is different threshold voltage. Indeed, the presence of gate potential indicates smaller threshold voltage than the ungated one. Now, we proceed to investigate the effects of semi- conductor electrodes on electronic transport. Note that we neglect all the interactions between current carriers and electron correlations are associated only with the Pauli principle. Then we can control the current func- tion the electrode/benzene/electrode system by contact- ing the benzene to the electrodes from different locations as a geometrical interference. B. Semiconductor electrode Having in examined hand, we the now consider metal/benzene/metal semiconduc- model junction within the AS tor/benzene/semiconductor and AB models. We first take two silicon electrodes in the AB framework. The semiconductor band gap com- pared with the metal electrodes is the most noticeable feature in the transmission plots of Fig.(9,11) as shown by the shadow region in each plot. In this region, the absence of states in the left electrode prevents electrons from injecting into the junction; also, there are no states for them to occupy once transmitted to the right electrode. A profound effect of semiconductor electrodes is the existence of threshold bias voltage Vt which is the minimum applied bias voltage needed to access states in either the valence or conduction band. in the semiconductor- molecule junction, is the molecule-induced semiconduc- tor surface states which can control charge transport across the junction37,38. Authors in Ref. [33] have shown that surface states can significantly enhance transport through molecular energy levels and addressed this to the role of the real part of the self-energy, Λ(E), which is non- negligible for semiconductor electrode. Fig.(7) shows how transmission function changes for different electrodes. It Another important issue 7 FIG. 7: (Color online) Transmissions function of junction with two silicon (top panel), titanium dioxide (middle panel) and gold (bottom panel) electrodes as a function of on-site energy of benzene molecule α and injected electron energy E. is clear that semiconductor electrode generally broads the high transmission region and this is also sensitive how molecule bonds to the semiconductor surface states. For example, the AS model transmission spectrum shows less symmetry than the AB model which is the direct conse- quence of density of states. In Fig.(8) we have depicted local density of states (LDOS) function for different elec- trodes. It is clear to see that LDOS of the AS model exhibits an asymmetry behavior versus energy. For de- tailed explain we refer the readers to the Ref. [33]. Alternating bond model(AB): In Fig.(9), we have plotted logarithmic scale of transmission function ver- FIG. 8: (Color online) Local density of states (LDOS) func- tion for junction with two silicon (top panel), titanium dioxide (middle panel) and gold (bottom panel) electrodes in the zero bias. The red lines are the energy levels of benzene. sus energy for three meta, ortho and para configurations, by considering 0, 1 and 2 surface states. As it can be seen from this figure transmission function shows equal anti-resonance points in comparison with the metal elec- trode. These points had been predicted by the graph- ical approach. Note, applying a gate potential only makes a tiny shift on the transmission probability and the anti-resonance points. However, in the ortho ar- rangement there is an anti-resonance point which is not shifted. Indeed, in all the three surface states the first anti-resonance peck occurs at the same point in gated and ungated cases. In the para configuration for the all three surface states, in the absence of gate potential there is not any anti-resonance nodes, which can be ad- 00.050.10.150.20.250.30.35−5−4−3−2−1012345LDOSEnergy(eV)00.050.10.150.20.250.30.35−505 lβ1l>lβ2llβ1l<lβ2lHOMOABLUMO00.511.522.53−5−4−3−2−1012345LDOSEnergy(eV) 00.511.522.53−505α>0α<0HOMOLUMOAS00.10.20.30.40.5−5−4−3−2−1012345LDOSEnergy(eV) 00.10.20.30.40.5−505HOMOLUMONA8 FIG. 9: (Color online) Transmission function versus configurations in 0, 1 and 2 surface states. The dotted (solid) lines corresponding to the gated (ungated) potential. FIG. 10: (Color online) The current-voltage (I-V) of the silicon/benzene/silicon junction for meta, ortho and para configurations in 0 ,1 and 2 surface states. The dotted (solid) lines corresponding to the gated (ungated) potential. dressed to the constructive superposition of the electron waves, but immediately after applying the gate potential its transmission shows an anti-resonance behaviour (same as metal electrode), particularly in the 1 surface state (dotted lines in the bottom panel of the three plots of Fig.(9). In Fig.(10), we have presented current as a func- tion of voltage for meta, ortho and para configurations and the three surface states. All the three current-voltage characteristics show stair-like feature. The notable effect which should be mentioned is that the silicon electrodes cause different threshold voltage Vt in comparison with gold electrodes in all the three meta, ortho and para con- figurations. Actually, in the presence of silicon electrode threshold voltage Vt occurs at lower bias. This effect is more profound in the ortho case. Furthermore, in the meta configuration for all the three surface states, cur- rent shows lower threshold voltage Vt in the gated case than the ungated one. By comparing all the three surface states, maximum magnitude of current amplitude is seen in the para configuration for the 0 surface state and the minimum one is seen in the ortho configuration for the 2 surface state. Alternating site model(AS): We now consider ti- tanium dioxide (T iO2) electrodes in the AS framework. In Fig.(11), we have plotted logarithmic scale of trans- mission function versus energy for three meta, ortho and para configurations in the presence and absence of gate potential which are shown in dotted and solid lines, re- spectively. For meta and ortho configurations in the top and mid- dle panels of Fig.(11), for all the three surface states, anti-resonance states happen in certain values of eigenen- ergies, due to the destructive quantum interference. The application of the gate potential produces a small change in the energy eigenvalues with respect to the ungated molecule and shifts the anti-resonance peaks. In Fig.(12), we have illustrated the current as a function of applied voltage for the three meta, ortho and para configurations −4−3−2−10123400.20.40.60.81Meta−4−3−2−10123400.20.40.60.8T(E)Ortho−4−3−2−10123400.51E(ev)Para WGG0 Surface state− AB model−4−3−2−10123400.51Meta−4−3−2−10123400.20.40.60.8T(E)Ortho−4−3−2−10123400.51E(ev)Para WGG1 Surface state− AB model−4−3−2−10123400.51Meta−4−3−2−10123400.51T(E)Ortho −4−3−2−10123400.51E(ev)Para WGG2 Surface state− AB model−6−4−20246−100−50050100Meta−6−4−20246−100−50050100OrthoI(µA)−6−4−20246−200−1000100200ParaV WGG0 Surface state− AB model−6−4−20246−100−50050100Meta−6−4−20246−50050OrthoI(µA)−6−4−20246−200−1000100200ParaV WGG1 Surface state− AB model−6−4−20246−100−50050100Meta−6−4−20246−100−50050100ParaV −6−4−20246−40−2002040OrthoI(µA)WGG2 Surface state− AB model9 FIG. 11: (Color online) Transmission function versus energy of the titanium-dioxide/benzene/titanium-dioxide junction for meta, ortho and para configurations in 0, 1 and 2 surface states. The dotted (solid) lines corresponding to the gated (ungated) potential. FIG. 12: (Color online) The current-voltage (I-V) of the titanium-dioxide/benzene/titanium-dioxide junction for meta, ortho and para configurations in 0 ,1 and 2 surface states. The dotted (solid) lines corresponding to the gated (ungated) potential. and the three surface states. For all the three meta, ortho and para configurations in the 0 surface state applying gate potential don't have any significant effects on the current profile. However, in the 1 surface state the ap- plied gate one shows the most different behaviour respect to the ungated case. The minimum of current amplitude is also seen in the 1 surface state. By comparing all the three meta, ortho and para configurations in the differ- ent surface states, one can easily find that the current has almost the same threshold voltage in the gated and ungated cases. IV. CONCLUSION We numerically investigated the coherent electron transport through a benzene molecule sandwiched be- tween metal or semiconductor electrodes. Two gener- alized tight-binding models, alternating site (AS) and alternating bond (AB) models, have been considered to represent the semiconductor leads. The well known Newns-Anderson model has been also used to mimic the metal leads as a comparison benchmark. Due to sym- metry breaking in the semiconductor leads surface states are formed. In this regard three kinds of surface states labeled as 0, 1 and 2 have been considered. The main goal of this report is to study the effect of semiconduc- tor electrode on the quantum interference characteristics of the benzene molecule. To this end, we have assumed three different terminal contact configurations (labeled as meta, ortho and para) and also a gate potential in or- der to verify the presence of quantum interference. Using the generalized Green's function technique based on the tight-binding model and the Landauer-Buttiker theory, we have calculated the transmission probability function and the current-voltage characteristics. −4−3−2−10123400.51Meta−4−3−2−10123400.51T(E)Ortho−4−3−2−10123400.51E(ev)Para WGG0 Surface state− AS model−4−3−2−10123400.51Meta−4−3−2−10123400.51T(E)Ortho−4−3−2−10123400.20.40.60.8E(ev)Para WGG1 Surface state− AS model−4−3−2−10123400.20.40.60.81Meta−4−3−2−10123400.20.40.60.8T(E)Ortho−4−3−2−10123400.51E(ev)Para WGG2 Surface state− AS model−6−4−20246−100−50050100Meta−6−4−20246−40−2002040OrthoI(µA)−6−4−20246−200−1000100200ParaV WGG0 Surface state− AS model−6−4−20246−40−2002040Meta−6−4−20246−20−1001020OrthoI(µA)−6−4−20246−40−2002040ParaV WGG1 Surface state− AS model−6−4−20246−100−50050100Meta−6−4−20246−20−1001020I(µA)Ortho−6−4−20246−200−1000100200VPara WGG2 Surface state− AS modelThe first remarkable feature is the existence of a bias voltage threshold Vt in the transmission probability func- tion and the current-voltage figures. This threshold is linked to the hlsemiconductor band gap and it is the min- imum applied bias voltage needed to access the states in either the valence or conduction band. The second observed behavior is that the profile of the transmis- sion probability function semiconductor electrodes shows an irregular pattern, in contrast to the metal electrode. This effect can be explained by molecular-level shifting imposed by adsorption to semiconductors and are very sensitive to how the molecule bonds to the semicon- ductor surface. Nevertheless, in the integrated quan- tities, such as the current-voltage characteristics, no such behaviour is observed. The third noticeable fea- ture is that the current magnitudes for the semiconduc- tor/benzene/semiconductor systems are higher than its metal counterpart which is linked to the broadening of transmission probability function. The gate potential can also create some anti-resonance peaks in the transmission probability and has a pro- found effect on the current magnitude in the meta configuration in both semiconductor and metal electrode cases. Moreover, we have explained how the quantum interference (QI) can create zeroes in the transmission function of the benzene molecule for all three electrode configurations using simple graphical schemas involving solely the topology of the molecule. We hope the results open up new possibilities for the design of single-molecule devices sandwiched between semicon- ductor leads based on quantum interference effects, for instance, switching devices that operate by com- bining destructive and constructive molecular structures. Appendix A: Calculation of self-energy ΣKD(E) 10 in Ref.[33] to find the solution step by step. We start by Eq.(7), ΣKD(E) = γ∗ 2 ΓKD(E), −1 is the surface green where gKD(E) = ((E) − HKD) function of the left/right electrode and HKD is their Hamiltonian c gKDγc = ΛKD(E) − i HKD = (A1)  0 ··· 0 ··· 0 ··· 0 β1 −α β2 ··· 0 β2 α ··· 0 ... ... . . . 0 α β1 0 β1 −α β2 0 0 β2 α β1 0 0 ... ... ...  In general the calculation of the Green's function of an operator H requires the preliminary diagonalization of H. However, for tridiagonal operators the calculation is straightforward. For operator H which has the tridiag- onal matrix form, the operatorE − H is also tridiagonal form. So one can write the Green's function diagonal matrix element as continued fraction. For our purpose which we are interested in surface Green's function one can achieve the following expression gKD(E) = E − α − (A2) 1 E + α − β2 1 β2 2 E − α + β2 1··· To perform the continued fraction sum, we recast the above equation as gKD(E) = 1 E − α − β2 1 E + α − β2 2 gKD(E) (A3) In order to have a self-contained and pedagogical manuscript, we have tried to go through the details of Eq.(8,9). In this regard, we adapt the procedure used after a simple calculation one can find Eq.(9). Now sub- stitution this solution into the self-energy it is easy to have the following expression ΣKD(E) γ2 = ΛKD(E) − i 2 E2 − α2 − β2 = ΓKD(E) 2 ±(cid:113)(cid:2)E2 − α2 − (β1 − β2)2(cid:3)(cid:2)E2 − α2 − (β1 + β2)2(cid:3) 1 + β2 2 (E − α) 2β2 (A4) having this equation, it is simple to find ΣN A(E), ΣAB(E) and ΣAS(E) equations. Here we For the NS model by setting α = 0 and β1 = β2 = β we have ΣN A(E) It is easy to see that ΣN A is complex for E2 − 4β2 < 0(E < 2β), but for E2−4β2 2β2 = E± √ γ2 . E < −2β and E > 2β is real and up to a sign it shows ΣN A(E → ∞) → 0. The real and complex be- haviour of ΣN A(E) can be referred to shifts and broaden- ing of molecular energy levels. So imaginary part of self- (cid:16) ΣN A(E) (cid:17) γ2 √ = ± 4β2−E2 2β2 energy Im gives spectral density (cid:112)(4β2 − E2) and the real part of self- √ as ΓN A(E) = γ2 β2 E2−4β2 energy gives broadening as ΛN A(E) 2β2 where ΘN A = Θ(−2β − E) − Θ(E − 2β). The same procedure holds for ΣAB(E) and ΣAS(E) equations and it is easy to ascertain. γ2 = E 2β2 +ΘN A 11 References 1 G. Cuniberti, G. Fagas and K. Richter, Introducing Molecular Electronics., Lect. Notes Phys. 680 (Springer, NewYork, 2005). 2 R.Baer and D. Neuhauser, J. Am. Chem. Soc. 124, 4200(2001). 3 R. Stadler, M. Forshaw and C. Joachim, Nanotechnology 14, 138 (2003). N. Agrat, Nano Lett. 8, 1673 (2008). 22 J. S. Kristensen, D. J, Mowbray, K. S. Thygesen, K. W.Jacobsen, J. Phys.: Condens. Matter. 20, 374101 (2008). 23 W. Tian, S. Datta, S. Hong, R. Reifenberger, J. I. Hender- son and C. P. Kubiak, J. Chem. Phys. 109, 2874 (1998). 24 R. M. Metzger, T. Xu and I. R. Peterson, J. Phys. Chem. 4 R. Stadler, S. Ami, M. Forshaw and C. Joachim, Nanotech- B. 105, 7280 (2001). nology 15, S115 (2004). 5 R. Stadler, K. S. Thygesen and K.W. Jacobsen, Nanotech- nology 16, S155 (2005). 6 T. A. Papadopoulos, I. M. Grace and C. J. Lambert, Phys. Rev. B 74, 193306 (2006). 7 D. M. Cardamone, C. A. Stafford and S. Mazumdar, Nano Lett. 6, 2422 (2006). 25 D. R. Ward, N. J. Halas, J. W. Ciszek, J. M. Tour, Y. Wu, P. Nordlander and D. Natelson, Nano Lett. 8, 919 (2008). 26 R. McCreery, J. Dieringer, A. O. Solak, B. Snyder, A. M. Nowak, W. R. McGovern and S. Duvall, J. Am. Chem. Soc. 125, 10748 (2003). 27 N. P.Guisinger, M. E.Greene, R. Basu, A. S. Baluch and M. C. Hersam, Nano Lett. 4, 55 (2004). 8 C. A. Stafford, D. M. Cardamone and S. Mazumdar, Nan- 28 N. P.Guisinger, N. L. Yoder and M. C. Hersam, Proc. Natl. otechnology 18, 424014 (2007). 9 S. H. Ke, W. Yang and H. U. Baranger, Nano Lett. 8, 3257 (2008). 10 F. Sols, M. Macucci, U. Ravaioli and K. Hess, Appl. Phys. Lett. 54, 350 (1989). Acad. Sci. U.S.A. 102, 8838 (2005). 29 P. G.Piva, G. A. DiLabio, J. L. Pitters, J. Zikovsky, M. Rezeq, S. Dogel, W. A. Hofer and R. A. Wolkow, Nature 435, 658 (2005). 30 T. Rakshit, G. C. Liang, A. W. Ghosh and S. Datta, Nano 11 W. Porod, Z. Shao and C. S. Lent, Appl. Phys. Lett. 61, Lett. 4, 1803 (2004). 1350 (1992). 31 T. Rakshit, G. C. Liang, A. W. Ghosh, M. C. Hersam and 12 W. Porod, Z. Shao and C. S. Lent, Phys. Rev. B 48, 8495 S. Datta, Phys. ReV. B. 72, 125305 (2005). (1993). 13 P. Debray, O. E. Raichev, P. Vasilopoulos, M. Rahman, R. Perrin and W. C. Mitchell, Phys. Rev. B 61, 10950 (2000). 14 P. Sautet and C. Joachim, Chem. Phys. Lett. 153, 511 (1988). 15 C. Patoux, C. Coudret, J. P. Launay, C. Joachim and A. Gourdon, Inorg. Chem. 36, 5037 (1997). 16 R. Stadler, Phys. Rev. B. 80, 125401(2009). 17 M. Magoga and C. Joachim, Pys. Rev. B. 56, 4722 (1997). 18 S. N. Yaliraki and M. Ratner, A. J. Chem. Phys. 109, 5036 (1998). 19 E. G. Emberly and G. Kirczenow, Phys. Rev. B. 58, 10911 (1998). 20 L. E. Hall, J. R. Reimers, N. S. Hush and K. Silverbrook, J. Chem. Phys. 112, 1510 (2000). 21 J. Hihath, C. R. Arroyo, G. Rubio-Bollinger, N. Tao, and 32 D. Nozaki, S. M. Avdoshenko, H. Sevinli, R. Gutierrez1 and G. Cuniberti, J. Phys.: Conf. Ser. 427, 012013 (2013). 33 M. G. Reuter, T. Hansen, T. Seideman,and M. A. Ratner, Journal of Physical Chemistry A. 113, 4665 (2009). 34 I. Tamm, Phys. Z. Sowjetunion 1, 733 (1933). 35 W. Shockley, W. Phys. ReV. 56, 317 (1939). 36 T. Markussen, R. Stadler and K. S. Thygesen, Nano Lett. 10, 4260 (2010). T. Markussen, R. Stadler and K. S. Thygesen, Chem. Chem. Phys. 13, 1431 (2011). 37 N. L. Yoder, N. P. Guisinger, M. C. Hersam, R. Jorn, C. C. Kaun and T. Seideman, Phys. Rev. Lett. 97, 187601 (2006). 38 L. H. Yu, N. G. Hackett, C. D. Zangmeister, C. A. Hacker, C. A. Richter and J. G. Kushmerick, J. Phys.: Condens. Matter 20, 374114 (2008)
1905.11521	1	1905	2019-05-27T21:48:26	Quantization of graphene plasmons	[ "cond-mat.mes-hall", "quant-ph" ]	In this article we perform the quantization of graphene plasmons using both a macroscopic approach based on the classical average electromagnetic energy and a quantum hydrodynamic model, in which graphene charge carriers are modeled as a charged fluid. Both models allow to take into account the dispersion of graphenes optical response, with the hydrodynamic model also allowing for the inclusion of non-local effects. Using both methods, the electromagnetic field mode-functions, and the respective frequencies, are determined for two different graphene structures. we show how to quantize graphene plasmons, considering that graphene is a dispersive medium, and taking into account both local and nonlocal descriptions. It is found that the dispersion of graphene's optical response leads to a non-trivial normalization condition for the mode-functions. The obtained mode-functions are then used to calculate the decay of an emitter, represented by a dipole, via the excitation of graphene surface plasmon-polaritons. The obtained results are compared with the total spontaneous decay rate of the emitter and a near perfect match is found in the relevant spectral range. It is found that non-local effects in graphene's conductivity, become relevant for the emission rate for small Fermi energies and small distances between the dipole and the graphene sheet.	cond-mat.mes-hall	cond-mat	a Quantization of graphene plasmons Beatriz A. Ferreira1,2, B. Amorim1,3, A. J. Chaves4, and N. M. R. Peres1,2∗ 1Department of Physics, Center of Physics, and QuantaLab, University of Minho, Campus of Gualtar, 4710-057, Braga, Portugal 2International Iberian Nanotechnology Laboratory (INL), Av. Mestre Jos´e Veiga, 4715-330 Braga, Portugal 3CeFEMA, Instituto Superior T´ecnico, Universidade de Lisboa, Av. Rovisco Pais, 1049-001 Lisboa, Portugal and 4Department of Physics, Instituto Tecnol´ogico de Aeron´autica, DCTA, 12228-900 Sao Jos´e dos Campos, Brazil In this article we perform the quantization of graphene plasmons using both a macroscopic ap- proach based on the classical average electromagnetic energy and a quantum hydrodynamic model, in which graphene charge carriers are modeled as a charged fluid. Both models allow to take into account the dispersion of graphenes optical response, with the hydrodynamic model also allowing for the inclusion of non-local effects. Using both methods, the electromagnetic field mode-functions, and the respective frequencies, are determined for two different graphene structures. we show how to quantize graphene plasmons, considering that graphene is a dispersive medium, and taking into account both local and nonlocal descriptions. It is found that the dispersion of graphene's opti- cal response leads to a non-trivial normalization condition for the mode-functions. The obtained mode-functions are then used to calculate the decay of an emitter, represented by a dipole, via the excitation of graphene surface plasmon-polaritons. The obtained results are compared with the total spontaneous decay rate of the emitter and a near perfect match is found in the relevant spectral range. It is found that non-local effects in graphene's conductivity, become relevant for the emission rate for small Fermi energies and small distances between the dipole and the graphene sheet. I. INTRODUCTION In many cases, light-matter interaction can be under- stood in a semi-classical picture, where matter is quan- tized and the electromagnetic field (EM) is treated classi- cally. This semi-classical approach holds when the num- ber of photons in the field is large or the light source is coherent. On the other hand, in order to understand the properties of a small number of photons the quantization of the EM field is required. Typical phenomena where the quantization of the EM field is necessary involve entan- glement, squeezed light, cavity electrodynamics, inter- action of photons with nano-mechanical resonators, and near-field radiative effects [1]. In near-field radiative effects, plasmonics emerges as a promising candidate to observe quantum effects of the electromagnetic radiation, an example being the Hong- Ou-Mandel interference of plasmons [2]. Many other quantum effects in plasmonics exist, such as the survival of entanglement, particle-wave duality, quantum size ef- fects due to reduced dimensions of metallic nanostruc- tures, quantum tunneling of plasmons (which are simul- taneously light and matter), and coupling of quantum emitters to surface plasmons [3 -- 11]. The exploration of quantum effects in plasmonics in unusual spectral ranges, such as the THz and the mid- IR, has been deterrent by the poor plasmonic response of noble metals in these regions of the electromagnetic spectrum. However, with the emergence of graphene plasmonics [12, 13] the possibility of exploring quantum ∗ Contact e-mail:[email protected] Figure 1. Schematic representation of the three systems con- sidered in this paper: a single graphene layer (left), and graphene double layer (center), and a graphene sheet near a perfect metal. The quantities n refer to the dielectric per- mittivity of the medium n and σg refers to the optical con- ductivity of graphene. effects in these yet unexplored spectral regions is a pos- sible prospect. Despite the fact that, at the time of writ- ing, quantum effects involving graphene plasmons remain illusive, the fact that graphene plasmons are character- ized by low losses [14 -- 16] boosts the above hope. In addition, graphene plasmons screened by a nearby metal (also called screened or acoustic plasmons) can be con- fined down to the atomic limit [17], which certainly opens the prospects of finding rich grounds for quantum plas- monics and nonlocal effects [18, 19]. Indeed, the idea of developing quantum optics with plasmons has already a long history [20] and quantization of localized plasmons in metallic nanoparticles was recently performed [21, 22]. The development of quantum theory of the electromag- netic field in the presence of dieletric media has a long his- tory and several approaches have been developed [23 -- 35]. These methods are typically based either on the quantiza- tion of the macroscopic classic energy[27], on the classical dyadic Green's function for the electric field[28] or on the diagonalization of Hopfield-type Hamiltonians[26]. The --quantization of evanescent EM waves [36, 37] and of the EM field in the vicinity of a metal [38] have also been considered in the past. In this paper, we perform the quantization of graphene plasmons, obtaining the plasmonic electromagnetic field mode-functions and, importantly, their normalization, when losses are neglected. These mode-functions are then used to study the interaction of graphene plas- mons with nearby quantum emitters and determine, in a very intuitive way, using Fermi's golden rule the spon- taneous decay rate of the emitter due to plasmon emis- sion. Thee quantization of graphene plasmons is per- formed in two ways: (i) A macroscopic approach, which starts from the classical time-averaged energy of the elec- tromagnetic field in a dielectric [24, 27, 39, 40]. This method allows for the inclusion of dispersion in the opti- cal response of graphene. (ii) A hydrodynamic approach, where graphene charge carriers are described in terms of an electronic fluid, which couples to the eletromag- netic field [41, 42]. The hydrodynamic approach allows not only for the inclusion of dispersion, but also for the inclusion of non-local effects in the optical response of graphene. The paper is organized as follows: in Section II, we present the general macroscopic approach for the quan- tization of the eletromagnetic field in dispersive, loss- less media and a normalization condition for the mode- functions is determined. In Section III, we present the quantum hydrodynamic model or graphene. We will see that when non-local effects are neglected, the result of the macroscopic approach is recovered. In Section IV, the plasmon dispersion relations, mode-functions and mode- function normalization for a single graphene layer and for a graphene-metal structures are obtained. In Section V, we use the quantized plasmon fields to compute the decay rate of a quantum emitter due to the spontaneous emis- sion of plasmons. The plasmon emission rate is compared with the total decay rate of the emitter, which is obtained from the complete dyadic Green's function for the electric field. The role of non-local response of graphene is ana- lyzed. Finally, we conclude with Section VI, commenting the obtained results and discussing future research paths. A set of appendices detailing the calculations is also pre- sented. II. MACROSCOPIC QUANTIZATION OF THE PLASMON ELECTROMAGNETIC FIELD In this section, we will describe how the plasmon fields can be quantized using the macroscopic classical energy of the electromagnetic field in a dispersive, lossless di- electric medium, a method first used in Refs. [24, 27, 40]. For electric and magnetic fields with a harmonic time dependence, E(r, t) = Eω(r)e−iωt + c.c. B(r, t) = Bω(r)e−iωt + c.c. (1) (2) 2 (4) (5) (6) (cid:90) (cid:18) close to a central frequency ω, the time-averaged classical electromagnetic energy in the presence of a dispersive, lossless dieletric is given by [39, 43] UEM(ω) = d3r 0E∗ ω(r) · ∂ ∂ω [ω¯r(r, ω)] · Eω(r) (cid:19) + 1 µ0 Bω2 , (3) where ¯r(r, ω) is the relative dielectric tensor of the medium, 0 and µ0 are, respectively, vacuum permittiv- ity and permeability. The idea of this method is to take the above equation as the quantum mechanical energy of a EM field eigenmode with frequency ω. We will work in the Weyl gauge, in which the scalar potential is set to zero φ = 0, such that the electric and magnetic fields are obtained only from the vector potential A as E(r, t) = − ∂A(r, t) , B(r, t) = ∇ × A(r, t). ∂t (cid:88) The vector potential is then expanded in modes as A(r, t) = αλe−iωλtAλ(r) + c.c., λ where αλ are amplitudes for the mode λ, with mode- function Aλ(r) and corresponding frequency ωλ. The mode-functions and frequencies are determined by solv- ing the non-linear eigenvalue problem ∇ × ∇ × Aλ(r) = ω2 λ c2 ¯r(r, ωλ) · Aλ(r), (7) with c vacuum's speed of light, which is just Amp`ere's law in the dielectric medium for the mode-function Aλ(r). Next, we assume that the total time-averaged energy for the vector potential Eq. (6) is given by UEM = UEM(ωλ)αλ2 . (8) (cid:88) λ The quantization of the theory is achieved by promoting the amplitudes αλ to quantum mechanical operators (cid:114) (cid:114) 2Lλ0ωλ 2Lλ0ωλ αλ → † λ → α (cid:104) † aλ, a λ(cid:48) (cid:105) aλ, † a λ, (9) (10) † where aλ (a λ) are bosonic annihilation (creation) opera- tors, which obey the usual equal-time commutation rela- tions = δλ,λ(cid:48), (11) and Lλ is a mode-length which is determined by demand- ing that the quantum mechanical Hamiltonian which is obtained from Eq. (8) by performing the replacement of Eqs. (9) and (10), (cid:88) λ H = U (ωλ) 4Lλ0ωλ (cid:16) † † λaλ + aλa a λ , (12) coincides with the Hamiltonian for a collection of quan- tum harmonic oscillators (cid:17) (cid:17) Imposing this condition, we have that the mode-length is † † λaλ + aλa a λ . (13) (cid:1). Using the mode-function 1 2 H = ωλ (cid:88) (cid:16) given by Lλ = U (ωλ)/(cid:0)20ω2 (cid:90) (cid:90) equation (7) to write d3r∇ × Aλ(r)2 = λ λ = ω2 λ c2 d3rA∗ (cid:90) (cid:90) (cid:18) d3rA∗ λ(r) · ∇ × ∇ × Aλ(r) (14) λ(r) · ¯r(r, ωλ) · Aλ(r), (cid:19) The mode-length can be written as Lλ = ωλ 2 ∂ ∂ω ¯r(r, ωλ) λ(r)· d3rA∗ ¯r(r, ωλ) + ·Aλ(r). (15) Notice that in the absence of dispersion, the second term vanishes and Lλ reduces to the usual norm in the pres- ence of a position dependent dielectric tensor ¯r(r, ωλ). Although this phenomenological method appears to be unjustified, it has actually been shown to be correct for the case in when the dielectric is modeled by a Lorentz oscillator [40]. We will see in the next section, that Eq. 15 remains valid within a quantum hydrodynamic model of graphene, even when non-local effects are included in the optical response. As a matter of fact Eq. (15) for the mode-length remains valid for any linear optical medium (including effects of dispersion, non-locality, inhomogene- ity and anisotropy) as long as losses can be neglected [44]. III. PLASMON QUANTIZATION WITHIN A QUANTUM HYDRODYNAMIC MODEL In this section, we will perform the quantization of graphene surface plasmons employing a hydrodynamic model. The hydrodynamic model treats both the elec- tron gas and the electromagnetic field using a classical picture and provides a simple and elegant way of in- cluding non-local effects [45]. Non-local effects are taken into account by including a pressure term in the Boltz- mann equation, that arises due to Pauli's exclusion prin- ciple. A detailed derivation of the hydrodynamic model for graphene can be found in [41, 42]. To illustrate the method, we choose the simple case of a single graphene sheet located at the plane z = 0, embedded by a static dielectric medium with relative dielectric constant ¯d(r). 3 A. Classical hydrodynamic Lagrangean and Hamiltonian The classical Lagrangean density for the hydrodynamic model of graphene can be written as: L = LEM + L2D hyd, (16) where LEM is the Lagrangean density of the electromag- netic field and L2D hyd describes the electronic fluid of graphene and its coupling to the electromagnetic field. Using once again the Weyl gauge, LEM is given by (∇ × A)2 , (∂tA) · ¯d · (∂tA) − 1 2µ0 LEM = 0 2 (17) where ¯d(r) is allowed to be position dependent, but is frequency independent. Within the hydrodynamic model, the electronic fluid of the graphene layer is de- scribed by the fluctuation, n, of the density around the equilibrium density, n0, and the displacement vec- tor υ = (υx, υy, 0), which should not be confused with the velocity field. In the Weyl gauge, L2D fluid is written as[41, 42] (cid:32) L2D hyd = δ(z) n0m (∂tυ)2 + mβ2n∇ · υ 1 2 + mβ2 2n0 n2 − n0e∂tυ · A (cid:19) , (18) where the δ-function δ(z) restricts the electronic fluid to the z = 0 2D plane, m is the Drude mass and β appears from the relation between the degeneracy pres- sure and the electronic density and depends on the band dispersion for the carriers (see Ref. [41]). In the ap- proximation of the linear dispersion for graphene elec- trons, the hydrodynamic parameters are given by [41]: n0 = k2 F /2, where kF is the Fermi wavenumber and vF the Fermi velocity of graphene. Equation (18) is the 2D equivalent Lagrangian for the hydrodynamic model presented in [46]. F /π, m = kF /vF and β2 = v2 Using the Euler-Lagrange equations for Eq. (18) with respect to A, we obtain ∇ × B = 1 c2 ¯d∂tE − µ0n0e∂tυδ(z), (19) which is nothing more than Amp`ere's law in the presence of a surface current given by s = −en0∂tυ. Jhyd (20) Using the Euler-Lagrange equations for Eq. (18) with respect to the fluid variables n and υ, we obtain the con- tinuity and Newton's second law with a diffusion term, which read respectively n0∇ · υ + n = 0, t υ + mβ2∇n = −en0E(z = 0), n0m∂2 (21) (22) from which we recognize the fluid electronic surface den- sity s = −en. ρhyd (23) Equations (19)-(22) correspond to the linearized hydro- dynamic model for graphene [41] (see also [47]). Notice that Eq. (21) has no dynamics. Therefore, we can use it to eliminate the field n, thus obtaining a new Lagrangean density L(cid:48) = LEM + L(cid:48) 2D hyd, (cid:20) with (cid:32) L(cid:48) 2D hyd = δ(z) 1 2 n0m (∂tυ)2 (24) (25) (cid:33) n0mβ2 (∇ · υ)2 − n0e∂tυ · A. − 1 2 This new Lagrangean is equivalent to the Eq. (16) as both lead to the same dynamics. Applying the Euler- Lagrange equations to Eq. (24) with respect to A leads to Eq. (19), while the equation obtained for υ reads − m∂2 t υ + mβ2∇ (∇ · υ) = eE(z = 0), (26) which can be obtained by combining Eqs. (21) and (22). From the Lagrangean density Eq. (24), we define the canonical momenta conjugate to A and υ, respectively, as Π = π = ∂L(cid:48) ∂ (∂tA) ∂L(cid:48) ∂ (∂tυ) = 0¯d∂tA, = n0m∂tυ − n0eA(z = 0). (27) (28) In terms of the variables A, Π, υ and π, the classical Hamiltonian obtained from Eq. (24) is given by (cid:90) (cid:18) 1 d3r (cid:32) H = (cid:90) + d2x Π · ¯−1 d · Π + 1 2µ0 (∇ × A)2 20 (π + en0A(z = 0))2 2n0m n0mβ2 (∇ · υ)2 + 1 2 (cid:19) (cid:33) . (29) B. Canonical quantization of hydrodynamic model In order to quantize the classical Hamiltonian Eq. (29), we start by introducing the eigenmodes of the coupled electronic fluid + electromagnetic field. Assuming, we have in-plane translational invariance, we write the vec- tor potential and fluid displacement variables as αq,λ(t)eiq·xAq,λ(z) + c.c., A(r, t) = (30) (cid:88) (cid:88) q,λ q,λ 1√ S 1√ S υ(x, t) = αq,λ(t)eiq·xυq,λ + c.c., (31) H = 1 2 2ω2 q,λ0Lq,λα∗ q,λαq,λ + c.c, (40) (cid:21) where S is the area of the graphene layer, αq,λ(t) = αq,λe−iωq,λt are mode amplitudes with, ωq,λthe mode fre- quency, and Aq,λ(z) and υq,λ are mode-functions, which, following from Eqs. (19) and (26), obey the equations q,λ0¯d(z) − 1 ω2 µ0 Dq × Dq× Aq,λ(z) = = −iωq,λδ(z)en0υq,λ, (cid:2)ω2 q,λ − β2q ⊗ q(cid:3) υq,λ = iωq,λen0Aq,λ(0). (32) (33) mn0 where we defined the differential operator Dq = iq + z∂z. From Eq. (33), we can write υq,λ = 1 en0 ¯σhyd g (q, ωq,λ) · Aq,λ(0), (34) g where ¯σhyd (q, ω) is the conductivity within the hydro- dynamic model, which we separate into transverse and longitudinal components as (cid:19) (cid:18) ¯1 − q ⊗ q q ⊗ q q2 q2 , ¯σhyd g (q, ω) = σhyd g,T (q, ω) + σhyd g,L (q, ω) respectively given by σhyd g,T (q, ω) = σhyd g,L (q, ω) = e2n0 m e2n0 m , i ω ω2 − β2q2 , iω 4 (35) (36) (37) where we identify D = e2n0/m as the Drude weight, which for graphene is given by D = e2vF kF /π. Notice that in the the limit q → 0, ¯σhyd (q, ω) reduces to the Drude model. Inserting Eq. (34) into Eq. (32), we obtain g Dq × Dq × Aq,λ(z) = q,λ ω2 c2 ¯r (q, z, ωλ) Aq,λ(z), (38) with the dieletric function, graphene electrons, being given by including screening by ¯r (q, z, ω) = ¯d(z) + i 0ω ¯σhyd g (q, ω) δ(z), (39) in agreement with Eq. (7). Inserting the expansions Eq. (30) and (31) into Eq. (29), and using the orthogonality properties of the mode-functions (Aq,λ(z), υq,λ) it is possible to write the Hamiltonian for the hydrodynamic model as (see Ap- pendix (B)) (cid:88) qλ with the mode-length defined here as Lq,λ = dzA∗ q,λ(z) · ¯d(z) · Aq,λ(z) + (cid:90) (cid:0)A∗ + ien0 20ωq,λ n0m 0 q,λ · υq,λ υ∗ q,λ · Aq,λ(0)(cid:1) . (41) Promoting the mode amplitudes to quantum mechanical operators as q,λ(0) · υq,λ − υ∗ (cid:115) (cid:115) 20ωq,λLq,λ 20ωq,λLq,λ q,λ(t) → α∗ αq,λ(t) → † a q,λ(t), (42) aq,λ(t), (43) † where a q,λ(aq,λ) are creation (annihilation) operators for the plasmon-polaritons, satisfying the usual equal-time commutation relations Eq. (11), the quantum Hamilto- nian for the hydrodynamic model becomes ωq,λ † a q,λaq,λ + aq,λa † q,λ . (44) (cid:88) qλ H = 1 2 (cid:104) (cid:105) We will now see that Eq. (41) can be recast in the same form as Eq. (41). In order to do so, we use Eq. (34) which allows to write Eq. (41) as Lq,λ = + e2 0n0 dzA∗ q,λ(z) · ¯d(z) · Aq,λ(z) βq2 q,λ(0) · q ⊗ q A∗ q2 e2 0n0 (ω2 − βq2)2 · Aq,λ(0). (45) It is easy to see that the above equation can also be written as (cid:32) Lq,λ = dzA∗ qλ(z) · ¯r(q, z, ωλ) (cid:33) · Aq,λ(z), (46) (cid:12)(cid:12)(cid:12)(cid:12)ω=ωλ + ωλ 2 ∂ ∂ω ¯r(q, z, ω) (cid:90) (cid:90) 5 where ¯r (q, z, ωλ) = ¯d (z) + (cid:88) (cid:96) i ¯σg(cid:96)(q, ω) 0ω δ(z − z(cid:96)), (48) where ¯d (z) is the dielectric constant of the medium, which we assume to be isotropic and a piecewise ho- mogeneous function of z, and (cid:96) labels the graphene layers which are located at the planes z = z(cid:96), with conductivity¯σg(cid:96)(q, ω). We model the conductivity of each graphene layer with Eq. (35), which when includ- ing losses becomes σg,T (q, ω) = D i ω + iγ , σg,L(q, ω) = D iω ω2 + iωγ − β2q2 , (49) (50) where D = 4EF σ0/ (π) is the Drude weight, with EF graphene's Fermi energy, σ0 = πe2/ (2h) and γ is a re- laxation rate. When determining mode-functions we will set γ = 0, but we allow for γ (cid:54)= 0, for the situation when ohmic losses are included in Section (V). The presence of the δ-functions in Eq. (48), implies that boundary con- ditions at each interface located at z = z(cid:96): z ×(cid:0)Eq,λ z ×(cid:0)Bq,λ (cid:0)z+ (cid:1) − Eq,λ (cid:1) − Bq,λ (cid:0)z+ (cid:96) (cid:0)z− (cid:0)z− (cid:96) (cid:1)(cid:1) = 0, (cid:1)(cid:1) = µ0Js,q,λ(z(cid:96)), (cid:96) (cid:96) (52) where Eq,λ (z) = iωq,(cid:96)Aq,λ(z) and Bq,λ (z) = Dq × Aq,λ(z) are the electric and magnetic fields correspond- ing to mode Aq,λ(z), and Js,q,λ(z(cid:96)) = ¯σg(cid:96)(q, ω)·Eq,λ (z(cid:96)) is the surface current in the graphene layer (cid:96). In addi- tion to the boundary conditions Eqs. (51) and (52), to determine of the plasmon modes one must impose that the fields decay for z → ±∞. Having determined the plasmon mode-function, Aq,sp(z), and dispersion, ωq,sp, the mode-length can be obtained from Eqs. (46) and (48) as (51) (cid:90) (cid:88) with ¯r(q, z, ω) given by Eq. (39). IV. DISPERSION RELATIONS AND MODE-FUNCTIONS OF GRAPHENE PLASMON IN TWO DIFFERENT STRUCTURES + q,sp(z) · Aq,sp(z) Lq,sp = i 0ωq,λ (cid:96) dzd(z)A∗ q,sp(z(cid:96))· ∂ A∗ ∂ω [ω¯σg(cid:96)(q, ω)]ω=ωq,sp ·Aq,sp(z(cid:96)). (53) We now wish to determine the dispersion relation and mode-functions of graphene plasmons in two different structures (see Fig. 1): a single graphene layer and a graphene sheet in the vicinity of a perfect metal. As in the previous section, we make use of the in-plane translational invariance of the structures being consid- ered. Therefore, the non-linear eigenvalue problem for the mode-functions, Eq. 7, can be written as Dq × Dq × Aq,λ(z) = q,λ ω2 c2 ¯r (q, z, ωλ) Aq,λ(z), (47) We will now analyse the different structures in detail. A. Single layer graphene We first discuss the simplest case of a single graphene sheet (see left panel of Fig. 1). The problem of finding the spectrum of the surface plasmons in a graphene sheet was first considered in [48] and a detailed derivation can be found in Refs. [49, 50]. We assume that the single layer of graphene is located at z = 0, with a encapsulating dielectric medium n = 2 for z > 0 and a medium n = 1 for z < 0, such that (cid:40) d(z) = 2, 1, z > 0 z < 0 . (54) d(z) = 6 We assume that the metal interface is located at z = −d and the graphene layer is located at z = 0. The dieletric constant is given by (cid:40) z > 0 2, 1, 0 > z > −d . (61) (cid:40) Noticing that the plasmon field should decay for z → ∞, the mode-function should have the form 1 u− A+ A+ 2 u+ 1 u+ 1,qeκ1z 2,qe−κ2z 1,qe−κ1z + A− Aq,sp(z) = z > 0 0 > z > −d (62) Notice that the presence of a perfect metal at z = −d implies that the tangential component of the electric field should vanish. Imposing the previous condition and the boundary conditions Eqs. (51) and (52) at z = 0, we obtain a homogeneous system of equations .  1 −1 −1 ξ2 − 1 0 eκ1d e−κ1d + i σg,L 1 κ1 κ1  = 0,  A+ 2 A+ A− 1 1 where ξ2 = 2 0ω . The dispersion relation of the κ2 screened plasmons is obtained by looking for the zero the determinant of the previous matrix, which leads to the condition for the dispersion relation 1 κ1 coth (κ1d) + 2 κ2 + i σg ω0 = 0, (64) The boundary conditions imply that the mode-function is given by (63) , z > 0 , 0 > z > −d z cosh (κ1 (z + d)) 2,qe−κ2z i q q sinh (κ1 (z + d)) + + q κ1,q sinh (κ1d) u+ sinh2 (κ1,qd)(cid:0)κ2 (cid:20) 1 sinh (2κ1,qd)(cid:0)κ2 2,q 2 + q2(cid:1) 1,q + q2(cid:1) + dκ1,q1 (cid:21) ω2 c2 The mode-length can be determined from Eq. 53, and we obtain Aq,sp(z) = + 1 2κ3 1,q 2 In order to determine the plasmon mode, we look for p- polarized solutions of the electric field (the electric field lies in the plane of incidence) in the form of evanescent waves for z → ±∞. In each piecewise homogeneous re- gion we have that Dq · Eq,λ(z) = 0. The mode-function must then take the form 2,qe−κ2,qz, 1,qeκ1,qz, z > 0 z < 0 , (55) Aq,sp(z) = where (cid:40) A+ 2 u+ 1 u− A− (cid:115) √ with cn = c/ introduced the vectors u± n,q = i κn,q = , (56) q2 − ω2 q,sp c2 n n the speed of light in medium n, and we q q ∓ q κn,q z. (57) Imposing the boundary conditions Eqs. (51) and (52), we obtain the following implicit relation for the surface plasmon dispersion: 1 κ1,q + 2 κ2,q + i σg(q, ωq,sp) 0ωq,sp = 0, (58) In general, Eq. (58) has no analytical solution, except in the simple case where 1 = 2, in which case its solution reduces to solving a quadratic equation. The corresponding mode-function can be written as (cid:40) and the mode-length is obtained according to Eq. (53) as (cid:0)κ2 2 + q2(cid:1) + (cid:0)κ2 1 + q2(cid:1) 1 2κ3 1 Lq,sp = + 2 2κ3 2 D 0 β2q2 (ω2 − β2q2)2 , where the last term is due to the dispersion in the graphene layer. We point out, that within the hydro- dynamic model used for conductivity of graphene, this contribution is only non-zero if non-local effects are also included, that is if β (cid:54)= 0. Aq,sp(z) = 2,qe−κ2,qz, u+ u− 1,qeκ1,qz, z > 0 z < 0 , (59) Lsp,q = 2 2κ3 + sinh2 (κ1,qd) D 0 β2q2 (ω2 − β2q2)2 , (65) (60) where, as in Eq. (60), the last term is due to the disper- sion of graphene. We note that the dispersion relation for the screened plasmons, Eq. (64), is the same one that would be ob- tained for the acoustic plasmons in a symmetric graphene double layer structure (center panel of Fig. 1). In this structure, we have two graphene layers located at z = 0 and z = −dgdl. The dieletric constant of the encapsulat- ing medium is given by B. Graphene-metal structure We now move to the more complex case of a graphene sheet near a perfect metal (see right panel of Fig. 1). d(z) = z > 0 1, 0 > z > −dgdl 3, −dgdl > z , (66) 2, Since the plasmonic modes should decay for z → ±∞, the plasmon mode-function must have the form A+ Aq,sp(z) = 2,qe−κ2z 2 u+ 1 u1,qe−κ1z + A− A+ A− 3 u− 3,qeκ3z 1 u1,qeκ1z z > 0 0 > z > −dgdl −dgdl > z . (67) Imposing the boundary conditions Eqs. (51) and (52) at z = 0 and z = d, we obtain the following homogeneous system of equations  1 −1 − 1 κ1 eκ1d eκ1dgdl − 1 1 κ1 −1 −e−κ3d e−κ1d e−κ1dgdl ξ3e−κ3dgdl 0 0 κ1 ξ2 0 0 1 κ1  = 0,   A+ 2 A+ A− 1 A− 1 3 0ω + i σgtop,L + i σgbot,L and ξ3 = 3 κ3 (68) where ξ2 = 2 0ω . The dis- κ2 persion relation is obtained from zeroing the determinant of the previous matrix. Since the system is composed of two graphene sheets the double layer structure has two dispersion branches, a low energy one -- the acoustic mode -- and a high energy one -- the optical mode. In the op- tical mode, the charge oscillations in the two graphene sheets are in-phase; whereas in the acoustic mode, the charge oscillations are out-of-phase. In the particular case where, we have a symmetry structure, with 2 = 3 and the conductivities of the top and bottom graphene layers are the same σgtop,L = σgbot,L = σg,L, the zero- ing of the determinant factorizes into two independent expressions (cid:18) κ1ddlg (cid:18) κ1ddlg 2 (cid:19) (cid:19) 2 1 κ1 tanh 1 κ1 coth for the optical mode, and + 2 κ2 + i σg,L ω0 = 0 (69) + 2 κ2 + i σg,L ω0 = 0 (70) for the acoustic one. Notice that the relation for the acoustic mode dispersion Eq. (70) coincides with the equation for the screened plasmon Eq. (64) provided ddlg = 2d. This fact can be understood in terms of image charges as depicted in Fig. 2. In the bottom panel of Fig. 3 we depict the loss func- tion for the graphene-metal system. Clearly, only one branch is seen, which coincides with the acoustic branch of the double layer graphene upon considering d equal to half that of the double layer system, as explained in Fig. 2. An alternative way to obtain the plasmon dispersion relation is to look for poles (or resonances in the pres- ence of losses) in the so called loss function[50] , which is defined as L(q, ω) = −Im [rp(q, ω)] , (71) where rp(q, ω) is the reflection coefficient of the structure in consideration for the p-polarization, and q and ω are 7 Figure 2. Comparison of the double layer graphene system and the graphene-metal case. Due to the image charges in the metal-graphene structure, the spectrum of the double layer graphene is equivalent to that of the graphene-metal system if we take into account that graphene is at at distance dgdl/2 from the metal, where dgdl is the interlayer distance in the double layer case. Therefore, the graphene-metal distance has to be d/2 to obtain the same spectrum as for the graphene double layer . For a full numerical equivalence of the two spectra it is also necessary to have 1 = 3 in the double layer geometry. the in-plane wavevector and frequency of the impingent radiation, respectively (see Appendix A). For a symmet- ric graphene double layer (1 = 3 and σgtop,L = σgbot,L) and neglecting losses γ → 0, this coefficient has poles at the solutions of Eqs. (70) and (69), as can be seen com- paring those equations with Eq. (A5). The loss function for the double layer graphene is depicted in the left panel of Fig. 3, as function of ω and of a dimensionless param- eter s = qc/ω which defines the dispersion relation of the single graphene layer, clad by two different dielectrics of dielectric functions 1 and 2, in the electrostatic limit by ω(s) = 4αEF 1 + 2 s, (72) where EF is the Fermi energy of graphene and α is the fine structure constant of vacuum. In the top panel of Fig. 3 two branches are clearly seen: a high energy one -- the optical branch -- and the acoustic branch at lower en- ergies. At high energies and high s the two branches merge and converge to the single layer branch. The reader may wonder why the lower branch starts at finite momentum. This happens due to the definition of the s parameter, which involves both the frequency and the real wavenumber q. This choice allows to clearly separate the two branches in the (ω, s) plane. V. APPLICATION: QUANTUM EMISSION CLOSE TO GRAPHENE STRUCTURES We will now apply the quantization of the plasmon modes in grahene structures to the problem of sponta- neous emission by a quantum emitter which is located above the structure. We model the quantum emitter as a two-level system embedded in medium 2 at position r0 = (0, 0, z0). The quantum emitter couples to the plas- monic field via dipolar coupling: Hsp-d = −d · Esp(r0), 8 surface plasmons in graphene is given by Fermi's golden rule [51, 52]: (cid:12)(cid:12)(cid:12)(cid:104)g; nq,sp + 1 d · Ee; nq,sp(cid:105)(cid:12)(cid:12)(cid:12)2 (cid:88) q Γsp = 2π δ (ω0 − ωq,sp) , (75) where ω0 is the transition frequency, g; nq,sp + 1(cid:105) rep- resents a final state with one more surface plasmon and the emitter in the ground state and e; nq,sp(cid:105) represents an initial state with nq,sp plasmons in graphene and the emitter in the excited state. The transition matrix ele- ment for spontaneous plasmon emission, when there are no surface plasmons in the initial state,is given by (cid:104)g; 1 d · Ee; 0(cid:105) = i dge · A∗ q,sp(z0), (76) With this result the transition rate reads 2S0Lq,sp (cid:115) ωq,sp (cid:12)(cid:12)dge · A∗ ωq,sp Lq,sp q,sp(z0)(cid:12)(cid:12)2 × × δ (ω0 − ωq,sp) , (77) = Nq dge2 e−2κ2,qz0× cos2 φ sin2 ψ + cos2 ψ , (78) q2 κ2 2,q (cid:33) (cid:90) Γsp = 1 4π0 d2q (cid:12)(cid:12)dge · A∗ q,sp(z0)(cid:12)(cid:12)2 (cid:32) × Using the mode-function we can write Figure 3. Loss function [Eq. (71)] for the p-polarization re- flection coefficient for: (Top panel) a graphene double layer structure; (Bottom panel) graphene near a perfect metal. The parameters use in the top panel are dgdl = 20 nm, 3 = 1, and for the bottom panel d = 10 nm. The remaining parameters used in both panels are: 1 = 3.9, 2 = 1, EF = 0.2 eV and γ = 10 meV. In both plots, non-local effects were neglected (β = 0). ψ is the angle the dipole makes with the axis perpendic- ular to graphene (z-axis), and φ is the azimuthal angle. The prefactor Nq is define as Nq = 1 for the single- layer graphene case and as Nq = sinh2 (κ1,qd) for the graphene+metal structure. Using the in-plane isotropy of the system, the momentum integration in Eq. (77) can be trivially performed, yielding: (cid:18) ∂ωq0,sp (cid:32) ∂q × (cid:19)−1 dge2 40 q2 κ2 2,q0 e−2κ2,q0 z0× (cid:33) sin2 ψ + 2 cos2 ψ , (79) Γsp = Nq0 q0ωq0,sp Lq0,sp with d = dgec† (cid:32) gce + h.c. (cid:115) ωq,sp i 2S0Lq,sp (cid:88) q Esp(r) = (73) (cid:33) Aq,sp(z)eiq·xaq,sp + h.c. (74) (cid:68) † where c g/e(cg/e) is the creation (annihilation) operator for the ground/excited state of the two-level system and dge = is the dipole matrix element. (cid:12)(cid:12)(cid:12) e (cid:12)(cid:12)(cid:12)d (cid:69) g The transition rate of an emitter due to emission of where q0 is the momentum of a surface plasmon, with frequency ω0, i.e., ω0 = ωq0,sp. So far the expression for Γsp is general. The differences arise from the particular forms of the dispersion ωq,sp, the mode length Lq,sp and the prefactor Nq. A. Decay rate for local conductivity We will now study the plasmon emission rate, when non-local effects are neglected, β = 0 in Eq. (50). We first focus on the case of a quantum emitter close to a single graphene layer and the same dielectric above 0102030400102030409 Figure 4. Decay rate of a quantum emitter close to a single graphene layer as a function of emitter frequency, for different emitter-graphene distances z0. The solid lines show the de- cay rate due to plasmon emission as evaluated with Eq. (80). The dashed lines show the total decay rate computed used Eq. (81). The parameters used are 1 = 2 = 1, EF = 0.3 eV, and cos2 ψ = 1/3. For the evaluation of the total decay rate a broadening of γ = 4 meV was used. The local form of graphene conductivity was used (β = 0). and bellow the graphene layer, 1 = 2 = . Using the analytic solution of Eq. (58) for this case, we can write Eq. (79) as (cid:34) (cid:18) (cid:34) (cid:19)2 (cid:18) 2αEF c d2 ge 40 2ω4 0 Γgsl sp = (cid:34) × sin2 ψ + 2κ−2 ω4 0 (cid:35) + ω2 0 c2 (cid:19)2 e−2κqz0 Lq0,sp (cid:35) × (cid:35) cos2 ψ . + ω2 0 c2 2αEF c (80) sp /Γ0, where Γ0 = 0/(cid:0)3π0c3(cid:1) the total decay rate of an emitter in In Figs. 4 and5, we plot the ratio Γgsl d2 geω3 vacuum, for different graphene-quantum emitter dis- tances z0 and dipole orientations. For comparison, we will display the ratio Γfull/Γ0, where Γfull is the total de- cay rate (including both plasmon and photonic losses) of the quantum emitter. The total decay rate can be com- puted from the knowledge of the reflection coefficients, which are incorporated into the dyadic EM Green's func- tion [50, 51, 53] (see Ref. [54] for a detailed study of the properties of an emitter near graphene using Dyadic Green's functions). For a dipole in medium n = 2, the total decay rate is given by (cid:18)(cid:90) ∞ Γfull Γ0 = 1 + 3 2 cos2 ψRe (cid:18)(cid:90) ∞ rpe−2κ2z0 dq k3 1 q (cid:18) q3 kz,1 rs − rp c2q2 ω2 0 dq k1 0 kz,1 + 3 4 sin2 ψRe (cid:19) (cid:19) (cid:19) , e−2κ2z0 (81) where rs/p are reflection coefficients the s/p- polarization and kn = ω/cn and kz,n are defined in Ap- pendix A. for √ A number of details are worth mentioning. There are two distinct behaviors of the decay rate Γfull. At low fre- quency the curves shoot up due to Ohmic losses, which are not included in Eq. (80). At intermediate frequencies, the curves develop a clear resonance due to the excitation of surface plasmons in graphene. Also the maximum of the resonances blue-shifts with the decrease of the dis- tance of the dipole to the graphene sheet. This behavior is easily explained remembering that the dispersion of the surface plasmons in single layer graphene is proportional to q. Since the distance z0 introduces a momentum scale q ∼ 1/z0, smaller z0 values correspond to higher q-values and higher energies of the resonant maximum. Equation (80) for the plasmon emission rate produces exactly the same resonance (same magnitude and same maximum of the resonance position) as Γfull, indicating that in this region, the decay rate of the quantum emit- ter is dominated by plasmon emission. This is further shown in Fig. 5, where losses were arbitrarily reduced in the evaluation of Γfull. We can also avoid the super- position of the Ohmic and surface plasmon contributions choosing either a larger Fermi energy or a larger distance from graphene to the metal. In Fig. 5 the ratio Γ/Γ0 is Figure 5. Decay rate of a quantum emitter close to a single graphene layer as a function of emitter frequency, for differ- ent dipole orientations: ψ = 0 and ψ = π/3. Solid lines show the emission rate [Eq. (80)] and the dashed lines show the total decay rate [Eq. (81)]. The parameters considered where z0 = 70 nm, EF = 0.4 eV, and 1 = 2 = 1. In the evalua- tion of the total decay rate a broadening factor of γ = 0.1 meV was used. For such small losses, the decay rate is com- pletely dominated by the plasmon emission. The local form of graphene conductivity was used (β = 0). smaller than the ones in Fig. 4 due to the larger distance of the dipole to the graphene sheet. An analytic expression for the plasmon emission rate can be also obtained for the case of graphene-metal struc- ture assuming that κ1d (cid:28) 1. In this limit, Eq. 64 can be approximately solved, yielding ωq,sp =(cid:112)4αdEF c/q. Plugging this result in Eq. 79, we obtain 10 Figure 7. Comparison between the plasmon emission rate of a quantum emitter close to a single layer of graphene using the local (β = 0) and non-local (hydr.) (β (cid:54)= 0) models for the graphene conductivity (see Eq. (37)). The used EF = 0.6 eV, 1 = 2 = 3.9, ψ = 0. Left panel : z0 = 7 nm. Right panel: z0 = 2 nm. The nonlocal effects causes an increase and blueshift of the resonance in the transition rate as the distance z0 between the quantum emitter and the graphene decreases. VI. CONCLUSIONS In this paper, we have performed the quantization of graphene plasmons, in the absence of losses, and applied the field quantization to the interaction of an emitter with doped graphene. The quantization was performed using both a macroscopic energy approach and a quan- tum hydrodynamic model, which allows for the inclusion of non-local effects in the EM response of graphene. The quantization approaches used allow for the determina- tion of the plasmon EM field mode-functions and, im- portantly, their normalization, which becomes non-trivial when dispersion is included. When comparing the total decay rate of a quantum emitter (as obtained using the full EM dyadic Green's function) with the decay rate due to quantum emission, it was shown that plasmon emission completely domi- nates the decay rate, for typical emitter-graphene sepa- rations and emitter frequencies. It was shown that non- local effects in the graphene response, become increas- ingly important for smaller graphene-emitter separations and smaller Fermi energies. The advantage of the quantization method developed in this work lies in its simplicity. The mode-functions are obtained from the solution of the Maxwell equations for the vector potential, and the normalization of the mode- functions can be expressed as a simply integral, which only involves the dielectric function of the medium. For situations when only a few modes contribute significantly to the physics, as in the case of quantum emission domi- nated by plamons, mode-functions allows to a much sim- pler and physically transparent description than the full EM dyadic Green's function. Since, the determination of the mode-functions only involves the solution of the clas- sical Maxwell equations, the quantization of the electro- magnetic field of more complexes structures can be eas- ily achieved. We also note that the procedure gives both the quantized form of the electric and magnetic fields. Figure 6. Density plot of the transition rate of a quantum emitter due to the emission of surface plasmons in a graphene- metal structure. The used parameters are z0 = 30 nm, EF = 0.2 eV, 1 = 3.9, and 2 = 1, ψ = 0. Making line cuts at constant d we can easily see the presence of a resonance in the dipole transition rate due to the excitation of graphene screened plasmons. sp (cid:39) d2 Γgm 4ε0 12 ε sinh2(κ1d) 4αdEF c × (cid:18) (ω0)2e−2κ2z0 Lsq(ωat) sin2 ψ + 2 q2 κ2 2 cos2 ψ (cid:19) . (82) This plasmon emission rate is shown in Fig. 6 as a func- tion of ω and the graphene-metal separation d. As in the case of a single graphene layer, for each d there is a well defined peak as function of ω. B. Decay rate with non-local effects We now focus on the role of non-local effects in the graphene conductivity play in the emitter decay rate. We will restrict ourselves to the case of a quantum emitter close to a single layer of graphene. In Fig. 7, we com- pare the transition rate of a quantum emitter calculated taking into account non-local effects β (cid:54)= 0 in Eq. (37), with the local case β = 0. The left panel shows the re- sult for EF = 0.6 eV, z0 = 7 nm, while the right panel shows the result forEF = 0.6 eV, z0 = 2 nm. We can see that the nonlocal effects play an important role at smaller emitter-graphene distances. As the distance be- tween quantum emitter and the graphene layer, z0, de- creases, non-locality leads to an increase and blueshift of the resonance in the transition rate as a function of the emitter frequency. We have also verified (not shown) that the non-local effects are also more prominent in the case of lower electronic densities. 050100150200Therefore, we could also study the enhancement of the spin relaxation. The only change would be the interac- tion Hamiltonian, which in this case would be of the form HI = −µ · B, where µ is the magnetic dipole moment of the emitter and B the quantized magnetic field of the surface plasmon. In possession of a quantized theory for graphene plas- mons, we have set the stage for the future discussion of other quantum effects involving these collective excita- tion made simultaneously of light and matter. ACKNOWLEDGMENTS B.A.F. and N.M.R.P. acknowledge support 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. Additionally, N. M. R. P. acknowledges COMPETE2020, PORTUGAL2020, FEDER and the Portuguese Foundation for Science and Technology (FCT) through projects PTDC/FIS- NAN/3668/2013 and POCI-01-0145-FEDER-028114. B. A. acknowledges the hospitality of CeFEMA where he was a visiting researcher during part of the time in which this work was developed. Appendix A: Reflection coefficients of graphene structures 11 1. Reflection coefficients for a single graphene layer The Fresnel problem for a single graphene sheet has been considered in Ref. [50]. Therefore we provide here the final results only and for simplicity we assume we are dealing with non-magnetic media. For an incoming wave from region 2, and for the s−polarization the reflection and transmission coefficient read rs = kz,2 − kz,1 − µ0ωσg,T kz,2 + kz,1 + µ0ωσg ts = 2kz,2 kz,2 + kz,1 + µ0ωσg,T whereas for the p−polarization we have , , (A1) (A2) (cid:114) 1 rp = − 2kz,1 − 1kz,2 − kz,1kz,2σg,L/ (ω0) 2kz,1 + 1kz,2 + kz,1kz,2σg,L/ (ω0) , (A3) In these equations, we have kz,n =(cid:112)nω2/c2 − q2 = iκn, 2kz,1 + 1kz,2 + kz,1kz,2σg,L/ (ω0) 2kz,21 . (A4) tp = 2 and σg,L/T is the optical longitudinal/transverse conduc- tivity of graphene, 0 and µ0 are the vacuum permitivity and permeability, respectively, q is the wavenumber along the graphene sheet, and ω is the frequency of the elec- tromagnetic radiation. 2. Metal-graphene reflectance coefficient The reflection coefficient for the p-polarization for this structure is given by rp = 1− 2kz,1k2 2 sin (kz,1d) kz,1 sin (kz,1d) (kz,2µ0ωσg,L + k2 2) + ikz,2k2 1 cos (kz,1d) (A5) n = nω2/c2, d is the graphene-metal distance, n − q2, where q is the wavenumber along the where k2 and κ2 graphene sheet. For the s−polarization we have n = k2 , In this appendix we provide the reflection coefficients needed for the evaluation of the full decay rate of a quan- tum emitter, Eq. (81). rs = −1 + 2kz,2 sin (kz,1d) sin (kz,1d) (kz,2 + µ0ωσg,T ) + ikz,1 cos (kz,1d) (A6) . Appendix B: Diagonalization of Hydrodynamic Hamiltonian 1. Orthogonality of mode-functions We start by showing that the mode-functions (Aq,λ(z), υq,λ) obey certain orthogonality conditions which will be useful. We start by writing the equations for the mode-functions within the hydrodynamic model, Eqs. (32) and (33), in matrix form as (cid:35)(cid:20) Aq,λ(z) (cid:21) δ(z) υq,λ (cid:17) = 0. (B1) (cid:34) ω2 q,λ0¯d(z) − 1 µ0 Dq × Dq× −iωq,λen0δ(z) iωq,λen0δ(z) q,λ − β2q ⊗ q ω2 n0m (cid:16) Let us now consider another mode λ(cid:48), with mode-function (Aq,λ(cid:48)(z), υq,λ(cid:48)), solution of (cid:34) ω2 q,λ(cid:48)0¯d(z) − 1 µ0 Dq × Dq× Taking the conjugate of Eq. (B4), we obtain Now we contract Eq. (B1) with z obtaining µ0 µ0 A (cid:105) n0m † q,λ(cid:48) † q,λ(cid:48)(z) υ Dq × Dq× Dq × Dq× −iωq,λ(cid:48)en0δ(z) −iωq,λen0δ(z) −iωq,λ(cid:48)en0δ(z) q,λ0¯d(z) − 1 q,λ(cid:48)0¯d(z) − 1 (cid:104) (cid:105)(cid:34) ω2 (cid:105)(cid:34) ω2 (cid:105)(cid:34) ω2 (cid:105) (cid:16) (cid:105)(cid:20) (ωq,λ + ωq,λ(cid:48)) 0¯d(z) q,λ(cid:48)0¯d(z) − 1 −iωq,λ(cid:48)en0δ(z) q,λ − ω2 ω2 q,λ(cid:48) Dq × Dq× 0¯d(z) † q,λ(cid:48) (cid:17) µ0 dz † q,λ(cid:48)(z) υ † q,λ(cid:48) A dz † q,λ(z) υ † q,λ A dz † q,λ(cid:48)(z) υ † q,λ(cid:48) A (cid:90) (cid:90) (cid:90) (cid:104) (cid:104) (cid:104) (cid:90) dz † q,λ(cid:48)(z) υ A (cid:104) (cid:90) (cid:104) Subtracting this last equation from Eq. (B3), we obtain , Eq. (B1) with and integrate both equations over (cid:16) iωq,λ(cid:48)en0δ(z) q,λ(cid:48) − β2q ⊗ q ω2 A δ(z) (cid:17) υq,λ(cid:48) † q,λ(z) υ (cid:35)(cid:20) Aq,λ(cid:48)(z) (cid:105) (cid:17) iωq,λen0δ(z) q,λ − β2q ⊗ q ω2 (cid:17) iωq,λ(cid:48)en0δ(z) q,λ(cid:48) − β2q ⊗ q ω2 (cid:104) (cid:16) (cid:16) † q,λ δ(z) δ(z) n0m n0m (cid:16) n0m iωq,λ(cid:48)en0δ(z) q,λ(cid:48) − β2q ⊗ q ω2 (cid:17) δ(z) 12 (cid:21) = 0. (B2) = 0, (B3) = 0. (B4) = 0. (B5) = 0. (B6) (cid:21) (cid:21) (cid:21) υq,λ υq,λ(cid:48) (cid:35)(cid:20) Aq,λ(z) (cid:35)(cid:20) Aq,λ(cid:48)(z) (cid:35)(cid:20) Aq,λ(z) (cid:20) Aq,λ(z) (cid:21) (cid:21) (cid:21)(cid:20) Aq,λ(z) υq,λ υq,λ υq,λ −i (ωq,λ − ωq,λ(cid:48)) en0δ(z) n0m (cid:16) i (ωq,λ − ωq,λ(cid:48)) en0δ(z) δ(z) q,λ − ω2 ω2 q,λ(cid:48) (cid:17) If ωq,λ (cid:54)= ωq,λ(cid:48), we can divide by ωq,λ − ωq,λ(cid:48) obtaining one of the orthogonality conditions: dz † q,λ(cid:48)(z) υ † q,λ(cid:48) A −ien0δ(z) ien0δ(z) n0m (ωq,λ − ωq,λ(cid:48)) δ(z) = 0. (B7) We can obtain an additional orthogonality relation. If we take the complex conjugate of Eq. (B3), replace q → −q (cid:16) −iωq,λ(cid:48)en0δ(z) q,λ(cid:48) − β2q ⊗ q ω2 (cid:17) δ(z) (cid:35)(cid:20) A∗ −q,λ(z) υ∗ −q,λ (cid:21) = 0. (B8) , integrate over z and take its complex conjugate, obtaining (cid:34) ω2 and λ → λ(cid:48), and using the fact that ωq,λ = ω−q,λ, we obtain µ0 iωq,λ(cid:48)en0δ(z) Dq × Dq× q,λ(cid:48)0¯d(z) − 1 (cid:104) † q,λ(z) υ q,λ(cid:48)0¯d(z) − 1 A n0m (cid:105) † q,λ µ0 iωq,λ(cid:48)en0δ(z) Dq × Dq× We now contract this equation with (cid:90) Contracting Eq. (B1) with(cid:2) At−q,λ(z) υt−q,λ (cid:90) dz(cid:2) At−q,λ(z) υt−q,λ dz(cid:2) At−q,λ(z) υt−q,λ (cid:90) dz(cid:2) At−q,λ(z) υt−q,λ (cid:90) (cid:3)(cid:34) ω2 (cid:3)(cid:34) ω2 (cid:3) (cid:16) (cid:3)(cid:20) (ωq,λ − ωq,λ(cid:48)) 0¯d(z) dz(cid:2) At−q,λ(z) υt−q,λ −i (ωq,λ + ωq,λ(cid:48)) en0δ(z) n0m For ωq,λ + ωq,λ(cid:48) (cid:54)= 0, we obtain a second orthogonality condition: Subtracting Eq. (B9) from (B10), we obtain q,λ0¯d(z) − 1 −iωq,λen0δ(z) q,λ − ω2 ω2 q,λ(cid:48) Dq × Dq× 0¯d(z) (cid:17) n0m µ0 −ien0δ(z) (cid:3) and integrating over z we obtain n0m (cid:16) (cid:17) −iωq,λ(cid:48)en0δ(z) q,λ(cid:48) − β2q ⊗ q ω2 (cid:17) iωq,λen0δ(z) q,λ − β2q ⊗ q ω2 (cid:16) (cid:16) i (ωq,λ + ωq,λ(cid:48)) en0δ(z) q,λ − ω2 ω2 q,λ(cid:48) δ(z) (cid:17) δ(z) (cid:21) δ(z) υq,λ (cid:35)(cid:20) Aq,λ(z) (cid:35)(cid:20) Aq,λ(z) (cid:21) (cid:20) Aq,λ(z) (cid:21) (cid:21)(cid:20) Aq,λ(z) υq,λ υq,λ (cid:21) = 0. = 0. (B9) = 0. (B10) (B11) (B12) ien0δ(z) n0m (ωq,λ − ωq,λ(cid:48)) δ(z) υq,λ = 0. 2. Hamiltonian in term of mode amplitudes 13 We now insert the expansion of the fields in terms of mode-functions, Eqs. (30) and (31), into the classical Hamil- tonian (29). Using Eqs. (27) and (28), we can write (cid:17) (cid:88) (cid:16) q,λλ(cid:48) H = 1 2 hλ,λ(cid:48)(q)α∗ q,λ(cid:48)(t)αq,λ(t) + hλ,λ(cid:48)(q)α−q,λ(cid:48)(t)αq,λ(t) + c.c., (B13) 0 dz † q,λ(cid:48)(z) υ A (cid:90) (cid:105) (cid:104) (cid:20) ωq,λ(cid:48)ωq,λ0¯d(z) + µ−1 † q,λ(cid:48) 0 Dq × Dq (cid:90) dz(cid:2) At−q,λ(cid:48)(z) υt−q,λ(cid:48) (cid:3) (cid:20) −ωq,λ(cid:48)ωq,λ0¯d(z) + µ−1 (cid:90) (cid:90) † q,λ(cid:48)(z) υ dz 0 0 Dq × Dq (cid:105)(cid:20) (ωq,λ(cid:48) + ωq,λ) 0¯d(z) (cid:104) dz(cid:2) At−q,λ(cid:48)(z) υt−q,λ(cid:48) (cid:3)(cid:20) (ωq,λ − ωq,λ(cid:48)) 0¯d(z) −ien0δ(z) † q,λ(cid:48) A −ien0δ(z) 0 υq,λ (cid:21) δ(z)(cid:0)n0mωq,λ(cid:48)ωq,λ + n0mβ2q ⊗ q(cid:1)(cid:21)(cid:20) Aq,λ(z) δ(z)(cid:0)−n0mωq,λ(cid:48)ωq,λ + n0mβ2q ⊗ q(cid:1)(cid:21)(cid:20) Aq,λ(z) (cid:21) (cid:21)(cid:20) Aq,λ(z) (cid:21)(cid:20) Aq,λ(z) δ(z)n0m (ωq,λ(cid:48) + ωq,λ) ien0δ(z) ien0δ(z) υq,λ υq,λ (cid:21) 0 , δ(z)n0m (ωq,λ − ωq,λ(cid:48)) υq,λ (B14) (B15) (cid:21) (B16) . (B17) where hλ,λ(cid:48)(q) = hλ,λ(cid:48)(q) = hλ,λ(cid:48)(q) = ωq,λ hλ,λ(cid:48)(q) = ωq,λ Using the mode-function equation (B1), we can write Using the orthogonality condition (B12) we conclude that hλ,λ(cid:48)(q) = 0. The orthogonality condition (B7) implies that hλ,λ(cid:48)(q) = 0, except when λ = λ(cid:48) (assuming there are no degeneracies in ωq,λ). Therefore, we see that the classical Hamiltonian can be written as in Eq. (40), with the mode-length being biven by (cid:90) (cid:104) (cid:105)(cid:20) 2ω2 Lq,λ = 1 20ω2 q,λ dz † q,λ(cid:48)(z) υ † q,λ(cid:48) A q,λ0¯d(z) −iωq,λen0δ(z) 2ω2 iωq,λen0δ(z) q,λδ(z)n0m , (B18) (cid:21)(cid:20) Aq,λ(z) (cid:21) υq,λ which can be written as Eq. (41). [1] G. S. Agarwal, Quantum Optics, 1st ed. (Cambridge Uni- versity Press, 2013). fet, G. Messin, and F. Marquier, Science Advances 2, e1501574 (2016). [2] R. W. Heeres, L. P. Kouwenhoven, and V. Zwiller, Nat. [9] S. Bozhevolnyi and N. Mortensen, Nanophotonics 6, 1185 Nano. 8, 719 (2013). (2017). [3] Z. Jacob, MRS Bulletin 37, 761 (2012). [4] M. S. Tame, K. R. McEnery, S. K. Ozdemir, J. Lee, S. A. Maier, and M. S. Kim, Nat. Phys. 9, 329 (2013). [5] D. Mart´ın-Cano, P. A. Huidobro, E. Moreno, and F. J. Garc´ıa-Vidal, in Modern Plasmonics, Vol. 4, edited by A. A. Maradudin, J. R. Sambles, and W. L. Barnes (Elsevier, New York, 2014) 1st ed., Chap. 12, p. 349. [6] J. S. Fakonas, A. Mitskovets, and H. A. Atwater, New J. Phys. 17, 023002 (2015). [7] J. M. Fitzgerald, P. Narang, R. V. Craster, S. A. Maier, and V. Giannini, Proceedings of the IEEE 104, 2307 (2016). [8] M.-C. Dheur, E. Devaux, T. W. Ebbesen, A. Baron, J.-C. Rodier, J.-P. Hugonin, P. Lalanne, J.-J. Gref- [10] D. Xu, X. Xiong, L. Wu, X.-F. Ren, C. E. Png, G.-C. Guo, Q. Gong, and Y.-F. Xiao, Advances in Optics and Photonics 10, 703 (2018). [11] A. I. Fern´andez-Dom´ınguez, S. I. Bozhevolnyi, and N. A. Mortensen, ACS Photonics 5, 3447 (2018). [12] L. Ju, B. Geng, J. Horng, C. Girit, M. Martin, Z. Hao, and H. A. Bechtel, X. Liang, A. Zettl, Y. R. Shen, F. Wang, Nat. Nano. 6, 630 (2011). [13] Z. Fei, G. O. Andreev, W. Bao, L. M. Zhang, A. S. Mcleod, C. Wang, M. K. Stewart, Z. Zhao, G. Dominguez, M. Thiemens, M. M. Fogler, M. J. Tauber, A. H. Castro-neto, C. N. Lau, F. Keilmann, and D. N. Basov, Nano Lett. 11, 4701 (2011). [14] Z. Fei, A. S. Rodin, G. O. Andreev, W. Bao, 14 A. S. McLeod, M. Wagner, L. M. Zhang, Z. Zhao, M. Thiemens, G. Dominguez, M. M. Fogler, A. H. C. Neto, C. N. Lau, F. Keilmann, and D. N. Basov, Nature 487, 82 (2012). [15] J. Chen, M. Badioli, P. Alonso-Gonz´alez, S. Thongrat- tanasiri, F. Huth, J. Osmond, M. Spasenovi´c, A. Cen- teno, A. Pesquera, P. Godignon, A. Zurutuza Elorza, N. Camara, F. J. G. de Abajo, R. Hillenbrand, and F. H. L. Koppens, Nature 487, 77 (2012). [16] A. Woessner, M. B. Lundeberg, Y. Gao, A. Principi, M. Carrega, K. Watanabe, T. Taniguchi, G. Vignale, M. Polini, J. Hone, R. Hillenbrand, and F. H. L. Kop- pens, Nat. Mat. 14, 421 (2015). [17] D. Alcaraz Iranzo, S. Nanot, E. J. C. Dias, I. Ep- stein, C. Peng, D. K. Efetov, M. B. Lundeberg, R. Par- ret, J. Osmond, J.-Y. Hong, J. Kong, D. R. Englund, N. M. R. Peres, and F. H. L. Koppens, Science 360, 291 (2018). [18] M. B. Lundeberg, Y. Gao, R. Asgari, C. Tan, B. Van Duppen, M. Autore, P. Alonso-Gonz´alez, A. Woessner, K. Watanabe, T. Taniguchi, R. Hillen- brand, J. Hone, M. Polini, and F. H. L. Koppens, Science 357, 187 (2017). [19] E. J. C. Dias, D. A. Iranzo, P. A. D. Gon¸calves, Y. Hajati, Y. V. Bludov, A.-P. Jauho, N. A. Mortensen, F. H. L. Koppens, and N. M. R. Peres, Phys. Rev. B 97, 245405 (2018). [31] H. T. Dung, L. Knoll, and D.-G. Welsch, Phys. Rev. A 57, 3931 (1998). [32] T. G. Philbin, New Journal of Physics 12, 123008 (2010). [33] G. W. Hanson, S. A. Hassani Gangaraj, C. Lee, D. G. An- gelakis, and M. Tame, Phys. Rev. A 92, 013828 (2015). [34] Z. Allameh, R. Roknizadeh, and R. Masoudi, Plasmonics 11, 875 (2016). [35] W. E. I. Sha, A. Y. Liu, and W. C. Chew, IEEE Jour- nal on Multiscale and Multiphysics Computational Tech- niques 3, 198 (2018). [36] C. K. Carniglia and L. Mandel, Phys. Rev. D 3, 280 (1971). [37] C. K. Carniglia, L. Mandel, and K. H. Drexhage, Journal of the Optical Society of America 62, 479 (1972). [38] P. Grossel, J.-M. Vigoureux, and R. Payen, Canadian Journal of Physics 55, 1259 (1977). [39] L. D. Landau and E. M. Lifshitz, Electrodynamics of Continuous Media, 2nd ed., Course of theoretical Physics (Pergamon Press, Oxford, UK, 1984). [40] M. Agu, Japanese Journal of Applied Physics 18, 2143 (1979). [41] A. J. Chaves, N. M. R. Peres, G. Smirnov, and N. A. Mortensen, Phys. Rev. B 96, 195438 (2017). [42] A. L. Fetter, Annals of Physics 81, 367 (1973). [43] R. Ruppin, Physics Letters A 299, 309 (2002). [44] B. Amorim and A. J. Chaves, forthcoming publication (2019). [20] D. E. Chang, A. S. Sørensen, P. R. Hemmer, and M. D. [45] T. Christensen, From classical to quantum plasmonics in Lukin, Phys. Rev. Lett. 97, 053002 (2006). three and two dimensions (Springer, 2017). [21] F. Alpeggiani and L. C. Andreani, Plasmonics 9, 965 (2014). [46] Y. O. Nakamura, Prog. Theo. Phys. 70, 908 (1983). [47] A. Lucas and K. C. Fong, J. Phys.: Condens. Matter 30, [22] V. G. Bordo, Journal of the Optical Society of America 053001 (2018). B 36, 323 (2019). [48] M. Jablan, H. Buljan, and M. Soljacic, Phys. Rev. B 80, [23] J. J. Hopfield, Phys. Rev. 112, 1555 (1958). [24] A. Alekseev and Y. P. Nikitin, Sov. Phys. JETP 23, 608 (1966). [25] G. S. Agarwal, Phys. Rev. A 11, 230 (1975). [26] B. Huttner and S. M. Barnett, Phys. Rev. A 46, 4306 (1992). [27] P. Milonni, J. Mod. Opt. 42, 1991 (1995). [28] T. Gruner and D.-G. Welsch, Phys. Rev. A 51, 3246 (1995). 245435 (2009). [49] Y. V. Bludov, A. Ferreira, N. M. R. Peres, and M. I. Vasilevskiy, International Journal of Modern Physics B 27, 1341001 (2013). [50] P. A. D. Gon¸calves and N. M. R. Peres, An Introduction to Graphene Plasmonics (World Scientific, 2016). [51] L. Novotny and B. Hecht, Principle of Nano-Optics (Cambridge University Press, New York, USA, 2006). [52] A. Archambault, F. Marquier, J.-J. Greffet, and [29] T. Gruner and D.-G. Welsch, Phys. Rev. A 53, 1818 C. Arnold, Phys. Rev. B 82, 035411 (2010). (1996). [30] R. Matloob, R. Loudon, S. M. Barnett, and J. Jeffers, Phys. Rev. A 52, 4823 (1995). [53] G. S. Agarwal, Phys. Rev. A 12, 1475 (1975). [54] F. H. L. Koppens, D. E. Chang, F. J. Garc´ıa De Abajo, and F. J. G. D. Abajo, Nano Letters 11, 3370 (2011).
1101.3731	1	1101	2011-01-19T16:36:50	Density of states of a dissipative quantum dot coupled to a quantum wire	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	We examine the local density of states of an impurity level or a quantum dot coupled to a fractional quantum Hall edge, or to the end of a single one-dimensional Luttinger-liquid lead. Effects of an Ohmic dissipative bath are also taken into account. Using both analytical and numerical techniques we show that, in general, the density of states exhibits power-law frequency dependence near the Fermi energy. In a substantial region of the parameter space it simply reflects the behavior of the tunneling density of states at the end of a Luttinger-liquid, and is insensitive either to the value of the dot-lead interaction or to the strength of dissipation; otherwise it depends on these couplings too. This behavior should be contrasted with the thermodynamic properties of the level, in particular, its occupancy, which were previously shown to depend on the various interactions in the system only through the corresponding Fermi edge singularity exponent, and thus cannot display any Luttinger-liquid specific power-law. Hence, we can construct different models, some with and some without interactions in the wire (but with equal Fermi edge singularity exponents), which would have very different level densities of states, although they all result in the same level population vs. energy curves.	cond-mat.mes-hall	cond-mat	Density of states of a dissipative quantum dot coupled to a quantum wire The Minerva Center, Department of Physics, Bar-Ilan University, Ramat-Gan 52900, Israel Moshe Goldstein and Richard Berkovits We examine the local density of states of an impurity level or a quantum dot coupled to a fractional quantum Hall edge, or to the end of a single one-dimensional Luttinger-liquid lead. Effects of an Ohmic dissipative bath are also taken into account. Using both analytical and numerical techniques we show that, in general, the density of states exhibits power-law frequency dependence near the Fermi energy. In a substantial region of the parameter space it simply reflects the behavior of the tunneling density of states at the end of a Luttinger-liquid, and is insensitive either to the value of the dot-lead interaction or to the strength of dissipation; otherwise it depends on these couplings too. This behavior should be contrasted with the thermodynamic properties of the level, in particular, its occupancy, which were previously shown to depend on the various interactions in the system only through the corresponding Fermi edge singularity exponent, and thus cannot display any Luttinger- liquid specific power-law. Hence, we can construct different models, some with and some without interactions in the wire (but with equal Fermi edge singularity exponents), which would have very different level densities of states, although they all result in the same level population vs. energy curves. PACS numbers: 73.23.Hk, 71.10.Pm, 73.20.Hb I. INTRODUCTION Understanding the behavior of low-dimensional elec- tronic systems has been one of the main challenges of experimental and theoretical physics in the last years. These systems are important not only as the basic build- ing blocks of nanoelectronic devices, but also for the in- tricate strongly-correlated phenomena they exhibit. An important subclass is that of metallic (gapless) one- dimensional systems, whose low energy dynamics is gov- erned not by Fermi liquid theory, but instead by the Lut- tinger liquid (LL) paradigm1. This description applies to a wide variety of experimental realizations, including nar- row quantum wires in semiconducting heterostructures, metallic nanowires, and carbon nanotubes. Closely re- lated are chiral LLs, formed at the edges of fractional quantum Hall effect (FQHE) systems2, and helical LLs, the edges of spin quantum Hall insulators3. The effect of impurities on these systems is interesting from both the applicative and fundamental points of view. These impu- rities could also be intentionally introduced, in the form of, e.g., quantum dots and anti-dots. Hence, there is no wonder that such questions have attracted much effort recently. However, most of these studies were restricted to investigation of transport phenomena1 -- 13, while other effects received much less attention14 -- 22. In this work we study probably the most basic exam- ple of such a system, namely, a single level in the vicinity of a fractional quantum Hall edge, or, equivalently23, a level attached to the end of a single LL wire14. We will refer to the two components (in both systems) as "dot" and "lead" respectively. We include in our treatment the effects of short range dot-lead interaction, as well as the influence of an Ohmic dissipative bath (e.g., elec- tromagnetic fluctuations in gate electrodes)16,24 -- 26. In a recent work21 we have studied the thermodynamic prop- erties of the model (e.g., the level population, entropy, and specific heat), and found that they are universal, in the sense that they depend on the various interac- tions in the model (intra-lead, dot-lead, and dot-bath) only through a single parameter, the Fermi edge singu- larity exponent. Thus, thermodynamics can neither be used to identify non-Fermi liquid behavior, nor to extract LL parameters. In this work we proceed to study, both analytically and numerically, the level density of states (LDoS), which may be probed by tunneling or absorp- tion spectroscopies. We find that the LDoS is sensitive to LL physics, even though its integral (times the Fermi function) gives the level occupancy, which is universal in the above sense. As we show below, the LDoS fea- tures power-law behavior near the Fermi energy. For not too strong interactions the exponent in this power-law is actually determined by LL physics alone, and is in- dependent of the level-lead and level-bath interactions. This and many other results derived below cannot be achieved using perturbative calculations19. The rest of this paper is organized as follows: In Sec. II we present our model, and apply to it the Anderson-Yuval Coulomb-gas (CG) expansion26 -- 31. We then proceed to analytic treatment of the LDoS in Sec. III, and to nu- merical calculations in Sec. IV. Finally, we summarize our findings in Sec. V. II. MODEL AND COULOMB-GAS EXPANSION The system is described by the Hamiltonian H = HD + HL + HDL + HB + HDB. The first term is the dot Hamiltonian HD = ε0d†d, with d† and d the level creation and annihilation operators, respectively, and ε0 the level energy. The second term is the lead Hamilto- 1 1 0 2 n a J 9 1 ] l l a h - s e m . t a m - d n o c [ 1 v 1 3 7 3 . 1 0 1 1 : v i X r a nian. It can be written in the form HL = v 4π ∞ Z−∞ [∂xφ(x)]2dx, (1) using chiral bosonic field φ(x) obeying the commutation relation [φ(x), φ(y)] = iπsgn(x− y), where v is the veloc- ity of excitations23. The level and the lead are coupled by: 2 HDL = t0d†ψ(0) + H.c. + U0(cid:0)d†d − 1 2(cid:1) √g π ∂xφ(0). (2) The two terms in this equation describe, respectively, dot-lead hopping (with t0 the tunneling matrix element), and local dot-lead interaction whose strength is U0. The electronic annihilation operator at the end of the lead can be written as ψ(0) = χeiφ(0)/√g/√2πa, where χ is a Majorana operator, a is a short distance cutoff (e.g., the lattice spacing), and g is the LL interaction parameter (g < 1 for repulsion, g > 1 for attraction). For a FQHE system with filling ν, g = ν for electron tunneling (i.e., a dot outside the FQHE bar). Finally, the level is coupled to a bath of harmonic oscillators16,24 -- 26 (describing, e.g., electromagnetic fluctuations in control gates), governed Kω. by HB = Pk ωka†kak. The dot-bath coupling can be 2(cid:1)Pk λk(a†k + ak). We assume written as HDB =(cid:0)d†d − 1 Ohmic dissipation, i.e., linear low-frequency behavior of the bath spectral function: JB(ω) ≡ Pk λ2 kδ(ω − ωk) = We examine this model employing the Anderson-Yuval CG expansion27. In this approach, any quantity of inter- est is expanded to all orders in t0. This results in a series of correlation functions, which need to be evaluated for vanishing t0.26 -- 31 The level-lead interaction gives rise to a potential at the end of the lead, which alternates be- tween U0/2 and −U0/2 whenever an electron tunnels in or out of the level. Similarly, the kth bath oscillators ex- perience a shift in its equilibrium position, proportional to λk. We thus have a sequence of Fermi edge singularity events32. The solution of this latter problem enables the calculation of all the terms in the series of correlation functions. Recently21 we have studied in this way the partition function Z of the model, whose derivatives with respect to the parameters of the system (for example, the level energy ε0 and the temperature T ) give us the thermo- dynamic properties (e.g., the level population, entropy, and specific heat). We were able to rewrite the series ex- pansion for Z in the form of a grand canonical partition function of a classical system of particles. These repre- sent hopping events generated by t0, and thus reside on the imaginary time axis of the original quantum model, which is a circle with circumference 1/T . Each particle is assigned a positive (negative) charge if it represents tun- neling of an electron from the lead to the dot (from the dot to the lead). Hence, there must be an even number of charges, which have to appear in alternating order of FIG. 1: (Color online) A typical term (with 2N = 6 t0-charges and s = −1) in the CG expansions for (a) the partition func- tion [Eqs. (3) -- (4)]; (b) the dot Green function for τ > τ ′ [Eqs. (5) -- (6)] (here s′ = 1 and M = 0 so a single t0-charge precedes τ ′, whereas there are 2M ′ = 4 t0-charges between τ ′ and τ ). Signs and positions of the charges are indicated, with t0-charges marked by thin green vertical lines, and d-charges by wide red ones. Level population as function of imaginary time is denoted by horizontal dashed blue lines. signs. The position of the ith particle is τi, and the sign of the charge of the first particle is denoted by s. The partition function then reads: Z = ∞ XN =0 s=±1 y2N 1/T Z0 dτ2N ξ τ2N−ξ Z0 dτ2N−1 ξ . . . τ3−ξ Z0 dτ2 ξ τ2−ξ Z0 dτ1 ξ e−SCG(s,{τi}), (3) The charges have a fugacity y = pΓ0ξ/π, where Γ0 = πt02ρL is the noninteracting level width [ρL = 1/(πv) is the corresponding lead local density of states], and ξ ∼ a/v is a short-time cutoff. The CG action is given by: SCG(s,{τi}) = 2N Xi<j=1 ~ei · ~ejVC (τj − τi)+ ε0" 1 − s + s 2T 2N (−1)iτi# . Xi=1 (4) The first term of this classical Hamiltonian describes an interaction between the particles, with VC (τ ) = ln{πT ξ/ sin[πTτ]}. This interaction is similar in form to 2D Coulomb interaction, and is the origin of the name "CG expansion". The charges are two component vec- tors, where the two components correspond to the ef- fects of the coupling with the lead and the bath, re- spectively. They are given by ~ei = s(−1)i−1~e0, where the squared-magnitude of the charges, to be denoted by αFES ≡ ~e02, is the Fermi edge singularity exponent of the model. It is defined by behavior of the zero- temperature correlator of d†ψ(0) with its Hermitian con- jugate, calculated at t0 = 0. This correlator decays as τ−αFES for long time τ . In our system we have found that ~e0 =(cid:16)1/√g − 2√gδeff/π,√K(cid:17), where δeff is the effective phase shift in the lead due to the dot-lead coupling21. It is equal to U0/(2v) in straightforward bosonization, but is more complicated in general. It may be extracted from, e.g., finite-size energy differences, which could be calcu- lated either numerically or analytically (via the Bethe ansatz)21. The other part of the CG action accounts for the energetic cost of ε0 per unit imaginary time for each interval in which the level is populated. Its form is analo- gous to an electric field applied to the classical system of 3 charges. A typical configuration is depicted in Fig. 1(a). A similar treatment can be given to the LDoS, which we shall denote by ρD(ω). It is equal to the imag- inary part of the level retarded Green function (mul- tiplied by −1/π). The retarded Green function is in turn the result of analytic continuation of the Matsub- ara Green function from the upper half of the complex frequency plane33. The latter Green function is defined by GD(τ − τ′) = −Tr{ Tτ e−H/T d(τ )d†(τ′)}/Z, where Tτ is the imaginary time ordering operator. Following the same methods as above, the numerator of this expres- sion can also be given a CG representation. This CG has the same form as Eqs. (3) -- (4), with two additional charges of sizes ±~ed, ~ed = ~e0 − (1/√g, 0), inserted at τ′ and τ , respectively. These charges correspond to the level creation and annihilation operators appearing in the definition of the Green function. In the following we will refer to these as "d-charges", to distinguish them form the other "t0-charges", which originate from the t0 term. The contribution of each such configuration is to be mul- tiplied by sgn(τ′ − τ ) to account for the Fermi statistics. Thus, for τ > τ′ the full CG expression for the dot Green function is: GD(τ > τ′) = − 1 Z ∞ y2N s=±1 XN =0 Zτ ′+ξ τ−ξ N−s′ XM=0 N−s′−M XM ′=0 1/T Zτ +ξ dτ2N ξ τ2(M +M ′ )+s′+2−ξ . . . Zτ +ξ dτ2(M+M ′)+s′+1 ξ × dτ2(M+M ′)+s′ ξ . . . τ2M +s′+2−ξ Zτ ′+ξ dτ2M+s′+1 ξ τ ′−ξ Z0 dτ2M+s′ ξ . . . τ2−ξ Z0 dτ1 ξ e−SCG,D(s,τ,τ ′,{τi}), (5) where s′ ≡ (1 − s)/2. The first 2M + s′ t0-charges occupy the interval [0, τ′], the following 2M′ charges reside in the interval [τ′, τ ], and the last 2(N − M − M′) − s′ t0-charges are in the interval [τ, 1/T ]. The classical action is given by: SCG,D(s, τ, τ′,{τi}) = ~e02 2N Xi<j=1 sisjVC (τj − τi) + ~e0 · ~ed 2N Xi=1 si [VC (τi − τ′) − VC (τi − τ )] − ~ed2VC (τ − τ′)+ siτi + τ − τ′# , ε0" 1 − s 2T − 2N Xi=1 (6) is the the sign of ith t0-charge where si = s(−1)i−1sgn(τi − τ′)sgn(τi − τ ). A typical configuration is shown in Fig. 1(b). Similar expressions hold for τ < τ′. Comparing the two CG expansions, the following ob- servation emerges: the CG expansion for the partition function contains only three parameters: Γ0, ε0, and αFES, while expansion for the Green function depends on g too (through ~ed). Hence, the different interaction types (i.e., interactions in the wire, the dot-wire inter- action, and the dot-bath coupling) affect the partition function through a single parameter, the Fermi edge sin- gularity exponent αFES. Thus, thermodynamic measure- ments cannot be used to distinguish between the different interaction types. In other words, one can construct very different models, whose interactions differ in strength and even in sign, which will have the same thermodynamic properties, provided Γ0, ε0, and αFES are indeed the same21. On the other hand, the LDoS, which depends explicitly on g, will exhibit different behavior for these different systems. Hence, it can be used to extract the strength of intra-wire interactions, as we show below. III. ANALYSIS OF THE LEVEL DENSITY OF STATES As noted in our earlier work21, the CG obtained here is identical to the original Anderson-Yuval expansion27 for the anisotropic single-channel Kondo model34, demon- strating that the models are equivalent14,16. Under this mapping the level population becomes the magnetization of the Kondo spin (plus one half). Hence, ε0 is analogous to a local magnetic field. Similarly, Jxy is related to Γ0, and Jz to αFES. The CG parameters obey the famous Kondo renormalization group (RG) equations27,34, which read, in our notations: dy d ln ξ dαFES d ln ξ = 2 − αFES 2 y, = −4y2αFES. (7) (8) O(pΓ0a/v). Thus, the system considered can be in one of two phases, a strong coupling (antiferromagnetic-Kondo like) phase and a weak coupling (ferromagnetic-Kondo like) phase. The transition occurs, for small Γ0, at αFES = 2 + In the weak-coupling phase the Coulomb charges form tightly-bound pairs. The level is thus effectively decoupled at low energies, resulting in its population being discontinuous as a function of ε0 at zero temperature14,16,18. In the strong-coupling phase free Coulomb charges proliferate. The impurity is well- coupled with the lead, so the level population is analytic in ε0, and could be extracted from the Bethe ansatz so- lution of the Kondo problem34. In particular, for small values of ε0 one has n(ε0) − 1/2 ∼ −ε0/TK, where TK = (v/a)(Γ0a/v)1/(2−αFES) is the Kondo temperature (effective level width, reducing to Γ0 in the noninteract- ing case)19. Hence, in this phase the population does not exhibit any nontrivial power-law dependence on ε0 or T . The same applies to other thermodynamic quantities. What are the implications of this on the LDoS? As we now show, we typically find that at zero temperature we have a power-law behavior ρD(ω) ∼ ωδ in the vicinity of the Fermi energy, i.e., when ω is much smaller than TK in the strong-coupling phase, and than the bandwidth ∼ v/a in the weak-coupling phase. The values of δ in the different regimes are summarized in Table I. It should be noted that when T > 0, or when the lead length L is finite, such power-law singularity will be smeared and become [max(ω, T, v/L)]δ. We will now consider each phase separately. 4 TABLE I: Summary of the analysis of Sec. III. In the vicinity of the Fermi energy (when ω and T are much smaller than TK in the strong-coupling phase, and than the bandwidth ∼ v/a in the weak-coupling phase) we have ρD(ω) ∼ [max(T, ω)]δ, with the values of the exponent δ in the different regimes denoted in the table below. When two values are given, the smaller one will give the dominant contribution35. Phase Strong-coupling Weak-coupling ω ≪ ε0 1/g − 1 1/g − 1 ω ≫ ε0 1/g − 1 αd FES − 1 or 1/g − 1 A. The strong-coupling phase Let us start from the strong-coupling phase. When ε0 is large enough (with respect to TK), CG charges must appear in tightly-bound pairs, since large intra-pair sep- aration is suppressed by the level energy, as dictated by the last term of Eq. (6). The two d-charges added in the calculation of the level Green function will also be accompanied by two screening t0-charges for the same reason. The resulting configuration should thus resemble Fig. 2(a). The leading contribution to the Green function will then come from the residual interaction of these par- tially screened d-charges, whose charges are (±1 times) ~e0 − ~ed = (1/√g, 0). Thus, for large τ − τ′ we have GD(τ − τ′) ∼ sgn(τ′ − τ )τ − τ′−1/g. This leads to ρD(ω) ∼ ω1/g−1, the usual tunneling density of states singularity at the end of a LL wire1,4. When ε0 is small (with respect to TK), renormalization effects are significant. In addition to the usual CG RG Eq. (8) representing the screening of interaction between t0-charges by pairs of nearby t0-charges (separated by ξ)27, one can write down similar equations for the flow of the Green function parameters30. It is easy to see that the coefficients of the logarithmic interaction between any two charges (either both t0-charges, both d-charges, or a mixed pair) are renormalized in the same way by the pairs of nearby t0-charges, i.e., d(~eµ · ~eν) d ln ξ = −2y2 (~eµ · ~e0 + ~e0 · ~eν) , (9) where µ, ν = 0, d. Thus, the combination ~e0 − ~ed2 = ~e02 + ~ed2 − 2~e0 · ~ed is invariant, and retains its initial value of 1/g. At the strong coupling fixed point y is large, so, by Eq. (8), ~e ∗d =pα∗FES = 0, where asterisks denote fixed-point values. Thus, ~e ∗0 = 1/√g, so that d 2 GD(τ′ − τ ) ∼ (τ′ − τ )−~e ∗ = (τ′ − τ )−1/g, resulting again in ρD(ω) ∼ ω1/g−1. Since this behavior holds at both large and small ε0 values, it should also apply at all intermediate values. Further support for this result is obtained by analysis of the particular case αFES = 1 (the Toulouse limit27), where the CG, and all thermodynamic properties, reduce to those of a noninteracting resonant level (for which g = 1, U0 = 0, and K = 0). A nontrivial (i.e., in- 5 teracting) realization of this condition, which still per- mits an exact calculation of the LDoS, is the case of no coupling to a bath (K = 0), but with g = 1/4 and a corresponding compensating value of the dot- lead interaction. Then ~ed = (−1, 0) = −~e0, so the d- charges have the same magnitudes as the corresponding t0-charges, but the opposite signs. Comparison of the corresponding CG expansions thus shows that the level Green function of the interacting system is equal to a two-particle Green function of the noninteracting reso- nant level h Tτ ψ(0, τ )d(τ )d†(τ′)ψ†(0, τ′)i, up to a factor of (2πa)−1sgn(τ′ − τ ). The operators at τ and τ′ in this noninteracting two-particle Green function are similar to the t0 term in the Hamiltonian, but with ψ(0) replaced by ψ†(0) to account for the signs of the d-charges in the interacting system. After a straightforward evaluation of this two particle Green function by Wick's theorem, we find for the LDoS the following expression: + 2 (10) 2πiT 2T(cid:17) Im(cid:26) ω − 2ε0 + 2iΓ0 × ω − 2ε0 (cid:19)(cid:21)(cid:27) , ω − ε0 + iΓ0 ρD(ω) = 2ρLa coth(cid:16) ω 2πiT (cid:19) − ψ(cid:18) 1 (cid:20)ψ(cid:18) 1 ε0 − iΓ0 2 − where ψ(z) is the digamma function36. For small ω and T we indeed recover the [max(T,ω)]3 behavior appro- priate for g = 1/4, for all values of ε0. Physically, the result is clear: in the strongly-coupled phase the level behaves, at low energies, as the last site of the lead, so its LDoS is similar to the local density of states near the end of a LL wire1,4, i.e., ρD(ω) ∼ ω1/g−1. Interestingly, not only dot-lead interactions, but even coupling to the bath does not modify this behavior. As a result of this, the LDoS exhibits a power-law behavior with exponent which depends only on the LL parameter g and not on αFES, i.e., on the interactions in the wire but not on the level-lead and level-bath couplings. This is in contrast with, e.g., the level occupancy, which de- pends on αFES but not on g, as discussed above. Below we also test these predictions numerically. B. The weak-coupling phase We now turn our attention to the weak-coupling phase. Here all the t0-charges are bound in pairs, so the high ε0 results discussed above (which should also hold in this phase) actually carry over to low values of ε0. It should however be remembered that then they compete with the contribution of the t0 = 0 term in the CG expansion (the weak coupling fixed point), the term representing the in- teraction of the unscreened d-charges, which gives the LDoS a contribution of the form ρpair(ω) ∼ ωαd FES−1, with αd FES ≡ ~ed2. This is simply the LDoS of a tunnel- decoupled level, broadened from a delta peak to a power- law by the Anderson orthogonalities in the wire32 and in the Ohmic bath24,25. Note that since the level is effec- FIG. 2: (Color online) Typical configuration of the CG ex- pansion for the dot Green function [Eqs. (5) -- (6); here τ > τ ′, s = −1, 2N = 6, s′ = 1, M = 0, and 2M ′ = 4, as in Fig. 1(b)] for (a) large ε0 in both phases (and possibly small ε0 in the weak-coupling phase35) -- d-charges are screened; (b) small ε0 in the weak-coupling phase -- d-charges are not screened. Notations are the same as in Fig. 1. See the text for further details. tively decoupled, the usual ω1/g−1 behavior at the end of a LL wire need not apply anymore. One could actually proceed to study higher order terms in the weak-coupling regime (and similarly, for large ε0 in the strong-coupling phase). For small t0 the lead- ing correction is dressing of the above-mentioned charge configurations by a series of pairs of close-by t0-charges (close-by since ε0 and/or αFES are large), as depicted in Fig. 2. One then has to sum over all the terms sim- ilar to Fig. 2(a) for large ε0 (i.e., for ω ≪ ε0 in the weak-coupling phase) or all the terms similar to Fig. 2(b) for small ε0 (i.e., for ω ≫ ε0). Since each pair has a very small dipole moment (due to the proximity of the charges), inter-pair interactions are negligible in a first approximation. This is actually an imaginary time variant of the noninteracting blip approximation (NIBA)24,25. The following argument can spare us the need of explicit calculations. For U0 = 0 and K = 0, the d-charges are noninteracting, ~ed = (0, 0). The sum over all the terms with pairs of nearby t0 charges would be the same (in the current approximation) as the Green function of a noninteracting system consisting of a level tunnel-coupled to a bath of noninteracting fermions with power-law local density of states ρL(ω) ∼ ωαFES−1 (with some appropriate high-energy cutoff and normalization). For the latter system the Green function can be easily evaluated to give G0 the dot self energy is33: D(iω) = [iω − ε0 − Σ0 D(iω)]−1, where D(iω) = (t0)2 Σ0 ∞ Z−∞ ρL(Ω) iω − Ω dΩ, (11) so that Σ0 D(iω) ∼ ωαFES−1. For small ε0 we indeed see that Σ0 D(ω) is subdominant with respect to the nonin- teracting contribution only if αFES > 2, which is exactly the condition for the weak-coupling phase for small t0. Then, exactly at ε0 = 0, a delta-function term appears at ω = 0 in the expression for the LDoS (similarly to the situation at t0 = 0), whose coefficient is determined by the requirement that the integral of the entire expression for the LDoS corresponding to G0 D(iω) is unity14. Before discussing nonzero U0 and K, it should be re- marked that this NIBA-like approximation exactly repro- duces the perturbative (in the tunneling t0) approach em- ployed in Ref. 19, and would lead to similar predictions for the behavior of the level population. Both approx- imations are justified only in the weak-coupling phase (or when ε0 is large), but not in the strong-coupling phase, where t0 grows under RG flow and thus cannot be treated perturbatively, similarly to the exchange Jxy in the equivalent Kondo problem34. In the strong-coupling regime perturbative results predict correctly the depen- dence of the Kondo temperature on t0 (again, just like in the Kondo model34) and the qualitative behavior of the LDoS for 1/2 < g < 1 and ε0 6= 0 (at U0 = 0 and K = 0), but deviate from our previous conclusions in many other respects. For example, as we discuss below, at nonzero U0 and K our NIBA calculations indicate that the ex- ponent in the power-law behavior of the LDoS at low energy may depend on these interactions too, in contrast with the situation in the strong-coupling phase, where the corresponding exponent depends only on the LL pa- rameter g, as shown above. Moreover, even for vanishing dot-lead and dot-bath interactions (the case treated in Ref. 19) the NIBA/perturbative expression given above does not agree with our previous analysis of the strong- coupling phase: for (a) 1/2 < g < 1 and ε0 = 0 or (b) g > 1 and any ε0, NIBA would suggest that the LDoS varies as ∼ ω1−1/g, i.e., with the opposite exponent to the one appearing in the density of states at the end of a LL wire. Perturbative results would thus imply that the LDoS may be enhanced for g < 1 or suppressed for g > 1, which is clearly at odds with both our previous results and the behavior of density of states at the lead edge. Moreover, integrating the perturbative LDoS leads to the prediction that the level population may have a power- law dependence on ε0 at the strong-coupling phase (for 1/2 < g < 2/3)19, which is in contrast with the analytical behavior expected from the exact mapping of our model onto the Kondo problem, as discussed above. To summa- rize, NIBA/perturbative (in t0) expressions do not hold in general in the strong-coupling phase. It may be noted that the numerical data of Ref. 19 does not cover these regimes of the strong coupling phase for which perturba- 6 tive calculations disagree with our previous analysis. Returning to the discussion of the NIBA approxima- tion for the weak-coupling phase, we will now treat the more general case, i.e., nonzero U0 and K (which was not addressed in Ref. 19). Again, screened d-charges terms [Fig. 2(a)] are dominant for ω ≪ ε0, whereas unscreened d-charges terms [Fig. 2(b)] are dominant for ω ≫ ε0.35 Let us start from the latter case. Now that there is interaction between d-charges, the CG ex- pression for the Green function GD(τ − τ′) contains an additional a factor of the form ∼ τ − τ′−αd FES (interac- tion of d-charges with the pairs of close-by t0-charges is negligible). Turning to the frequency domain, the LDoS of the unscreened d-charges contribution [Fig. 2(b)] will thus be the convolution of the LDoS ρ0 D(ω) associated with G0 D(iω) from the previous paragraph, with a func- tion ρpair(ω) ∼ ωαd FES−1, which is simply the LDoS of a decoupled (t0 = 0) level as discussed above, i.e., ω (12) ρD(ω) = 2sgn(ω) ρ0 D(ω − Ω)ρpair(Ω)dΩ, Z0 at T = 0. Thus, the ωαd FES−1 behavior of the decoupled level will survive for vanishing ε0, due to the delta peak in G0 D(iω). This behavior actually applies to the entire ω ≫ ε0 region, which is exactly where the contribution of terms similar to Fig. 2(b) is important. Higher order corrections will give extra powers of ωαFES−2, which are subleading since αFES > 2. Similar considerations apply to the screened d-charges contribution [Fig. 2(a)] when ω ≪ ε0. For U0 6= 0 and/or K 6= 0, two corrections are due. The first correc- tion takes into account the factors coming from the in- teraction between each d-charge and the neighboring t0- charge, which are power-laws in the time domain. Since the d-t0 charges form tightly-bound pairs, we can take the limits of integration over their separation to infinity. They thus yield factors of ∞ Z0 (cid:18) ξ τ(cid:19)~e0·~ed e−ε0τ dτ ξ = Γ(1 − ~e0 · ~ed) (ε0ξ)1−~e0·~ed , (13) where Γ(z) is the gamma function36, and ~e0·~ed = (αd FES+ αFES − 1/g)/2. The correction is then the ratio between this expression and its value at ~ed = 0. Apart from this constant factor, one must compensate for the fact that the inter-pair interaction of these two d-t0 pairs gives the Green function a factor which varies as τ − τ′−1/g (since ~ed−~e02 = 1/g), instead of the τ − τ′−αFES dependence used in the calculation of G0 D(ω) should be convoluted here with ∼ ω1/g−αFES−1. For ω ≪ ε0 the LDoS will thus retain the ω1/g−1 behavior of the strong-coupling phase. To conclude the discussion of this NIBA-type approx- imation, a general expression for the LDoS which inter- polates between large and small ω/ε0 limits35 is of the D. Hence, ρ0 100 0 Γ ) 0 ( 10-1 D ρ 10-2 0.3 0.25 0.2 0 Γ D ρ 0.15 0.1 0.05 (a) (b) ≠0 g=1, U0 g=3/2, U0=0 ≠0 g=3/4, U0 0.25 0.2 0 Γ D ρ 0.15 100 0 Γ ) 0 ( 10-1 D ρ 10-2 0.1 0.05 101 Γ 0/T 102 101 Γ 0/T 102 7 ≠0 g=1, U0 g=3/2, U=0 ≠0 g=3/4, U0 0 -15 -10 -5 0.3 0.25 100 0 ω/Γ0 (c) 0 10-1 0.2 Γ ) 0 ( 5 10 0 -15 -10 -5 ≠0 g=1, U0 ≠0 g=2, U0 g=2/3, U0 ≠0 0.3 0.25 100 0 10-1 0.2 Γ ) 0 ( 5 10 0 ω/Γ0 (d) ≠0 g=1, U0 ≠0 g=2, U0 g=2/3, U0 ≠0 D 10-2 0 Γ D ρ ρ 0.15 D 10-2 0 Γ D ρ ρ 0.15 10-3 0.1 0.05 101 Γ 0/T 102 10-3 0.1 0.05 101 Γ 0/T 102 0 -15 -10 -5 0 ω/Γ0 5 10 0 -15 -10 -5 0 ω/Γ0 5 10 FIG. 3: (Color online) MC results for the LDoS as a function of frequency (measured with respect to the Fermi energy) for three different models, all with αFES = 2/3 and T /Γ0 = 0.1. The three curves in each panel correspond to three different values of g: g = 1 (black pluses), g > 1 (green X's), and g < 1 (red circles), as indicated in the legends. The insets present the the LDoS at the Fermi energy as a function of the temperature by symbols (with the same code as before), together with a cyan line showing best-fit to the expected power-law behavior ρD(0) ∼ T 1/g−1 (only the prefactor is fitted). The different panels correspond to different values of ε0 and K: (a) ε0 = 0 and K = 0; (b) ε0 = 2Γ0 and K = 0; (c) ε0 = 0 and K = 1/6; (d) ε0 = 2Γ0 and K = 1/6. In each case the value of U0 was chosen according to the requirement αFES = 2/3. form of Eq. (12), but with ρpair(Ω) replaced by ρ′pair(Ω) ∼ Ωαd FES−1(cid:12)(cid:12)(cid:12) ε0 Ω − 1(cid:12)(cid:12)(cid:12) All the low-energy results of this section are summarized in Table I. αFES+αd FES−1/g . (14) IV. NUMERICAL CALCULATIONS In this section we present the results of numerical cal- culations, verifying the conclusions of our previous anal- ysis, i.e., that the LDoS at low energies features a power- law behavior with the power determined by LL physics only (in the strong coupling phase), although, as we have shown before21, its integral (the level population) is uni- versal, and cannot be used to extract LL parameters. To calculate the LDoS we used classical Monte-Carlo (MC) simulations on the CG expansion of dot Green function28. The MC update procedure used is similar to the one employed recently for the closely-related con- tinuous time quantum MC algorithm37. After obtain- ing the imaginary-time Green function it was Fourier- transformed to Matsubara frequencies, followed by ana- lytic continuation to real frequencies using the Pad´e ap- proximant technique38. This yields the retarded Green function, whose imaginary part is proportional to the LDoS33. Below we present data in the non-perturbative strong-coupling region, which confirms the results of our previous analysis. Actually, in the weak-coupling phase (which is accessible analytically through the NIBA-like 1 0.8 0.6 0.4 0.2 D n α FES 1.40 1.15 1 0.82 0.70 ≠0 g=1, Ud g≠1, Ud=0 g≠1, Ud ≠0 0 -30 -20 -10 10 20 30 ε 0 0/Γ0 FIG. 4: (Color online) DMRG results for the average level occupancy nD as a function of its energy, ε0, for a discrete model of the lead [Eqs. (15) -- (16)]. The curves on which the symbols reside (which serve as guides to the eye) correspond to the various αFES values, where the larger αFES the nar- rower the curve and vice versa. The widest curve has similar parameters to those used in the MC simulations presented in Fig. 3. On each curve there are three different choices of U and Ud (all giving the same αFES value), denoted by the three different symbol types. See the text for further details. approximation) MC simulations are not efficient, since there CG charges are rare and averaging very slow. In this sense, our analysis and numerical calculations are complementary. The results presented in the different panels Fig. 3. The values of g, U0 and K are varied, in a way which keeps αFES constant at a value of 2/3. Hence, the occupancies as functions of ε0 are the same, as we also verify below (actually, the CG representation would predict exactly identical occupations). The LDoS curves are, however, markedly different: depending on whether g > 1, g < 1, or g = 1, they have a maximum, a minimum, or no special feature near the Fermi energy, respectively. In the inset we demonstrate that in all cases the LDoS at the Fermi energy exhibits power-law dependence on temperature, ρD(0) ∼ T 1/g−1, as found in the previous section (cf. Table I). For the sake of completeness, we will repeat here some of our previous data on the level occupancy21. Since the MC simulation is based on the CG, to have an in- dependent check of the universality of the level popu- lation we employed the density matrix renormalization group (DMRG)39 algorithm, using block-sizes of up to 256. DMRG is also better suited to ground state cal- culations, and thus complements the necessarily finite- temperature MC in this respect. The model used is a half-filled tight binding chain with nearest-neighbor in- teractions. It is described by the Hamiltonian: 8 N−1 HL = Xi=1 htc†i ci+1 + H.c. + U(cid:16)c†i ci − 1 2(cid:17)(cid:16)c†i+1ci+1 − 1 2(cid:17)i (15) where c†i (ci) is the electronic creation (annihilation) op- erator at the ith site of the wire (i = 1 . . . N , with L = N a), and t and U are, respectively, nearest- neighbor hopping amplitude and interaction strength. The low energy physics of this model is known to be governed by LL theory for not too large interactions U < 2t),1 with g = π/[2 cos−1(−U/2t)] and (i.e., v/(2at) = πp1 − (U/2t)2/[2 cos−1(U/2t)]. The dot is still governed by the same HD = ε0d†d, and is coupled to the lead through: 2(cid:1)(cid:16)c†1c1 − 1 2(cid:17) HDL = tdc†1d + H.c. + Ud(cid:0)d†d − 1 td, Ud are related to the corresponding parameters of the continuum version Eq. (2) by t0 = td√a, and U0 = Uda, a being the lattice spacing. We have pre- viously shown that boundary conformal field theory ar- guments and the Bethe ansatz solution yield that here (16) δeff = tan−1(Ud/√4t2 − U 2).21 The level population is plotted in Fig. 4 as a function of ε0. Different curves correspond to different values of αFES, as indicated in the legend. On each such curve there are three types of symbols, denoting DMRG data on three different models: (i) g = 1 (i.e., U = 0) but nonzero Ud; (ii) g 6= 1 (nonzero U ) but Ud = 0; (iii) both U and Ud are nonzero. All the models are without cou- pling to the bath (K = 0). The values of U and Ud in each model were chosen so as to give the same value of αFES for each curve. For model (iii) we used U = ±0.5t, with sign opposite to that of model (ii). In all cases we chose td to get Γ0 = 10−4t and used N = 100v/(at) sites. The results clearly show that the occupancy is indeed univer- sal, depending only on αFES, and not on the strengths or signs of the interactions (U and Ud). It should be noted that the widest curve has similar parameters to those used in the MC simulations (Fig. 3). V. CONCLUSIONS To summarize, we have studied, both analytically and numerically, the LDoS of a level coupled to a LL and to an Ohmic bath over the entire parameter space. We have found that in general it exhibits a power-law dependence at low energies. In large parts of the phase space this is just the power-law behavior of the tunneling density of states at the end of a LL wire. Thus, a measurement of the LDoS there can be used to extract the value of the LL interaction parameter g. In other regions it is also affected by level-lead and level-bath interactions. In any case the LDoS is explicitly sensitive to the value of the LL parameter g, although the LDoS determines the level population, which was found before to be universal21, and thus not to feature any LL-specific power-law. Acknowledgments We would like to thank Y. Weiss for his invaluable help with the DMRG calculations, A. Schiller for many useful 9 suggestions, and Y. Gefen for discussions. M.G. is sup- ported by the Adams Foundation of the Israel Academy of Sciences and Humanities. Financial aid from the Israel Science Foundation (Grant 569/07) is gratefully acknowl- edged. 1 A.O. Gogolin, A.A. Nersesyan, and A.M. Tsvelik, Bosonization and Strongly Correlated Systems, (Cam- bridge University Press, Cambridge, 1998); T. Giamarchi, Quantum Physics in One Dimension, (Oxford University Press, Oxford, 2003). 2 For a review see: A.M. Chang, Rev. Mod. Phys. 75, 1449 (2003), and references cited therein. 3 For a review see: M. Konig, H. Buhmann, L.W. Molenkamp, T. Hughes, C.-X. Liu, X.-L. Qi, and S.-C. Zhang, J. Phys. Soc. Jpn. 77, 031007 (2008), and refer- ences cited therein. 4 C.L. Kane and M.P.A. Fisher, Phys. Rev. Lett. 68, 1220 (1992); Phys. Rev. B 46, R7268 (1992); 46, 15233 (1992). 5 A. Furusaki and N. Nagaosa, Phys. Rev. B 47, 3827 (1993). 6 A. Furusaki, Phys. Rev. B 57, 7141 (1998). 7 O.M. Auslaender, A. Yacoby, R. de Picciotto, K.W. Bald- win, L.N. Pfeiffer, and K.W. West, Phys. Rev. Lett. 84, 1764 (2000). 8 H.W.Ch. Postma, T. Teepen, Z. Yao, M. Grifoni, and C. Dekker, Science 293, 76 (2001). 9 A. Komnik and A.O. Gogolin, Phys. Rev. Lett. 90, 246403 (2003); Phys. Rev. B 68, 235323 (2003). 10 Yu.V. Nazarov and L.I. Glazman, Phys. Rev. Lett. 91, 86, 67012 (2009); Proceedings of FQMT '08, Physica E 42, 610 (2010). 22 F. Elste, D.R. Reichman, and A.J. Millis, Phys. Rev. B 81, 205413 (2010). 23 A non-chiral LL with a boundary can mapped onto a chiral LL by unfolding the decoupled (Bogolubov-transformed) right- and left-movers [M. Fabrizio and A.O. Gogolin, Phys. Rev. B 51, 17827 (1995)]. 24 A.J. Leggett, S. Chakravarty, A.T. Dorsey, M.P.A. Fisher, A. Garg, and W. Zwerger, Rev. Mod. Phys. 59, 1 (1987). 25 U. Weiss, Quantum Dissipative Systems (World Scientific, Singapore, 1999). 26 A. Kamenev and Y. Gefen, Phys. Rev. B 54, 5428 (1996); arXiv:cond-mat/9708109. 27 P.W. Anderson and G. Yuval, Phys. Rev. Lett. 23, 89 (1969); G. Yuval and P.W. Anderson, Phys. Rev. B 1, 1522 (1970); P.W. Anderson, G. Yuval, and D.R. Hamann, ibid. 1, 4464 (1970). 28 K.D. Schotte and U. Schotte, Phys. Rev. B 4, 2228 (1971). 29 P.B. Wiegmann and A.M. Finkelstein, Zh. Eksp. Teor. Fiz. 75, 204 (1978) [Sov. Phys. JETP 48, 102 (1978)]. 30 Q. Si, and G. Kotliar, Phys. Rev. B 48, 13881 (1993). 31 M. Fabrizio, A.O. Gogolin, and P. Nozi`eres, Phys. Rev. B 126804 (2003). 51, 16088 (1995). 11 D.G. Polyakov and I.V. Gornyi, Phys. Rev. B 68, 035421 32 P. Nozi`eres and C.T. De Dominicis, Phys. Rev. 178, 1097 (2003). (1969). 12 I.V. Lerner, V.I. Yudson, and I.V. Yurkevich, Phys. Rev. 33 G.D. Mahan, Many-Particle Physics (Kluwer, New York, Lett. 100, 256805 (2008). 2000). 13 M. Goldstein and R. Berkovits, Phys. Rev. Lett. 104, 34 A.C. Hewson, The Kondo Problem to Heavy Fermions 106403 (2010). 14 A. Furusaki and K.A. Matveev, Phys. Rev. Lett. 88, 226404 (2002). 15 M. Sade, Y. Weiss, M. Goldstein, and R. Berkovits, Phys. Rev. B 71, 153301 (2005); 16 K. Le Hur and M.-R. Li, Phys. Rev. B 72, 073305 (2005). 17 Y. Weiss, M. Sade, M. Goldstein, and R. Berkovits, Phys. Stat. Sol. (b) 243, 399 (2006); Y. Weiss, M. Goldstein, and R. Berkovits, J. Phys.: Condens. Matter 19, 086215 (2007); 18 Y. Weiss, M. Goldstein, and R. Berkovits, Phys. Rev. B 75, 064209 (2007); 76, 024204 (2007); 77, 205128 (2008). 19 P. Wachter, V. Meden, and K. Schonhammer, Phys. Rev. (Cambridge University Press, Cambridge, 1993). 35 When 1/g < αd FES (i.e., in the presence of strong dot- lead and/or dot-bath interactions), interaction between d- charges is large enough to make the screened d-charges contribution [Fig. 2(a)] dominant even for small ε0. Using similar methods one can show that then the LDoS will vary as ω1/g−1, just like for ω ≪ ε0. 36 M. Abramowitz and I.A. Stegun, Handbook of Mathemat- ical Functions (Dover, New York, 1964). 37 P. Werner, A. Comanac, L. de' Medici, M. Troyer, and A.J. Millis, Phys. Rev. Lett. 97, 076405 (2006); P. Werner and A.J. Millis, Phys. Rev. B 74, 155107 (2006). 38 H.J. Vidberg and J.W. Serene, J. Low Temp. Phys. 29, B 76, 125316 (2007). 179 (1977). 20 G.A. Fiete, W. Bishara, C. Nayak, Phys. Rev. Lett. 101, 39 U. Schollwock, Rev. Mod. Phys. 77, 259 (2005); K.A. Hall- 176801 (2008); Phys. Rev. B 82, 035301 (2010). 21 M. Goldstein, Y. Weiss, and R. Berkovits, Europhys. Lett. berg, Adv. Phys. 55, 477 (2006).

Subsets and Splits