* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Electron and nuclear spins in semiconductor
Quantum mechanics wikipedia , lookup
History of subatomic physics wikipedia , lookup
Quantum chromodynamics wikipedia , lookup
Relational approach to quantum physics wikipedia , lookup
Probability amplitude wikipedia , lookup
Electromagnetism wikipedia , lookup
Quantum field theory wikipedia , lookup
Introduction to gauge theory wikipedia , lookup
Renormalization wikipedia , lookup
Fundamental interaction wikipedia , lookup
Quantum potential wikipedia , lookup
Theoretical and experimental justification for the Schrödinger equation wikipedia , lookup
Quantum electrodynamics wikipedia , lookup
Nuclear physics wikipedia , lookup
Quantum entanglement wikipedia , lookup
Hydrogen atom wikipedia , lookup
Mathematical formulation of the Standard Model wikipedia , lookup
Quantum vacuum thruster wikipedia , lookup
History of quantum field theory wikipedia , lookup
Old quantum theory wikipedia , lookup
Condensed matter physics wikipedia , lookup
EPR paradox wikipedia , lookup
Bell's theorem wikipedia , lookup
Quantum state wikipedia , lookup
Spin (physics) wikipedia , lookup
Relativistic quantum mechanics wikipedia , lookup
Geometrical frustration wikipedia , lookup
Electron and nuclear spins in semiconductor quantum dots A dissertation presented by Jacob Jonathan Krich to The Department of Physics in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the subject of Physics Harvard University Cambridge, Massachusetts July 2009 c 2009 - Jacob Jonathan Krich All rights reserved. Thesis advisor Author Bertrand I. Halperin Jacob Jonathan Krich Electron and nuclear spins in semiconductor quantum dots Abstract The electron and nuclear spin degrees of freedom in two-dimensional semiconductor quantum dots are studied as important resources for such fields as spintronics and quantum information. The coupling of electron spins to their orbital motion, via the spinorbit interaction, and to nuclear spins, via the hyperfine interaction, are important for understanding spin-dynamics in quantum dot systems. This work is concerned with both of these interactions as they relate to two-dimensional semiconductor quantum dots. We first consider the spin-orbit interaction in many-electron quantum dots, studying its role in conductance fluctuations. We further explore the creation and destruction of spin-polarized currents by chaotic quantum dots in the strong spin-orbit limit, finding that even without magnetic fields or ferromagnets (i.e., with time reversal symmetry) such systems can produce large spin-polarizations in currents passing through a small number of open channels. We use a density matrix formalism for transport through quantum dots, allowing consideration of currents entangled between different leads, which we show can have larger fluctuations than currents which are not so entangled. Second, we consider the hyperfine interaction between electrons and approximately 106 nuclei in two-electron double quantum dots. The nuclei in each dot collectively form an effective magnetic field interacting with the electron spins. We show that a procedure originally explored with the intent to polarize the nuclei can also equalize the effective magnetic fields of the nuclei in the two quantum dots or, in other parameter regimes, can iii iv cause the effective magnetic fields to have large differences. Abstract Contents Title Page . . . . . . . . . . . . . Abstract . . . . . . . . . . . . . . Table of Contents . . . . . . . . . Citations to Previously Published Acknowledgments . . . . . . . . . Dedication . . . . . . . . . . . . . . . . . . . . . . . . . Work . . . . . . . . 1 Introduction 1.1 Spin-orbit interaction . . . . . . 1.2 Time-reversal symmetry . . . . . 1.3 Random matrix theory . . . . . . 1.4 2DES and quantum dots . . . . 1.5 Organization of thesis . . . . . . 1.6 Summary of new results obtained . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . in this . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i iii v vii viii xi . . . . . . 1 4 8 11 20 33 37 2 Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 3 Spin polarized current generation from quantum dots without magnetic fields 3.1 Introduction . . . . . . . . . . . . . . . . . . . 3.2 Setup and symmetry restrictions . . . . . . . 3.3 Random matrix theory . . . . . . . . . . . . . 3.4 Dephasing . . . . . . . . . . . . . . . . . . . . 3.5 Finite bias and temperature . . . . . . . . . . 3.6 Discussion . . . . . . . . . . . . . . . . . . . . 4 Fluctuations of spin transport through chaotic quantum dots with 4.1 Introduction . . . . . . . . . . . . . 4.2 Setup . . . . . . . . . . . . . . . . 4.3 Random matrix theory . . . . . . . 4.4 Discussion . . . . . . . . . . . . . . 40 . . . . . . spin-orbit . . . . . . . . . . . . . . . . . . . . . . . . . . . . v . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 52 53 59 64 68 70 . . . . 72 73 74 80 90 vi 5 Inhomogeneous nuclear spin flips 5.1 Introduction . . . . . . . . . . . . . . . . . . . 5.2 Electronic states of the double dot with N=2 5.3 Nuclear spin flip location . . . . . . . . . . . 5.4 Nuclear states . . . . . . . . . . . . . . . . . . 5.5 Discussion . . . . . . . . . . . . . . . . . . . . Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Preparation of non-equilibrium nuclear spin states in quantum dots . . . . . 92 92 94 96 97 101 103 A Appendix to Chapter 2 114 A.1 Further cubic spin-orbit terms . . . . . . . . . . . . . . . . . . . . . . . . . . 114 A.2 Refitting var g data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 A.3 Value of γ in GaAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 B Appendix to Chapter 3: Quaternions 119 C Appendix to Chapter 3: Spin polarization forbidden with M = Nφ = 1 121 D Appendix to Chapter 3: Spin polarization from dephasing 124 Bibliography 127 Citations to Previously Published Work The material of chapters 2, 3, and 5 have appeared in the following forms: “Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots,” Jacob J. Krich and Bertrand I. Halperin, Phys. Rev. Lett. 98, 226802 (2007), cond-mat/0702667; “Spin-polarized current generation from quantum dots without magnetic fields,” Jacob J. Krich and Bertrand I. Halperin, Phys. Rev. B 78, 035338 (2008), arXiv:0801.2592; “Inhomogeneous nuclear spin flips,” M. Stopa, J. J. Krich, and A. Yacoby, arXiv:0905.4520. Chapters 4 and 6 are based on manuscripts in preparation. Electronic preprints (shown in typewriter font) are available on the Internet at the following URL: http://arXiv.org vii Acknowledgments This thesis has been the influenced by many people who supported, advised, and entertained me these last six years. I cannot thank them all, but I’ll try. Foremost, I cannot imagine a better choice for my thesis supervisor than Bert Halperin. Though he initially played hard to get, he has been dedicated to guiding me both in understanding physics and also in how to be a physicist. I can only hope to have picked up a fraction of his unerring physical intuition and conscientious attention to precision and detail. Bert will always serve for me as the model scientist, with deep physical insights and great respect for those around him. My road to mesoscopic physics began in the spring of 2004, when I took Charlie Marcus’ mesoscopic physics course. Bert, while getting his caffeine fixes, often expounded on and explained subtleties of the quantum Hall effect. Amir Yacoby made regular guest appearances, challenging us to be more creative in imagining the implications of the phenomena we discussed. Charlie so exuded enthusiasm for the subject that it was inevitable I found myself drawn to mesoscopics for my thesis work. Though he failed to make me an experimentalist, he gave me my field. Charlie and Rick Heller, both on my orals committee, have provided me with much useful advice on physics, grad school, and life. Rick was also a pleasure to work with as a research and teaching supervisor, and I greatly benefited from conversations with his group, including Rob Parrott, Jay Vaishnav, Jamie Walls, Tobias Kramer, Florian Mintert, and Jamal Sakhr. In the last eighteen months I have enjoyed a fruitful collaboration dedicated to solving the mysteries of the so-called Zamboni effect. Conversations with Misha Lukin and Amir Yacoby were always enlightening and provided useful direction. Mike Stopa started me on this work and helped out in several different places in my time at Harvard; he remains a font of knowledge about all things GaAs. Jake Taylor’s clarity of thought and expression viii Acknowledgments ix are extraordinary. Michael Gullans has taught me a great deal about physics and how to fearlessly pursue it, and I look forward to more fruitful collaboration. I additionally thank Bert, Charlie, and Misha for serving on my thesis committee. I have had the great privilege and pleasure of sharing an office with Emmanuel Rashba, whose comments and advice have become an essential part of my week. The hallways of the condensed matter theory group have provided a number of people whose insights and company I have enjoyed, including Hansres Engel, Ilya Finkler, Ari Turner, Jiang Qian, Naomi Chang, Caio Lewenkopf, Adrian del Maestro, David Pekker, Mark Rudner, Izhar Neder, Michael Levin, and Wesley Wong. Hakan Tureci helped me out with the numerics of semiclassical simulations at a crucial point. Able experimentalists have worked with me throughout my time at Harvard, and I would particularly like to thank Dominik Zumbühl and Jeff Miller, for setting me straight about spin-orbit coupling in quantum dots, and Sandra Foletti, Hendrik Bluhm, David Reilly, and Christian Barthel for doing the same on all matters hyperfine. My classmates have been great friends and colleagues, from puppet show to defense. Thanks go to Subhy Lahiri, Abram Falk, Mark Romanowsky, Rodrigo Guerra, Megha Padi, Jeremy Munday, Josh Boehm, Jon Gillen, Esteban Real, Phil Larochelle, Jihye Seo, Yi-Chia Lin, and Alex Wissner-Gross. The life of a physics graduate student is made immeasurably easier by Sheila Ferguson, who provides advice for navigating the whole system and an abiding concern about each of us. I have had the honor of being supported by the Fannie and John Hertz Foundation, which gave me the freedom and flexibility to move at my own pace. For the last year, I thank the support from Bert (and the NSF) and from teaching a great course on energy x Acknowledgments technology with Mike Aziz. Though its influence is not apparent in this thesis, the most rewarding and educational part of my experience at Harvard has been with the Harvard Energy Journal Club, and I particularly thank Kurt House, Mark Winkler, Ernst van Nierop, Alex Johnson, Suni Shah, David Romps, Jason Rugolo, Kate Dennis, and Josh Goldman for their gleeful engagement with all things energy. For always keeping me entertained, informed, distracted, and wanting more, I must thank Kevin Drum, without whom this thesis might have been completed a year earlier. Through means hidden and overt, my parents fostered both my tendency to ask the great scientific question, “why?” and the self-conscious awareness that all things can and should be optimized. They have inspired, supported, and encouraged me during the formation of this thesis and throughout my life. Abby has been the best sister possible, even selflessly moving to Somerville to make the last year of my PhD better. She has cared deeply about my work, always wanting to know what I’m studying, even going so far as to read the first unintelligible draft of my first paper. Her dedication to making the world a better place has stimulated my own shift of interests in the direction of energy science. I formally joined Harriet and John Lenard’s family in my third year at Harvard, but I felt they had already welcomed me before I ever arrived in Cambridge. Their encouragement and enthusiasm have been incomparable. It is a counterfactual the truth of which we can never be sure we have determined correctly, but I do not believe I could have completed this thesis without the unconditional support, love, and inspiration of Patti. To Patti xi Chapter 1 Introduction This thesis is concerned with objects small enough that quantum mechanics is essential to understand their behavior, but large enough to contain many particles, so statistics are important. Such systems are called mesoscopic, and they combine many of the features of both atomic and bulk systems. Mesoscopic physics has been a rich subject for over twenty years and continues to bring out fundamental science and an ever-expanding set of proposed applications. We will consider some of the smallest and largest mesoscopic systems. This may not sound like a large range, but the differences between quantum dots on the scale of 100 nm and those on the scale of 5 µm can be profound. Quantum dots are carved out from a two-dimensional electron system, and the larger variety retain many of the properties of that two-dimensional system, including the relative unimportance of Coulomb interactions. As experimentalists learn to cool electrons to lower temperatures, ever larger quantum dots can be considered mesoscopic, the requirement being that the electrons maintain quantum coherence across the dot. In the smallest case, single electrons can be trapped inside a quantum dot, producing an artificial atom. This system has received attention in part because 1 2 Chapter 1: Introduction it was hoped that it would avoid many of the stochastic issues of larger mesoscale devices, making relatively easily manipulable artificial atoms for applications such as quantum information. At least in the GaAs system, however, such single-electron quantum dots are still firmly mesoscopic, due to the hyperfine interaction between the electron spin and millions of nuclear spins. While this may not be the ideal situation for technological purposes, it provides a fascinating system for probing fundamental interactions between a small number of electrons and a large number of nuclei. This thesis is divided into two parts. Chapters 2-4 concern the spin-orbit interaction (SOI) in large quantum dots while chapters 5 and 6 concern the hyperfine coupling of electron and nuclear spins in two-electron double quantum dots. In the case of the large quantum dots, we begin in chapter 2 by discussing the effects of the oft-neglected cubic Dresselhaus SOI on transport through quantum dots, showing that the cubic Dresselhaus term is important for understanding the conductance fluctuations in the devices studied; combined with experiments, these results imply that the Dresselhaus coupling in GaAs is about a third of its most commonly cited value. We move in chapters 3 and 4 to discussing the use of the SOI to manipulate the spin-polarization of currents. We use random matrix theory to determine the typical amounts of spin-polarization that can be expected by sending (charge or spin) currents through chaotic quantum dots with strong spin-orbit coupling. We use a density matrix formalism for the currents, which allows consideration of currents entangled between the channels of two leads; such entangled currents produce larger fluctuations than currents without such entanglement. To lay the groundwork for these three chapters, we begin in section 1.1 by providing an introduction to the SOI, especially in the conduction band of common zincblende semiconductors. We continue in section 1.2 to detail some relevant effects of time-reversal symmetry and the formal properties of operators in time-reversal invariant systems. With this foundation constructed, we discuss in section Chapter 1: Introduction 3 1.3 the random matrix theory of scattering, which is the primary tool used for exploring the effects of the SOI in mesoscopic transport in chapters 2-4. For the second part of the thesis, we put the SOI aside and focus on another interaction of electron spins, this time with nuclear spins. Chapters 5 and 6 form the beginning of a theoretical picture of how to control aspects of the nuclear spin configuration by manipulation of the electron spin states in a double quantum dot with two electrons. We show that, for some parameter regimes in the model we construct, the effective magnetic fields produced by the nuclei in the two quantum dots can be made to equalize with each other, while in other parameter regimes these effective magnetic fields can be made to diverge. In section 1.4, we introduce this material by describing double quantum dots and the hyperfine interaction with nuclear spins. A leitmotif in this thesis is quantum coherence and its loss. As an electron exchanges energy with phonons, other electrons, or stray fields, its wavefunction necessarily becomes a superposition with the states of all these other systems, which can effectively destroy its quantum coherence. All of the systems considered in this thesis interact with a robust environment, which causes the systems to eventually behave more classically. The dephasing time is the time required for these decohering interactions to occur.When the characteristic time a particle spends in the system, the dwell time, is on the order of the dephasing time, or shorter, quantum effects on the transport properties through the device become important. One of the appeals of technology using the spin degree of freedom is that it does not interact as strongly with its environment as does the electron charge, so it possibly can retain coherence for useful periods of time [160]. And even in cases where the electron is fully dephased, coherence of the nuclear system can be important, as discussed in chapters 5 and 6. 4 1.1 Chapter 1: Introduction Spin-orbit interaction One of the interesting and useful features of electron spin is that it is inert when compared with electrical charge. It is relatively easy to control charged particles with small electrical fields, and these are now easily produced in nanoscale environments. Spin, however, does not couple directly to electrical fields, so it is both harder to manipulate than electrical charge and resistant to various forms of environmental decoherence. The latter property makes the spin degree of freedom useful for such things as quantum information. For manipulating spins, then, we must concern ourselves with relatively weak couplings. The most direct way to interact with spin is via a magnetic field. The Hamiltonian ~ in a magnetic field is for spin angular momentum S Hs = g e ~ ~ B · S, 2mc (1.1) where g ≈ 2 is the electron g-factor, −e is the electron charge, m is the mass of the electron, and c is the speed of light. We define the Bohr magneton µB = e~/2mc, where ~ is Planck’s constant. Special relativity requires that a particle moving with velocity v ≪ c in an electric ~ experiences an effective magnetic field B ~ =E ~× field E ~v c + O(v 2 /c2 ). This magnetic field couples to the spin, producing a Hamiltonian Hso = ge ~ ~ E×~ p · S, 4m2 c2 (1.2) where we have put an extra factor of 2 in the denominator, to account for the relativistic effect of Thomas precession [70]. The same result can be derived from expansion of the Dirac equation in the nonrelativistic limit [129]. This Hamiltonian shows the coupling of the momentum (or orbit) of the electron with its spin, when in the presence of an electric ~ → −S ~ and E ~ → E; ~ this will be explored more formally field. Upon time reversal ~ p → −~ p, S Chapter 1: Introduction 5 in Section 1.2. We thus see that the spin-orbit Hamiltonian is time-reversal invariant, unlike a Hamiltonian containing an ordinary magnetic field. In solid state systems, we are mostly interested in electrons in the conduction band (or holes in the valence band), which move with respect to the strong electric fields of the ionic cores of the crystal lattice. In this case, the electric field is a function which oscillates on the length scale of the lattice constant, while electronic states of long wavelength overlap a large number of ionic cores. In a crystal, Bloch’s theorem says that the single-particle ~ wavefunctions can be written ψ = eik·~r uk,n (~r), where ~r is the position, uk,n (~r) is a highly oscillatory function periodic on the unit cell, n is a band index, and ~k is the wavevector confined to the first Brillouin zone. In the zincblende semiconductors which are the focus of this work, the conduction band has a twofold degeneracy at k = 0. This double degeneracy behaves like the electron spin degree of freedom, except with renormalized mass and g-factor. It is convenient to approximate the full Hamiltonian of the solid state system by an effective Hamiltonian for the conduction band quasiparticles, commonly called electrons. We can then write a generic Hamiltonian for the spin-orbit coupling in the conduction band of the form Hso,2 = X ∞ X µ γnml kxn kym kzl σµ , (1.3) µ={x,y,z} n,m,l=0 µ where γnml is a constant, and σµ are the standard Pauli spin matrices, acting on the double degeneracy of the band (that is, equivalent to the spin degree of freedom). This expression does not immediately look helpful, but it is usefully simplified by considering the symmetries ~ → −S. ~ Since spinof the system. As mentioned above, under time reversal, ~k → −~k and S µ orbit coupling does not violate time-reversal symmetry (TRS), we immediately require γnml to be zero unless n+m+l is odd. If the crystal has inversion symmetry, then H(~k) = H(−~k), which requires that n, m, l are all even. We thus immediately see that breaking inversion 6 Chapter 1: Introduction symmetry is required for spin-orbit coupling in the conduction band of such a crystal, so bulk crystals with the diamond lattice, such as silicon, do not have conduction band spin-orbit coupling in the absence of external fields. A large class of experimentally relevant semiconductor alloys, most notably GaAs, has the zincblende structure, which breaks inversion symmetry. In most such semiconductors, we are interested in the properties of states with wavevector k ≪ π/a, where a is the lattice constant, so it is natural to consider the terms which are of lowest order in k in the spin-orbit interaction of Eq. 1.3. Formally, the zincblende structure has point group Td [157], and there are no terms linear in k in Eq. 1.3 consistent with the point group. The leading terms were worked out by Dresselhaus, so the coupling is generally called Dresselhaus coupling. It has the form [44] Hd = γσx kx (ky2 − kz2 ) + cyclic permutations. (1.4) In the two-dimensional structures that are the subject of this thesis, the electrons are confined to an interface between GaAs and AlGaAs. For a sufficiently small electron density and temperature, the conduction band electrons occupy only one quantum state in the confinement direction, and thus form a quasi-2D electron system (2DES). The AlGaAs acts as a barrier, so the electrons mostly occupy the GaAs crystal. For heterostructures grown in the (001) direction, the structure has its symmetry group broken down to C2v or D2d , for symmetric and asymmetric confinement, respectively [157], which allows more spin-orbit terms in Eq. 1.3. In particular, it allows spin-orbit terms linear in k, which will be the dominant contributions for low electron density. The most important term comes from Eq. 1.4 by simply replacing all of the powers of kz by their expectation values across the confinement wavefunction. This gives Hd,2 = γ 2 kz (σy ky − σx kx ) + σx kx ky2 − σy ky2 , (1.5) Chapter 1: Introduction 7 where hkz i = 0, since the wavefunction is not moving in the z-direction. The linear term in Eq. 1.5 is called the linear Dresselhaus SOI. We can roughly estimate the magnitude of √ the cubic terms relative to the linear ones by considering kx ≈ ky ≈ kF , where kF = 2πn is the 2D Fermi momentum, with n the electron density. Then the cubic terms have scale γkF3 and the linear terms have scale γ kz2 kF , so we can neglect the cubic terms when kF2 / kz2 ≪ 1. If we model the confinement as an infinite square well, then the second subband becomes occupied when kF2 ≥ 3kz2 ; for an infinite triangular well[42], the second subband is occupied when kF2 & 2kz2 . Thus, if the quantum well is doped close to filling the second subband, the cubic Dresselhaus interaction can be as important as the linear terms. The cubic terms are, nonetheless, often neglected; some of their effects are considered in chapter 2. If the confinement in the growth direction lacks inversion symmetry, either from the heterostructure growth process or from the application of an electric field, there is a further contribution to the SOI from this structure inversion asymmetry (SIA). This gives a spin-orbit interaction which is linear in k, commonly referred to as the Rashba coupling, of the form Hr = α ~σ × ~k · ẑ, (1.6) where ẑ is the confinement direction. The groups C2v and D2d actually permit additional cubic spin-orbit terms than the cubic terms in Eq. 1.5, but they are generally weaker (See Appendix A and Ref. [157]). There are many ways to think about the spin-orbit interaction. One of the most convenient is to think of it as a velocity-dependent magnetic field coupled to the electron spin. That is, ~ ~k) · S/~. ~ Hso = gµB B( (1.7) 8 Chapter 1: Introduction In the case with only the Rashba and linear Dresselhaus SOI terms, it is convenient to consider the SOI in a related fashion as a spin-dependent vector potential. In this case, the Hamiltonian can be written as ~2 k2 + α(k2 σ1 − k1 σ2 ) + β(k1 σ2 + k2 σ1 ) + V (~r) 2m ~2 k2 = + (α + β)k2 σ1 + (β − α)k1 σ2 + V (~r), 2m Hl = (1.8) (1.9) where the components are with respect to the axes ê1 = [11̄0], ê2 = [110], and β = γ kz2 . We can complete the squares to rewrite this as Hl = i2 ~2 h m m m k1 + 2 (β − α)σ2 , k2 + 2 (β + α)σ1 + V (~r) − 2 (α2 + β 2 ), 2m ~ ~ ~ (1.10) where the linear spin-orbit terms look just like a vector potential, except dependent on spin, as described in Ref. [5]. Since the vector potential does not depend on position, it would at first appear to have no effect on the dynamics of the electrons. It is, however, non-abelian. That is, if we try to gauge away this constant vector potential with the unitary transformation [5] U = exp ir1 σ2 ir2 σ1 − 2λ1 2λ2 , (1.11) where λ1,2 = ~2 /2m(β ∓ α) is the distance required for the spin to precess around the effective magnetic field, we can successfully cancel the effects of the spin-orbit coupling to first order in 1/λ1,2 , but we will introduce effects at higher order in 1/λ1,2 [5]. Some implications of this theory, and its modifications when the cubic Dresselhaus SOI is included, are the subject of chapter 2. 1.2 Time-reversal symmetry Time-reversal symmetry plays such an important role in this work that it is worth defining its properties carefully. This discussion follows that in Mehta [105]. The time- Chapter 1: Introduction 9 reversal operation T is antiunitary [130] and for any basis can be written in the form T = UC (1.12) where U is a unitary operator and C is the complex conjugation operator. We require that for any states |αi, |βi,1 hT β|T αi = hα|βi . (1.13) For any operator O, we can define its time-reverse OR by hβ| O |αi = hT α| OR |T βi , (1.14) OR = T O† T −1 = U CO† CU −1 = U OT U −1 . (1.15) and it is easy to show that Since reversing time twice should have no physical effects, we further require that T 2 = α11 with |α| = 1. Then U CU C = U U ∗ = α11, so by unitarity U = αU T = α(αU ), (1.16) (1.17) so α = ±1, corresponding to integral or half-integral angular momentum. If we change the basis of states by sending |ψi → R |ψi with R a unitary transformation, then T becomes RU CR−1 = RU RT C, that is, U → RU RT . In all that follows, we choose a standard form for the time-reversal operator T and then limit ourselves to basis transformations that leave T invariant. In the case that α = 1 in Eq. 1.17, one solution is to have Ui = 11, where the subscript i refers to integer spin. To illustrate the importance of this 1 Note that since T is antiunitary, it can only be applied to kets, so hT β| is defined as the dual of T |βi. 10 Chapter 1: Introduction choice, note that it is often stated that Hamiltonians of spinless time-reversal invariant systems are symmetric (i.e., real) [105], but all Hamiltonians are Hermitian and thus unitarily similar to diagonal matrices with purely real components. The important claim is that if we limit ourselves to bases in which the time-reversal operator is T = C (i.e., Ui = 11), then all spinless time-reversal invariant Hamiltonians are simultaneously symmetric, as Eq. 1.15 implies that H = H R = H T . We see that this choice of representation of T restricts the permissible basis transformations R to be orthogonal transformations; if we use a general unitary (not orthogonal) basis transformation, then H will no longer be symmetric, though it will still of course be time-reversal invariant. In the half-integer spin case, where α = −1, then Uh = −UhT , and Uh is antisymmetric, where the subscript h indicates half-integer spin. An antisymmetric unitary matrix must have even dimension[105], and one can choose Uh to be the 2N × 2N block diagonal 0 1 repeated on the diagonal. That is, matrix with −1 0 0 −1 Uh = 0 0 .. . We recognize 0 1 −1 0 0 . . . 0 0 0 . . . 0 0 1 . . . . 0 −1 0 . . . .. .. .. . . . . . . 1 0 (1.18) as iσy where σy is the standard Pauli matrix, so the time reversal of the operator O for a spin-1/2 system takes the usual form OR = σy OT σy . Once we fix Uh in the form of Eq. 1.18, we are allowed unitary basis transformations R that preserve Uh , i.e., where Uh = RUh RT . The set of all such matrices R is called the symplectic group Sp(N ). Chapter 1: Introduction 1.3 11 Random matrix theory When presented with a relatively simple quantum system, such as a hydrogenlike atom or a parabolic potential, it is useful to solve for the eigenvalues and eigenfunctions of the Hamiltonian to determine the dynamics of the system. For many more complicated systems, such as electrons in a quantum dot in a 2DES (described in Sec. 1.4), the spectrum and eigenfunctions depend sensitively on the exact shape of the quantum dot (e.g., on the particular voltages applied to all of the gates defining the dot) and rapidly change as that shape is modified. It is thus not only impractical to solve for the spectrum of such Hamiltonians but also unhelpful, as the important, reproducible properties of the system will be those common to many particular quantum dot shapes. It has proven helpful in just this circumstance to throw in the towel and not attempt to solve for the system’s properties in detail, but rather to assume that the Hamiltonian has effectively random entries, subject to the overall symmetries of the system. For spinless noninteracting electrons confined to a so-called chaotic quantum dot with timereversal symmetry, we begin the random matrix theory (RMT) by noting that the Hamiltonian must be a real symmetric matrix. The most important energy scale for the singleparticle system is the mean level spacing (i.e., the inverse of the density of states), which in the effective mass approximation is ∆ = 2π~2 /2m∗ A, where A is the area of the quantum dot and m∗ is the effective mass. We can then consider the properties of Hamiltonians H drawn from the Gaussian Orthogonal Ensemble in which the matrix elements Hij are drawn independently from the distribution invariant under orthogonal basis transformation, which gives the unique probability distribution for the symmetric Hermitian matrix H [105] trH 2 P (H) dH ∝ exp − 2σ 2 dH, (1.19) 12 Chapter 1: Introduction a) b) w N M y x Figure 1.1: a) Section of an infinite quasi-1D wire. b) scattering region (shaded) connected to two quasi-1D wires. The wire on the left has N open modes while the wire on the right has M open modes. where the volume element dH is dH = N Y dHij , (1.20) i≤j and the standard deviation is σ = √ 2N ∆/π, where N is the dimension of H. This method was first introduced by Wigner [156] for studying heavy nuclei, and many interesting results are reviewed in Refs. [105, 12]. In this work, we shall be interested in an equivalent formulation for open quantum systems, most easily described in terms of scattering matrices. 1.3.1 Random scattering matrix theory Consider an ideal quasi-one dimensional wire in a 2DES. It is oriented in the x- direction and has a width w, as illustrated in Figure 1.1a. For spinless electrons propagating in the wire, the Schrödinger equation is ~2 2 − ∇ + V (y) ψ(x, y) = Eψ(x, y), 2m (1.21) where V (y) is zero for |y| < w/2 and infinite otherwise. This is a free particle in the x-direction and the classic particle in a box in the y-direction, with solutions ψ(x, y) = r nπy m cos e±ikx x , 2π~2 kx w w (1.22) Chapter 1: Introduction 13 where n is an integer, E = ~2 [kx2 + (nπ/w)2 ]/2m, and the state has been normalized to carry a probability current per unit energy of 1/h (in particular, the current is independent of kx ). For fixed energy E, there are N= $r 2mEw2 ~2 π 2 % (1.23) propagating states (i.e., states with kx real), where ⌊.⌋ is the integer part. These independent states are called modes or channels, and we denote them by ψn± (E), where the ± indicates the left- or right-moving state. Consider two such ideal wires with N , M open channels at energy E connected adiabatically to a scattering region such as a quantum dot, as shown in Fig. 1.1b, and let K = N + M . The incoming states can be written as superpositions of all propagating wavefunctions with momenta directed toward the scattering region, as ψin = N X j=1 + αLj ψLj + M X − αRi ψRi , (1.24) i=1 where L/R indicates the wavefunctions in the wire on the left, right respectively, and αLi , − αRi are constants. The outgoing states can similarly be written as superpositions of ψLi + and ψRi with coefficients βLi , βRi . The solution to the full quantum mechanical scattering problem for the region is then given by the matrix relation βL1 αL1 .. .. . . βLN αLN =S , β α R1 R1 .. .. . . βRM αRM (1.25) where S is a K × K complex matrix. By conservation of flux, S must be unitary. The space of all unitary matrices of dimension K is compact [66](p. 69); it is thus mathematically 14 Chapter 1: Introduction possible to take the uniform distribution over all unitary matrices of dimension K, which is called the circular unitary ensemble (CUE). We now consider the time-reversal operation. To use the convention for integerspin systems that Ui = 11, we must express our states with respect to the basis of cos(kx x) and sin(kx x). Then eikx x = cos(kx x) + i sin(kx x), and under time reversal we recover the ± ∓ → α∗ ψRn . Assuming that the system and thus the S-matrix is standard result that αψRn not modified by time reversal, we find ∗ ∗ αL1 βL1 .. . . = S .. , ∗ α∗RM βRM so we can multiply by S † and take the complex conjugate to find αL1 βL1 .. T .. . = S . . βRM αRM (1.26) (1.27) For this spinless time-reversal invariant system, S = S T , just as we found for the Hamiltonian in Section 1.2. The space of symmetric unitary matrices is also compact, and we can define the uniform distribution of all such matrices, called the circular orthogonal ensemble (COE). It is called the circular orthogonal ensemble because the distribution is invariant on multiplication by any orthogonal matrix, which is precisely the set of basis transformations that preserve the representation of the time-reversal operator, as shown in Section 1.2. If we consider the spin degree of freedom, then each of the wavefunctions of Eq. 1.22 becomes a two-component spinor, and the S-matrix becomes a 2K × 2K complex unitary matrix. Each 2 × 2 matrix in S can be expressed as a linear combination of 112 , σ1 , σ2 , and σ3 , where the σi are the standard Pauli spin matrices, and we will let σ0 ≡ 112 . The most convenient choice is to express each 2 × 2 matrix as q = q 0 σ0 + q 1 iσ1 + q 2 iσ2 + q 3 iσ3 for Chapter 1: Introduction 15 complex numbers q j . Such 2 × 2 matrices are called quaternions. The 2K × 2K complex matrix S can then be written as a K × K quaternion matrix. A quaternion has three conjugates, q ∗ = q 0∗ σ0 + q 1∗ iσ1 + q 2∗ iσ2 + q 3∗ iσ3 (1.28) q R = q 0 σ0 − q 1 iσ1 − q 2 iσ2 − q 3 iσ3 (1.29) q † = q R∗ = q ∗R , (1.30) respectively called the complex conjugate, quaternion dual, and Hermitian conjugate. For examples, see Appendix B. The Hermitian conjugate gives the same result as the Hermitian conjugate of the equivalent complex matrix, but the same is not true of the complex conjugate. The quaternion dual is denoted q R because it is precisely the time-reversal operation for a spin-1/2 particle, sending each of the spin components to its opposite. For a matrix of quaternions Q, we define (Qij )∗ = Q∗ij (1.31) (Qij )R = QR ji (1.32) Q† = QR∗ = Q∗R . (1.33) If Q is the quaternion representation of the complex matrix Qc , then QR is the quaternion representation of the complex matrix Uh QTc Uh−1 with Uh from Eq. 1.18. That is, the quaternion dual is precisely the time-reversal operation for spin-1/2 systems, which is the reason quaternions are useful to introduce. The uniform distribution over all quaternion unitary self-dual matrices is called the circular symplectic ensemble (CSE), which gets the name because the distribution is invariant under symplectic transformations, precisely those shown in Section 1.2 to preserve the form of the time-reversal operator. These circular ensembles of scattering matrices not only exist but are also quite useful in describing physical systems. Large quantum dots, if you’ll pardon the oxymoron, 16 Chapter 1: Introduction are those on the micron scale, generally containing hundreds to thousands of electrons. For such quantum dots with irregular boundaries, a plunger gate applied to the side of the dot can sufficiently change the shape of the dot so as to effectively scramble the wavefunctions inside the dot [30], thus making an easily obtained ensemble of quantum dots as a function of plunger-gate voltage. The essential physics is that for sufficiently complicated wavefunctions, the transport properties through the quantum dot are essentially random functions, subject only to the symmetries of the system, such as time reversal. The transport will be sufficiently chaotic to be described by RMT when the dwell time of particles inside the dots is long compared to the time required to interact with the boundary, also known as the bounce time or the Thouless time ET = L/v, where L is the typical linear size of the dot and v is the particle velocity, generally taken to be the Fermi velocity. If we want an ensemble of physical S-matrices to correspond to the circular ensembles (CUE, COE, or CSE), all the K modes connected to the scattering region must be coupled with ideal contacts (otherwise the reflection coefficients would generally be larger than the transmission coefficients; this case corresponds to what is called the Poisson kernel rather than the CUE [12]). For an ensemble of such chaotic quantum dots without time-reversal symmetry, such as those with an external magnetic field, the CUE is a natural guess. For systems with TRS, if there is no spin-orbit coupling, then the transport through the dot will not mix spin directions, and the transport will be effectively two copies of the COE results. When the different spin channels are strongly mixed with each other, for example by the spinorbit interaction, the CSE can be appropriate. Using the CSE requires that the spin-orbit coupling be strong enough that by the time an electron exits the dot its spin has rotated sufficiently that it is uncorrelated with its original orientation. The spin-orbit interaction can be characterized by a spin-orbit time, which is the time required for the spin-orbit interaction to rotate a spin by π. The CSE is a good ensemble to use if the dots are chaotic Chapter 1: Introduction 17 and the mean dwell time is much larger than the spin-orbit time in the material. This limit is, in practice, difficult to meet in GaAs quantum dots without making the dwell time an appreciable fraction of the dephasing time, though technologies for ever lower temperatures (increasing the dephasing time) make this regime increasingly accessible. The spectral and transport properties of ensembles of quantum dots have been shown to be well-modeled by suitable versions of random matrix theory (modified to include dephasing, which will be discussed further in chapter 3) [30, 56, 68, 171, 172]. A microscopic justification for random matrix theory for disordered metals is discussed in the review by Efetov [48]. There is a connection between the statistics of the closed-system Hamiltonians and the open-system scattering matrices [152]. Namely, if an ensemble of closed quantum dots with Hamiltonians taken from the Gaussian unitary ensemble is connected to 1D wires by ideal contacts, then the scattering matrices will be distributed according to the CUE [57]. 1.3.2 Landauer formula Having set up the scattering matrix, it is now straightforward to present one of the key results of transport theory in one-dimension (the wires are quasi-one dimensional), often referred to as the Landauer formula [91, 69, 23, 24]. Consider the case of a scattering region connected to two quasi-1D wires, as in Figure 1.1b, with the wires each connected by reflectionless contacts to large reservoirs with fixed chemical potentials µL,R in the left, right respectively. For convenience, we consider the case of spinless electrons at zero temperature. All of the right-moving states in the left lead with energy less than µL are occupied, and similarly for all the left-moving states in the right lead. In the noninteracting system, the many-particle wavefunctions are Slater determinants, and it is natural to describe them using a density matrix formalism on the single-particle wavefunctions. A density matrix is 18 Chapter 1: Introduction generally of the form w(ǫ) = |ψǫ i hψǫ |, so from the rule that |ψout i = S |ψin i , (1.34) wout (ǫ) = S(ǫ)win (ǫ)S † (ǫ), (1.35) we find where the density matrix win (ǫ) represents the states at energy ǫ ingoing toward the scattering region and wout (ǫ) represents the outgoing states at energy ǫ from the scattering region. All the ingoing states in the left lead of energy less than µL are fully occupied, and similarly for the right lead, so win (ǫ) = PL Θ(µL − ǫ) + PR Θ(µR − ǫ), (1.36) where Θ(x) is the unit step function. Thus, the net right-moving current in the left lead is −e IL = h Z dǫ tr PL [win (ǫ) − wout (ǫ)]PL , (1.37) where PL is the projection matrix onto the N modes of the left lead. The factor of e comes from considering the electric (rather than probability) current, and the 1/h arises because we normalized the states to carry a probability flux of 1/h per unit energy.2 Similarly, the R net right-moving current in the right lead is IR = dǫ tr[PR (win − wout )PR ]e/h. Since current is conserved, we must have IL = IR , which follows immediately from the unitarity of S. 2 The choice of normalization in Eq. 1.22 is not arbitrary but rather comes from considering the 1D density of states and velocity. That is, the current carried per unit energy by a 1D system is the velocity v(ǫ) times the density of states ρ(ǫ). By simple arguments, the density of right-moving states in a spinless 1D system Chapter 1: Introduction 19 We then find a total current Z n o −e I =IL = dǫ tr PL [win (ǫ) − Swin (ǫ)S † ] h Z o n −e dǫ tr PL Θ(µL − ǫ) − PL S(ǫ)[PL Θ(µL − ǫ) + PR θ(µR − ǫ)]S † . = h (1.40) (1.41) For ǫ > µL , µR , the integrand is clearly zero. For ǫ < µL , µR , the integrand is zero by unitarity of S, since PL + PR = 11K . Without loss of generality, let µL ≥ µR , so −e I= h Z µL µR dǫ tr[PL − PL S(ǫ)PL S(ǫ)† ]. (1.42) If S(ǫ) is a constant over this range of energy (i.e., in the linear response regime [40]), −e (µL − µR )tr(PL − PL SPL S † ) h e2 I = VLR tr(PL − PL SPL S † ), h I= (1.43) (1.44) where VLR is the voltage between the left and right reservoirs. We then have an expression for the conductance G = I/VLR , which is the Landauer formula, though not in the way it is usually written down. For that, we need to express S in terms of transmission and reflection matrices ′ r t S= , t r′ (1.45) where r is the N × N reflection matrix of the left lead, r ′ is the M × M reflection matrix of the right lead, t is the M × N transmission matrix from left to right, and t′ is the N × M of free fermions is r m 2π 2 ~2 ǫ 1 , = 2π~v(ǫ) 1 ρ(ǫ) = 2 (1.38) (1.39) for velocity v. It is thus clear that the velocity cancels out of the current carried per unit energy. The current per unit energy is then i = ρ(ǫ)v(ǫ) = h1 , which is the choice made in the main text. 20 Chapter 1: Introduction transmission matrix from right to left. By unitarity of S, rr † + t′ t′† = 11N . We then see that Eq. 1.44 is e2 VLR tr(11N − rr † ) h e2 e2 = VLR tr(t′ t′† ) = VLR tr(t† t), h h I= which is the usual Landauer formula, giving G = e2 † h tr(t t). (1.46) (1.47) The key insight of the Landauer formula is that the conductance through a device is proportional to the probability that an electron incident from the left will exit to the right, which is the meaning of tr(t† t). In the theory of mesoscopic transport, theorists like to consider ideal 1D wires coupled adiabatically by reflectionless contacts into the scattering region, as used in the discussion here. In practice, experiments are performed with quantum dots separated from the large 2DES by a quantum point contact (QPC), which is not a long 1D wire. The central constriction of the QPC is the closest the experiments generally come to a 1D wire. But the conductance through that QPC is quantized [149], just as for the ideal 1D wire, so we apply the theory, with generally good agreement. Differences in predictions for some conduction features were studied in Refs. [60, 162, 6], among others. 1.4 2DES and quantum dots The work in this thesis concerns electrons confined into a quasi-two dimensional world at the interface of two semiconductors, such as GaAs/Al0.3 Ga0.7 As. Though more expensive than the standard silicon heterostructures that are the workhorses of modern computers, the GaAs/AlGaAs system provides a number of advantages. The similarity of the lattice constants of GaAs and AlGaAs allow the interface to be grown with nearly atomic perfection. Modulation doping, where the dopants are displaced by tens or hundreds of nanometers from the interface, allow the creation of low density electron systems at the Chapter 1: Introduction 21 interface without strong scattering from the ionized dopants[65]. Such a system with all of the electrons occupying the ground state wavefunction in the confinement direction is called a two dimension electron system (2DES). These systems can have long Fermi wavelengths, on the order of 50 nm, depending on the choice of electron density, and mean free paths on the order of tens of microns, allowing devices to be built firmly in the ballistic limit, i.e., where the device size is smaller than the mean free path. A metal gate placed on the top of the semiconductor wafer can modulate the electron density beneath, entirely depleting the electrons with a gate voltage of only a fraction of a volt. This allows the sculpting of the lateral confinement potential of the 2DES, limiting the electrons to move in large 2D regions, small 1D wires, or confined regions called quantum dots. Quantum dots containing as few as one electron are made routinely, and larger quantum dots containing hundreds to thousands of electrons are also studied. The important feature of a quantum dot is that its length be less than both the electron mean free path and dephasing length, so the coherent properties of the system confined to the quantum dot (and not interacting with impurities) can be probed. Within the tiny world of quantum dots, this thesis considers the largest and the smallest specimens. In the large, micron-scale quantum dots, the electrons are well-modeled as non-interacting in the framework of Landau’s Fermi liquid theory [109]. It is in this regime that we can treat the transport properties of noninteracting electrons using the random scattering matrices of Section 1.3. In the other limit of small, few-electron quantum dots, the Coulomb repulsion energy is much larger than the kinetic energy, so we can study single electrons in their orbital ground states. This situation is ideal for gaining access to the spin degree of freedom, as we will now elaborate. 22 1.4.1 Chapter 1: Introduction Double quantum dots Here we are interested in a double quantum dot system, that is, two immediately adjacent quantum dots, as illustrated in Fig. 1.2. We will consider the situation in which there is one orbital quantum state energetically accessible in each dot. By controlling the voltages on gates around the edges of the dots, the two-particle ground state can be shifted from having one electron in each dot to having both electrons in one dot. In the case that both electrons occupy the single orbital on the right dot, they must form a spin-singlet to satisfy the Pauli exclusion principle, where we neglect any spin-orbit splitting for simplicity. The gate-controlled energy splitting between the (1,1) and (0,2) states (where (NL ,NR ) indicates the number of electrons on the left, right dots) is ǫ, called the detuning. There are four allowed (1,1) states, and we use the singlet/triplet basis to describe them. Combined with the single energetically accessible (0, 2)S state, these five states form the relevant electronic space for all that follows. As shown in Fig 1.3a, finite tunnel coupling between the two dots allows hybridization of the (0, 2)S and (1, 1)S states. An external magnetic field Bext will split off the triplet states, giving the energy diagrams shown in Fig. 1.3b. For typical GaAs quantum dots, the energy splitting to the next orbital states is ∼ 1 meV, and the tunnel coupling γc is ∼ 10 µeV [147]. With the basis {(0, 2)S , (1, 1)S , T+ , T0 , T− }, the Hamiltonian can be written −ǫ γ c γc 0 , Hc = (1.48) −Ez 0 EZ where Ez = |g∗ µB Bext | is the Zeeman energy, with g∗ the effective g-factor for conduction band electrons, g∗ = −0.44 in GaAs [155], and µB the Bohr magneton. Chapter 1: Introduction 23 a) gate depleted region Ohmic contact AlGaAs 2DEG GaAs b) I DO T I QP C I QP C 200 nm Figure 1.2: a) Schematic of a double quantum dot device in GaAs/AlGaAs 2DES, showing gates on the top of the chip which can deplete the electron gas beneath them. b) SEM micrograph of a double quantum dot, showing the locations of the two dots. Figure indicates a device with current passing through it, unlike those discussed in this work. Adapted from [64]. (0,2) (0,2) S S (1,1) T− (1,1) S 0 (1,1) S Energy Energy Ez (1,1) (1,1) 0 S T0 (1,1) S −Ez (1,1) T+ (0,2) (0,2) S 0 ε S 0 ε ε̃ Figure 1.3: a) Energy levels of the lowest lying charge states in the two-electron double quantum dot, as a function of the detuning ǫ between the (1, 1) and (0, 2) charge states, including tunnel coupling γc between the dots. b) Energy levels, including spin, in the two-electron double quantum dot in the presence of an external magnetic field. 24 Chapter 1: Introduction The tunneling process conserves spin, so only the (1, 1)S state can tunnel over to the (0, 2) charge state, since the (0, 2) triplet state requires one of the electrons to occupy an energetically inaccessible orbital state. This produces the phenomenon of Pauli blockade [111], in which the system gets stuck in one of the (1, 1) triplet states even in a situation where (0, 2)S is energetically favorable. Pauli blockade is very useful experimentally, as it converts the spin information (i.e., whether the state is singlet or triplet) into charge information (i.e., whether both electrons are in the right dot or one in each). The charge state of the dots can be measured using an adjacent quantum point contact charge sensor [53, 75]. It is useful to diagonalize the upper 2 × 2 matrix in Eq 1.48 to find the adiabatic singlet states. Following Taylor[146], we write them as E S̃ = cos θ |(1, 1)S i + sin θ |(0, 2)S i E G̃ = − sin θ |(1, 1)S i + cos θ |(0, 2)S i , where θ = arctan and the energies are 12 (−ǫ ± 1.4.2 2γ p c ǫ − 4γc2 + ǫ2 ! , (1.49) (1.50) (1.51) p ǫ2 + 4γc2 ), with |S̃i the lower energy state [147]. Hyperfine coupling There are, of course, processes that can provide further off-diagonal matrix ele- ments in Eq. 1.48. These can include spin-orbit coupling and cotunneling processes, but the ones considered in this work are from the hyperfine interaction between the electron and nuclear spins. Not all materials, of course, have nuclear spins, but GaAs is blessed (or cursed) with an abundance of them, with gallium consisting of two stable isotopes, and 71 Ga, and arsenic having just one stable isotope, 75 As, 69 Ga all three of which have spin Chapter 1: Introduction 25 3/2 nuclei. The conduction band in GaAs is formed from atomic s-orbitals, which have an overlap with the atomic nuclei. Thus, the electron spin dipole overlaps with the nuclear dipole, forming the Fermi contact hyperfine interaction between the electron and nuclei. For a single electron interacting with many nuclear spins, the hyperfine Hamiltonian is [1] Hhf = v0 X ′ ~ k,β )S ~ · I~k,β , Aβ δ(~r − R ~2 (1.52) k,β ~ k,β is the position of the kth spin of species β, ~r is the where β is the nuclear species, R ~ is the electron spin, I~k,β is the kth nuclear spin of species β, A′ is the electron position, S β hyperfine coupling constant, which depends on the type of nucleus, and v0 is the unit cell volume (which contains two nuclei in GaAs). Writing Eq. 1.52 as an effective spin Hamiltonian, taking matrix elements with the electron spatial wavefunction, gives ~ 2S ~ nuc ·B ~ 2 v0 X ′ ~ = ∗ Aβ ψ(Rk,β ) I~k,β . g µB ~ Hhf = g∗ µB ~ nuc B (1.53) (1.54) k,β ~ nuc has the form of a magnetic field coupled to the electron spin, in analogy We see that B with Eq. 1.1. For conduction band electrons, the wavefunction can be written ψ(~r) = u(~r)f (~r) where u(~r) is a highly oscillatory periodic function on the crystal lattice and f (~r) ~ kβ )|2 and normalize is smoothly varying on the scale of the lattice. We let dβ = |u(R R |f (~r)|2 = v0 . The dβ vary because the electron generally has greater weight on the As sites than the Ga sites. From Ref. [113] dAs = 98 Å−3 , dGa = 58 Å−3 . The number of As atoms per unit cell is xAs =1, and x69 Ga = 0.6 and x71 Ga = 0.4 [113]. We 26 Chapter 1: Introduction ~ nuc as can rewrite B ~ nuc = B v0 X ′ ~ 2 ~ A x d ( R ) f Ik,β β β k β g ∗ µB ~ (1.55) k,β = X k,β where ~ 2 ~ bβ f (R k ) Ik,β /~. bβ = v0 ∗ g µ B A′β xβ dβ . (1.56) (1.57) Using v0 = a3 /4 with a = 5.63 Å the GaAs lattice constant defined with eight atoms per unit cell [101] gives the result [113] b75 As = −1.84 T, b69 Ga = −0.91 T, b71 Ga = −0.78 T, using g∗ = −0.44. If all of the nuclei are polarized in the ẑ direction, then X ~ nuc = 3 B bβ ẑ = −5.3 T ẑ, 2 (1.58) β so the fully polarized nuclear system exerts a 5.3 Tesla magnetic field on the conduction electrons in GaAs. We can also express the couplings as Aβ ≡ A′β xβ dβ v0 = g∗ µB bβ , which incorporates the rapidly oscillating part of the wavefunction and abundance of the nuclei, so it gives the energy scale that couples only to the smooth envelope function of the conduction electrons. The GaAs energy scales are A75 As = −47 µeV, A69 Ga = −23 µeV, A71 Ga = −20 µeV. Chapter 1: Introduction 27 Thus, the typical energy scale for the GaAs hyperfine interaction is Atot = −90 µeV. The negative sign indicates an anti-ferromagnetic interaction between electron and nuclear magnetic moments. This discussion has been general for conduction band electrons, in or out of quantum dots, but has considered only a single electron. In the case of a double quantum dot ~1 and S ~2 . We consider only two orbital with two electrons, there are two electron spins, S wave functions, ψL/R (~r), with L/R indicating the left/right dot. We can write the Hamiltonian in a simplified form by considering the single-particle electron wavefunctions ψ(~r1 ) R to consist only of the envelope-function (normalized so |ψ(~r)|2 = 1), without the highly oscillatory Bloch function. In that case, we can use the A coupling constant, and if we consider only one species of nuclei, we can write the hyperfine Hamiltonian as a sum of terms like Eq. 1.52. Hhf = Atot i v0 X h ~ k )S ~1 · I~k + δ(~r2 − R ~ k )S ~2 · I~k . δ(~ r − R 1 ~2 (1.59) k Since the orbital wave functions in the quantum dots are non-degenerate ground states of a confining potential, we can take them to be real. The spatial wavefunctions ψR (~r), ψL (~r) are not necessarily orthogonal. Taking matrix elements of Hhf with the double dot electronic basis states as in Eq. 1.48, we find 00 0 0 1 Hhf = √ 2 2 A A† B , (1.60) 28 Chapter 1: Introduction where √ RL RR LL RR I− − I− 2 I− − hR|Li I− √ RR LL A = 2 hR|Li IzRR − IzRL 2[Iz − Iz ] √ RR − I RL RR − I LL I 2 hR|Li I+ + + + √ LL RR LL RR 0 2[Iz + Iz ] I− + I− LL RR LL RR B = I+ + I+ , 0 I− + I− √ LL + I RR − 2[I LL + I RR ] 0 I+ + z z where v0 I~AB ≡ Atot ~ X ~ k )ψB (R ~ k ), I~k ψA (R (1.61) k for A, B either R(ight) or L(eft), which is correct to first order in the wavefunction overlap hR|Li = hψR |ψL i. We see from the second column of A that transitions from the (1, 1)S to ~ = (I~LL − I~RR )/2. the three triplet states T+ are mediated by the nuclear difference field D There are direct transitions from (0, 2)S to the triplet states as long as there is finite overlap between the wavefunctions. The first entry of A shows that (0, 2)S can transition to T+ RR ) or flipping one in the barrier (I RL ). either by flipping a spin in the right dot (hR|Li I− − These overlap terms are generally small, so the usual course is to set hR|Li = I~RL = 0, giving the simpler Hamiltonian 0 0 0 0 0 √ 0 0 D+ − 2Dz −D− 1 √ Hhf = √ 0 D− 2Sz S− 0 , 2 √ 0 − 2Dz S+ 0 S− √ 0 −D+ 0 S+ − 2Sz (1.62) where S~ = (I~LL + I~RR )/2. Eq. 1.62 shows that in this limit there is no direct hyperfine coupling from (0, 2)S to any of the triplets, as the absence of wavefunction overlap hR|Li Chapter 1: Introduction 29 means there is no way to move an electron from one dot to the other by a hyperfine transition. If we include the tunnel coupling γc between left and right, then there are hyperfine processes for the (0, 2)S state which are second-order in γc /ǫ. We also see that ~ couples the singlet states to the triplet states and the sum field S~ the difference field D couples the triplet states to each other. The total Hamiltonian in this subspace of 5 states is the sum of Eqs. 1.48 and 1.62. S̃, T0 subspace In the limit that ǫ < 0, |ǫ/γc | ≫ 1, the states S̃ and T0 come close to degeneracy, and the Hamiltonian for just that subspace in the basis {S̃, T0 } is approximately −J(ǫ) −Dz HST0 = , −Dz 0 (1.63) where the exchange splitting J(ǫ) is the hyperfine-free energy difference between S̃ and p T0 . In the model of Eq. 1.48, J(ǫ) = 12 (−ǫ + ǫ2 + 4γc2 ) ≈ γc2 / |ǫ|. We see from Eq. 1.63 that the S̃ and T0 states are coupled by the difference in the z-components of the effective magnetic field induced by the nuclei in the left and right dots. Eq. 1.63 has been written down assuming that the quantization axis for the electron spin is along the external field ~ ∗ µB | ≈ 1.3 mT [124]. B~ext . This is a good approximation as long as Bext ≫ |S/g The {S̃, T0 } subspace has been extensively studied for quantum information applications [64, 116, 85]. Since both electronic states have zero angular momentum along the external field, the states are unaffected to leading order by fluctuations of the external field, which can give them long coherence times. The nuclear field is the dominant source of dephasing. Experimentally, the ensemble coherence time T2∗ is measured [116] by initializing the system at large ǫ in the (0, 2)S state. Then ǫ is reduced rapidly compared to the 30 Chapter 1: Introduction hyperfine field scales but slowly compared to γc , which separates the two electrons, one in each dot. The goal is to reduce ǫ far enough that J(ǫ) . Dz , so the {S̃, T0 } superposition evolves due to Dz , as indicated by Eq. 1.63. The detuning ǫ is held at this value for a time τs and then ǫ is ramped quickly back to its original value. An adjacent charge sensor [53] can detect whether the system made a transition to the T0 state, thus staying in the (1, 1) charge state due to Pauli blockade. This cycle is repeated many times at each τs , accumulating a singlet-return probability PS (τs ), which shows a Gaussian decay with τs , with decay constant T2∗ [116]. ~ evolves slowly on the timescale on which the electron experThe hyperfine field D iments are performed [146], so the effects of Dz in Eq. 1.63 can be removed by standard spin-echo techniques [116]. These techniques show that the coherence time T2 of the {S̃, T0 } space is greater than 1 µs [116]. If, however, no spin-echo techniques are used, each element of the ensemble of measurements is performed with a different value of Dz , so the Rabi oscillations are lost in ensemble averaging, giving an ensemble coherence time T2∗ of approximately 15 ns [124]. If Dz could be reduced, then the {S̃, T0 } space would increase in value as a quantum information resource, as it would not require spin-echo procedures to maintain coherence, simplifying the necessary pulse sequences. Dynamic nuclear polarization - the S̃, T+ subspace There is a second important cycle in these double quantum dots, which we call the dynamic nuclear polarization (DNP) cycle. It uses the avoided crossing between S̃ and T+ , i.e., near where EZ ≈ J(ǫ). With EZ large enough that the other states are separated by a large energy gap, the effective Hamiltonian near this degeneracy in the basis {S̃, T+ } Chapter 1: Introduction 31 is HST+ −J(ǫ) = √θ D− cos 2 √θ D+ cos 2 . −Ez + Sz (1.64) We define ǫ̃ to be the value of the detuning ǫ such that J(ǫ̃) = EZ − Sz , as marked in Figure 1.3b. The contact hyperfine interaction is rotationally invariant and thus conserves total angular momentum, so flipping an electron spin up on transitioning from S̃ to T+ must involve lowering a nuclear spin, which is clear from HST+ . This controlled flip of the nuclear spin makes it possible to polarize the nuclei using a pulse sequence similar to the T2∗ cycle described above. There are a few variants on the DNP sequence [117, 124, 55], one of which is illustrated in Figure 1.4. If the slow sweep were truly adiabatic, this cycle should flip one nuclear spin each time it is repeated. As the nuclear system polarizes, Sz decreases, increasing the total effective magnetic field felt by the electrons and thus also increasing ǫ̃. For typical single-electron quantum dots, the wavefunction strongly interacts with ≈ 106 nuclei in each dot [146]. Repeating this DNP cycle at 4 MHz for a second or two should then be able to flip all of the nuclei down. In practice, polarizations of only about 1% are observed [117, 123, 124]. That is, the shift of ǫ̃ after running the DNP cycle is consistent with a nuclear field strength of approximately 80 mT, while the fully polarized system would have a magnetic field of 5.3 T, as in Eq. 1.58 [116]. Similar processes using transport through the vertical quantum dots rather than a gate-controlled pulse sequence have succeeded in producing nuclear polarizations of approximately 40% [11]. These nuclear polarizations are, of course, not static. If the electrons are not in a singlet configuration, the dominant decay mode is believed to be the electron-mediated nuclear-nuclear coupling [1, 38, 165, 123]; otherwise, the dominant decay mode is the nuclear dipole-dipole coupling, which causes the 32 Chapter 1: Introduction Adiabatic nuclear polarization Prepare Singlet T- T0 (0,2)S Energy T+ (0,2)S ε~ 0 ε Rapid Adiabatic Passage (0,2)S T- T0 t=0 Energy T+ (0,2)S εS ε 0 Slow Adiabatic Passage (0,2)S t=τS T- T0 Energy T+ (0,2)S 0 εF ε Figure 1.4: Schematic of the DNP cycle, adapted from [117]. This figure has the energy levels rotated 45◦ from those in Figure 1.3, which corresponds to ǫ/2 added to the Hamiltonian of Eq. 1.48. At top, the system is loaded in the (0, 2)S state. At middle, the detuning ǫ is moved rapidly (compared to the magnitude of D− ) past ǫ̃, not allowing transition into the T+ state. At bottom, ǫ is increased slowly, flipping an electron and nuclear spin on transitioning from singlet to T+ . Chapter 1: Introduction 33 polarization to diffuse out from the quantum dots to the surrounding material [114]. The decay time for the dipole-dipole diffusion is on the order of 10 seconds [123], so it should not be limiting the polarization in the lateral double dot devices to only 1%. It is possible that something is sending D− to zero, so the avoided crossing itself closes; that is, the probability of transitioning to the T+ state decreases as the DNP cycle is repeated, but we do not yet know. In chapter 6 we present a small piece of the puzzle indicating that such a suppression of D− may occur in some cases. The subject of dynamic nuclear polarization became even more interesting when it was reported that performing the DNP cycle increased the measured T2∗ by a factor of 70 to about 1 µs, which implies a drastic reduction of Dz [124]. Mike Stopa had an idea for a force which could cause this, which is detailed in chapter 5. An implication of this work is that loading in the T+ state and transferring to the S̃ state should produce the opposite effect, causing a large increase in |Dz |. When Sandra Foletti and Hendrik Bluhm in Amir Yacoby’s group tried this experiment, they indeed found results consistent with a large |Dz | [55]. However, they also found a large |Dz | even when executing the usual DNP cycle. There are more experiments on the subject ongoing as this thesis is being written, but the interface between the DNP cycle and the nuclear spins is quite rich, as will be discussed further in chapter 6. 1.5 Organization of thesis Chapters 2-4 concern spin-orbit coupling in many-electron quantum dots while chapters 5-6 concern hyperfine interactions between electron and nuclear spins in two electron double quantum dots. We give a brief overview of each chapter. 34 1.5.1 Chapter 1: Introduction Chapter 2 We consider the effects of the cubic Dresselhaus spin-orbit interaction on the con- ductance fluctuations in transport through a quantum dot with a magnetic field. This topic follows experiments of just this type performed by Dominik Zumbühl and collaborators on a set of four nominally chaotic quantum dots of two different sizes [171, 172]. A random matrix theory explanation of the results was provided by Jan-Hein Cremers, Piet Brouwer, and Vladimir Fal’ko, which was in good agreement with the experimental results [39]. Cremers’ theory, however, started from the assumption that only the linear Dresselhaus and Rashba SOI were important. As mentioned in section 1.1, this is generally a good approximation for sufficiently low density electron systems. There is, however, reason to believe that the cubic Dresselhaus term should be anomalously important in quantum dot systems. As noted in Eq. 1.10, the linear spin-orbit terms appear as a constant non-abelian vector potential, and so can be gauged away to first order in the inverse spin-orbit lengths, L/λ1,2 , where L is the linear size of the quantum dot [5]. This implies that the relevant comparison for the cubic and linear spin-orbit terms is not of the order of kF2 / kz2 , as discussed in section 1.1, but rather λ1,2 kF2 /L kz2 . For sufficiently small quantum dots, then, the cubic SOI effects should be more important than the linear SOI effects. We perform semiclassical simulations of chaotic quantum dots, modeled as twodimensional billiards, to estimate the magnitude of the cubic Dresselhaus contribution to the conductance fluctuations. We combine these results with Cremers’ RMT formulation and refit Zumbühl’s data. The functional form of the fitting functions is unchanged by the inclusion of the cubic Dresselhaus contribution, but the interpretation of the parameters is modified to include both the cubic and linear SOI. As a result, the refits cannot improve on those in the original papers, but an interesting conclusion results nonetheless: the most Chapter 1: Introduction 35 widely cited value for the Dresselhaus SOI constant, γ of Eq. 1.4, is incompatible with our results and the conductance fluctuations measured by Zumbühl. It turns out that the value of γ in GaAs is disputed within a factor of three. The most commonly cited value in recent years, 27.5 eVÅ3 , comes from Ref. [157], but there have been an array of theoretical and experimental results over the last thirty years. These results are summarized in Table A.1 in appendix A, which has been updated since its original publication [88] to include more recent results [86, 97]. Our results are consistent with |γ| approximately equal to 10 eVÅ3 or less, which is supported by a number of experiments, though by no means all. Our results imply both that γ should be on the lower end of theoretical and experimental results and that the cubic Dresselhaus SOI can have an anomalously important role for spin-flip processes in quantum dots. 1.5.2 Chapter 3 The generation of spin-polarized currents without magnetic fields or ferromagnets is an important goal for the field of spintronics [160], and there have been a number of proposals for specific devices that produce such spin-polarized currents using SOI. We consider the spin polarization produced by sending a charge current through a chaotic quantum dot with strong SOI in a two-terminal geometry. The nonintuitive production of spin polarization from a charge current violates no requirements of TRS, except in one special case. We rederive a theorem which stipulates that with TRS, no spin polarization can be produced in the exit lead if it has only one open channel. Using random scattering matrix theory, we find that having just two open channels in the right lead and one in the left produces an rms spin polarization of 45%. We include the effects of dephasing, finite temperature, and finite bias to quantify how much the spin polarization is reduced. Interestingly, we find that dephasing removes 36 Chapter 1: Introduction the restriction that having only one channel forbids spin polarization in the right lead. Since such chaotic quantum dots have been studied in laboratories around the world for decades, we believe that the only reason these spin polarizations have not been observed is that there is, as of now, no good way to measure spin polarizations in small currents. When such measurement techniques improve, these currents will likely be found to be generic features of mesoscopic devices with strong spin-orbit interaction. 1.5.3 Chapter 4 In the converse case, it is interesting to consider the effects of sending a spin- polarized current or a pure spin current through a quantum dot with SOI, as might occur in a spin field effect transistor [136]. We use a similar random matrix theory and density matrix formulation as in chapter 3, which allows us to consider all the usual methods for spin-conductance studies. This formalism is flexible enough, however, to also allow us to study currents with arbitrary entanglement between the orbital channels, which has not previously been considered. Currents entangled between the channels of the two leads show larger fluctuations than currents without such entanglement. 1.5.4 Chapters 5 and 6 We move to double quantum dots with two electrons to study the hyperfine inter- action with the nuclei. Motivated by experiments whose implications are, as of this writing, still uncertain, we consider the forces that a dynamic nuclear polarization cycle imposes on the distribution of Overhauser fields in the double quantum dot. In chapter 5, we study a model in which the electron wavefunction is taken to be uniform inside a spherical box, so we can consider each quantum dot as effectively having a single large nuclear magnetic moment. In that model, we explore feedback between the Overhauser difference field Dz Chapter 1: Introduction 37 and the probability that a nuclear spin will flip on the left or right dot during the DNP cycle. We find that for the standard DNP cycle in GaAs, there is a feedback force in the direction of reducing |Dz |. In chapter 6, we consider a more sophisticated semiclassical model for the nuclear dynamics during the DNP cycle, including a treatment of the Landau-Zener-like sweep through the |S̃i-|T+ i avoided crossing and an inhomogeneous electron wavefunction. We find a rich parameter space which can send |Dz | strongly to zero or, alternatively, can send |Dz | to large values. Exactly how these results relate to the experimental findings is an area of active research. 1.6 1.6.1 Summary of new results obtained in this research Chapter 2 We have introduced the cubic Dresselhaus spin-orbit interaction (SOI) for the first time into calculations of the conductance fluctuations of transport through a quantum dot. We find that the cubic Dresselhaus effects should be large enough to be inconsistent with experimental results [171, 172] unless the GaAs Dresselhaus coupling constant γ is about a third of its most commonly cited value. These results provide evidence, added to that of several experimental and theoretical works, in favor of a smaller value of γ. Understanding the value of γ is essential for designing spintronic devices using the SOI in GaAs. 1.6.2 Chapter 3 We show that unpolarized currents incident on quantum dots with spin-orbit cou- pling will produce spin-polarized currents in most geometries. We quantify the expected values of spin-polarization using random matrix theory, and found expected spin polariza- 38 Chapter 1: Introduction tions up to 45%. We include the effects of dephasing, finite temperature, and bias, each of which reduces the expected spin polarizations. Such devices could provide useful sources of spin-polarized currents for spintronic devices without magnetic fields or ferromagnets. 1.6.3 Chapter 4 Here, we extend the results of chapter 3 to arbitrary charge or spin currents inci- dent on the quantum dot, finding the average and fluctuations of charge and spin currents produced, with and without time reversal symmetry. This work also details a useful improvement in formalism from the standard method of S-matrix based transport calculations. By considering the input currents using a density matrix, it is possible to consider entangled currents, which are shown to produce larger fluctuations than incoherent currents. Understanding such fluctuations is important for designing spintronic devices using the SOI. 1.6.4 Chapters 5 and 6 Here, we try to understand how nuclear spins can be manipulated by dynamic nuclear polarization (DNP) with the hyperfine interaction in experiments with two-electron double quantum dots. In particular, we wish to understand circumstances where Dz , the difference in z-components of the Overhauser fields produced by nuclei in the two dots, can increase or decrease after many electronic cycles. Chapter 5 details one effect of the electronic and nuclear structure which could cause the DNP cycle to reduce Dz . In chapter 6, we develop a more sophisticated model for the evolution of the nuclear spins. We break the nuclei into groups, each with constant hyperfine coupling to the electrons. This allows treating each group as having a single large nuclear spin, which we model semiclassically. ~ are When all three components of the total difference nuclear magnetic field, D, zero, the system is in a “dark state” and does not evolve. We simulate the system numer- Chapter 1: Introduction 39 ically and find that for some parameter regimes, the system is attracted to the vicinity of such fixed points; for other parameters these dark states are only metastable. In the latter case, the system tends to produce a large Dz . This is the first theoretical work indicating that large Dz can be produced in this system. The features of the equations of motion that cause these behaviors are still poorly understood. This work is important for evaluating whether a quantum computer using GaAs quantum dots can ever be built. Chapter 2 Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots Jacob J. Krich and Bertrand I. Halperin Physics Department, Harvard University, Cambridge, Massachusetts Abstract We study effects of the oft-neglected cubic Dresselhaus spin-orbit coupling (i.e. ∝ p3 ) in GaAs/AlGaAs quantum dots. Using a semiclassical billiard model, we estimate the magnitude of the spin-orbit induced avoided crossings in a closed quantum dot in a Zeeman field. Using previous analyses based on random matrix theory, we calculate corresponding effects on the conductance through an open quantum dot. Combining our results with an experiment on an 8 µm2 quantum dot [D. M. Zumbühl et al., Phys. Rev. B 72, 081305 (2005)] suggests that 1) the 3 GaAs Dresselhaus coupling constant γ is approximately 9 eVÅ , significantly 3 less than the commonly cited value of 27.5 eVÅ , and 2) the majority of the spin-flip effects can come from the cubic Dresselhaus term. Control over electron spin in semiconductors has promise for quantum computing 40 Chapter 2: Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 41 and spintronics. In such applications, it is essential to understand how the transport of an electron through a circuit affects its spin; i.e., we must understand spin-orbit coupling (SO). In technologically important III/V semiconductor heterostructures, SO originates in the asymmetry of the confining potential (called the Rashba term), which can be controlled by gates, and in bulk inversion asymmetry of the crystal lattice (called the Dresselhaus term). In quasi-2D systems, the Dresselhaus term has two components, one linear in the electron momentum and the other cubic. The cubic Dresselhaus term (CD) is usually neglected, as it is generally smaller than the linear contribution. Datta and Das proposed a spin-field-effect transistor (SFET) for quasi-1D ballistic wires with Rashba coupling [41]. Schliemann, Egues, and Loss proposed an SFET that can operate in diffusive quasi-2D systems based on tuning the Rashba and linear Dresselhaus (LD) terms to be equal in strength, which produces long spin lifetimes, neglecting CD [136]. The strengths of the SO terms are difficult to measure independently, but a full understanding of their strengths is crucial to making such devices. Additionally, in confined systems such as quantum dots, some effects of the linear SO terms are suppressed [63], and it is important to know the magnitude of the CD contribution, which could limit or even prevent the functioning of spintronic devices. We characterize the strength of CD in a confined system by its effect on avoided crossings in an in-plane magnetic field Bk that couples only to the electron spin. With no SO, each eigenstate can be written as a product of orbital state |αi and spin quantized along Bk . Eigenstates |α ↑i and |β ↓i become degenerate when ǫα − ǫβ = EZ , where ǫα,β are the orbital energies and EZ is the Zeeman energy,1 but SO leads to avoided crossings. In the first half of this Letter, we estimate the CD contribution to the avoided 1 This neglects other spin-flip processes, e.g., hyperfine coupling, which in GaAs is much smaller than SO effects for quantum dots on the micron scale. Chapter 2: Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 42 crossings, which can be larger than the linear terms’ contribution since the latter are suppressed for small Bk [63]. In the second half of this Letter, we relate these avoided crossings in closed quantum dots to the mean and variance of the conductance when the quantum dot is connected to ideal leads. We compare these predictions to the results of Zumbühl et al. [172, 171] and find that agreement is possible only if the CD coupling constant in GaAs γ 3 is considerably less than the frequently cited value of 27.5 eVÅ [83, 157] from k · p theory. A smaller value of γ has also been suggested by experiments [126, 77] and band structure studies [132, 133, 31]. Even with this smaller value of γ, we find that CD is the dominant spin-flip mechanism in the sample considered. We consider conduction electrons in a 2D electron system (2DES) grown on a (001) surface of a III/V semiconductor confined to a small area by a potential V (r). We use an effective Hamiltonian H= (p − Aso )2 γ + 3 (p22 − p21 )(p × σ) · ê3 2m 2~ 1 + V (r) + gµB B · σ 2 (2.1) where p = P − eA/c, P is the canonical momentum, A is the vector potential from the perpendicular magnetic field, σ is the vector of Pauli matrices, m is the effective mass, Aso = ê1 ~σ2 /2λ1 − ê2 ~σ1 /2λ2 is the effective SO vector potential, which contains both the LD and Rashba SO terms, and λ1,2 are the (linear) SO lengths [5, 39] We choose a coordinate system with axes ê1 =[110], ê2 =[11̄0], and ê3 =[001̄]. The second term is the CD.2 In a system of linear size L, the linear SO terms can be gauged away to first order in L/λ by the unitary transformation H → U HU † ≡ H′ where U = exp(ir · Aso ) [5]. 2 We ignore other cubic terms, which are allowed by symmetry, but are estimated to be small. See Appendix A. Chapter 2: Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 43 Expanding to leading order in L/λ, H′ = + 1 (p − a⊥ − ak )2 + bZ + bZ⊥ 2m γ (p2 − p21 )(p × σ) · ê3 + V (r), 2~3 2 (2.2) where a⊥ = (~σ3 /4λ1 λ2 )[ê3 × r], ak = (~/6λ1 λ2 ) × (x1 σ1 /λ1 + x2 σ2 /λ2 )[ê3 × r], bZ = gµB B · σ/2, and bZ⊥ = −gµB (B1 x1 /λ1 + B2 x2 /λ2 )σ3 /4. When we apply a Zeeman field, we can treat each induced degeneracy as a two-level system, assuming the SO matrix elements ǫso are much less than the single spin mean level spacing, ∆ = 2π~2 /mA, with m the conduction band effective mass and A the dot area. The magnitude of the avoided crossings at the Fermi energy is given by ǫso = |hα ↑| Hso |β ↓i|, where ǫα − ǫβ = EZ and ǫα = EF . We want to find the rms value of ǫso . Following Ref. [63], for a closed chaotic dot we may write Λ2 ≡ (ǫso /∆)2 as Λ2 = |(Hso )α↑,β↓ |2 δ(ǫα − ǫβ − EZ )δ(ǫα − EF ), X αβ (2.3) where the overbar indicates ensemble averaging and Oa,b ≡ ha| O |bi. We rewrite Eq. 2.3 as in Ref. [63] using the t-dependent representation of the delta function and interaction picture operators, and, after summing over β, find Λ2 = Z ∞ −∞ dt −iωZ t so (t)Hso † (0) |αi, e hα| HSF SF ∆2π~ (2.4) where ωZ = EZ /~ is the Zeeman frequency, |αi is a typical orbital eigenstate with ǫα ≈ EF , so ≡ h↑| Hso |↓i is the spin-flip part of Hso . We consider CD alone, Hc = (γ/2~3 )(p2 − and HSF 2 p21 )(p1 σ2 − p2 σ1 ), and estimate its contribution to Λ, which we call Λc . We estimate Λc semiclassically using a billiard model for the quantum dot, where the matrix element in Eq. 2.4 is replaced by the corresponding expectation value for a classical particle moving at the Fermi velocity vF starting at a random point in phase space. Semiclassical methods using SO have been rigorously justified [14] and used for studying 2D Chapter 2: Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 0.8 0.6 0.6 c 0.8 a) 0.4 Λ Λ c 44 b) 0.4 [010] [100] [110] [010] 0.2 0 0 20 40 E (µeV) 0.2 [100] 0 0 60 20 0.8 0.6 0.6 c 0.8 c) 0.4 0.2 0 0 40 E (µeV) 60 40 EZ (µeV) 60 Z Λ Λ c Z d) 0.4 0.2 20 40 EZ (µeV) 60 0 0 20 Figure 2.1: Normalized rms avoided crossing Λc due to cubic Dresselhaus spin-orbit coupling as a function of Zeeman energy EZ for four billiards with in-plane magnetic field along the 3 indicated directions, with γ = 8.5 eVÅ [31]. Insets show the billiard shapes with crystal axes. Solid (dashed) lines indicate specular (diffuse) boundary conditions. a) has a mixed phase space with small regions of regular trajectories, b) is a stadium billiard, c) is similar to the dot in Ref. [172], with diffuse boundaries to ensure chaos, and d) is a square with diffuse scattering from the top and bottom and specular scattering from the sides (see text). electron SO effects [167]. We consider B⊥ = 0 for these simulations. Each of (2 − 3) × 105 such trajectories is followed for an equal amount of time, which is generally about 300 bounces total in the forward and backward directions. Increasing the number of trajectories R ′ c (t)Hc† (t′ ) for 100 or bounces does not change the results. We calculate dte−iωZ (t−t ) HSF SF random initial times t′ on each trajectory as a function of ωZ , and their average gives Λ2c when multiplied by the appropriate prefactors. We add a damping function to the integrand that sends it smoothly to zero as t approaches the simulation cutoff. We consider four billiard shapes and, for specificity, choose parameters corresponding to the largest, highest density dot in Ref. [172], with A = 8 µm2 and n = 5.8×1015 m−2 . We use g = 0.44 and m = 0.067me where me is the electron mass. Fig. 2.1(a-c) shows the 3 resulting Λc (EZ ) for three orientations of Bk , with γ = 8.5 eVÅ . For other choices, Λc Chapter 2: Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 45 scales linearly with γ. For our method to be valid, we must have ∆ ≪ EZ ≪ ET , where √ ET = ~vF / A is the Thouless energy. For the case discussed here, ∆ = 0.9 µeV and ET ≈ 80 µeV. We can understand the approximate scale of Λc by using a simpler, unphysical billiard. Consider an Lx × Ly rectangle with specular reflections from the sides and diffuse scattering from the top and bottom. At each collision with a diffuse wall, we choose the tangential momentum from a uniform distribution on [-pF , pF ] [54]. This choice gives a correct weighting for diffuse scattering and maintains detailed balance. In such a billiard, for Bk k x̂, in the limit EZ → 0, Λ2c EZ →0 γ2 = ∆2π~7 Z ∞ −∞ dtpy p2x (t)py p2x (0) , (2.5) and we can break each trajectory into segments between collisions with the top/bottom walls. Along each segment, p2x and py are constant, and the particle takes time t = mLy / |py | to move from one end of the segment to the other, so we can rewrite Eq. 2.5 as * + ∞ X γ 2 mLy 2 2 n 2 Λc = (−1) px,n , |py (0)| px (0) ∆2π~7 EZ →0 n=−∞ (2.6) which we can evaluate explicitly, since the p2x,i are uncorrelated between segments. The particle begins moving in a random direction with P(px,0 ) = π −1 (p2F − p2x,0 )−1/2 , where P is the probability density on [−pF , pF ]. Since the diffuse boundaries in this billiard are the P n top and bottom, P(px,i6=0 ) = 1/2pF . We regularize the infinite sum by ∞ n=−∞ (−1) = 0, and, noting that ET = ~pF /mLy , we find Λ2c = 4γ 2 p6F /(45π 2 ∆ET ~6 ). For the parameters in Fig. 2.1, this gives Λc (EZ → 0) = 0.678. Finite values of EZ are not amenable to such simple treatment, but simulations of this billiard appear in Fig. 2.1d, where the results for Bk k x̂, shown by the solid trace, approach the analytic prediction for EZ → 0. Avoided crossings have not yet been directly measured in chaotic dots, but our calculations can be related to experiments measuring the conductance g through a quantum Chapter 2: Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 46 dot by Zumbühl et al. [171, 172]. To make this comparison, we need a connection between avoided crossings in a closed dot and properties of the dot with leads attached. Cremers et al., using random matrix theory (RMT), worked out a similar connection for dots with only LD and Rashba SO [39]. We point out that CD can be added easily into the predictions of Cremers et al. without changing their expressions for hgi and var g by reinterpreting one of their RMT energy scales to include both linear and cubic SO terms. We now elaborate. In Ref. [39], the chaotic quantum dot is connected to two ideal leads with N ≫ 1 open channels, giving a scattering matrix from the circular orthogonal ensemble. They treat the magnetic field and SO with a stub model [20] in which the stub has the M × M ′ , given by association to H′ in Eq. 2.2 (without CD), as perturbation Hamiltonian, HRMT ′ HRMT = ∆h iA0 (x11 + a⊥ σ3 ) + iak (A1 σ1 + A2 σ2 ) 2π i −b · σ + b⊥ Bh σ3 ) , (2.7) D E where Ai , i = 0, 1, 2, are real antisymmetric matrices with tr Ai ATj = δij M 2 , Bh is a real symmetric matrix with tr Bh2 = M 2 , M ≫ 1 is the number of channels in the stub, and x, a⊥ , ak , b, and b⊥ are dimensionless parameters, with x corresponding to B⊥ , b to the Zeeman field, and a⊥ , ak , and b⊥ to the similarly named terms in Eq. 2.2 (without CD). Dephasing is included by setting Neff = N + 2π~/τφ ∆, with τφ the dephasing time. Expressions are then obtained for hgi and var g as functions of x, a⊥ , ak , b, b⊥ , and Neff to leading order in 1/Neff [39]. Zumbühl et al. use these results to fit their data. Without CD, the correspondence between Eqs. 2.7 and 2.2 gives the following Chapter 2: Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 47 mapping from physical parameters to RMT parameters: x2 = πκET /∆(4πΦ/Φ0 )2 b = πEZ /∆ a2⊥ = πκET /∆(A/λ2SO )2 (2.8) a2k = a2⊥ κ′ [(L1 /λ1 )2 + (L2 /λ2 )2 ] b2⊥ = πκ′′ (EZ )2 /ET ∆(A/λ2SO ) where Φ is the magnetic flux through the quantum dot, Φ0 = h/e is the flux quantum, √ λSO = λ1 λ2 , L1,2 are the linear dimensions of the roughly rectangular dot, oriented along ê1,2 , and κ, κ′ , and κ′′ are geometric factors of order unity [39, 5]. We add CD to this theory by noting that, as a random matrix, the CD in Eq. 2.2 has the same symmetry as the ak term in Eq. 2.7, i.e., it contains only σ1 and σ2 Pauli matrices. By making the simplest ′ assumption of no correlation between the cubic and linear terms, we include CD in HRMT by setting a2k = a2k,l + a2k,c , where ak,l is the Rashba and LD contribution, given by Eq. 2.8, ′ contains the SO part of the Hamiltonian of and ak,c is the CD contribution. Since HRMT the closed quantum dot, we relate ak,c to Λc by finding the rms spin-flip matrix element (with spins quantized along Bk ) due to ak,c , giving ak,c = 2πΛc . Including the CD term ′ in this way lifts the constraint that ak ≪ a⊥ [5], similar to spatially varying SO in HRMT strengths [21]. Zumbühl et al. observe weak anti-localization (WAL) in only one of the GaAs/AlGaAs heterostructure quantum dots they study [171, 172], and that dot gives the best defined values of the RMT parameters; we use it for the discussion of our results. The other dots do not contradict this discussion. The dot that displays WAL has area A = 8 µm2 and electron density n = 5.8 × 1015 m−2 . The dot has N = 2 and Neff = 13.9 [172]. Zumbühl et al. measure var g as a function of Bk with time reversal symmetry Chapter 2: Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 48 broken by a small B⊥ . They fit to the expression of Cremers et al. [39], with ak (and all parameters except κ′′ ) fixed to the value determined from the hgi data. We redo the fits to the var g data, constraining only ak ≥ ak,c , with ak,c from our simulations, and τφ fixed to the value determined from hgi. From Fig. 2.1, a typical value of Λc in all our billiard shapes 3 is 0.4, giving ak,c = 2.5 (8.1) for γ = 8.5 (27.5) eVÅ (recalling that Λc ∝ γ). We find that a value of ak,c ≈ 2.5 is compatible with the experimental data. However, if we require that ak ≥ 8.1, the fits to the data become markedly worse (see Appendix A). Zumbühl et al. also measure hgi as a function of B⊥ , which they use to determine ak , finding ak = 3.1 [171, 172, 170]. Since a2k = a2k,l + a2k,c , we must have ak ≥ ak,c , so 3 3 we conclude that γ = 27.5 eVÅ is inconsistent with these results, while γ = 8.5 eVÅ is consistent with the data. So both hgi and var g data indicate that γ should be closer to 9 3 3 eVÅ than 28 eVÅ 3 . Moreover, even with the smaller value of γ, the CD term gives the dominant contribution to ak . There are only a few other experiments pertaining to the value of γ in GaAs. The best, most direct study is the Raman scattering in a GaAs/AlGaAs quantum well 3 by Richards et al. in which they found γ = 11.0 eVÅ [126]; the same group also found 3 3 γ = 16.5 eVÅ in a different sample [77]. A recent experimental value of γ = 28 eVÅ from transport measurements [106] is less direct, includes the CD only as a density-dependent renormalization of the LD, and assumes the Rashba coupling is independent of gate voltage. Theoretical work has indicated that γ is smaller in AlGaAs/GaAs heterostructures and superlattices than it is in bulk GaAs [132, 133, 99], so it is possible that experiments are 3 not probing the bulk Dresselhaus coupling, though Ref. [31] predicts γ = 8.5 eVÅ in bulk GaAs. We include in Appendix A a table with experimental and theoretical values of γ in 3 Even using Λc = 0.2, the lowest value in Fig. 2.1, gives ak,c = 4.1 for the larger value of γ, which is still inconsistent with the experiment. Chapter 2: Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 49 GaAs. Strictly speaking, our calculations are not directly applicable to the hgi data of Zumbühl et al., as our calculations assume EZ ≫ ∆, and hgi is measured with Bk =0. We do not believe, however, that ak,c changes significantly as Bk → 0; similarly, Cremers et al. consider ak to be constant for all Bk [39]. We believe that ak,c (Bk = 0) can be estimated by simply averaging our results from the different field directions in the limit Bk → 0. Our reinterpretation that a2k = a2k,l + a2k,c requires, of course, that ak,l be less than 3.1 in the experiment of Zumbühl et al. This reduction of ak,l can be absorbed into the geometric parameter κ′ (which was set to 1 without fitting in Zumbühl et al.) without affecting any of the physical parameters, τφ , λSO , found by Zumbühl et al.. Reducing κ′ is reasonable, as Ref. [39] predicts κ′ = 1/3 for a circular diffusive system. Since ∆ ∝ A−1 and ET ∝ A−1/2 n1/2 , we can see that if the thickness of the 2DES does not change with density, ak,l ∝ A7/4 n1/4 , while ak,c ∝ Λc ∝ A3/4 n5/4 . We therefore expect that CD should be relatively more important in small, high density dots, precisely the ones likely to be useful for producing an SFET. In summary, we have used billiard simulations to estimate the effect of the cubic Dresselhaus term on avoided crossings in a closed chaotic quantum dot. These results are related to the conductance through a dot with ideal leads attached. The CD plays a strong and previously ignored role in observed transport properties through quantum dots. Our calculations suggest that 1) the Dresselhaus SO coupling constant, γ, in GaAs/AlGaAs 3 3 heterostructures has a value near 9 eVÅ and not the frequently cited value of 27.5 eVÅ , and 2) even with this smaller value of γ, in the experiments considered the cubic Dresselhaus term provided the bulk of the spin-flip portion of the SO Hamiltonian, which had previously been assigned to the effects of linear SO terms. The value of γ in this technologically important system deserves further study. Chapter 2: Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots 50 Acknowledgments The authors acknowledge a careful reading by Charlie Marcus and helpful conversations with Dominik Zumbühl, Hakan Tureci, Mike Stopa, Emmanuel Rashba, Jeff Miller, Subhaneil Lahiri, Eric Heller, and Hans-Andreas Engel. The work was supported in part by the Fannie and John Hertz Foundation and NSF grants PHY01-17795 and DMR05-41988. Chapter 3 Spin polarized current generation from quantum dots without magnetic fields Jacob J. Krich and Bertrand I. Halperin Physics Department, Harvard University, Cambridge, Massachusetts Abstract An unpolarized charge current passing through a chaotic quantum dot with spinorbit coupling can produce a spin polarized exit current without magnetic fields or ferromagnets. We use random matrix theory to estimate the typical spin polarization as a function of the number of channels in each lead in the limit of large spin-orbit coupling. We find rms spin polarizations up to 45% with one input channel and two output channels. Finite temperature and dephasing both suppress the effect, and we include dephasing effects using a new variation of the third lead model. If there is only one channel in the output lead, no 51 Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 52 spin polarization can be produced, but we show that dephasing lifts this restriction. 3.1 Introduction The generation and control of spin polarized currents, in particular without mag- netic fields or ferromagnets, is a major focus of recent experimental and theoretical work. This includes the spin Hall effect, which produces spin currents transverse to an electric field in a two-dimensional electron system (2DES) with spin-orbit coupling, with spin accumulation at the edges [49]. Similarly, the magnetoelectric effect [93, 47, 7] produces a steady state spin accumulation when an electric field is applied to a 2DES with spin-orbit coupling. The accumulation can be uniform [148, 67] in the case of uniform Rashba spinorbit coupling [26] or at the edges of a channel in either the Rashba model [34, 32, 94] or with spin-orbit coupling induced by lateral confinement [72, 161]. Experiments have observed current induced spin polarization in n-type 3D samples [78] and in 2D hole systems [59, 139] with spin polarization estimated to be up to 10% [139]. Further work suggests a spin polarized current can be produced in quantum wire junctions [80, 61], by a quantum point contact (QPC) [140, 51], in a carbon nanotube [74], in a ballistic ratchet [135], in a torsional oscillator [87], in vertical transport through a quantum well [100], or in disordered mesoscopic systems [46]. Here we show that generating a polarized current from an unpolarized current is a generic property of scattering through a mesoscopic system with spin-orbit coupling. We propose using many-electron quantum dots (outside the Coulomb blockade regime) with spin-orbit coupling to produce partially spin polarized currents without magnetic fields or ferromagnets. Due to the complicated boundary conditions of the quantum dot, we do not solve for the spin polarization in terms of any particular spin-orbit coupling model, Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 53 geometry, and contact configuration. We estimate the effect for a ballistic system in the limit of strong spin-orbit coupling by performing a random matrix theory (RMT) calculation for the spin polarization, allowing consideration of realistic quantum dot devices robust to details of shape and contact placement. Finely tuned systems should be able to exceed these polarizations, but these results provide a useful benchmark for whether a particular tuned system is better than a generic chaotic one. We use a density matrix formalism throughout, which allows us to develop straightforwardly a spin-conserving dephasing probe, using a new variant of the third lead technique for accounting for dephasing. Dephasing and finite temperature both reduce the expected polarization. Without dephasing, we find that if there is only one outgoing channel then no spin polarization is possible, which was first shown simultaneously by Zhai and Xu [169], and Kiselev and Kim [81]. Interestingly, with dephasing, spin polarization can be produced with only one outgoing channel. The case of polarized input currents will be discussed elsewhere (see chapter 4). Analogous calculations have been performed by Bardarson, Adagideli, and Jacquod in a four-terminal geometry, to study the transverse spin current produced by an applied charge current [10]. 3.2 Setup and symmetry restrictions We consider non-interacting electrons in a quantum dot with two attached leads connected to large reservoirs. For any electron current entering from the leads, we can describe the output state in the leads in terms of the S-matrix of the dot, including any tunnel barriers between the leads and the dot. We assume negligible spin-orbit coupling in the leads and consider the lead on the left (right) to have N (M ) spin-degenerate channels at least partially open at the Fermi energy, and let K = N + M . As usual, the channel Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 54 Figure 3.1: A special model quantum dot with N = 1 channels in the left lead and M = 2 channels in the right lead. A skew scatterer sends all ẑ spins to the top channel and all −ẑ spins to the bottom channel. The shaded area in the bottom channel has Rashba spinorbit coupling of precisely the strength to rotate a down spin at the Fermi energy to an up spin, thus producing a perfectly spin polarized exit current from any input current, while respecting time reversal symmetry. wavefunctions are normalized so all channels have the same flux. The S-matrix S is a 2K×2K unitary matrix of complex numbers. For spin 1/2 particles with spin-orbit coupling, however, it is convenient to consider S to be a K × K matrix of 2× 2 matrices. We represent P these 2 × 2 matrices using quaternions, where a quaternion q = q (0) 112 + i 3µ=1 q (µ) σµ , where σµ are the Pauli matrices and q (µ) ∈ C. We give a brief introduction to quaternions in Appendix B. The quaternion representation is convenient, as the time reversal operation for a scattering matrix can simply be written as S → S R , where R gives the quaternion dual (which takes the transpose and sends q (1,2,3) → −q (1,2,3) , see Appendix B) [105]. The S-matrix of a system with time reversal symmetry (TRS) is self-dual. If win (wout ) is the K × K quaternion density matrix of the incoming (outgoing) current, wout = Swin S † . The density matrix describing the unpolarized incoherent combination of all N incoming channels is win = 1 11N 2N 0M (3.1) That is, win = PL /2N where PL is the projection onto the channels of the left lead. We choose trwin = 1/2, due to the quaternion trace convention (see Appendix B), so win represents one incident electron. Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 55 The Landauer-Büttiker formula gives the conductance in terms of the S-matrix ′ [25]. We write the K × K quaternion S-matrix as rt rt ′ with r (r ′ ) being the N × N (M × M ) reflection matrix and t (t′ ) the M × N (N × M ) transmission matrix. Then we write the Landauer-Büttiker formula in units of 2e2 /h as G = tr(tt† ), (3.2) = tr(PR SPL S † ), = 2N tr(PR Swin S † ), = 2N tr(PR wout ), where PR is the projection onto the channels of the right lead. Since win is normalized to represent one input particle entering the system, g = 2tr(PR wout ) is the probability for that particle to exit through the right lead. The conductance is N times this probability, so we call g the conductance per channel in the left lead. Similarly, we define a vector spin conductance [169] (i.e., exit spin current divided ~ s in units of e/2π as by voltage) G ~ s = tr(~σ tt† ), G (3.3) = 2N tr(~σ PR wout ). Then g~s = 2tr(~σ PR wout ) (3.4) is the spin conductance per channel in the left lead. Hence, gµs is the µ-component of the spin polarization of the exit current times the probability of exiting into the right lead. Thus, the spin polarization of the current in the right lead is p~ = g~s /g, with |p| ≤ 1. We can, of course, construct g, g~s , and p~ using only the S-matrix and not the density matrices win and wout . The density matrix approach, however, gives the flexibility Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 56 to consider arbitrarily correlated states of incoming current and also to look for arbitrary correlations in the outgoing current (see chapter 4). We will also use it to straightforwardly derive a method of accounting for non-magnetic dephasing in a device with spin-orbit coupling. To complete the translation to the standard notation of conductances, we consider sending up- or down-polarized electrons into a sample and collecting either up- or downpolarized electrons, giving a conductance matrix [107] G↑↑ G↑↓ G= G↓↑ G↓↓ (3.5) with the total charge conductance being G = G↑↑ +G↑↓ +G↓↑ +G↓↓ . Gσ,σ′ is the conductance for an input current of spin σ ′ and an exit current of spin σ, for σ, σ ′ =↑, ↓. We translate the quaternion representation into the standard notation by noting that the up-polarized incoherent input current has input density matrix w↑in = 1+σ3 2N PL . The output density matrix 3 † is w↑out = Sw↑in S † and the portion representing the output in the right lead is t 1+σ 2N t . The Landauer-Büttiker formula gives, in units of 2e2 /h, out 3 G↑↑ =N tr(PR 1+σ 2 w↑ ), 3 1+σ3 † =tr( 1+σ 2 t 2 t ). (3.6) 3 1+σ3 † G↓↑ =tr( 1−σ 2 t 2 t ), (3.7) 3 1−σ3 † G↑↓ =tr( 1+σ 2 t 2 t ), (3.8) 3 1−σ3 † G↓↓ =tr( 1−σ 2 t 2 t ), (3.9) Similarly, from which we see that G = tr(tt† ), which is the usual Landauer-Büttiker formula [25]. Though there are several proposed spin-orbit coupled systems that demonstrate spin polarization from unpolarized input, in many cases the effect is subtle [140, 51, 73, Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 57 135, 115]. Here we give a simple, idealized thought experiment to show that time reversal symmetry does not forbid generating a spin polarized current from an unpolarized input current. Consider a system with N = 1 and M = 2, as illustrated in Fig. 3.1. All input electrons are incident on a perfect skew scatterer [49], which sends spins quantized in the +z direction into exit channel 1 and spins quantized in the −z direction into exit channel 2. Exit channel 2 has a region with Rashba spin-orbit coupling [26] which is precisely of the strength and length necessary to rotate −z spins to +z. Thus, all spins incident from the left lead exit with their spins up, and the system respects TRS, since skew scattering and Rashba spin-orbit interaction are each time reversal symmetric. We illustrate by constructing S explicitly. We can express the scattering matrix for this thought experiment (up to an overall phase) in the 6 × 6 and 3 × 3 representations as S= ≡ 1 2 0 0 1 0 0 0 0 0 0 0 −1 0 0 1 0 0 0 0 0 0 0 0 0 0 eiθ 0 1 0 0 0 0 0 −eiθ 0 0 0 0 1−σz −σx −iσy 1+σz 0 eiθ (σx −iσy ) , σx +iσy −eiθ (σx −iσy ) 0 (3.10) (3.11) where θ ∈ [0, 2π) and r and t have been determined by the above description, while the rest of the matrix is given by TRS and unitarity. The unpolarized input quaternion density 1/2 in matrix is w = , giving 0 0 wout 0 0 0 1 = Swin S † = 0 1 + σz , 0 4 0 0 1 + σz (3.12) so Eq. 3.4 gives g~s = p~ = ẑ, as stated above. We now prove that having at least two channels in the outgoing lead is essential. That is, for a dot with TRS and K channels in attached leads, if an unpolarized equally Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 58 weighted incoherent current is sent into N = K−1 of the channels, then the spin polarization in the remaining channel must be zero. This result has been shown before [169, 81], but the quaternion formalism with density matrices makes it particularly transparent, so we include the proof here. We start with win = 1 11N 2N 11K − PK , = 2N 0 (3.13) where PK is the projection onto the K th channel. The quaternion scattering matrix satisfies S = S R since TRS is unbroken, and wout = 11K − SPK S † SS † − SPK S † = . 2N 2N (3.14) Note that S = S R implies both S † = S ∗ and Sii ∈ C for i = 1 . . . K (see Eq. B.3). out ](µ) , where [q](µ) is the µUsing Eq. 3.4, the spin conductance is gµs = 2i[wKK out has no quaternion part, then g s = 0. component of the quaternion q. In particular, if wKK µ We have out wKK = ∗ 1 − SKK SKK , 2N (3.15) ∗ out ∈ R and g~s = 0. This proof applies with channels that are and SKK SKK is real, so wKK fully open or have tunnel barriers, as it requires only that the S-matrix satisfy TRS and unitarity, which are unchanged by tunnel barriers. We note further that if K > 2 then 1) the reflected current in any of the K − 1 input channels can be spin polarized, and 2) if the input current goes through less than K − 1 channels, then the remaining channels can have a spin polarization, as shown in the example of Fig. 3.1. Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 59 Figure 3.2: Generic device with one input channel. Unpolarized current consisting of equal parts spin-up and -down components is incident on a chaotic quantum dot from a single channel on the left. The irregularly shaped region in the middle is a quantum dot in the strong spin-orbit regime. The spin-up incident current has a probability to exit into each of the three channels of the right lead with a different spin direction in each channel, indicated by the direction of the dark arrows. Similarly for the spin-down incident current, indicated by the light arrows. The exit channels are shown spatially separated, for convenience. There is also a probability to scatter back into the left lead (not shown). The total spin current in the exit lead is given by summing the probabilities and directions of the six spin polarizations shown, which is the meaning of Eq. 3.4 and is indicated by the arrow on the right. 3.3 Random matrix theory The device illustrated in Fig. 3.1 cannot easily be made, but realistic devices with spin-orbit coupling will show a similar (albeit weaker) spin current generation. An intuitive picture for the generation of spin currents from an applied voltage in a realistic device is to consider the case with N = 1 and M > 1. A current of spin-up electrons incident from the left has some probability to exit in each of the M channels of the right lead, each associated with a spin polarization direction; in the strong spin-orbit limit, these spin polarizations can have arbitrary directions. The same is true for a current of spin-down electrons incident from the left. As illustrated in Fig. 3.2, a charge current entering from the left is a combination of these spin-up and spin-down currents. The spin polarization of the current in the right lead is given by the sum of the spin polarizations of each of the 2M currents in the channels of the right lead, weighted by the probability of the particle to Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 60 enter that channel. Despite the presence of time reversal symmetry, there is no requirement that this polarization sum to zero, as seen explicitly in the example of Figure 3.1. In the chaotic strong spin-orbit limit, these 2M vectors are (almost) uncorrelated, and their sum will generically be nonzero. As M or N are increased, however, there are more independent vectors contributing to the sum, which generally brings the sum closer to zero. We formalize this intuition using random matrix theory. We estimate the expected spin polarization in realistic situations by using random scattering matrix theory. We assume that the mean dwell time τd = 2π~/K∆ of particles in the dot is much greater than the Thouless time τTh = Ld /vF , where K is the number of fully open orbital channels attached to the dot, ∆ = 2π~2 /mA is the mean orbital level spacing, m is the effective mass, A is the area of the dot, vF is the Fermi velocity, and Ld is a typical length scale of the quantum dot. We further assume the strong spin-orbit limit, where the spin-orbit time τso is much less than τd . Since τd = mA/~K, for a sufficiently large A, even a material with “weak” spin-orbit coupling will be in the strong spin-orbit limit. The crossover from weak to strong spin-orbit coupling in chaotic quantum dots has been studied in the K ≫ 1 limit in the context of adiabatic spin pumping [138]. For dots with strong spin-orbit coupling, we assume that the S-matrix is chosen from the uniform distribution of unitary matrices subject to TRS, called the circular symplectic ensemble (CSE) [105, 12]. We find the root mean square (rms) magnitude of the spin conductance on averaging over the CSE, which gives the typical spin conductance magnitude to be expected from chaotic devices. Such an averaging can be realized in practice by small alterations of the dot shape [171, 172]. By symmetry, gµs = 0 for µ = 1, 2, 3. Using Eq. 3.4, we evaluate D E (gs )2 = 4 tr(σµ PR Swin S † )tr(σµ PR Swin S † ) , (3.16) Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 61 where we sum over µ. We use the technique for averaging over the CSE described by Brouwer and Beenakker in Section V of Ref. [18]. We need just two generic averages, which we will use repeatedly. The first is of the form hF1 (S)i = tr(ASBS † ) , where A and B are con- stant K × K quaternion matrices and the average is taken over S chosen from the CSE of K × K quaternion self-dual matrices. Then[18] 1 [2tr(A)tr(B) − tr(AB R )]. (3.17) 2K − 1 The second average we need is hF2 (S)i = tr(ASBS † )tr(CSDS † ) where A, B, C, D are hF1 i = constant K × K quaternion matrices and AB = AD = CB = CD = 0. We find[18] hF2 i = 1 (K − 1)[4trAtrBtrCtrD + tr(AC)tr(BD)] Λ − [trAtrCtr(BD) + tr(AC)trBtrD] , (3.18) where Λ = K(2K − 1)(2K − 3). Using Eq. 3.18, we find (gs )2 = 3 M (M − 1) , NΛ (3.19) where we used tr(σµ PR ) = 0, tr(PR2 ) = M , trwin = 1/2, and tr (win )2 = 1/4N . Note that when M = 1, (gs )2 = 0, consistent with the general symmetry. If we are interested in the mean square polarization of the exit current, p2 = s 2 −2 , we can approximate it by (gs )2 / hgi2 . This approximate form is useful for (g ) g analytical progress and will be compared to numerical results. Using Eq. 3.17, hgi = 2M , 2K − 1 (3.20) which, combined with Equation 3.19, gives 2 p ≈ 3(M − 1)(2K − 1) . 4M N K(2K − 3) (3.21) Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 62 1 s 2 1/2 <g> 2 1/2 <(g ) > <p > 0.4 0.5 0.2 N=1 N=2 N=3 N=4 0 1 3 5 M 7 0 1 3 5 M 7 1 3 5 M 7 Figure 3.3: Numerical (symbols) and analytical (lines) results for normalized mean conductance hgi, rms spin conductance gs , and rms spin polarization p of current exiting a chaotic quantum dot with N (M ) channels in the entrance (exit) lead. An average over 60000 S-matrices from the CSE was performed for each data point. The lines are from Eqs. 3.19–3.21. We study the approximation (gs )2 g−2 ≈ (gs )2 / hgi2 numerically. We choose a 2K × 2K complex Hermitian matrix from the Gaussian unitary ensemble [105] and find the unitary matrix U which diagonalizes it. We multiply columns of U by random phases, map U into a K × K matrix of quaternions, and construct unitary self-dual S by setting S = U U R , giving S chosen from the CSE [57]. Figure 3.3 shows the numerical and analytical results, which agree quantitatively for g2 and (gs )2 and qualitatively for p2 . The largest percentage disagreement for 2 p is 7%. Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 1 63 s 2 1/2 <g> 2 1/2 <(g ) > <p > 0.4 0.75 0.2 Nφ=1 Nφ=2 N =3 φ 0.5 1 3 5 M 7 0 1 3 5 M 7 1 3 5 M 7 Figure 3.4: Model with spin-conserving dephasing lead. Numerical (symbols) and analytical (lines) results for normalized mean conductance hgi, rms spin conductance gs , and rms spin polarization p of current exiting a chaotic quantum dot with N = 1 channel in the entrance lead. M and Nφ are the numbers of channels in the exit and dephasing leads, respectively. An average over 60000 S-matrices from the CSE was performed for each data point. The lines are from Eqs. 3.28 and 3.29. Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 64 3.4 Dephasing We add dephasing to this setup using the dephasing voltage probe technique [24, 17, 9]. We add a fictitious voltage probe drawing no current with Nφ = 2π~/∆τφ fully open orbital channels, so the time for escape into the dephasing lead is the dephasing time τφ . If the dephasing process is spin independent, then it is appropriate to conserve the spin of the reinjected electrons, and we extend the third lead model to allow this. If, however, dephasing processes relax the spin, then it is appropriate for the dephasing lead to reinject unpolarized current, preserving only electron number. We consider both of these models, with emphasis on the first, as it is new in this work. In our formulation, we explicitly model reinjection of electrons from the fictitious voltage lead by modifying win to include incoherent reinjection from the dephasing lead. In either model of dephasing, the reinjection matches the total charge current absorbed by the dephasing lead, but distributes the charge current evenly between the channels and removes the correlations. In the spin-conserving case, the reinjection also preserves the spin current. Consider ηµ = tr(σµ Pφ Sw0in S † ), where µ = 0, 1, 2, 3, σ0 = 112 , Pφ is the projection operator onto the dephasing lead’s channels, and w0in is the input density matrix. Then 2η0 is the probability for a particle to enter the dephasing lead, and 2~ η is the spin conductance into the dephasing lead, which is proportional to the spin current into the dephasing lead. We reinject from the dephasing lead with 0K w1φ = = c1µ σµ Pφ , c1µ σµ 11Nφ (3.22) where we sum over repeated index µ. To preserve both spin and charge currents, we set c1µ = ηµ /Nφ . Some of this reinjected current reflects back into the dephasing lead, so it must be reinjected again. We define a 4 × 4 complex matrix Ξµν = tr(σν Pφ Sσµ Pφ S † ), which gives the charge/spin current in the dephasing lead due to this reinjection. Defining Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 65 win = w0in + wφin , this procedure gives wφin = ∞ X wnφ n=1 = Pφ σµ tr(σν Pφ Sw0in S † ) ∞ X (Ξn−1 )µν n=1 Nφn = Pφ σµ tr(σν Pφ Sw0in S † )(Nφ δµν − Ξµν )−1 , (3.23) where we sum over repeated indices µ, ν = 0, 1, 2, 3. This result holds for any input current, not just the unpolarized incoherent w0in discussed here. We approximate wφin by replacing Ξµν with its average in Eq. 3.23, similar to Eq. 3.21. Using Eq. 3.17, hΞµν i = Nφ [2(Nφ − 1)δµ0 δν0 + δµν ], 2Kφ − 1 (3.24) where Kφ = N + M + Nφ . We further replace tr(σν Pφ Sw0in S † ) by its average, D E δν0 Nφ , tr(σν Pφ Sw0in S † ) = 2Kφ − 1 (3.25) which gives win ≈ w0in + Pφ /2K. (3.26) Note that Eq. 3.26 satisfies unitarity only on average; the total probability of exiting either through the right or left lead equals 1 only on average. In the case of a spin-relaxing dephasing lead, Eq. 3.23 becomes wφin = Pφ tr(Pφ Sw0in S † ) . Nφ − tr(Pφ SPφ S † ) (3.27) Interestingly, approximating by taking averages separately of the numerator and denominator gives the same answer as in Eq. 3.26. As is clear from Eq. 3.26, the approximations of Eqs. 3.24 and 3.25 result in unpolarized reinjection from the the dephasing lead. Averaging these terms separately Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 66 removes the coherence between the injected spin and the reinjected spin, so they separately average to spinless quantities. As a result, this approximation reproduces the effect of a spin-relaxing dephasing lead. In fact, we find that Eq. 3.26 matches the exact numerical results for a spin-relaxing lead better than it matches the exact results for a spin-conserving lead, which is consistent with this understanding of the approximations in Eqs. 3.24 and 3.25. Using Eqs. 3.26 and 3.18, we find 3M (gs )2 ≈ N Λφ M − 1 + Nφ M 2 + N (M − 1) K2 , (3.28) where Λφ = Kφ (2Kφ − 1)(2Kφ − 3). Note that Eq. 3.28 predicts an output spin polarization in all cases with Nφ > 0, including when there is only one outgoing channel, M = 1. Numerical results for the spinconserving dephasing lead, using Eq. 3.23, are shown in Fig. 3.4, and they show an rms spin conductance in agreement with Eq. 3.28 except for Nφ = M = 1. In the special case of Nφ = M = 1, an exact treatment shows that g~s = 0, contrary to Eq. 3.28, even with arbitrary tunnel barriers between the leads and the sample, as shown in Appendix C. Appendix D contains explicit examples of dephasing-induced spin polarization in the case M = N = 1, in both the spin-conserving and spin-relaxing cases. Numerical results for the spin-relaxing dephasing lead are not shown in Fig. 3.4, but fall very close to the corresponding lines for the analytic results, even for the case Nφ = M = 1. (A finite spin-polarization is allowed in this case, with a dephasing lead that relaxes spin.) It may seem surprising that dephasing can produce spin polarization in the case M = 1, where none would be produced in the absence of dephasing. Clearly, the presence of the dephasing lead violates the conditions of the proof in Section 3.2 that M = 1 implies g~s = 0, but it is not obvious that dephasing will nonetheless produce polarization. Indeed, a single Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 67 channel spin-conserving dephasing lead does not, as shown in Appendix C. Nevertheless, we may consider whether the production of spin polarization with M = 1 is an artifact of the particular third lead dephasing models studied here or a real consequence of inelastic dephasing processes. To further explore this question, we have considered two further variants of the third lead dephasing model. One could modify the spin-conserving dephasing model to have Nφ dephasing leads each with one channel, each separately reinjecting the same charge/spin that it absorbs. In this model, too, we find that a nonzero g~s can be produced for Nφ > 1 (results not shown). Brouwer and Beenakker modified the third lead dephasing model to make dephasing uniform in phase space by placing a tunnel barrier with transparency Γ between the dephasing lead and the dot, with Γ → 0 and Nφ → ∞ while maintaining ΓNφ = 2π~/∆τφ [19]. The S-matrix is then not drawn from the CSE, and simple analytical results in the spin-orbit coupled system are challenging. We have studied the spin-conserving variant of this model numerically and find that for fixed τφ , it gives qualitatively similar results to the simpler model described above; in particular it also gives a nonzero spin current when M = 1 (results not shown). All variants of third lead dephasing models considered here show dephasing-induced spin currents with M = 1. Without a microscopic model of dephasing, however, we cannot rule out the possibility that this effect is an artifact of third lead dephasing models in general. This dephasing-induced polarization is worthy of further study and will be important to consider when designing devices based on the lack of spin polarization with M = 1 [168]. Finally, returning to the single dephasing lead with Γ = 1, we estimate p2 ≈ (gs )2 / hgi2 , as in Sec. 3.3, where we modify hgi to include the dephasing lead. Using Eqs. Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 68 3.17 and 3.26, we have hgi ≈ 2M Kφ . K(2Kφ − 1) (3.29) We estimate p2 , using Eqs. 3.28 and 3.29. Comparison of these approximations to numerical evaluations is shown in Fig. 3.4. Again we find that the numerical and analytical results agree qualitatively, except when Nφ = M = 1. 3.5 Finite bias and temperature If the source-drain bias V or the temperature T is large enough, the polarization will be further suppressed by electrons of different energy feeling uncorrelated scattering matrices. This effectively increases the number of orbital channels, which decreases the residual polarization. First consider infinitesimal bias at temperature T . Adapting Datta [40], we take win (ǫ) = − ∂f 1 11N ∂ǫ 2N 0M (3.30) where f (ǫ) is the Fermi distribution. If the scattering matrix for particles of energy ǫ is S(ǫ), then wout (ǫ) = S(ǫ)win (ǫ)S † (ǫ). We approximate S(ǫ) as correlated only within energy intervals of scale given by the level broadening due to escape into the leads ∆′ = ∆K/2 (see Ref. [68] for an equivalent treatment). That is, we take D E D E † † Sab (ǫ)Scd (ǫ′ ) = ∆′ δ(ǫ − ǫ′ ) Sab (ǫ)Scd (ǫ) , (3.31) and D E † † Sab (ǫ)Scd (ǫ′ )Sef (ǫ)Sgh (ǫ′ ) D ED E † † = Sab (ǫ)Sef (ǫ) Scd(ǫ′ )Sgh (ǫ′ ) D E † † + ∆′ δ(ǫ − ǫ′ ) Sab (ǫ)Scd (ǫ)Sef (ǫ)Sgh (ǫ) , (3.32) Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 69 which are valid only for T ≫ ∆′ , which is often true for chaotic quantum dots. For T ≈ ∆′ , S(ǫ)S † (ǫ) can be calculated using the random Hamiltonian method [152]. We need an average over a new function, h(ǫ, ǫ′ ) = f ′ (ǫ)f ′ (ǫ′ )tr[AS(ǫ)BS † (ǫ)]tr[CS(ǫ′ )DS † (ǫ′ )], where AB = AD = CB = CD = 0 and f ′ = ∂f /∂ǫ. We evaluate the average of h with the K × K quaternion matrix S(ǫ) chosen from the CSE along with Eq. 3.32, giving, Z dǫ dǫ′ h(ǫ, ǫ′ ) = 4 trAtrBtrCtrD (2K − 1)2 = 4 trAtrBtrCtrD (2K − 1)2 ∆′ + Λ + Z dǫ f ′ (ǫ)2 (K − 1)[4trAtrBtrCtrD + trACtrBD] − trAtrCtrBD − trACtrBtrD ∆′ (K − 1)[4trAtrBtrCtrD + trACtrBD] − trAtrCtrBD − trACtrBtrD . 6T Λ (3.33) Using Eq. 3.33 in place of Eq. 3.18, we evaluate (gs )2 as above, which simply multiplies Eq. 3.19 by ∆′ 6T . Also, hgi is unaffected by temperature, so Eq. 3.21 is also multiplied by ∆′ /6T . When dephasing and temperature are both included, the level broadening has a component due to dephasing (effectively due to escape into the dephasing lead) so ∆′ = ∆(K/2 + Nφ /2) [68]. Eq. 3.28 is then multiplied by ∆′ /6T , and Eq. 3.29 is unchanged. If the temperature is small but the source-drain bias Vsd is large compared to ∆′ , then we can repeat this calculation with win (ǫ) = [Θ(ǫ) − Θ(ǫ − Vsd )] 1 11N 2N Vsd 0M , (3.34) where Θ is the unit step function. Using the equivalent of Eq. 3.33, this multiplies Eqs. 3.19, 3.21, and 3.28 by ∆′ /Vsd . Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 70 3.6 Discussion This spin polarization should be able to be produced and detected experimentally. Even quantum dots in n-type GaAs/AlGaAs heterostructures have been observed to have sufficiently strong spin-orbit coupling to approach the RMT symplectic limit [171, 172]. If the spin-orbit coupling is not strong enough for the S-matrices of the dot to be drawn from the CSE, the spin polarization predicted here will be reduced but should still be present. In a given material with fixed spin-orbit coupling strength, a sufficiently large quantum dot will be well described by the CSE, with a possible increase in dephasing rate as the dot size increases. Section 3.5 shows that the rms spin polarization goes down as p (∆′ /Vsd ) ∝ (AVsd )−1/2 , so as the dot is made larger to enter the strong spin-orbit limit, the bias range where the results are observable decreases. Thus, the effects predicted in this paper are most likely to be observable in a material with inherently strong spin-orbit coupling, such as p-type III/V heterostructures. If a measurement technique or application is only sensitive to spin polarization along a particular axis, then the rms predictions for the µ-component of the polarization and √ spin conductance are only 3 times smaller than the results stated above, since (gµs )2 = s 2 (g ) /3 and p2µ = p2 /3. We have shown that quantum dots with spin-orbit coupling can generate spin polarized currents without magnetic fields or ferromagnets, except in the case of only one outgoing channel, when such a device can only produce a spin current if dephasing is present. These mesoscopic fluctuations can be large enough to give appreciable spin currents in devices with a small number of propagating channels. Even if the spin-orbit coupling is weak, a sufficiently large device will show these effects. Dephasing generally decreases the spin polarization, except in the case of one outgoing channel, where spin polarization cannot Chapter 3: Spin polarized current generation from quantum dots without magnetic fields 71 be produced in the absence of dephasing. Acknowledgments We acknowledge helpful discussions with Caio Lewenkopf, Emmanuel Rashba, Ilya Finkler, Philippe Jacquod, and particularly Ari Turner, who suggested the method of choosing matrices from the CSE. We note the use of the Quaternion Toolbox for MATLAB, created by S. J. Sangwine and N. Le Bihan. This work was supported in part by the Fannie and John Hertz Foundation and NSF grants PHY-0646094 and DMR-0541988. Chapter 4 Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling Jacob J. Krich Physics Department, Harvard University, Cambridge, Massachusetts Abstract As devices to control spin currents using the spin-orbit interaction are proposed and implemented, it is important to understand the fluctuations that spin-orbit coupling can impose on transmission through a quantum dot. Using random matrix theory, we estimate the typical scale of transmitted charge and spin currents when a spin current is injected into a chaotic quantum dot with strong spin-orbit coupling. These results have implications for the functioning of the spin transistor proposed by Schliemann, Egues, and Loss. We use a density matrix formalism appropriate for treating arbitrary input currents and indicate its connections to the widely used spin-conductance picture. We further consider the case of currents entangled between two leads, finding larger fluctuations. 72 Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 4.1 73 Introduction There has been much recent progress in the creation and control of spin currents. There have been demonstrations and proposals for producing spin-polarized currents both with [79, 141, 51, 10, 89] and without time reversal symmetry (TRS) [96, 58]. Recent progress in measuring and controlling the spin-orbit coupling in semiconductor heterostructures [86, 92, 144] promises to enable a range of spintronic applications relying on the spin-orbit interaction. As such devices are considered and developed, it is important to understand the role of coherent mesoscopic fluctuations in these systems. In this paper, we consider the effects of injecting either a spin-polarized current or a pure spin current into a two-dimensional ballistic region with strong spin-orbit coupling and consider the scale of the fluctuations of charge and spin currents transmitted through such a device. For example, these effects could be important for the Schliemann-Egues-Loss spin field effect transistor (SFET) proposal [136]. In such a SFET, spin-polarized electrons are injected into a region (e.g., a diffusive wire or a quantum dot) with spin-orbit coupling. In the “on” state of the device, the Rashba [26] and k-linear Dresselhaus [44, 157] spinorbit couplings are tuned to be equal, and the spin polarization does not decay as the electrons cross the region, but instead undergoes a controlled rotation [136]. In the “off” state, the Rashba and k-linear Dresselhaus strengths are tuned to be different, and the spin polarization is lost while traversing the region due to the random spin rotations experienced by electrons traversing different trajectories through the dot. Ideally, the on state has a fully spin-polarized current exiting the device and the off state has no spin polarization in the exit current. For coherent 2D quantum systems, however, the decay of the spin current in the off state relies on having a sufficiently large number of channels to average together. In the 1D limit, with two ideal one-dimensional wires, each having only one propagating mode, Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 74 a fully spin-polarized current injected into the first wire is a pure state, so the transmitted current must have a spin pointing in some direction; this fact implies that no reduction in spin-polarization is possible in the coherent 1D limit. Other limitations to the SFET proposal have been simulated by Shafir et al. [137]. In this paper we discuss the general problem of coherent propagation of currents through quantum dots, focusing on the relationship of incident to exit spin-polarization of the currents. For the case of 2D ballistic chaotic scattering regions with strong spin-orbit interaction, we use random matrix theory to give analytic results for the expected values of spin-polarization in the exit currents. Once we can describe the ingoing current in terms of a density matrix, all of the conclusions will follow. Thus, the problem is generally broken into two parts: first, find the relevant input density matrix for the system of interest; second, propagate that density matrix to find the output currents and polarizations. We choose the density matrix formalism to describe the input currents to the quantum dots, as it is flexible enough to describe any current in the noninteracting system. As an important example, we describe how to construct the density matrices representing currents produced from potentials applied to (possibly spin-split) reservoirs. We go beyond this model and also consider injection of spin currents entangled between the two leads, finding larger fluctuations in this case. Similar work in a three-terminal geometry was considered in Ref. [3]. The case of unpolarized input currents was considered in chapter 3. 4.2 Setup We consider a quantum dot attached to two ideal leads through quantum point contacts (QPCs). There are N , M open spin-degenerate channels in the left, right QPCs, respectively, and we let K = N + M . We take a basis for the propagating states in the ideal Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 75 leads normalized to unit flux in each channel, as usual. We consider noninteracting spin 1/2 particles which are coherently scattered by the quantum dot, which we describe using an S-matrix. Given a density matrix w representing the current into the dot from the K channels, the output current is described by density matrix wout = SwS † . With K open channels, the S-matrix S can be represented by a 2K × 2K matrix of complex numbers. In systems with time reversal symmetry, it is convenient to consider S to be a K ×K matrix of 2×2 matrices. Any 2×2 matrix can be written as a linear combination of the four Pauli matrices, but it is convenient to consider the basis {σ 0 , iσ 1 , iσ 2 , iσ 3 }, where the σ i are the Pauli spin matrices. In this basis, a 2 × 2 matrix q = q 0 σ 0 + i~ q · ~σ , with q 0 , ~q ∈ C, which is also called a quaternion [105]. Then q is defined to have a complex conjugate q ∗ = q 0∗ σ 0 + i~ q ∗ · ~σ , dual q R = q 0 σ 0 − i~ q · ~σ , and Hermitian conjugate q † = q R∗ . The Hermitian conjugate is the same as the standard Hermitian conjugate of a complex matrix, but the complex conjugate is not the same. For an S-matrix of quaternions, we define complex conjugate (S ∗ )ij = (Sij )∗ , dual (S R )ij = (Sji )R , and Hermitian conjugate S † = S R∗ . This representation is convenient because for time reversal invariant systems, P S = S R . The quaternion representation has the standard convention that tr(S) = i Sii0 , which is half of the trace of the equivalent complex matrix. 4.2.1 Constructing w from chemical potentials Consider for the moment not two leads attached to the dot but K leads, each with one open channel and connected to its own reservoir with adiabatic, reflectionless contacts. Modeling the reservoirs as paramagnetic, each reservoir can be spin-split along its own quantization axis with each spin band separately in equilibrium, having its own chemical potential µνm , where m ∈ {1 . . . K} labels the channel and ν ∈ {0, x, y, z} indicates the charge and spin potentials [151, 145]. There has been some confusion [52] on the consistency of 76 Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling defining this chemical potential, so we give an example. If reservoir m is spin-split along axis x̂, then µ0m is the average chemical potential in the reservoir, 2µxm is the chemical potential difference between spin-up and spin-down electrons quantized along x̂, and µy,z m = 0. In general, if the quantization axis is n̂ and the chemical potential difference along that axis is 2µs , then µi = µs (n̂ · î). Such spin-split chemical potentials can be realized, for example, by optical excitation in heterostructures, in an environment with inelastic relaxation much faster than spin relaxation [151, 98, 79]. We assume the leads have negligible spin-orbit coupling and spin relaxation, so there is a well-defined spin current in the leads. In the absence of inelastic processes, we can consider the current carried by particles with energy ǫ. For simplicity, we assume the number of open channels does not vary over the range of ǫ considered here. Then the particle-currents flowing in from each channel are represented by the quaternion density matrix w̃ nm (ǫ) = δnm [f (ǫ − µ0n ) − ~σ · ~ µsn f ′ (ǫ − µ0n )], (4.1) where f (ǫ) is the Fermi function at temperature T , and we assume that µsn < max(T, ∆), where ∆ is the mean orbital level spacing in the quantum dot without leads attached, and the prime indicates the derivative with respect to ǫ. The charge current in the nth channel of particles with energy ǫ is e jn0 (ǫ) = 2tr{Pn [w̃(ǫ) − w̃out (ǫ)]} , h (4.2) where −e is the electron charge, h is Planck’s constant, and Pn is the projection matrix onto the nth channel (i.e., (Pn )ab = δan δbn ). Similarly, the spin-current in the nth channel is jni (ǫ) = 2tr{Pn σ i [w̃(ǫ) − w̃out (ǫ)]} e . 2π (4.3) Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 77 We choose units in which e = h = 2π, so Eq. 4.3 can describe both charge and spin currents if we let σ 0 be the identity. The currents are the physical objects in the system, and we note that the currents are unaffected by adding any multiple of the identity to w̃(ǫ), since w̃out = S w̃S † and S is unitary. We can thus use the density matrix to represent the currents, but we do not need to maintain trw = 1 or even that w has positive eigenvalues. In the case where there are only two leads, we can subtract f (ǫ − µ02 ) from w, giving 0 0 ′ 0 µs1 f ǫ − µ1 − f ǫ − µ2 − f ǫ − µ1 ~σ · ~ w(ǫ) = 0 ′ 0 ′ 0 µs1 −f ǫ − µ δµ − f ǫ − µ1 ~σ · ~ ≈ −f ′ −f ′ 0 0 ǫ − µ2 ~σ ǫ − µ2 ~σ ·µ ~ s2 ·~ µs2 , (4.4) (4.5) where µ0 = (µ01 + µ02 )/2 and δµ0 = µ01 − µ02 . Note that if δµ0 = 0 then the average chemical potential in both leads is the same, so no net charge flows and w is traceless. If we consider an energy range in which the S-matrix does not vary (i.e., the linear response regime [40], where δµν < {T, ∆}), then we can represent the currents by integrating over energy in the density matrix, giving 0 s ~1 δµ + ~σ · µ w= ~σ · µ ~ s2 (4.6) and jnν = 2tr[Pn σ ν (w − wout )]. (4.7) Spin-polarized injection from ferromagnetic contacts does not immediately map onto the chemical potential formalism. It is clear that if a ferromagnet is in equilibrium with a wire, connected by adiabatic contacts, it will not produce a spin current in the wire, since adiabaticity requires that the lowest energy levels remain filled. For practical injection 78 Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling of spin-polarized currents from a ferromagnet to a normal metal system, a tunnel barrier at the contact is the most common form of non-adiabaticity [122, 96]. We can consider a situation where the ferromagnet injects into a semiconductor, which serves as the reservoir for a wire connected to our quantum dot. If we consider the case where the semiconductor has an energy relaxation time τe much shorter than the spin relaxation time τs , then the spin-polarized current injected from the ferromagnet into the reservoir can relax to two independent distributions with a spin-split chemical potential. This is the same assumption used for optical excitation of spin-split chemical potentials. We can then use the formulation in terms of potentials as described above. The tunnel barrier at the ferromagnet introduces a second complication, as it implies that the ingoing current in the wire contains particles injected directly from the reservoir and also particles reflected from the scattering region and reflected back from the barrier. The input density matrix thus needs to be determined self-consistently, including the effects of both reflections. Such effects can be included systematically, by using the Poisson kernel [12] rather than the circular ensemble described below and also including the TRS-breaking effects of the ferromagnetic scattering. For a sufficiently large reservoir in the semiconductor, this reflection can represent a small perturbation to the input currents, and the procedure described below will be a good approximation. 4.2.2 Connection to spin conductances We can write a generalized Büttiker-type conductance equation [23] jlν = X k,ρ ρ ν Gνρ lk µk − 2Ml µl , (4.8) where Gνρ lk is the conductance from lead k to lead l and spin ρ to ν and 2Ml is the number of modes, including spin, in lead l. The absence of equilibrium charge or spin currents (since Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 79 there is no spin-orbit coupling in the leads) implies X k Gν0 lk − 2Ml = 0. (4.9) Further, the conservation of charge current implies that X G0ν lk = 2Mk δν0 . (4.10) l Specializing to the case of two leads with N and M modes in the left and right leads with potentials µνL , µνR , respectively, we can express Gνρ lk simply in terms of the S-matrix. Setting µνR = 0 and µνL = δνα , Eq. 4.8 gives ν jR = Gνα RL . Eq. 4.6 says that w = σα 11N 0M (4.11) ν = 2tr(σ ν P Sσ α P S † ) = = σ α PL , and by Eq. 4.7 we have jR R L να ν α † Gνα RL . Similarly, GRR = 2tr(σ PR Sσ PR S ). If the system is time reversal invariant, then S = S R , which imposes some relations between the different conductance matrix elements. Since tr(AR ) = tr(A), we have the Onsager-like relations ν ρ ρν Gνρ lk = h h Gkl , (4.12) for k, l = R, L where hν = (1, −1, −1, −1). We thus see that we can express all of the Gνρ ij in terms of traces over appropriate density matrices multiplying S-matrices. We will consider the current in the right lead associated with the input density matrix w, defined as ν jw ≡ 2tr[σ ν PR (SwS † − w)], (4.13) which is proportional to the outgoing current in the right lead after injection represented by w, where the sign is chosen so that outgoing currents to the right are positive. Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 80 4.2.3 Purity of w We will see that the coherence properties of the currents are important, so it is interesting to consider when w represents a pure state. Ordinarily, density matrices are defined (with quaternion trace convention) so 2trρ = 1, and ρ is pure if ρ2 = ρ. In our open system, normalization is a choice, and we set 2trw = t, where t gives the total current incident on the dot. We can also add any multiple of the identity to w without affecting the physical currents. Taking both these factors into account, w represents a pure state only if there is a real number α such that w − α11 2tr(w − α11) 2 = w − α11 . 2tr(w − α11) (4.14) This condition implies that the K × K quaternion matrix w represents a pure state only if 1. w2 = tw, 2. w2 = −t 2K−1 w, or 3. w is invertible and ∃α ∈ R such that w−1 = 4.3 w−[t−2α(K−1)]11 −α[t−α(2K−1)] . Random matrix theory Though for any particular quantum dot it is difficult to determine the full scat- tering matrix exactly, if there is a small number of open channels in the leads connected to the dot, mesoscopic fluctuations should produce an appreciable spin polarization in the exit current. We can understand this by considering that the current from one of the input channels has some probability to exit into each of the M exit channels after undergoing some spin rotation. In the chaotic strong spin-orbit limit, there is no correlation between the entry and transmitted spin polarizations. Though on average the transmitted spin polarization is zero, in any particular case there will still be some residual polarization in some Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 81 direction in the exit lead. When there is only a small number of channels in the entrance and exit, these residual polarizations can be large. We will find the root mean square spin currents in the right lead by averaging over the ensemble of coherent cavities with strong spin-orbit coupling. These fluctuations are due to mesoscopic interference effects inside the quantum dots.We are primarily interested in the time-reversal invariant case, but we will present results valid with and without TRS. We consider coherent elastic scattering of noninteracting electrons with no spinrelaxation in the leads. We consider the chaotic limit for the quantum dot, in which the electron dwell time τd = 2π~/K∆ is much longer than the Thouless time τTh = Ld /vF , where Ld is a typical linear distance across the dot, vF is the Fermi velocity, and ∆ = 2π~2 /mA is the mean orbital level spacing in the quantum dot, with m the effective mass and A the area of the dot. We further assume the strong spin-orbit limit, where the spinorbit time τso is much less than τd . We assume that all of the channels have perfect coupling into the quantum dot. We are interested in the properties of the current in the right lead. For an input ν , we define the outgoing current density matrix w, in addition to jw ν jout = 2tr σ ν PR SwS † , (4.15) and the current due only to the input state ν jin = 2tr(σ ν PR w) (4.16) ν = j ν − j ν . The charge current is j 0 and the spin current is ~ s = so jw jw . We define jw out w in ~ 0 . A small number pw = ~jw /jw jw . The polarization of the current in the right lead is ~ of parameters of the input current are sufficient to describe the effects of any w in a two- Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 82 terminal configuration. In particular, we define t = 2trw C = 2tr(w 2 ) (4.17) (4.18) ν R Dν = 2tr(σ ν PR w R ) = (jin ) (4.19) E ν = 2tr(σ ν PR w R w R ) (4.20) F ν = 2tr(σ ν PR w R PR σ ν w R ), (4.21) where superscript R is the quaternion dual, t is the total flux incident on the dot, C is a measure of the coherence of the current, Dν gives the incident charge/spin current from the right lead, and E ν and F ν are more measures of coherence. By adding a multiple of 11 to w, we can choose D0 = 2tr(PR w) = 0, and all results below assume this choice. Note that if current is incident only from the left lead, then Dν = E ν = F ν = 0. We take averages over the uniform ensemble of all S-matrices in the strong spinorbit limit, either with TRS (called the circular symplectic ensemble – CSE) or without TRS (called the circular unitary ensemble – CUE) [105, 12]. Such averaging is readily performed experimentally by small changes of the shape of a quantum dot [171]; the root mean square (rms) fluctuations also give a typical value to be expected for any one chaotic dot. An external magnetic field can easily break TRS, moving between these ensembles. A convenient formalism for performing such averages was worked out by Brouwer and Beenakker [18]. From that work, we need two averages. In the quaternion representation, for f1 = tr(ASBS † ) for A, B constant K × K quaternion matrices, 1 [2trAtrB − tr(AR B)] 2K − 1 1 = trAtrB. K hf1 iCSE = (4.22) hf1 iCUE (4.23) The other average we need is of f2 = tr(ASBS † )tr(ASBS † ) for A, B constant Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling K × K quaternion matrices. We find [18] n 1 {K − 1} 8[trA]2 [trB]2 + 2tr[A2 ]tr[B 2 ] + 4[tr(AB R )]2 hf2 iCSE = 2ΛS o − 8tr[A]tr[B]tr[AB R ] − 2tr[AAB R B R ] n − 2[trA]2 tr[B 2 ] + tr[A2 ][trB]2 − 4tr[A]tr[AR B 2 ] − 4tr[B]tr[A2 B R ] o R R R R R + 4tr[A]tr[B]tr[AB ] + tr[AB AB ] + tr[AAB B ] hf2 iCUE = 83 (4.24) 1 4K(trA)2 (trB)2 + Ktr(A2 )tr(B 2 ) − tr(A2 )(trB)2 − (trA)2 tr(B 2 ) , (4.25) ΛU where ΛS = K(2K − 1)(2K − 3) and ΛU = K(4K 2 − 1). ν . Using Eq. 4.22, We consider the mean and fluctuations of jw ν ν ν hjw i = hjout i − jin ν hjout i = δν0 2tM Dν − δS , 2K − δS 2K − 1 (4.26) (4.27) where δS = 1 for averages over the CSE and δS = 0 for averages over the CUE. The relevant ν = j ν − hj ν i, which satisfy fluctuations to study are of ∆jw w w D E ν 2 ν 2 ν (∆jw ) = jout − hjout i2 (4.28) Using Eqs. 4.24 and 4.25, we find 1 ν 2 jout = M δν0 4M t2 (K − δS ) − 2M C + 4δS E 0 Λ − M t2 + (K − δS )(2M C + 2δS Dν 2 ) (4.29) − δS [E 0 (2K − 1) − F ν ] , ν 2 does not where Λ = ΛS , ΛU for the CSE, CUE, respectively. We note that jout CUE depend on D, E, or F . Combining this result with Eq. 4.27, ν 2 1n M t2 (1 + δS ) 2 ν 0 jout − hjout i = M δν0 − 2M C + 4δS E + 2M C(K − δS ) Λ K − δS /2 o 2K 2 − 3K + 2 − M t2 + δS [(Dν )2 − E 0 (2K − 1) − F ν ] 2K − 1 (4.30) Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 84 Eq. 4.30 is the main result of this work, and we will now look at its implications in some special cases. First, an arbitrarily polarized current incident from the left lead, as can be readily created by optical methods. Second, a pure spin current uniformly distributed between the leads. Third, a pure state pure spin current, with entanglement between the currents incident from each lead. Case 1: Spin-polarized current For any current incident exclusively from the left, the total current t and the parameter C are sufficient to describe mean and rms currents in the right lead. We consider the input current represented by w1 = 1 11N (tσ0 + ~s · ~σ ) 2N 0M (4.31) where ~s is the polarization magnitude and direction of the input spin current. Note that t can be positive, negative, or zero, depending on the direction of the charge current through the device. For |~s| = |t|, the current is fully polarized. For the density matrix of Eq. 4.31, C = (t2 + s2 )/2N , and D = E = F = 0. Applying Eq. 4.27, the mean spin current in the right lead is zero and the average charge 0 0 current is jw = 2tM/(2K − δS ). The reduction of jw as TRS is broken (δS → 0) is the signature of weak antilocalization [12, 13, 29]. The rms spin current in the right lead is M [(M − δS )t2 + (K − δS )s2 ] s2 jw =3 . NΛ (4.32) The fluctuations in the charge current are D 0 ∆jw 2 E = M [4M N − δS (4M − 1/N )]t2 + (1 − δS /N )s2 Λ (4.33) In the case of an unpolarized charge current, s = 0, with TRS, spin current in the exit lead is forbidden when M = 1 due to the combined effects of time reversal symmetry Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 85 1 <j0 > <(js )2>1/2 w <(ps )2>1/2 w w 0.8 0.6 0.4 CUE N=2 0.2 0 N=1 N=2 N=3 N=4 2 4 M 6 8 2 4 M 6 8 2 4 M 6 8 Figure 4.1: For the fully spin-polarized current represented by Eq. 4.31 with t = s = 1 and time reversal symmetry, comparison of numerical (symbols) and analytical (lines) results for the mean charge current (left), rms spin current (middle) and rms spin polarization (right) in the exit lead, where M (N ) is the number of channels in the exit (entrance) lead. An average over 50 000 S-matrices from the CSE was performed for each data point. The lines are from Eqs. 4.26, 4.32, and 4.35. The right panel shows that the expected spin polarization in the exit lead is still appreciable, even for several open modes in each of the leads. Also shown are the equivalent CUE results with N = 2, showing that the rms spin polarization is nearly unchanged by breaking TRS in this case. Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 86 and unitarity [89, 81, 169], as can be seen in Eq. 4.32. We can consider a pure spin current incident from the left by setting t = 0. In that case, we see that D E M (N − δ )s2 S 02 ∆jw = , NΛ (4.34) showing the scale of charge currents produced from the pure spin current. Similar effects have recently been proposed to measure the spin conductance in a three-terminal geometry 2 0 i [3]. We note that hjw CSE = 0 if N = 1, showing that a pure spin-current incident from a single channel cannot produce a net charge current in the other channels. This is the time reversed statement of the theorem that with TRS a charge current cannot produce a spin-polarized current when M = 1. 0 . It We can further consider the spin-polarization of the exit current, p~w = ~jw /jw is clear that h~ pw i = 0, just as h~jw i = 0, but there is some rms spin polarization of the exit s 2 0 2 / jw , we can use the above results to find current. If we approximate p2w ≈ jw t2 (M − δS ) + s2 (K − δS ) . p2w ≈ 3(K − δS /2)2 Λt2 M N (4.35) To test this approximation, we found p2w by numerically averaging over the CSE. Matrices drawn from the CSE were chosen by diagonalizing matrices from the Gaussian Unitary Ensemble, as described in Ref. [89]. Results are shown in Fig. 4.1 for the case t = s = 1, and it is clear that Eq. 4.35 agrees very well with the numerical results (right panel). The case shown in the figure is the relevant one for the Schliemann-Egues-Loss SFET, in which a fully polarized spin current is incident from one lead. In the off state, which relies on large spin-orbit coupling, the spin polarization in the exit lead is supposed to be zero. We see in Fig. 4.1 that even for several open channels in each lead, we expect to find an appreciable spin polarization in the output, limiting off-state function. Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 87 1 0.16 0.14 0.8 <(j0 )2>1/2 w 0.12 <jz > <(∆ js )2>1/2 w w 0.6 0.1 0.08 0.4 0.06 0.04 CUE N=2 0.2 N=1 N=2 N=3 N=4 0.02 0 2 4 6 M 8 0 2 4 6 M 8 2 4 6 M 8 Figure 4.2: For the pure spin current represented by Eq. 4.36, comparison of numerical (symbols) and analytical (lines) results for the rms charge current (left), mean spin current (middle) and rms spin current fluctuations (right) in the exit lead, where M (N ) is the number of channels in the exit (entrance) lead. An average over 50 000 S-matrices from the CSE was performed for each data point. The lines are from Eqs. 4.37–4.40. The left panel shows that this pure spin current should still be expected to produce significant charge currents, with a nonmonotonic dependence on the number of open channels N and M . Also shown are the equivalent CUE results with N = 2. Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 88 Case 2: Pure spin current from both leads We consider a pure spin current incident from both leads, represented by the density matrix σz 11N 2N w2 = σz −11M 2M . (4.36) This density matrix represents a spin current of +ẑ incident from the left and a spin current of −ẑ incident from the right, which together are an incident pure spin current from left to right with polarization +ẑ. In this case, t = 0, C = K/2M N , Dν = (0, 0, 0, 1), E ν = (1/2M, 0, 0, 0), and F ν = (1, −1, −1, 1)/2M . Though the mean value of the charge current is zero, since it is as likely for the charge current to flow in as out, the spin current can produce a mean square charge current D 02 jw E N2 + M2 K 1 − δS . = Λ MNK (4.37) We note that when M = N = 1, j02 CSE = 0, showing that no charge current can be produced. This result is another implication of the theorem that, with time reversal symmetry, a spin current incident in one channel cannot produce a charge current, combined with a simpler result that coherence and time reversal symmetry forbid spin-to-charge reflection in a single channel. The spin current in the right lead is a combination of the incident spin current, the reflected spin current from the right and the transmitted spin current from the left. Together, these give a mean spin current of i jw = (0, 0, 1 − δS 1 ). 2K − 1 (4.38) Thus, with TRS, the spin current in the right lead is, on average, reduced from 1. In the case M = N = 1, this reduction removes 1/3 of the spin current that began in the lead. Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 89 The fluctuations around the mean are K(K − δS ) 1 − δS MNK = (4.39) NΛ z 2 (M − 1)K 2 1 3M N K 2 ∆jw = K + δS − K(M + 2 − )+ + (K − δS /2)N Λ M 2M 2 K x,y 2 ∆jw (4.40) These results, along with confirming numerical simulations, are shown in Fig. 4.2. Case 3: Pure state pure spin current We consider entanglement between the currents in the two leads, which is beyond the standard chemical potential formulation of transport. In particular, consider a pure state spin current entangled between both leads, rather than the mixed state spin current of case 2. With M = N = 1, we consider w3 = 1 2 σz σx − iσy σx + iσy −σz (4.41) This state has, as in case 2, a pure spin current +ẑ incident from the left and a pure spin current −ẑ incident from the right, but the off-diagonal terms of w3 indicate that the currents are entangled. The density matrix formalism easily allows consideration of such offdiagonal correlations between the channel currents. The entanglement could be produced by passing a current through a beamsplitter produced from quantum dots[110, 134, 84], feeding into the two channels or from spin injection by optical orientation using entangled photons. The density matrix w3 represents a pure state by condition 3 of section 4.2.3 with α = −1/2. In this scenario, t = 0, C = 3, Dν = (0, 0, 0, 1), E ν = (3/2, 0, 0, 1), and F ν = (1, −1, −1, 1)/2. This should be compared with case 2 in the M = N = 1 limit, which is the same except C = 1 and E ν = (1/2, 0, 0, 0). Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 90 The most significant difference from case 2 is that coherence between the channels allows a charge current to be produced, even when M = N = 1 with TRS. There is still no 2 1/2 2 1/2 0 i 0 mean charge current, but the rms fluctuations are hjw CSE = 0.41, hjw iCUE = 0.45. This rms charge current is much larger than the results of case 2, even away from M = N = 1, (see Fig. 4.2, left, and Eq. 4.37) indicating that the entangled spin current is better able to couple into charge current than is the incoherent spin current. We have normalized w3 i to have jw = (0, 0, 1 − δS /3), as in case 2. We find fluctuations around the mean of 2 1/2 2 1/2 i i i h∆jw CSE = (0.58, 0.58, 0.62), h∆jw iCUE = (0.63, 0.63, 0.63). With TRS, the fluctuations are larger along the polarization axis, but not markedly so. The total spin polarization s 2 1/2 s 2 1/2 fluctuations are ∆jw = 1.03, ∆jw CUE = 1.10 which is larger than the mean CSE s , showing that coherence between the current and equal in scale to the input current jin channels significantly enhances the mesoscopic fluctuations; this should be compared with Fig. 4.2 (right panel). Such large fluctuations entail a significant loss of knowledge of the quantization axis of the spin current, so the initially z-polarized current can exit polarized in many directions. 4.4 Discussion Mesoscopic fluctuations of spin current on passing through a chaotic ballistic quan- tum dot can produce large fluctuations in spin-polarization, charge currents from pure spin currents, and spin currents from charge currents [89]. These predictions for mean and rms currents will be modified by dephasing and effects of the energy dependence of the S-matrix. Dephasing processes can be readily added to this model using the third lead method [24, 17, 9], as detailed in chapter 3. Dephasing generally reduces the fluctuations in charge and spin currents and also removes the symmetry that forbids charge or spin Chapter 4: Fluctuations of spin transport through chaotic quantum dots with spin-orbit coupling 91 currents at certain values of M and N with TRS. If the ingoing current contains particles with energies varying over a large enough range, the energy dependence of the S-matrix must be considered as well. The S-matrix is generally correlated on the energy scale of the level broadening of the quantum dot eigenstates, approximately ∆′ = ∆K/2 + γφ /2, where γφ is the dephasing rate [68, 89]. If the incident particles have energies that differ by a large amount compared to the level broadening ∆′ , as can happen at sufficiently large temperatures or δµν , then the mesoscopic fluctuations are suppressed, as there are effectively more open channels for particles passing through the dot. Mesoscopic fluctuations producing spin polarized exit currents could be important for operation of a Schliemann-Egues-Loss SFET. To avoid this impact on the off-state polarization, such a device should have many scattering regions in parallel or operate in a regime with sufficiently large temperature, bias, or dephasing so as to reduce these mesoscopic effects. Acknowledgments We acknowledge useful discussions and a careful reading of this manuscript by Bert Halperin. We acknowledge helpful conversations with Caio Lewenkopf, Emmanuel Rashba, Ilya Finkler, and Ari Turner. This work was supported in part by the Fannie and John Hertz Foundation and NSF grant PHY-0646094. Chapter 5 Inhomogeneous nuclear spin flips M. Stopa,1 J. J. Krich,2 and A. Yacoby2 1) Center for Nanoscale Systems, Harvard University, Cambridge, MA 2) Physics Department, Harvard University, Cambridge, Massachusetts Abstract We discuss a feedback mechanism between electronic states in a double quantum dot and the underlying nuclear spin bath. We analyze two pumping cycles for which this feedback provides a force for the Overhauser fields of the two dots to either equilibrate or diverge. Which of these effects is favored depends on the g-factor and Overhauser coupling constant A of the material. The strength of the effect increases with A/Vx , where Vx is the exchange matrix element, and also increases as the external magnetic field Bext decreases. 5.1 Introduction Hyperfine interaction with the host nuclei in nanoscale GaAs systems, while rel- atively weak, can nevertheless limit the electron coherence time and thereby complicate strategies to implement quantum information and quantum computing schemes in these systems [75, 112, 158, 95]. Conversely, ever-increasing control of angular momentum transfer between electrons and nuclei in a range of materials enables numerous applications precisely 92 Chapter 5: Inhomogeneous nuclear spin flips 93 because of the environmental isolation of the nuclear system. These include applications to quantum information processing employing NMR [150]. From the perspective of fundamental physics, experiments on few-electron systems with controllable coupling to the nuclear many-body system uncover a fascinating arena of new phenomena with ramifications for theoretical physics and engineering [64]. Experiments on double quantum dots with electron number N = 2 have uncovered and exploited an intriguing phenomenon called the “Pauli blockade” [111] in which two electrons with parallel spins are forbidden from combining in one dot by the exclusion principle. In transport or in gate pulsing, even when such a transition becomes energetically favorable, it can only proceed via a spin “flip-flop” process in which angular momentum is exchanged with the local nuclei. Repeating the spin transfer, however, modifies the character of the nuclear spin distribution. One metric for the nuclear state is the difference between the total Overhauser fields of each dot. These are the effective Zeeman fields which the electrons experience due to the hyperfine interaction. Several recent experiments addressed transfer of angular momentum from the electron system to the nuclear bath through various pumping cycles. One experiment claims that under a specific, repeated pulsing sequence (see below) [124] the polarizations in the two dots tend to equilibrate; a phenomenon which has been numerically reproduced [121, 125]. However, another similar experiment claims to find a large difference induced between the Overhauser fields of the two dots [55]. The theory which we describe here does not claim to explain either experiment. Here we describe a force toward either equalizing or inducing differences between the Overhauser fields in the two dots. The direction of this force depends on the spin of the initial electron state (i.e. the direction of the electron-nuclear “flip-flop” process) as well as on the sign of the product of the g-factor g and the Overhauser coupling constants Aβ , where β is the nuclear species. Assuming GaAs (A69 Ga = −23 µeV , A71 Ga = −20 µeV Chapter 5: Inhomogeneous nuclear spin flips Energy E (µeV) 94 10 0 0.8 c 0.6 b Τ+ ~ ε Ψ a c/b 1.3 c/b a,b,c -10 0.4 1.1 0.2 0.0 0.0 0.5 1.0 1.5 detuning ε (mV) Figure 5.1: Top: electronic states near the (1,1) to (0,2) stability diagram transition. Crossing of Ψ and T + at ε̃ becomes anti-crossing in presence of transverse Overhauser field gradient. Bottom: overlap of the Ψ state with S(0, 2) (a), with |L↑ R↓ i (b) and with |L↓ R↑ i (c). Parameters: Bext = 0.2 T , γ = 1.2 µeV , ∆ = 1000, EC = 0.6meV , Vx = 1µeV . and A75 As = −47 µeV , g = −0.44 [113]), we describe two pulse sequences which differ in the choice of the initial electron state, which consequently have a force tending to cause the Overhauser fields in the two dots to equilibrate or to diverge. 5.2 Electronic states of the double dot with N=2 We calculate the electronic states of the two electron (N = 2) double dot within the Hund-Mulliken formalism [22, 103] developed for the hydrogen molecule. We focus on the regime in the charge stability diagram [116] where the charge states (NL , NR ) = (1, 1) and (0, 2) are close to degeneracy, with NL , NR the numbers of electrons on the left, right dots. Typically, in this method, eigenstates of total spin, singlets and triplets, are employed as basis states. However, since we wish to study the inhomogeneous Overhauser effect due to different effective magnetic fields in the two dots, we choose a basis which diagonalizes, at the single particle level, the z-component of this inhomogeneous field and in which the Chapter 5: Inhomogeneous nuclear spin flips 95 spatial dependence of nuclear spin flips induced by electronic spin “flops” is transparent. The basis is: {ξn } ≡ {|R↑ R↓ i , |L↑ R↓ i , |L↓ R↑ i , |L↑ R↑ i}, where L and R indicate the orbital states of the left and right dot, the arrows denote spin direction.1 Two remaining states of the Hund-Mulliken model, |L↓ R↓ i and |L↑ L↓ i are not relevant to our analysis, the former due to high Zeeman energy and the latter being far away in the charge stability diagram. Note that |R↑ R↓ i is the standard S(0, 2) state and |L↑ R↑ i is the standard T+ state. The hyperfine Hamiltonian for two electrons is properly written: M Hhf = vA X [δ(r1 − Rm )S1 · Im ⊗ 1 + 1 ⊗ δ(r2 − Rm )S2 · Im ] ~2 m (5.1) where ri and Si are operators in the subspace of electron i (first quantized representation) and m is summed over a total of M nuclei (typically M ∼ 106 ); and where v is the volume per nucleus. We assume, for simplicity, a single nuclear species with spin 1/2. Then, constraining the maximum Overhauser field to be 5.3 T [146] leads to an average coupling constant A = −270 µeV . We incorporate the matrix elements of Hhf from Eq. 5.1 in our basis {ξn } into the Hund-Mulliken Hamiltonian which gives (upper triangle of Hermitian matrix shown): |R↑ R↓ i EC − ε H= |L↑ R↓ i IzLR − |L↓ R↑ i IzRR hL|Ri IzLL − IzRR +γ IzLR |L↑ R↑ i − IzRR hL|Ri + γ RR I+ hL|Ri − Vx RR I+ IzRR − IzLL LL I+ LR I+ IzRR + IzLL + EZ , (5.2) where ε is the potential “detuning” [116], γ is the tunneling coefficient and EZ ≡ gµB Bext is the Zeeman energy with µB the Bohr magneton and Bext the external magnetic field, √ More√traditionally, the linear combinations S(1, 1) ≡ [|L↑ R↓ i − |L↓ R↑ i]/ 2 and T0 (1, 1) ≡ [|L↑ R↓ i + |L↓ R↑ i]/ 2, are employed. 1 96 Chapter 5: Inhomogeneous nuclear spin flips defining the z-direction. We have taken the orbital energies of L and R to be zero for simplicity. We include only two Coulomb terms: the charging energy EC ≡ VRRRR − VRLRL and the exchange matrix element Vx ≡ VLRRL .2 Equation 5.2 is written to leading order in the overlap, hL|Ri, of the non-orthogonal single particle basis. Higher order terms (O(| hL|Ri |2 )) occur due to the normalization of the basis states [22]. Note that the matrix elements of H in this electronic basis remain operators in the Hilbert space of the nuclear coordinates:3 M A X ∗ αβ ~ I ≡v ψ (Rm )ψβ (Rm )I~m 2~ m=1 α (5.3) where α, β ∈ {L, R}. Note that previous researchers have typically ignored the transition LR , which we see from Eq. 5.2 can lead to a direct transition between |R R i and term I+ ↑ ↓ |L↑ R↑ i and causes a spin flip where the two wavefunctions overlap, in this case in the barrier. While such terms could be experimentally important for large Bext , i.e. where |R↑ R↓ i and |L↑ R↑ i anti-cross deep in the (0,2) regime, we will, in this paper, concentrate on spin flipflop processes occurring entirely in the left or right dot. We therefore consider the simpler Hamiltonian with the terms proportional to the L − R overlap omitted. 5.3 Nuclear spin flip location The crucial feature of Eq. 5.2 is that the |L↑ R↑ i state is coupled to |L↑ R↓ i via RR ) and it is coupled to |L R i by a a term which flips a nuclear spin in the right dot (I+ ↓ ↑ LL ) . In the absence of flip-flop coupling to term that flips a nuclear spin in the left dot (I+ the |L↑ R↑ i state, the upper left 3x3 matrix in Eq. 5.2 (see also yellow highlighted region of 2 Coulomb are defined in the usual way in our two state basis, α, β, γ, δ ∈ {L, R}: R R matrix ∗elements Vαβγδ ≡ dr1 dr2 ψα (r1 )ψβ∗ (r2 )V (r1 , r2 )ψγ (r1 )ψδ (r2 ). 3 We have also used the identity: S · Im = Sz Imz + [S− Im+ + S+ Im− ]/2. Chapter 5: Inhomogeneous nuclear spin flips 97 Fig. 5.2) has a ground state, which we denote: |Ψi = a(ε) |R↑ R↓ i + b(ε) |L↑ R↓ i + c(ε) |L↓ R↑ i . (5.4) As shown in figure 5.1, at large (positive) ε, |Ψi → |R↑ R↓ i ≡ S(0, 2) and at large negative ε, |Ψi becomes an unequal superposition of |L↑ R↓ i and |L↓ R↑ i. Even when Vx > |hIzRR −IzLL i|, the inhomogeneous Overhauser effect will produce a preference for either the |L↑ R↓ i or the |L↓ R↑ i component of Ψ (see figure 5.1), with the electron down spin preferentially located on the dot with smaller Iz . In the first electron pulsing sequence which we describe, the electron state is initialized at large positive ε into |Ψi ≈ S(0, 2) and detuning is swept approximately adiabatically through the Ψ - |L↑ R↑ i anti-crossing. The position of this anticrossing, ε̃, is determined by the energy of |L↑ R↑ i ≡ T+ (see figure 5.1) which is determined by Bext . Insofar as b(ε̃) 6= c(ε̃), a transition from Ψ to |L↑ R↑ i will preferentially induce a nuclear spin flip (down) on the side with the larger Iz . This tends to equilibrate the values of IzRR and IzLL . In the second pulse sequence the electrons are initialized into |L↑ R↑ i and the transition occurs into Ψ. The same feedback mechanism now preferentially causes nuclear spins to flip up, but still on the side with the larger Iz , thus leading to a tendency for |IzLL − IzRR | to grow. Both of these sequences can be experimentally implemented [116, 55]. Our further analysis focuses mainly on the first pulse sequence. Note that the preceding argument depends on the sign of A which in turn depends on the sign of g and the sign of the effective Overhauser field which, for Ga and As, are anti-parallel to the nuclear spins [113]. 5.4 Nuclear states To further analyze the Hamiltonian, Eq. 5.2, it is helpful to introduce a simplified basis for the nuclear states in which all of the nuclei are either in the left or right dot and 98 Chapter 5: Inhomogeneous nuclear spin flips all within a given dot interact equally with the electron. In other words, |ψL (r)|2 is taken as a constant within a spherical “box” of some volume, V. In this model, which we refer to as the “box model,” the squares of the total angular momenta Iα2 are conserved, where P I~α ≡ (vA/V) m∈α I~m , and where α ∈ {L, R}. Thus, the electrons essentially interact with two composite nuclear spins, one on the left and one on the right. The nuclear state basis is {IL , IR , ILz , IRz } (where Iα (Iα + 1) is the eigenvalue of (I~αα )2 and Iαz is the eigenvalue of Izαα ). Finally, for given IL , IR , it is convenient to transform to the basis of ∆ ≡ ILz − IRz and s ≡ ILz + IRz . In this basis the z-components of the nuclear operators have non-zero matrix elements on the diagonal blocks, but the raising and lowering operators connect different (∆, s) subspaces (see Fig. 5.2). The strength of the narrowing force depends on the ratio r ≡ c/b at ε̃. This depends on ∆ and on Bext . For example, smaller Bext results in smaller (or more negative) ε̃, where, as shown in Fig. 5.1, the ratio c/b increases (for ∆ > 0). Exactly how large c/b can get depends on Vx which, in the example of Fig. 5.1, we have set to 1 µeV.4 In Fig. 5.3 we plot the value of r(ε̃) as a function of Bext for various values of ∆. The key point is that r(ε̃) increases monotonically with ∆ (cf. yellow highlighted region of Fig. 5.2), however it also decreases monotonically with Bext (and hence ε̃). Interestingly, because the |R↑ R↓ i state is coupled equally to |L↑ R↓ i and |L↓ R↑ i, the value of b/c is independent of γ. We note that the flip-flop process naturally also depends on the rate at which ε is swept since, in order to be adiabatic and remain on the lower branch of the Ψ - |L↑ R↑ i anti-crossing the ε variation must be sufficiently slow. More generally, the character of the state evolution can be examined as a Landau-Zener tunneling problem, as in chapter 6 or 4 Self consistent electronic structure calculations show (Stopa, unpublished) that for lateral double quantum dots, Vx can range from 250 µV to less than 1 µV . Here, we have chosen the lower value. Chapter 5: Inhomogeneous nuclear spin flips ∆−1,s−1 EC-ε γ* γ* 0 0 0 0 0 γ ∆−1 Vx* 0 0 0 0 0 γ 0 Vx 0 −∆−1 0 s-1+EZ 0 0 0 IR+ 0 0 0 0 0 99 ∆,s ∆+1,s−1 0 0 0 0 EC-ε γ* γ* 0 0 0 0 IRγ 0 0 0 0 γ ∆ V x* 0 Vx −∆ 0 0 0 0 0 0 0 0 0 0 0 0 IL- 0 0 0 0 0 0 0 s+EZ 0 0 0 0 0 0 0 0 EC-ε γ* γ* 0 0 0 0 0 γ ∆+1 V x* 0 0 0 0 0 γ 0 0 IL+ 0 0 0 0 Vx −∆+1 0 s-1+EZ Figure 5.2: Hamiltonian for three sectors of the nuclear difference quantum number (∆ − 1, s−1), (∆, s), (∆+1, s−1). In the above, ±∆±1 and s±1 are shorthand for (vA/V)(±∆±1) and (vA/V)(s ± 1) respectively. else evaluated numerically [142]. The evolution of the full nuclear state is complex and the experimental manifestations of that evolution are ambiguous. Nevertheless, as a possible baseline for more detailed studies of the nuclear evolution, we describe a simple, incoherent model which results in narrowing of the distribution of ∆. If we assume that the system is in the well-defined state |Ψi ⊗ {IL , IR , ILz , IRz } and the detuning is moved quickly to ε̃ and held there for time τ , we can compute, by time dependent perturbation theory, the probability for a nuclear spin to flip in the right dot as: ΓR (ILz , IRz → ILz , IRz − 1) ≡ ΓR (s, ∆ → s − 1, ∆ − 1) τ2 RR |hILz IRz − 1|I− |ILz IRz i|2 ~2 A2 Ω2R− τ 2 2 = |b| 4~2 = (5.5) where we have suppressed the IL , IR dependence for brevity and where the matrix elements of the ladder operators are given by the well-known formulas: Ωα± ≡ hIα , Iαz ± p 1|I± |Iα , Iαz i = Iα (Iα + 1) − Iαz (Iαz ± 1). Similarly, the flip probability in the left dot is proportional to the c component of Ψ ΓL (s, ∆ → s − 1, ∆ + 1) = A2 Ω2L− τ 2 2 |c| . 4~2 (5.6) 100 Chapter 5: Inhomogeneous nuclear spin flips Ψ−component ratio c/b 1.10 ∆ = 200 ∆ = 600 ∆ = 1000 ∆ = 1400 ∆ = 1800 1.08 1.06 1.04 1.02 0.2 0.4 0.6 0.8 1.0 Magnetic Field B [T] Figure 5.3: The wave function ratio r ≡ c/b evaluated at the Ψ − T + crossing point, ε̃, as a function of Bext for various values of the nuclear spin z-component difference ∆. r(ε̃) is monotonically increasing with ∆ and decreasing with Bext . If we denote the probability distribution for the nuclear state (at fixed IL , IR ) as W (s, ∆), then the condition for W to be stable in its dependence on ∆ can be written (cf. Fig. 5.4a): W (s + 1, ∆ + 1)ΓL (s + 1, ∆ + 1) = W (s, ∆)ΓR (s, ∆) Ω2 (s, ∆) |b(∆)|2 W (s, ∆ + 1) = W (s, ∆) 2 R− ΩL− (s, ∆ + 1) |c(∆ + 1)|2 (5.7) where we have assumed that W (s) ≈ W (s + 1) and we have used the fact that b and c depend very weakly on s (only through the s-dependence of ε̃). Recursion relation Eq. 5.7 can be solved iteratively and the influence of the narrowing force evaluated. In Fig. 5.4 we have plotted W (∆) computed with the ratio ΩR− /ΩL− set to unity to show only the narrowing from the inhomogeneous Overhauser effect described here with the same electronic parameters as in Fig. 5.1, and with IL = IR = 1000;5 including the Ω’s induces more narrowing. For comparison we show the T → ∞ thermal distribution of ∆, averaged over s, also for IL = IR = 1000. Inset (a) shows the ratio of the root-mean-square (rms) ∆ in the thermal distribution, σT , to the rms ∆ with the 5 We assume Nnuc ∼ 106 nuclei in each dot leading to an average spin ∼ √ Nnuc ∼ 1000. Chapter 5: Inhomogeneous nuclear spin flips distribution W 6 (b) I (a) s Rz 7 6 ∆,s 5 ΓR σΤ/σ -3 8x10 101 4 ΓL ∆+1,s+1 ∆+1,s-1 3 4 2 0.2 0.4 0.6 ∆ Bext (T) ILz Bext=0.05,0.10,...,0.75 T high temperature thermal distr. 0 0 500 1000 1500 ∆=ILz-IRz 2000 Figure 5.4: (main) Reduced distribution W (∆, s), calculated from Eq. 5.7 (solid lines), for Vx = 1 µeV as a function of ∆ for various Bext = 0.05, 0.10, ..., 0.75 T (lower fields have narrower W ); and thermal W (∆) (dashed), averaged over s, all with IL = IR = 103 . Inset (a) narrowing factor σT /σ(Bext ) versus Bext . Inset (b) Illustration of ILz − IRz plane. ∆ and s are the diagonal coordinates, with ∆ ≡ ILz − IRz . narrowing force at varying Bext , σ(Bext ). A substantial narrowing of W (∆) results from the inhomogeneous Overhauser effect. 5.5 Discussion Experimentally, the |Ψi to |L↑ R↑ i pulse sequence polarizes only about 1% of the nuclei, even when running sufficient cycles to flip all of the nuclei [116]. This saturation of the nuclear polarization is still an open problem. A recent article by Yao [164] discusses a model similar to that described herein. In that paper, no mechanism for stopping the flip-flop process is proposed when the pumping continues (as it does in experiments) beyond ∼ 105 cycles. In our model, polarization will saturate when both ILz = −IL and IRz = −IR , implying that s = −IL − IR . However, the resulting distribution of ∆ will then mirror the difference in the initial distributions of IL and IR , and hence will show no narrowing of 102 Chapter 5: Inhomogeneous nuclear spin flips W (∆). Thus our box model can qualitatively explain the narrowing effect or the saturation, but not both. We believe that a full understanding of these phenomena depends on the variable coupling of the electron wave function to different groups of nuclei, so that conservation of the magnitudes of two spins, I~L and I~R , is not required. Such a model with multiple interacting composite nuclear spins, incorporating the narrowing effect described here as well as the Landau-Zener tunneling behavior near ε̃, in some parameter regimes shows the potential to send |∆| → 0 while reducing the spin flip probability, slowing the growth of total polarization; for other parameters, |∆| grows large despite the narrowing force described here, as described in chapter 6. Acknowledgments We thank B. I. Halperin, M. Gullans, J. Taylor, M. Lukin, S. Foletti, H. Bluhm, Y. Tokura, D. Reilly and M. Rudner for valuable conversations. We thank E. Rashba for a critical reading of the manuscript. We thank the National Nanotechnology Infrastructure Network Computation Project for computational support. We gratefully acknowledge support from the Fannie and John Hertz Foundation, NSF grants PIF-0653336 and DMR05-41988 and the ARO. Chapter 6 Preparation of non-equilibrium nuclear spin states in quantum dots M. Gullans,∗1 J. J. Krich,∗1 J. M. Taylor,† B. I. Halperin,∗ A. Yacoby,∗ M. Stopa,‡ and M. D. Lukin∗ ∗ Department of Physics, Harvard University, Cambridge, MA Department of Physics, Massachusetts Institute of Technology, Cambridge, MA Joint Quantum Institute, University of Maryland and National Institute of Standards and Technology, College Park, MD ‡ Center for Nanoscale Systems, Harvard University, Cambridge, MA † † Abstract We develop a technique to model nuclear spin dynamics in a double quantum dot system undergoing adiabatic pumping. Our model, while semiclassical, allows to explore a wide range of parameters. In different parameter regimes, we find the system exhibits either a strong reduction or a large growth in the difference Overhauser field produced by the nuclei in the two dots. The study of non-equilibrium dynamics of nuclei in solids has a long history [2] and has become particularly relevant as nanoscale engineering and improvements in control 1 Contributed equally to this work. 103 104 Chapter 6: Preparation of non-equilibrium nuclear spin states in quantum dots allow to probe mesoscopic collections of nuclear spins [166, 43, 131, 112, 85, 15, 90]. This control has direct applicability to quantum information science, where nuclear spins are often a main source of dephasing [64]. The hope and goal of developing an understanding of electronic control of nuclei is to usher in an era of mesoscopic state engineering, in which nuclear spins can be programmed to be useful resources [82], as indicated in recent experiments [124, 55]. Many groups have developed specific theoretical techniques to solve subsets of this intractably hard problem, including semiclassical solutions for the central spin [33, 4], cluster and diagramatic expansion techniques for short time non-equilibrium behavior [159, 165, 37], and exact solutions for small systems, relevant in the case of homogenous coupling to the electron [121, 125, 164, 143]. In this paper, we present a model for nuclear spin dynamics in a double quantum dot undergoing dynamic nuclear polarization (DNP). We include the inhomogeneous coupling of nuclear spins to the electrons [35], incorporate the full electron spin dynamics, and determine the long time dynamics for the nuclear spins while containing parameters motivated by experiments. Our treatment of nuclear spin evolution is semiclassical, which gives results for the distribution of sum and difference Overhauser fields of the nuclei in the two quantum dots, but does not include the quantum corrections required to understand T2 decoherence processes. Numerical simulations show that, for some parameters, the distribution of the Overhauser difference field between the dots is suppressed. For other parameters, including those closer to experimental situations, simulations show apparent instabilities leading to the growth of large Overhauser difference fields. An electron spin confined in a quantum dot interacts with the lattice nuclear spins through the contact hyperfine interaction. DNP experiments operate near the crossing of the electronic singlet s with the spin-polarized triplet T+ state. A typical example (Fig. 6.1a) is Chapter 6: Preparation of non-equilibrium nuclear spin states in quantum dots a b |T− |T− |T0 Energy Energy |T+ Dz D+ Bext S− |T0 D − S− S − J | t=T |S |T+ |S J | t=−T ε d |T− ∆ Dz |T− Λ+ |T0 |T+ |T+ |S |T− D+ D− |T+ |S |T− Λ |T0 |T− ΓR |T0 S+ |T+ |S |T0 D− Sz |S |T− Γ |T0 Dz ∆- 105 |T0 D− S− |T+ |S |S D+ |T+ Figure 6.1: a) Schematic of two-electron energy levels as a function of detuning ε between (1,1) and (0,2) charge states. Arrows indicate adiabatic sweep through avoided crossing (pink) and rapid sweep back to (0,2) with reload (green). b) Spin-flip pathways between the s and T+ states as the exchange energy J(ε) is swept through the crossing, showing the nuclear operators involved in each path. Each pathway is a term in D̃− in Eq. 6.2. c) Annular approximation to the electron wavefunction in the double dot. d) Key processes contributing to Eq. 6.6. an adiabatic sweep through the s-T+ degeneracy, followed by a non-adiabatic return to (0,2) and reset of the electronic state via coupling to leads. To model this type of experiment, we first derive an effective two-level Hamiltonian to describe the system near the crossing of the singlet and lowest energy triplet state, T+ , then solve the time dynamics. For a double quantum dot with two electrons, we can write the Hamiltonian for the lowest energy (1, 1) and (0, 2) electron states, where (n, m) indicates n (m) electrons in the left (right) dot. If ψd (r) is the single-particle envelope wave function on dot d = l,r (for the left, right dot), the hyperfine coupling constant for the nuclear spin at rkd is P 2 1/2 gkd = ahf v0 |ψd (rkd )|2 and the rms Overhauser energy is Ω = ( k gkd ) , where ahf is the hyperfine coupling constant, and v0 is the volume per nuclear spin. We assume that Ω is the same in each dot and choose units such that Ω = −~g∗ µB = 1, where g∗ is the electron effective g-factor and µB is the Bohr magneton. We introduce two collective nuclear spin 106 Chapter 6: Preparation of non-equilibrium nuclear spin states in quantum dots P operators to denote the Overhauser fields in the left (L) and right (R) dots, L = k gkl Ikl P and R = k gkr Ikr and further define S = (L + R)/2, D = (L − R)/2, where Ikd is the angular momentum of the kth nucleus on dot d. In the basis {|si , |T+ i , |T0 i , |T− i}, where the Tm are the (1, 1) triplet states, the Hamiltonian is [146] √ vD+ − 2vDz −vD− −J(ε) √ v D− −Bext + Sz S− / 2 0 H= √ . √ √ − 2vDz S+ / 2 0 S− / 2 √ 0 S+ / 2 Bext − Sz −vD+ √ where Bext is an external magnetic field, and v = v(ε) = cos θ(ε)/ 2. The parameters cos θ(ε), the overlap of the adiabatic singlet state with the (1, 1) singlet state, and J(ε), the splitting between s and T0 , are both functions of the energy difference ε between the (1, 1) and (0, 2) charge states. We consider the nuclei to be spin-3/2 of a single species in a frame rotating at the nuclear Larmor frequency. Assuming that J, Bext ≫ 1, we perform a formal expansion in the inverse electron Zeeman energy operator m̂ = (Bext −Sz )−1 . We apply a unitary transformation that rotates the quantization axis of the triplet states to align with Bext − S and find, to second order in J, m̂ Heff where v(ε)D̃+ −J(ε) + hs = , v(ε)D̃− −Bext + hT 2v 2 † v2 D̃z D̃z − D̃− D̃+ , J J + Bext − Sz 1 hT = Sz − (S− S+ m̂ + m̂S− S+ ), 4 1 1 2 D̃− = D− + m̂S− Dz − m̂S− m̂D+ − m̂ S− S+ m̂D− , 4 4 1 D̃z = Dz − S+ m̂ D− + S− m̂D+ . 2 hs = − (6.1) (6.2) Chapter 6: Preparation of non-equilibrium nuclear spin states in quantum dots 107 We develop a model for the evolution of the nuclear spin density matrix after one pair of electrons has passed through the system. By coarse graining this evolution over a cycle we can derive a master equation for the nuclear spins. The electron system is prepared in |si at large negative t = −T . We identify the (nuclear spin) eigenstates of the operator ˆ D̃ ˆ 2 D̃ + − , labeled |D⊥ i with eigenvalues D⊥ . After the crossing, the state is unchanged or ˆ |D i. ′ i ⊗ |T i where |D ′ i ≡ D −1 D̃ flips an electron and nuclear spin to the state |D⊥ + − ⊥ ⊥ ⊥ ˆ D̃ ˆ ′ i is an eigenstate of D̃ 2 We note that |D⊥ − + with eigenvalue D⊥ . The problem is now reduced to finding Landau-Zener solutions for each independent two-level system |D⊥ i⊗|si, ′ i ⊗ |T i, where we approximate ĥ → hD | ĥ |D i, ĥ ′ ′ |D⊥ + s T → hD⊥ | ĥT |D⊥ i, valid to ⊥ s ⊥ leading order in 1/J and 1/N . We model the actual sweep of ε by a linear sweep of J so q J(t) = −β 2 t + Bext , where β = 12 |dJ(ε)/dt| |t=0 . We approximate that v(ε) is constant, valid in the limit of large tunnel coupling and assume β ≪ Bext to ensure the applicability of Eq. 6.1. ˆ D̃− After one cycle, the state is |Ψi = (cS |si + cT D |T i) |D⊥ i. For β 2 T ≫ 1, the flip ⊥ probability is pf = 1 − exp(−2πω 2 ), where ω = vD⊥ /β, and cS = φS ≈ φT ≈ p Z 1 − pf exp(−iφS ), cT = √ pf exp(−iφT ) T hS dt −T Z t0 −T (6.3) hS dt + (T − t0 )hT + φAD (ω), where the crossing occurs at a time t0 ≈ Sz /β 2 . We include in φT the phase picked up by following the adiabat, φAD . We approximate φAD by interpolating between the limits ω = v D˜⊥ /β → 0 and ω → ∞, giving [153] φAD where τ = T β. 2 τ 2 = 2πω + pf ω 1 − 2π + log − π/4 , ω2 2 108 Chapter 6: Preparation of non-equilibrium nuclear spin states in quantum dots We move from the independent two-level systems to the general case by noting that the components of |Ψi depend only on the eigenvalue D⊥ and on the polarization Sz (which we approximate as commuting). Thus we can write an operator pˆf = P D⊥ pf (D⊥ ) |D⊥ i hD⊥ |, and similarly for φ̂S , φ̂T . The nuclear spin density matrix after each cycle is given by tracing over the electronic states. Rather than solve for the exact dynamics of the nuclear density matrix–still an intractably hard computational problem for any reasonable number of nuclear spins–we instead adopt an approximate solution to the problem using the P-representation for the density matrix as a integral over products of spin coherent states. From the thermal distribution, we choose such a spin coherent state and evolve it, where we interpret expectation values h...i as being taken in that state. The ensemble of such trajectories represents the physical system [4]. We organize this calculation by noting that the components of the Landau-Zener model (φS , φT , pf , D̃± ) are only functions of L and R. A spin coherent state is entirely described by its expectation values iid = hIid i. For the jth spin on the left dot, we expand the discrete time difference hIjl in − hIjl in−1 after n and n − 1 cycles in the small parameter gjl , giving an evolution equation where X dijl = gjl Pl,µ i[∂gjl Lµ , Ijl ] = gjl Pl × ijl , dt µ Pl = (6.4) 1 [h1 − pf i h∇l φS i + hpf i h∇l φT i − Im(γl )] , 2T where ∇l = (∂Lx , ∂Ly , ∂Lz ) and γl = D̃+ pf D̃− D̃+ ∇l D̃− , (6.5) and similarly for ijr , Pr , and γr , with L replaced by R. The factorization of expectation values is a natural consequence of our spin-coherent state approximation, as it explicitly Chapter 6: Preparation of non-equilibrium nuclear spin states in quantum dots 109 prevents entanglement between spins. Thus we have an effective, semi-classical picture of nuclear spins precessing and being polarized by their interaction with the electron spin, integrated over one cycle. We approximate the electron wavefunction as a piecewise-flat function with M levels, which we refer to as the annular approximation, as illustrated in Fig. 6.1c. We define P In = j∈n ij where the sum is over all nuclei with the same hyperfine coupling to the electron. Since gj is identical for all j ∈ n, we can simply replace i with I in Eq. 6.4. This is convenient, because In2 is a conserved quantity, so we can study the evolution of M ≪ N p spins. Additionally, the typical size of In is now ∼ N/M ≫ 1. This allows us to replace the spin-coherent states used above with semi-classical spins, and makes taking expectation values straightforward: all quantum operators can be replaced by their expectation values directly. −1 To illustrate, to first order in m0 = Bext , for d = l, r, Pd =pf λ (Λ+ ẑ − Λ0 S⊥ ) + m0 Γ0 pf Dz ẑ × D 2πω 2 (6.6) h β2 + ΓR pf ∇d φAD ∓ Γ0 Im (γl − γr ) 4πv 2 i + (1 − pf λ/2)(∆0 Dz ẑ + ∆− D⊥ ) where the top sign applies for d = l, D⊥ = (Dx , Dy , 0), S⊥ = (Sx , Sy , 0), λ = 1 − t0 /T gives the shift in the location of the crossing, and ∆0 , ∆− , Λ+ , Λ0 , ΓR , and Γ0 are constants depending on the details of the pulse cycle. To leading order in m0 , Im(γl − γr ) = 2(D × 2 . It is clear from Eqs. 6.4-6.6 that all dynamics stop if D = 0. As indicated in ẑ)pf /D⊥ Fig. 6.1d, the Γ0 term originates in the hyperfine flip-flop, the ∆0 and ∆− terms are the off-resonant effects of coupling from the singlet state to the T0 and T− states, respectively, Λ0 comes from coupling between the T+ and T0 states, and Λ+ comes from Knight shifts due to occupation of the T+ state. To leading order in m0 , for a pulse sequence consisting 110 Chapter 6: Preparation of non-equilibrium nuclear spin states in quantum dots of only the Landau-Zener sweep, with instantaneous eject and reload, the parameters have values ∆0 = 2v 2 J(t) c Λ+ = 1/4, Γ0 = 2πv 2 fc , β2 ≈ m0 , ∆− = v2 J(t) + Bext c ≈ m0 /4 Λ0 = m0 /4 ΓR = f c where fc = 1/2T is the cycle frequency and h.ic indicates an average taken over a full cycle; these values can be modified readily by changing the details of the pulse cycle, while leaving the Landau-Zener portion unchanged. The simulations shown below were performed with the equations of motion correct to second order in m0 with ψd (r) a 2D Gaussian. Taking v 2 ≈ 1/2, we estimate that for experiments performed with Bext = 10 mT with T = 25 ns [124], m0 ≈ 0.13, Γ0 ≈ 0.20, but the ∆ and Λ terms depend on the rest of the cycle. In each of the simulations, we choose initial magnitudes and directions of the spins In by a procedure equivalent to choosing initial directions for each of the Nn spin-3/2 nuclei in the P nth annulus and evaluating In = j∈n ij explicitly. The relationship between simulation time and laboratory time depends on the details of the pulse cycle, including pauses and reloads not considered explicitly here, but simulation time is roughly in units of (Ωgmax )−1 , where gmax ≈ 2Ω/ahf is the largest value of gk , so t = 400 is approximately 10 ms. When the number of annuli M is chosen to be small, the system moves rapidly to its maximally polarized state, with In ≈ −In ẑ for all n. (This does not correspond to all of the nuclei being polarized, which also requires In = 3Nn /2.) In the limit that M = 1, this polarization is the only result, showing that inhomogeneity of the electron wavefunction is essential for interesting phenomena involving Dz . The annular approximation should correctly describe the nuclear dynamics for a time scale given by the inverse of the difference between the gj of adjacent annuli. For any M , when ∆0 = 0, the system rapidly saturates Chapter 6: Preparation of non-equilibrium nuclear spin states in quantum dots 111 (i.e., pf → 0) without any statistical change in the distribution of Dz ; coupling from the singlet to the T0 state is an essential ingredient in all of the effects discussed below. By tuning the parameters of this model, two features of particular note are observed. First, as illustrated in Fig. 6.2a, we show an ensemble of trajectories in which D rapidly reduces toward zero. For the parameters of Fig. 6.2a, the standard deviation of Dz was reduced by a factor of 28. We remark that as D → 0, the singlet state does not mix with the triplets and nuclear spin dynamics stop. Until something (outside this model, such as nuclear dipole-dipole coupling) restores D, the DNP process is shut off, limiting the total nuclear polarization that can build up. While not shown in Fig. 6.2a, we observe a dramatic reduction of the total |D|, not just Dz , consistent with this qualitative observation. There have been a number of attempts to explain [125, 164, 143] the reported reduction in |Dz | [124], none of which has taken into account both the coupling to the T0 state and the inhomogeneity of the electron wavefunction. We only observe a narrowing of the Dz distribution in these simulations with both of these factors. Refs. [164, 143] claim that the force to reduce |Dz | originates in the hybridization of the singlet and T0 states before anti-crossing with T+ , which can be seen in the m̂S− Dz term in D̃− in Eq. 6.2. While we find that this effect can help reduce |Dz |, the strongest narrowing occurs in simulations with sufficiently large ∆0 and Λ0 , regardless of the change D− → D̃− . We find that the most significant factors for strong narrowing of the Dz -distribution are ∆0 /Γ0 ≈ Λ0 /Γ0 ≈ 1 − 2 and Λ+ /Γ0 . 1. A similar phenomenon can be seem from modeling the dynamics as a diffusion in Dz [62]. We now consider two prototypical pulse sequences motivated by experiments, one with small m0 = 0.01 and τ = 8, the other with larger m0 = 0.05 and smaller τ = 4, with β held constant. In both cases, over 90% of the trajectories display a growth in |Dz | as shown in Fig. 6.2b. This increase in |Dz | indicates that the spin flips are occurring predominantly 112 Chapter 6: Preparation of non-equilibrium nuclear spin states in quantum dots Figure 6.2: (a) Simulations with √ M = 100, m0 = 0.05, ∆0 /Γ0 = 1, ∆− /Γ0 = 0.25, Λ+ /Γ0 = 0.5, Λ0 /Γ0 = 1, ΓR /Γ0 = 1, β = 2π, τ = 4. At bottom, hSz ie is the median value of Sz at each time step in an ensemble of 1000 trajectories. Gray shaded region shows the 84th and 16th percentiles. Above, with independent y-axis, is similar h|Dz |ie . Thin red line is a single trajectory. For these parameters, the trajectories have |Dz | → 0 quickly, without time for strong polarization. (inset) Mean of the maximum value of |Dz | reached on each trajectory for the same parameters as in (b) (open circles) except M varied between 20 and 160, with 5000 trajectories per point. Closed circles show similar results with m0 = 0.05, τ = 4 and all other parameters scaled appropriately. The physical system has M → N ≈ 106 , so we interpret this as an instability to large |Dz |. (b) Simulations as in (a) except m0 = 0.01, ∆0 /Γ0 = 0.78, ∆− /Γ0 = 0.19, Λ+ /Γ0 = 5.8, Λ0 /Γ0 = 0.08, τ = 7.8. Results are presented as in (a), except each trajectory was shifted so that its maximum |Dz | occurs at time zero. The thin line at top shows the fraction of trajectories contributing to the ensemble at each time. 4.5% of the trajectories, which do not show this peak in |Dz|, are not included. Approximately 10% of the trajectories show behavior similar to that shown in the thin red line, where |Dz | is reduced initially and then goes unstable to large |Dz |. Chapter 6: Preparation of non-equilibrium nuclear spin states in quantum dots 113 in one dot. We interpret these results as showing a continuing increase of |Dz |, where the peak is an artifact of the annular approximation. Near the peak, many of the annular spins artificially reach their maximal polarization, at which point they should be broken into more annuli. Similar trajectories with different M show the maximum value of |Dz | increasing with M (Fig. 6.2 inset). The physical cause of this increase in |Dz | is not clear, but it is associated with both ∆0 /Γ0 and Λ+ /Γ0 being sufficiently large. This could be the same phenomenon as seen in Ref. [55], though some other effect is required to produce the large |Dz | ≈ Bext of that work. Future work will discuss the relationship between this effect and the shape of the electron wavefunction, in particular its inhomogeneity. There are a variety of outcomes possible in this model, including crossover regions between the Dz distribution narrowing and spreading described here. Simulations performed with parameters intended to approximate experiments are in this crossover regime. These simulations do not include such effects as the nuclear dipole-dipole coupling, so we expect them to describe accurately the effects of pumping and coherence for times shorter than the dipole-dipole diffusion time scale. Spin diffusion processes should oppose the buildup of polarization, return Dz to its thermal values, and, since they do not conserve In2 , can also break up the annulus spins, possibly allowing larger |Dz | to be produced than seen in these simulations. Acknowledgements We thank S. Foletti, H. Bluhm, C. Barthel, C. M. Marcus, and M. Rudner for valuable conversations. We gratefully acknowledge support from the Fannie and John Hertz Foundation, Pappalardo, NSF grants PIF-0653336 and DMR-05-41988, and the ARO. Appendix A Appendix to Chapter 2 A.1 Further cubic spin-orbit terms The following discussion was prepared with major input from Hans-Andreas Engel, and we are grateful for his contribution. In 3D zincblende crystals, the lowest order in k spin-orbit coupling term for conduction band electrons consistent with the crystal symmetries is HDresselhaus = γ[σx kx (ky2 − kz2 ) + c.p.] (A.1) where c.p. indicates cyclic permutations and x, y, z are the cubic crystalline directions [44]. When electrons are confined on a (001) plane, kz2 can be sent to hkz2 i and the 3D Dresselhaus term produces the 2D linear and cubic Dresselhaus terms, H001 = γhkz2 i(σy ky − σx kx ) + γ(σx kx ky2 − σy ky kx2 ). (A.2) Asymmetric confinement in the (001) plane also produces spin-orbit coupling of the Rashba type [26], HRashba = α(ky σx − kx σy ). 114 (A.3) Appendix A: Appendix to Chapter 2 115 These three terms together give the effective Hamiltonian for 2D conduction electrons in Eq. 2.1 of the main text. In a 2D zincblende system on a (001) surface, there are further cubic spin-orbit terms originating from both bulk and structure inversion asymmetries [28]. The extra contribution from bulk inversion asymmetry is γ1 (kx3 σx − ky3 σy ) while the cubic Rashba terms are α1 (kx3 σy − ky3 σx ) + α2 (ky kx2 σx − kx ky2 σy ). Winkler showed that γ1 (which he calls b6c6c 53 ) is much less than γ [157]. Specifically, using his formulae and parameters for GaAs gives γ/γ1 ≈ 103 for kF = 0.19 nm−1 . In a similar calculation, von Allmen considered an axial cubic Dresselhaus term of the form γ2 k2 (kx σx − ky σy ) and showed that γ/γ2 ≈ 20–400, depending on the model used [154]. We are unaware of estimates of the magnitudes of α1 and α2 in GaAs. Yang and Chang recently showed that in a 15 nm-wide GaAs/AlGaAs quantum well, the nonlinear Rashba terms do not become significant until considerably higher electron densities than considered in this paper [163]. An approximation of the cubic Rashba terms can also be given using the 8 × 8 Kane model, which considers the s-type conduction band and the p-type valence bands. The model is parameterized by the band gap E0 , the energy of the split-off holes ∆0 , and the matrix element P of the momentum (times ~/m0 , where m0 is the free electron mass) between the s- and p-type states. When an external electric potential is applied across the sample, the conduction band Hamiltonian contains a term Hext = λ σ · (k × ∇V ) . (A.4) For a system confined along the z-direction, one can take the expectation value hHext i along that direction. Noting that the only contribution of the confinement field is ∝ h∇z V i and for λ a constant, one obtains the k-linear Rashba Hamiltonian HRashba . If we 116 Appendix A: Appendix to Chapter 2 consider higher order corrections to λ, we can write λ = λ(0) + λ(1) k2 . The λ(1) term gives a term HRashba,c = α3 k2 (ky σx − kx σy ), which gives the axial (C∞v ) approximation to the cubic Rashba terms. Using third-order perturbation theory [157, 108], (0) λ P2 1 1 = − . 3 E02 (E0 + ∆0 )2 (A.5) Fifth-order perturbation theory gives [50] (1) λ =− P 4 ∆0 24E03 + 41E02 ∆0 + 26E0 ∆20 + 6∆30 9E04 (E0 + ∆0 )4 2 . (A.6) 4 In GaAs, these yield λ(0) = 5.3 Å and λ(1) = −870 Å . Because this estimate does not include edge effects, it only provides a rough estimate for the coupling constants α, α3 . We can, however, use Eqs. A.5 and A.6 to estimate the ratio of the strengths of the linear and cubic Rashba terms. For k = 0.19 nm−1 corresponding to the sample considered in the main text, 1 k2 λ(1) ≈ . (0) 17 λ (A.7) We are interested, however, in the relative sizes of the cubic Dresselhaus and the cubic Rashba. In the sample considered, the linear Dresselhaus and the linear Rashba strengths appear to be almost the same [106], and we can estimate that the linear Dresselhaus is about twice the strength of the cubic Dresselhaus (i.e., hkz2 i/kF2 ≈ 2, using the results in Ref. [106]), giving γ/α3 ≈ 8. It is possible that a more careful calculation of the cubic Rashba strengths, including band offsets and the full crystal symmetry, would modify these results substantially, but these results indicate that the cubic Dresselhaus term considered in the main text is the dominant cubic contribution to the spin-orbit coupling of GaAs quantum dots. Appendix A: Appendix to Chapter 2 117 Regardless of their strengths, all of these cubic terms contribute to ak similarly to the cubic Dresselhaus term discussed in the main text. We neglect the other cubic contributions, but their inclusion would only increase the constraint on γ. A.2 Refitting var g data Fig. A.1 shows the Zumbühl et al. var g versus Bk data discussed in the main text [172]. Also shown are the original curvefit and our refit with ak constrained to be greater than 8.1. This constrained fit clearly does not match the data. We thank Dominik Zumbühl for making the data available to us. −5 7 x 10 Data Original Fit a||≥8.1 6.5 var(g) (e2/h)2 6 5.5 5 4.5 4 3.5 3 0 1 2 3 4 5 6 7 B|| (T) Figure A.1: Variance in conductance as a function of in-plane field for quantum dot of area A = 8 µm2 and electron density n = 5.8 × 1015 m−2 , from Ref. [172]. A small perpendicular magnetic field is applied to break time reversal symmetry. The original fit to the data is shown along with our refit constraining only ak ≥ 8.1 and τφ = 0.39 ns (the value found from hgi). A.3 Value of γ in GaAs We include here a table of values of γ from both experiment and theory. 118 Appendix A: Appendix to Chapter 2 3 Table A.1: Magnitude of Dresselhaus spin-orbit coupling constant γ in GaAs (eVÅ ). All results are in bulk GaAs unless otherwise specified. Year Ref |γ| ’83 [102] 24.5 ’83 ’92 [8] [45] 20.9 26.1 ’93 [127] 23.5 ’94 ’95 ’96 [76] [77] [126] 34.5 16.5±3 11.0 ’03 [106] 28±4 ; 31±3 ’09 [86] 5.0 ’84 [36] 9 ; 8.5 ’84 ’85 ’86 [128] [16] [99] 19 ; 30 28 27.57 ; 21.29 ’88 ’88 ’90 ’92 [120] [27] [118] [154] 14.0 14.9 ; 28.2 24.12 18.3 ; 7.6 ; 36.3 ’93 ’94 [104] [132] 19.8 8.9 ’95 ’96 ’96 ’03 [133] [119] [83] [157] 8.5 24.21 10 ; 27.5 27.48 ’05 ’06 ’09 [71] [31] [97] 23.6 8.5 5.7 ; 8.3 Technique/Comments EXPERIMENT Optical orientation using Dyakonov-Perel (DP) spin relaxation time. Conduction electrons in p-type GaAs. Same as [102]. GaAs/AlGaAs heterostructure magnetoconductance measurements. Sets kz2 → 0 instead of hkz2 i. Raman scattering from 180Å GaAs quantum well (QW). Does not include Rashba term. Same as [127]. No Rashba term. Raman scattering from GaAs/AlGaAs heterostructure. Raman scattering from asymmetric GaAs/AlGaAs QW. More data than earlier. GaAs/AlGaAs heterostructure magnetoconductance measurements. GaAs/AlGaAs superlattice, transient spin grating measurement THEORY 14 × 14 k · p theory ; Linear Muffin Tin Orbitals (LMTO). Room temperature band gap. 8 × 8 ; 14 × 14 k · p models. 14 × 14 k · p. 14 × 14 k · p. Bulk GaAs ; Ga.65 Al.35 As/GaAs heterostructure. 14 × 14 k · p. LMTO ; 16 × 16 k · p. 14 × 14 k · p. 14 × 14 k · p in bulk GaAs; bulk Al0.35 Ga0.65 As ; GaAs/AlGaAs superlattice. k · p. sp3 s∗ tight binding (TB) model of a GaAs/AlAs superlattice. TB model of 100 Å GaAs/AlAs QW. 14 × 14 k · p. sp3 s∗ TB model ; k · p. 14 × 14 k · p theory. Notes that value is reduced to 19.6 in higher order perturbation theory (p. 74). TB model used to refine 14 × 14 k · p. Quasiparticle self-consistent GW method (QPscGW). Pseudopotentials. Appendix B Appendix to Chapter 3: Quaternions A quaternion q is a 2 × 2 matrix of complex numbers q = q (0) 112 + i 3 X q (µ) σµ , µ=1 where σµ are the Pauli matrices and q (µ) ∈ C. We define three conjugates of q: P q ∗ = q (0)∗ 112 + i q (µ)∗ σµ , P q R = q (0) 112 − i q (µ) σµ , q † = q R∗ complex conjugate: quaternion dual: and Hermitian conjugate: 119 (B.1) 120 Appendix B: Appendix to Chapter 3: Quaternions If the 2×2 matrix quaternion, a b c d is expressed as a quaternion q, then in the 2 × 2 notation for the q∗ = d∗ −c∗ −b∗ a∗ d −b qR = −c a ∗ ∗ a c q† = . b∗ d∗ (B.2) It is clear that q † corresponds to the usual definition of Hermitian conjugation, but complex conjugation is not equivalent. For a K × K matrix of quaternions Q we similarly define (Q∗ )ij = Qij ∗ , QR ij = QjiR , Q† = (Q∗ )R , (B.3) where again Hermitian conjugation of a K × K quaternion matrix corresponds to the usual Hermitian conjugation of the equivalent 2K × 2K complex matrix, but complex conjugation is not equivalent. By convention, the trace of the quaternion matrix Q is trQ = X (0) Qii , (B.4) i which accords with the usual definition of the trace of the equivalent complex matrix (since the Pauli matrices have zero trace), except that the quaternion trace is a factor of 2 smaller. Appendix C Appendix to Chapter 3: Spin polarization forbidden with M = Nφ = 1 Consider a quantum dot with N input channels, 1 output channel and 1 spinconserving dephasing channel. We show here that if we send an unpolarized incoherent current in the N channels and measure the spin polarization in the output channel, then as long as TRS is unbroken, there can be no spin polarization in the measured channel, independent of the transparency of the contacts. From Eq. 3.23, we have w = w0 + wφ where wφ = Pφ σµ tr(σν Pφ Sw0 S † )(Nφ δµν − Ξµν )−1 , (C.1) out ](µ) , where P th channel and gµs = 2tr(σµ PK wout ) = 2[wKK K is the projection onto the K (the output channel), and [q](µ) is the µ-component of the quaternion q. The outgoing density matrix is wout = w0out + wφout where w0out = Sw0 S † and 121 122 Appendix C: Appendix to Chapter 3: Spin polarization forbidden with M = Nφ = 1 wφout = Swφ S † . For convenience of notation, the outgoing channel has index K and the dephasing lead channel has index 1, so Pφ = P1 . Now w0 = 1 (11 − PK − P1 ), 2N (C.2) 1 ∗ ∗ (1 − SKK SKK − SK1 S1K ). 2N (C.3) so (w0out )KK = As in Sec. 3.2, the first two terms are real, but the third term can have a quaternion ∗ is exactly canceled by (wout ) component. We show that SK1 S1K KK in the final result for φ wout . We have tr(σν P1 Sw0 S † ) = hσ ν 2N i(0) ∗ ∗ (1 − S1K SK1 − S11 S11 ) (C.4) and ∗ , Ξµν = tr(σν P1 Sσµ P1 S † ) = δµν S11 S11 (C.5) where the last equality follows because S11 commutes with σµ , since S11 ∈ C. Together, these give wφ = P1 σµ σµ 2N (1 ∗ − S S ∗ ) (0) − S1K SK1 11 11 ∗ 1 − S11 S11 (C.6) with implied summation over µ. For q a quaternion, σµ [σµ q](0) = q, so P1 wφ = 2N ∗ S1K SK1 1− ∗ 1 − S11 S11 . (C.7) Then wφout KK 1 = 2N ∗ SK1 S1K ∗ S∗ SK1 S1K SK1 1K − ∗ 1 − S11 S11 . (C.8) ∗ in Eq. C.3. The second term is real, since S = S R , The first term in Eq. C.8 cancels SK1 S1K R ∈ C. Thus, wout is real and g~s = 0. Since this result relies only so SK1 S1K = SK1 SK1 KK Appendix C: Appendix to Chapter 3: Spin polarization forbidden with M = Nφ = 1 123 on the unitarity and self-duality of the S-matrix, it is true independent of the transparency of the dephasing contact. Note, however, that if Nφ > 1, then g~s can be nonzero in this theory, even when M = 1. Appendix D Appendix to Chapter 3: Spin polarization from dephasing Here we give explicit examples of dephasing-induced spin polarization in the case M = N = 1 for both the spin-conserving and spin-relaxing dephasing leads. The key to the operation of the dephasing leads is that they break coherence between different spin states. The spin-relaxing dephasing lead breaks the coherence between the different spins within a single channel and can thus produce a spin conductance with Nφ = 1. The spin-conserving dephasing lead does not break coherence between modes within a single channel and thus requires Nφ = 2 to produce a spin polarization. In the case of a spin-relaxing dephasing lead, consider an example of an S-matrix 124 Appendix D: Appendix to Chapter 3: Spin polarization from dephasing 125 in the complex representation with M = N = Nφ = 1: 0 0 1 S= 0 0 0 0 0 0 0 α α 0 α −α 0 1 0 0 0 0 0 0 , 0 0 α α α 0 0 0 −α 0 0 0 0 (D.1) √ where α = 1/ 2. An up spin incident from the left exits as an up spin to the right. A down spin incident from the left enters the dephasing lead as an x-polarized spin. It is reinjected as a dephased equal combination of up- and down-polarized spins, which then exit 50% as up spins to the left and 50% as down spins to the right. This results in g = 3/4, g~s = ẑ/4, and ~ p = ẑ/3. We can obtain this result formally, applying Eq. 3.27 to find wφ = Pφ /4. Note that we divide by 2 because we are using the complex rather than quaternion representation. Applying Eq. 3.4, g~s = ẑ/4. Note that if the dephasing lead had preserved the spin of its absorbed current, it would have reinjected x-polarized spins, which would have all exited as spin down to the right, giving g = 1 and g~s = 0, as Appendix C proves must happen. For the spin-conserving dephasing lead, consider a quantum dot with M = 1 = N 126 Appendix D: Appendix to Chapter 3: Spin polarization from dephasing and Nφ = 2 with complex S-matrix 0 0 0 0 0 0 1 0 0 0 0 0 S= 0 α α 0 0 0 0 −α α 0 0 0 0 0 −α 1 0 0 0 0 0 0 0 α 0 0 0 0 0 0 0 0 0 0 1 0 0 α 0 0 0 0 0 α . 0 0 −1 0 0 0 0 0 (D.2) Again, an up spin incident from the left exits as an up spin to the right. A down spin incident from the left enters the dephasing lead as a superposition of spin up in mode 1 and mode 2. This spin is conserved but the phase between the modes is broken. After reinjection, the current reflects back into the dephasing lead, this time as a superposition of spin down in both modes. Again the spin is conserved but dephased, and it exits as equal parts spin up to the left and spin down to the right, which again gives g = 3/4, g~s = ẑ/4, and ~ p = ẑ/3. Using Eq. 3.23, wφ = Pφ /4, just as in the previous example. Bibliography [1] A. Abragam. The Principles of Nuclear Magnetism. Oxford, 1962. [2] A. Abragam and M. Goldman. Principles of dynamic nuclear polarisation. Rep. Prog. Phys., 41(3):395–467, 1978. [3] I. Adagideli, J. H. Bardarson, and Ph Jacquod. Electrical probing of the spin conductance of mesoscopic cavities. J Phys : Condens Matter, 21(15):155503–, 2009. [4] K. A. Al-Hassanieh, V. V. Dobrovitski, E. Dagotto, and B. N. Harmon. Numerical modeling of the central spin problem using the spin-coherent-state p representation. Phys. Rev. Lett., 97(3):037204, 2006. [5] I. L. Aleiner and Vladimir I. Fal’ko. Spin-orbit coupling effects on quantum transport in lateral semiconductor dots. Phys. Rev. Lett., 87(25):256801, 2001. [6] T. Ando. Quantum point contacts in magnetic fields. Phys. Rev. B, 44(15):8017–8027, Oct 1991. [7] A. G. Aronov, Yu. B. Lyanda-Geller, and G. E. Pikus. Spin polarization of electrons by an electric current. Sov. Phys. JETP, 73:537–541, 1991. [8] A. G. Aronov, G. E. Pikus, and A. N. Titkov. Spin relaxation of conduction electrons in p-type III-V compounds. Sov. Phys. JETP, 57(3):680–687, 1983. [9] Harold U. Baranger and Pier A. Mello. Effect of phase breaking on quantum transport through chaotic cavities. Phys. Rev. B, 51(7):4703–4706, Feb 1995. [10] J. H. Bardarson, İ. Adagideli, and Ph. Jacquod. Mesoscopic spin Hall effect. Phys. Rev. Lett., 98(19):196601, 2007. [11] Jonathan Baugh, Yosuke Kitamura, Keiji Ono, and Seigo Tarucha. Large nuclear Overhauser fields detected in vertically coupled double quantum dots. Phys. Rev. Lett., 99(9):096804, 2007. [12] C. W. J. Beenakker. Random-matrix theory of quantum transport. Rev. Mod. Phys., 69(3):731–808, Jul 1997. [13] G. Bergmann. Weak anti-localization–an experimental proof for the destructive interference of rotated spin 1/2. Solid State Commun., 42(11):815–817, June 1982. 127 128 Bibliography [14] Jens Bolte and Stefan Keppeler. Semiclassical time evolution and trace formula for relativistic spin-1/2 particles. Phys. Rev. Lett., 81(10):1987–1991, Sep 1998. [15] A. S. Bracker, E. A. Stinaff, D. Gammon, M. E. Ware, J. G. Tischler, A. Shabaev, Al. L. Efros, D. Park, D. Gershoni, V. L. Korenev, and I. A. Merkulov. Optical pumping of the electronic and nuclear spin of single charge-tunable quantum dots. Phys. Rev. Lett., 94(4):047402, Feb 2005. [16] M Braun and U Rössler. Magneto-optic transitions and non-parabolicity parameters in the conduction band of semiconductors. Journal of Physics C: Solid State Physics, 18(17):3365–3377, 1985. [17] P. W. Brouwer and C. W. J. Beenakker. Effect of a voltage probe on the phasecoherent conductance of a ballistic chaotic cavity. Phys. Rev. B, 51(12):7739–7743, Mar 1995. [18] P. W. Brouwer and C. W. J. Beenakker. Diagrammatic method of integration over the unitary group, with applications to quantum transport in mesoscopic systems. J. Math. Phys., 37(10):4904–4934, 1996. [19] P. W. Brouwer and C. W. J. Beenakker. Voltage-probe and imaginary-potential models for dephasing in a chaotic quantum dot. Phys. Rev. B, 55(7):4695–4702, Feb 1997. [20] P. W. Brouwer and M. Büttiker. Charge-relaxation and dwell time in the fluctuating admittance of a chaotic cavity. Europhys. Lett., 37(7):441, 1997. [21] P. W. Brouwer, J. N. H. J. Cremers, and B. I. Halperin. Weak localization and conductance fluctuations of a chaotic quantum dot with tunable spin-orbit coupling. Phys. Rev. B, 65(8):081302(R), Feb 2002. [22] Guido Burkard, Daniel Loss, and David P. DiVincenzo. Coupled quantum dots as quantum gates. Phys. Rev. B, 59(3):2070–2078, Jan 1999. [23] M. Büttiker. Four-terminal phase-coherent conductance. 57(14):1761–1764, Oct 1986. Phys. Rev. Lett., [24] M. Büttiker. Coherent and sequential tunneling in series barriers. IBM J. Res. Dev., 32:63, 1988. [25] M. Büttiker, Y. Imry, R. Landauer, and S. Pinhas. Generalized many-channel conductance formula with application to small rings. Phys. Rev. B, 31(10):6207–6215, May 1985. [26] Yu A Bychkov and E I Rashba. Oscillatory effects and the magnetic susceptibility of carriers in inversion layers. J. Phys. C, 17(33):6039–6045, 1984. [27] M. Cardona, N. E. Christensen, and G. Fasol. Relativistic band structure and spinorbit splitting of zinc-blende-type semiconductors. Phys. Rev. B, 38(3):1806–1827, Jul 1988. Bibliography 129 [28] X. Cartoixa, L.-W. Wang, D.Z.-Y. Ting, and Y.-C. Chang. Higher-order contributions to Rashba and Dresselhaus effects. Phys. Rev. B, 73(20):205341, 2006. [29] Sudip Chakravarty and Albert Schmid. Weak localization: The quasiclassical theory of electrons in a random potential. Physics Reports, 140(4):193–236, July 1986. [30] I. H. Chan, R. M. Clarke, C. M. Marcus, K. Campman, and A. C. Gossard. Ballistic conductance fluctuations in shape space. Phys. Rev. Lett., 74(19):3876–3879, May 1995. [31] Athanasios N. Chantis, Mark van Schilfgaarde, and Takao Kotani. Ab initio prediction of conduction band spin splitting in zinc blende semiconductors. Phys. Rev. Lett., 96(8):086405, 2006. [32] A. V. Chaplik, M. V. Entin, and L. I. Magarill. Spin orientation of electrons by lateral electric field in 2D system without inversion symmetry. Physica E, 13(2-4):744–747, March 2002. [33] Gang Chen, Doron L. Bergman, and Leon Balents. Semiclassical dynamics and long-time asymptotics of the central-spin problem in a quantum dot. Phys. Rev. B, 76(4):045312, 2007. [34] S. Chesi and G. F. Giuliani. Spin polarization effects in a two dimensional mesoscopic electronic structure with Rashba spin-orbit and lateral confinement. cond-mat/0701415. [35] H. Christ, J. I. Cirac, and G. Giedke. Quantum description of nuclear spin cooling in a quantum dot. Phys. Rev. B, 75(15):155324, 2007. [36] N. E. Christensen and M. Cardona. Splitting of the conduction bands of GaAs for k along [110]. Solid State Comm., 51:491–493, 1984. [37] W. A. Coish and Daniel Loss. Hyperfine interaction in a quantum dot: Non-Markovian electron spin dynamics. Phys. Rev. B, 70(19):195340, 2004. [38] W. A. Coish and Daniel Loss. Singlet-triplet decoherence due to nuclear spins in a double quantum dot. Phys. Rev. B, 72(12):125337, Sep 2005. [39] Jan-Hein Cremers, Piet W. Brouwer, and Vladimir I. Fal’ko. Weak localization and conductance fluctuations in a quantum dot with parallel magnetic field and spin-orbit scattering. Phys. Rev. B, 68(12):125329, 2003. [40] Supriyo Datta. Electronic Transport in Mesoscopic Systems. Cambridge University Press, 1995. [41] Supriyo Datta and Biswajit Das. Electronic analog of the electro-optic modulator. Appl. Phys. Lett., 56(7):665–667, 1990. [42] John H. Davies. The Physics of Low-Dimensional Semiconductors: An Introduction. Cambridge University Press, 1993. 130 Bibliography [43] David C. Dixon, Keith R. Wald, Paul L. McEuen, and M. R. Melloch. Dynamic nuclear polarization at the edge of a two-dimensional electron gas. Phys. Rev. B, 56(8):4743–4750, Aug 1997. [44] G. Dresselhaus. Spin-orbit coupling effects in zinc blende structures. Phys. Rev., 100(2):580–586, Oct 1955. [45] P. D. Dresselhaus, C. M. A. Papavassiliou, R. G. Wheeler, and R. N. Sacks. Observation of spin precession in GaAs inversion layers using antilocalization. Phys. Rev. Lett., 68(1):106–109, Jan 1992. [46] Mathias Duckheim and Daniel Loss. Mesoscopic fluctuations in the spin-electric susceptibility due to Rashba spin-orbit interaction. Phys. Rev. Lett., 101:226602, 2008. [47] V. M. Edelstein. Spin polarization of conduction electrons induced by electric current in two-dimensional asymmetric electron systems. Solid State Commun., 73:233–235, 1990. [48] K. B. Efetov. Random matrices and supersymmetry in disordered systems. Applications of Random Matrices in Physics, edited by E. Brezin, V. Kazakov, P. Wiegmann, and A. Zabrodin, Springer, p. 95, 2006. [49] Hans-Andreas Engel, Emmanuel I. Rashba, and Bertrand I. Halperin. Handbook of Magnetism and Advanced Magnetic Materials, volume 5, pages 2858–2877. Wiley, 2007. [50] Hans-Andres Engel. unpublished. [51] Mikio Eto, Tetsuya Hayashi, and Yuji Kurotani. Spin polarization at semiconductor point contacts in absence of magnetic field. J. Phys. Soc. Jpn., 74:1934, 2005. [52] Qing feng Sun, X. C. Xie, and Jian Wang. Persistent spin current in nanodevices and definition of the spin current. Phys. Rev. B, 77(3):035327, 2008. [53] M. Field, C. G. Smith, M. Pepper, D. A. Ritchie, J. E. F. Frost, G. A. C. Jones, and D. G. Hasko. Measurements of coulomb blockade with a noninvasive voltage probe. Phys. Rev. Lett., 70(9):1311–1314, Mar 1993. [54] M. Fliesser, G. J. O. Schmidt, and H. Spohn. Magnetotransport of the Sinai billiard. Phys. Rev. E, 53(6):5690–5697, Jun 1996. [55] S. Foletti, H. Bluhm, D. Mahalu, V. Umansky, and A. Yacoby. Universal quantum control in two-electron spin quantum bits using dynamic nuclear polarization. unpublished. [56] J. A. Folk, S. R. Patel, S. F. Godijn, A. G. Huibers, S. M. Cronenwett, C. M. Marcus, K. Campman, and A. C. Gossard. Statistics and parametric correlations of coulomb blockade peak fluctuations in quantum dots. Phys. Rev. Lett., 76(10):1699–1702, Mar 1996. Bibliography 131 [57] Peter Forrester. Log-Gases and Random Matrices. unpublished, Ch. 2, 2005. [58] S.M. Frolov, A. Venkatesan, W. Yu, S. Luescher, W. Wegscheider, and J.A. Folk. Electrical generation of pure spin currents in a two-dimensional electron gas. Phys. Rev. Lett., 102:116802, 2009. [59] S.D. Ganichev, S.N. Danilov, Petra Schneider, V.V. Belkov, L.E. Golub, W. Wegscheider, D. Weiss, and W. Prettl. Electric current-induced spin orientation in quantum well structures. J. Magn. Magn. Mater., 300:127–131, 2006. [60] L. I. Glazman, G. B. Lesovik, D. E. Khmel’nitskii, and R. I. Shekhter. Reflectionless quantum transport and fundamental ballistic-resistance steps in microscopic constrictions. JETP Lett., 48:238, 1988. [61] M. Governale and U. Zülicke. Spin accumulation in quantum wires with strong Rashba spin-orbit coupling. Phys. Rev. B, 66(7):073311, Aug 2002. [62] M. Gullans, J. J. Krich, J. M. Taylor, B. I. Halperin, A. Yacoby, M. Stopa, and M. D. Lukin. in preparation. [63] B. I. Halperin, Ady Stern, Yuval Oreg, J. N. H. J. Cremers, J. A. Folk, and C. M. Marcus. Spin-orbit effects in a GaAs quantum dot in a parallel magnetic field. Phys. Rev. Lett., 86(10):2106–2109, Mar 2001. [64] R. Hanson, L. P. Kouwenhoven, J. R. Petta, S. Tarucha, and L. M. K. Vandersypen. Spins in few-electron quantum dots. Rev. Mod. Phys., 79(4):1217, 2007. [65] Thomas Heinzel. Mesoscopic Electronics in Solid State Nanostructures. Wiley, 2003. [66] Roger A. Horn and Charles R. Johnson. Matrix Analysis. Cambridge University Press, Cambridge, 1990. p. 69. [67] Zhian Huang and Liangbin Hu. Controllable kinetic magnetoelectric effect in twodimensional electron gases with both Rashba and Dresselhaus spin-orbit couplings. Phys. Rev. B, 73(11):113312, 2006. [68] A. G. Huibers, S. R. Patel, C. M. Marcus, P. W. Brouwer, C. I. Duruöz, and J. S. Harris. Distributions of the conductance and its parametric derivatives in quantum dots. Phys. Rev. Lett., 81(9):1917–1920, Aug 1998. [69] Y. Imry. Directions in Condensed Matter Physics. World Scientific, 1986. [70] J. D. Jackson. Classical Electrodynamics. Wiley, third edition, 1998. [71] J.-M. Jancu, R. Scholz, E. A. de Andrada e Silva, and G. C. La Rocca. Atomistic spin-orbit coupling and k dot p parameters in III-V semiconductors. Phys. Rev. B, 72(19):193201, 2005. [72] Yongjin Jiang and Liangbin Hu. Kinetic magnetoelectric effect in a two-dimensional semiconductor strip due to boundary-confinement-induced spin-orbit coupling. Phys. Rev. B, 74(7):075302, 2006. 132 Bibliography [73] Yongjin Jiang and Liangbin Hu. Symmetry properties of spin currents and spin polarizations in multiterminal mesoscopic spin-orbit-coupled systems. Phys. Rev. B, 75(19):195343, 2007. [74] Zhan-Feng Jiang, Jian Li, Shun-Qing Shen, and W. M. Liu. Spin polarized current in a junction of zigzag carbon nanotube. arxiv:0709.1566. [75] A. C. Johnson, J. R. Petta, J. M. Taylor, A. Yacoby, M. D. Lukin, C. M. Marcus, M. P. Hanson, and A. C. Gossard. Triplet-singlet spin relaxation via nuclei in a double quantum dot. Nature, 435(7044):925–928, June 2005. [76] B. Jusserand, D. Richards, H. Peric, and B. Etienne. Raman scattering on modulationdoped quantum wells: Intrinsic spin splitting of the GaAs conduction band. Surface Science, 305:247–250, 1994. [77] Bernard Jusserand, David Richards, Guy Allan, Catherine Priester, and Bernard Etienne. Spin orientation at semiconductor heterointerfaces. Phys. Rev. B, 51(7):4707– 4710, 1995. [78] Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Awschalom. Current-induced spin polarization in strained semiconductors. Phys. Rev. Lett., 93(17):176601, 2004. [79] J. M. Kikkawa and D. D. Awschalom. Lateral drag of spin coherence in gallium arsenide. Nature, 397(6715):139–141, January 1999. [80] A. A. Kiselev and K. W. Kim. T-shaped ballistic spin filter. Appl. Phys. Lett., 78(6):775–777, 2001. [81] A. A. Kiselev and K. W. Kim. Prohibition of equilibrium spin currents in multiterminal ballistic devices. Phys. Rev. B, 71(15):153315, 2005. [82] D. Klauser, W. A. Coish, and Daniel Loss. Nuclear spin dynamics and Zeno effect in quantum dots and defect centers. Phys. Rev. B, 78:205301, 2008. [83] W. Knap, C. Skierbiszewski, A. Zduniak, E. Litwin-Staszewska, D. Bertho, F. Kobbi, J. L. Robert, G. E. Pikus, F. G. Pikus, S. V. Iordanskii, V. Mosser, K. Zekentes, and Yu. B. Lyanda-Geller. Weak antilocalization and spin precession in quantum wells. Phys. Rev. B, 53(7):3912–3924, Feb 1996. [84] Avinash Kolli, Simon C. Benjamin, Jose Garcia Coello, Sougato Bose, and Brendon W. Lovett. Large spin entangled current from a passive device. New J. Phys., 11(1):013018, 2009. [85] F. H. L. Koppens, K. C. Nowack, and L. M. K. Vandersypen. Spin echo of a single electron spin in a quantum dot. Phys. Rev. Lett., 100(23):236802, 2008. [86] J. D. Koralek, C. P. Weber, J. Orenstein, B. A. Bernevig, Shou-Cheng Zhang, S. Mack, and D. D. Awschalom. Emergence of the persistent spin helix in semiconductor quantum wells. Nature, 458:610–613, 2009. Bibliography 133 [87] Alexey A. Kovalev, Liviu P. Zârbo, Y. Tserkovnyak, G. E. W. Bauer, and Jairo Sinova. Nanomechanical spin-polarizer. Phys. Rev. Lett., 101:036401, 2008. [88] Jacob J. Krich and Bertrand I. Halperin. Cubic Dresselhaus spin-orbit coupling in 2D electron quantum dots. Phys. Rev. Lett., 98(22):226802, 2007. [89] Jacob J. Krich and Bertrand I. Halperin. Spin-polarized current generation from quantum dots without magnetic fields. Phys. Rev. B, 78(3):035338, 2008. [90] C. W. Lai, P. Maletinsky, A. Badolato, and A. Imamoglu. Knight-field-enabled nuclear spin polarization in single quantum dots. Phys. Rev. Lett., 96(16):167403, 2006. [91] R. Landauer. Electrical resistance of disordered one-dimensional lattices. Phil. Mag., 21:863, 1970. [92] V. Lechner, L. E. Golub, P. Olbrich, S. Stachel, D. Schuh, W. Wegscheider, V. V. Bel’kov, and S. D. Ganichev. Tuning of structure inversion asymmetry by the δ-doping position in (001)-grown GaAs quantum wells. tt arxiv:0903.1232. [93] L. S. Levitov, Yu. V. Nazarov, and G. M. Éliashberg. Magnetoelectric effects in conductors with mirror isomer symmetry. Sov. Phys. JETP, 61(1):133–137, 1985. [94] Jun-Feng Liu, Zhi-Cheng Zhong, Lei Chen, Dingping Li, Chao Zhang, and Zhongshui Ma. Enhancement of polarization in a spin-orbit coupling quantum wire with a constriction. Phys. Rev. B, 76:195304, 2007. [95] Daniel Loss and David P. DiVincenzo. Quantum computation with quantum dots. Phys. Rev. A, 57(1):120–126, Jan 1998. [96] Xiaohua Lou, Christoph Adelmann, Scott A. Crooker, Eric S. Garlid, Jianjie Zhang, K. S. Madhukar Reddy, Soren D. Flexner, Chris J. Palmstrom, and Paul A. Crowell. Electrical detection of spin transport in lateral ferromagnet-semiconductor devices. Nat Phys, 3(3):197–202, March 2007. [97] Jun-Wei Luo, Gabriel Bester, and Alex Zunger. Full-zone spin splitting for electrons and holes in bulk GaAs and GaSb. Physical Review Letters, 102(5):056405, 2009. [98] I. Malajovich, J. M. Kikkawa, D. D. Awschalom, J. J. Berry, and N. Samarth. Coherent transfer of spin through a semiconductor heterointerface. Phys. Rev. Lett., 84(5):1015–1018, Jan 2000. [99] F. Malcher, G. Lommer, and U. Rössler. Electron states in GaAs/Ga1-xAlxAs heterostructures: Nonparabolicity and spin-splitting. Superlattices Microstruct., 2(3):267–272, 1986. [100] A.G. Mal’shukov and C.S.Chu. Spin orientation and spin-Hall effect induced by tunneling electrons. Phys. Rev. B, 76:245326, 2007. [101] Michael P. Marder. Condensed Matter Physics. Wiley, 2000. 134 Bibliography [102] V. A. Marushchak, M. N. Stepanova, and A. N. Titkov. Spin relaxation of conduction electrons in moderately doped gallium arsenide crystals. Sov. Phys. Solid State, 25(12):2035–2038, 1983. [103] D. P. Mattis. Theory of Magnetism. Springer Series in Solid-State Sciences No. 17, Vol. I, Sec. 4.5. Springer, New York, 1988. [104] H. Mayer and U. Rössler. Nonparabolicity in the conduction band of II-VI semiconductors. Solid State Comm., 87:81–84, 1993. [105] Madan Lal Mehta. Random Matrices. Elsevier, 2004. [106] J. B. Miller, D. M. Zumbühl, C. M. Marcus, Y. B. Lyanda-Geller, D. GoldhaberGordon, K. Campman, and A. C. Gossard. Gate-controlled spin-orbit quantum interference effects in lateral transport. Phys. Rev. Lett., 90(7):076807, 2003. [107] Branislav K. Nikolić and Satofumi Souma. Decoherence of transported spin in multichannel spin-orbit-coupled spintronic devices: Scattering approach to spin-density matrix from the ballistic to the localized regime. Phys. Rev. B, 71(19):195328, 2005. [108] P. Nozières and C. Lewiner. J. Phys. (Paris), 34:901, 1973. [109] Philippe Nozieres and David Pines. Theory of Quantum Liquids. Westview Press, 1999. [110] W. D. Oliver, F. Yamaguchi, and Y. Yamamoto. Electron entanglement via a quantum dot. Phys. Rev. Lett., 88(3):037901, Jan 2002. [111] K. Ono, D. G. Austing, Y. Tokura, and S. Tarucha. Current rectification by pauli exclusion in a weakly coupled double quantum dot system. Science, 297(5585):1313– 1317, August 2002. [112] Keiji Ono and Seigo Tarucha. Nuclear-spin-induced oscillatory current in spinblockaded quantum dots. Phys. Rev. Lett., 92(25):256803, Jun 2004. [113] D. Paget, G. Lampel, B. Sapoval, and V. I. Safarov. Low field electron-nuclear spin coupling in gallium arsenide under optical pumping conditions. Phys. Rev. B, 15(12):5780–5796, Jun 1977. [114] Daniel Paget. Optical detection of NMR in high-purity GaAs: Direct study of the relaxation of nuclei close to shallow donors. Phys. Rev. B, 25(7):4444–4451, Apr 1982. [115] C A Perroni, D Bercioux, V Marigliano Ramaglia, and V Cataudella. Rashba quantum wire: exact solution and ballistic transport. J. Phys.: Condens. Matter, 19(18):186227, 2007. [116] 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. Coherent manipulation of coupled electron spins in semiconductor quantum dots. Science, 309(5744):2180–2184, September 2005. Bibliography 135 [117] J. R. Petta, J. M. Taylor, A. C. Johnson, A. Yacoby, M. D. Lukin, C. M. Marcus, M. P. Hanson, and A. C. Gossard. Dynamic nuclear polarization with single electron spins. Phys. Rev. Lett., 100(6):067601, 2008. [118] P. Pfeffer and W. Zawadzki. Conduction electrons in GaAs: Five-level k·p theory and polaron effects. Phys. Rev. B, 41(3):1561–1576, Jan 1990. [119] P. Pfeffer and W. Zawadzki. Five-level k·p model for the conduction and valence bands of GaAs and InP. Phys. Rev. B, 53(19):12813–12828, May 1996. [120] G. E. Pikus, V. A. Marushchak, and A. N. Titkov. Spin splitting of energy bands and spin relaxation of carriers in cubic III-V cystals (review). Sov. Phys. Semicond., 22:115–124, 1988. [121] Guy Ramon and Xuedong Hu. Dynamical nuclear spin polarization and the Zamboni effect in gated double quantum dots. Phys. Rev. B, 75(16):161301, 2007. [122] E. I. Rashba. Theory of electrical spin injection: Tunnel contacts as a solution of the conductivity mismatch problem. Phys. Rev. B, 62(24):R16267–R16270, Dec 2000. [123] D. J. Reilly, J. M. Taylor, J. R. Petta, C. M. Marcus, M. P. Hanson, and A. C. Gossard. Exchange control of nuclear spin diffusion in a double quantum dot. arxiv:0803.3082. [124] D. J. Reilly, J. M. Taylor, J. R. Petta, C. M. Marcus, M. P. Hanson, and A. C. Gossard. Suppressing spin qubit dephasing by nuclear state preparation. Science, 321:781, 2008. [125] Hugo Ribeiro and Guido Burkard. Nuclear state preparation via Landau-ZenerStueckelberg transitions in double quantum dots. Phys. Rev. Lett., 102:216802, 2009. [126] D. Richards, B. Jusserand, G. Allan, C. Priester, and B. Etienne. Electron spinflip Raman scattering in asymmetric quantum wells: Spin orientation. Solid-State Electron., 40:127–131, 1996. [127] D. Richards, B. Jusserand, H. Peric, and B. Etienne. Intrasubband excitations and spin-splitting anisotropy in GaAs modulation-doped quantum wells. Phys. Rev. B, 47(23):16028–16031, Jun 1993. [128] U. Rössler. Nonparabolicity and warping in the conduction band of GaAs. Solid State Comm., 49:943–947, 1984. [129] J. J. Sakurai. Advanced Quantum Mechanics. Addison-Wesley, 1967. p. 68. [130] J. J. Sakurai. Modern Quantum Mechanics. Addison-Wesley, 1994. [131] G. Salis, D. T. Fuchs, J. M. Kikkawa, D. D. Awschalom, Y. Ohno, and H. Ohno. Optical manipulation of nuclear spin by a two-dimensional electron gas. Phys. Rev. Lett., 86(12):2677–2680, Mar 2001. 136 Bibliography [132] Paulo V. Santos and M. Cardona. Comment on observation of spin precession in GaAs inversion layers using antilocalization. Phys. Rev. Lett., 72(3):432, Jan 1994. [133] Paulo V. Santos, M. Willatzen, M. Cardona, and A. Cantarero. Tight-binding calculation of spin splittings in semiconductor superlattices. Phys. Rev. B, 51(8):5121–5129, Feb 1995. [134] Daniel S. Saraga and Daniel Loss. Spin-entangled currents created by a triple quantum dot. Phys. Rev. Lett., 90(16):166803, 2003. [135] Matthias Scheid, Andreas Pfund, Dario Bercioux, and Klaus Richter. Coherent spin ratchets: A spin-orbit based quantum ratchet mechanism for spin-polarized currents in ballistic conductors. Phys. Rev. B, 76(19):195303, 2007. [136] John Schliemann, J. Carlos Egues, and Daniel Loss. Nonballistic spin-field-effect transistor. Phys. Rev. Lett., 90(14):146801, 2003. [137] Ehud Shafir, Min Shen, and Semion Saikin. Modulation of spin dynamics in a channel of a nonballistic spin field effect transistor. Phys. Rev. B, 70(24):241302, 2004. [138] Prashant Sharma and Piet W. Brouwer. Mesoscopic effects in adiabatic spin pumping. Phys. Rev. Lett., 91(16):166801, Oct 2003. [139] A. Yu. Silov, P. A. Blajnov, J. H. Wolter, R. Hey, K. H. Ploog, and N. S. Averkiev. Current-induced spin polarization at a single heterojunction. Appl. Phys. Lett., 85(24):5929–5931, 2004. [140] P. G. Silvestrov and E. G. Mishchenko. Polarized electric current in semiclassical transport with spin-orbit interaction. Phys. Rev. B, 74(16):165301, 2006. [141] D. Stich, J. Zhou, T. Korn, R. Schulz, D. Schuh, W. Wegscheider, M. W. Wu, and C. Schüller. Effect of initial spin polarization on spin dephasing and the electron g factor in a high-mobility two-dimensional electron system. Phys. Rev. Lett., 98(17):176401, 2007. [142] M. Stopa. unpublished. [143] M. Stopa, J. J. Krich, and A. Yacoby. Inhomogeneous nuclear spin flips. May 2009. arxiv:0905.4520. [144] M. Studer, G. Salis, K. Ensslin, D. C. Driscoll, and A. C. Gossard. Gate-controlled spin-orbit interaction in a parabolic GaAs/AlGaAs quantum well. arxiv:0903.0920. [145] Hong X. Tang, F. G. Monzon, Friso J. Jedema, Adnrei T. Filip, Bart J. van Wees, and Michael L. Roukes. Semiconductor Spintronics and Quantum Computation, chapter 2, page 68. Springer-Verlag, 2002. [146] J. M. Taylor, J. R. Petta, A. C. Johnson, A. Yacoby, C. M. Marcus, and M. D. Lukin. Relaxation, dephasing, and quantum control of electron spins in double quantum dots. Phys. Rev. B, 76(3):035315, 2007. Bibliography 137 [147] Jacob M. Taylor. Hyperfine interactions and quantum information processing in quantum dots. PhD thesis, Harvard University, 2006. [148] Maxim Trushin and John Schliemann. Anisotropic current-induced spin accumulation in the two-dimensional electron gas with spin-orbit coupling. Phys. Rev. B, 75(15):155323, 2007. [149] B. J. van Wees, H. van Houten, C. W. J. Beenakker, J. G. Williamson, L. P. Kouwenhoven, D. van der Marel, and C. T. Foxon. Quantized conductance of point contacts in a two-dimensional electron gas. Phys. Rev. Lett., 60(9):848–850, Feb 1988. [150] L. M. K. Vandersypen and I. L. Chuang. NMR techniques for quantum control and computation. Rev. Mod. Phys., 76(4):1037–1069, Jan 2005. [151] Martin Y. Veillette, Cristina Bena, and Leon Balents. Spin precession and oscillations in mesoscopic systems. Phys. Rev. B, 69(7):075319, Feb 2004. [152] J. J. M. Verbaarschot, H. A. Weidenmuller, and M. R. Zirnbauer. Grassmann integration in stochastic quantum physics: The case of compound-nucleus scattering. Physics Reports, 129(6):367–438, December 1985. [153] N. V. Vitanov and B. M. Garraway. Landau-Zener model: Effects of finite coupling duration. Phys. Rev. A, 53(6):4288–4304, Jun 1996. [154] Paul von Allmen. Conduction subbands in a GaAs/Alx Ga1−x As quantum well: Comparing different k·p models. Phys. Rev. B, 46(23):15382–15386, Dec 1992. [155] Claude Weisbuch and Claudine Hermann. Optical detection of conduction-electron spin resonance in GaAs, Ga1−x Inx As, and Ga1−x Alx As. Phys. Rev. B, 15(2):816–822, Jan 1977. [156] E. P. Wigner. On the statistical distribution of the widths and spacings of nuclear resonance levels. Mathematical Proceedings of the Cambridge Philosophical Society, 47:790–798, 1951. [157] Roland Winkler. Spin-Orbit Coupling Effects in Two-Dimensional Electon and Hole Systems. Springer, Berlin, 2003. [158] W. M. Witzel and S. Das Sarma. Quantum theory for electron spin decoherence induced by nuclear spin dynamics in semiconductor quantum computer architectures: Spectral diffusion of localized electron spins in the nuclear solid-state environment. Phys. Rev. B, 74(3):035322, 2006. [159] W. M. Witzel and S. Das Sarma. Wavefunction considerations for the central spin decoherence problem in a nuclear spin bath. Phys. Rev. B, 77(16):165319, 2008. [160] S. A. Wolf, D. D. Awschalom, R. A. Buhrman, J. M. Daughton, S. von Molnar, M. L. Roukes, A. Y. Chtchelkanova, and D. M. Treger. Spintronics: A spin-based electronics vision for the future. Science, 294(5546):1488–1495, November 2001. 138 Bibliography [161] Yanxia Xing, Qing-feng Sun, Liang Tang, and Jiangping Hu. Opposite spin accumulations on the transverse edges by the confining potential. Phys. Rev. B, 74:155313, 2006. [162] A. Yacoby and Y. Imry. Quantization of the conductance of ballistic point contacts beyond the adiabatic approximation. Phys. Rev. B, 41(8):5341–5350, Mar 1990. [163] W. Yang and Kai Chang. Nonlinear Rashba model and spin relaxation in quantum wells. Phys. Rev. B, 74(19):193314, 2006. [164] Wang Yao. ‘Measurement-free’ quantum feedback control of nuclear field squeezing in quantum dot. 2009. arxiv:0905.2460. [165] Wang Yao, Ren-Bao Liu, and L. J. Sham. Theory of electron spin decoherence by interacting nuclear spins in a quantum dot. Phys. Rev. B, 74(19):195301, 2006. [166] Go Yusa, Koji Muraki, Kei Takashina, Katsushi Hashimoto, and Yoshiro Hirayama. Controlled multiple quantum coherences of nuclear spins in a nanometre-scale device. Nature, 434(7036):1001–1005, April 2005. [167] Oleg Zaitsev, Diego Frustaglia, and Klaus Richter. Role of orbital dynamics in spin relaxation and weak antilocalization in quantum dots. Phys. Rev. Lett., 94(2):026809, 2005. [168] Feng Zhai, Kai Chang, and H. Q. Xu. Spin current diode based on an electron waveguide with spin-orbit interaction. Appl. Phys. Lett., 92(10):102111, 2008. [169] Feng Zhai and H. Q. Xu. Symmetry of spin transport in two-terminal waveguides with a spin-orbital interaction and magnetic field modulations. Phys. Rev. Lett., 94(24):246601, 2005. [170] D. M. Zumbühl. personal communication. [171] D. M. Zumbühl, J. B. Miller, C. M. Marcus, K. Campman, and A. C. Gossard. Spinorbit coupling, antilocalization, and parallel magnetic fields in quantum dots. Phys. Rev. Lett., 89(27):276803, Dec 2002. [172] D. M. Zumbühl, J. B. Miller, C. M. Marcus, D. Goldhaber-Gordon, Jr. J. S. Harris, K. Campman, and A. C. Gossard. Conductance fluctuations and partially broken spin symmetries in quantum dots. Phys. Rev. B, 72(8):081305(R), 2005.