Spin-orbit coupling and strong correlations in ultracold Bose gases
DISSERTATION
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University
By William S. Cole, Jr., B.S. Graduate Program in Physics
The Ohio State University 2014
Dissertation Committee: Professor Nandini Trivedi, Advisor
Professor Mohit Randeria
Professor P. Chris Hammel
Professor Richard J. Furnstahl c Copyright by
William S. Cole, Jr.
2014 Abstract
The ability to create artificial gauge fields for neutral atoms adds a powerful new dimension to the idea of using ultracold atomic gases as “quantum simulators” of models that arise in conventional solid state physics. At present, cold atom experiments are able to simulate orbital magnetism and a certain kind of spin-orbit coupling at scales which are quite difficult to achieve in the solid state. This takes us beyond the realm of simulation into questions about new states of matter which might only be possible in cold atom experiments. Indeed, as a byproduct of finding methods to simulate traditional gauge potentials for neutral gases, it has been realized that gauge potentials with no solid state analog can also be created and finely tuned to manipulate the few- and many-body physics of bosons and fermions in remarkable new ways.
Motivated along these lines, in this dissertation I address several issues related to bosons with spin-orbit coupling. After an introduction to synthetic gauge fields in general I describe
Rashba spin-orbit coupling specifically and how it modifies the usual behavior of bosons in the continuum and in a harmonic trapping potential. Following this, I study the effects of interactions induced by combining spin-orbit coupling with an optical lattice potential. I use a variety of theoretical tools to characterize and classify the exotic phases which emerge.
In the weakly interacting limit, mean-field theory suggests Bose-Einstein condensed states with the possibility of magnetically ordered phases arising through interference between macroscopically occupied single particle states. In the deep lattice limit, the ground state is a Mott insulator with magnetic order driven by bosonic superexchange. I address the Mott transition, as well as the nature of superfluid states near the Mott transition, and introduce a novel “slave-boson” approach to understand these results. Finally, using some of the
ii insight developed for this general case, I restrict my attention to a more experimentally relevant implementation of SOC, where I can use numerically exact calculations to justify the mean-field results.
In the final Chapter, I address a problem which perhaps seems, at first glance, to be unrelated to bosons with Rashba SOC. The problem is identifying the ground states of spinless hardcore bosons on a frustrated honeycomb lattice. The connection between these problems is the delicate question of how bosons condense in a flat band. In that section I provide numerical support to a recently proposed scenario of “statistical transmutation,” where an approximate ground state of the hard core bosons can be obtained from a mean-
field approximation in terms of composite fermions.
iii Acknowledgments
First and foremost, I am deeply thankful to my advisor Nandini Trivedi for her guidance along my winding Ph.D. path. Her enthusiasm for physics and her breadth of interests are inspiring. I would also like to give special thanks to Mohit Randeria, who often served a similar role, and together the examples they have set as teachers and mentors I can only hope to emulate.
I am also particularly indebted to Arun Paramekanti and Shizhong Zhang for their collaboration on much of the work presented in this dissertation, and for enlightening dis- cussions over the past few years.
The condensed matter community at Ohio State has been a stimulating, nurturing environment, full of talented post-docs and students from whom I have learned a great deal. Special thanks go to Yen Lee Loh, who helped me to really get my bearings in the folklore of condensed matter theory. Also I must thank Eric Duchon and Weiran Li for being ideal officemates, along with several other fellow students (Onur, Nganba, Mason,
Tim, Daniel, Jim, John) who have helped to guide my own development as a physicist.
That this thesis was completed at all is a great credit to my wife Alexis. Her unfailing support and encouragement have kept me afloat through every challenge. I have also been blessed with a wonderful, loving family who sparked my initial interest in math and science and continue to fan the flames.
iv Vita
May, 2008 ...... B.S. in Physics, University of Central Florida, Orlando, FL September, 2008 – present ...... Graduate Teaching and Research Assis- tant, and Ohio State Presidential Fellow, the Ohio State University, Columbus, OH
Publications
William S. Cole, Shizhong Zhang, Arun Paramekanti, Nandini Trivedi, Bose Hubbard Models with Synthetic Spin-Orbit Coupling: Mott Insulators, Spin Textures and Superfluidity, Phys. Rev. Lett. 109 085302 (2012)
Anamitra Mukherjee, William S. Cole, Nandini Trivedi, Mohit Randeria, Patrick Wood- ward, Theory of strain controlled magnetotransport and stabilization of the ferromagnetic insulating phase in manganite thin films, Phys. Rev. Lett. 110 157201 (2013)
O. Nganba Meetei, William S. Cole, Mohit Randeria, Nandini Trivedi, Novel magnetic state in d4 Mott insulators, arXiv:1311.2823
Zhihao Xu, William Cole, Shizhong Zhang, Mott-superfluid transition for spin-orbit-coupled bosons in one-dimensional optical lattices, Phys. Rev. A 89 051604(R) (2014)
William S. Cole, Nandini Trivedi, Statistical transmutation of hard core bosons on the frus- trated honeycomb lattice, in preparation
Shizhong Zhang, William S. Cole, Arun Paramekanti, Nandini Trivedi, Synthetic gauge fields in optical lattices, in preparation for Annual Reviews of Cold Atoms and Molecules
Fields of Study
v Major Field: Physics
vi Table of Contents
Page Abstract...... ii Acknowledgments...... iv Vita...... v List of Figures ...... ix List of Tables ...... x
Chapters
1. Introduction, motivations, and outline ...... 1 1.1 Spin-orbit coupling...... 2 1.2 Ultracold bosons ...... 4 1.2.1 The Bose-Hubbard model...... 5 1.3 Synthetic gauge fields ...... 6 1.3.1 Berry’s connection and the adiabatic principle ...... 8 1.3.2 Raman induced spin-orbit coupling...... 12 1.3.3 “Realistic” Rashba spin-orbit coupling...... 15
2. Exotic Bose matter from nonabelian gauge fields...... 18 2.1 Bose-Einstein condensation with spin, and with spin-orbit coupling . . . . . 18 2.2 The Rashba hamiltonian...... 21 2.2.1 Cylindrical coordinates ...... 23 2.2.2 The Rashba hamiltonian with a harmonic trapping potential . . . . 26 2.3 Adding interactions...... 27 2.4 An exact two-body ground state ...... 29
3. The Bose-Hubbard model with spin-orbit coupling...... 32 3.1 Spin-orbit coupling in a periodic optical potential...... 34 3.2 Model hamiltonian...... 36 3.3 Weak-coupling approximation...... 36 3.4 Strong-coupling approximation ...... 42 3.5 Mean-field theory of the superfluid-insulator transition...... 51 3.5.1 Description of the mean-field method...... 51 3.5.2 Results ...... 54 3.6 Slave-boson theory...... 56 3.7 Exact numerics in the one-dimensional limit...... 60
vii 4. Hard core bosons on the frustrated honeycomb lattice ...... 63 4.1 Hard core bosons and Jordan-Wigner fermions ...... 63 4.2 Model and methods ...... 67 4.3 Comparing the ground-state energies of hard core bosons and fermions . . . 70 4.4 Monte Carlo results ...... 72 4.4.1 Variational bounds on the ground state energy...... 72 4.4.2 One-body density matrix ...... 72 4.5 Discussion and conclusions...... 73
Bibliography...... 75
Appendices
A. Numerical solution of the Rashba hamiltonian in a harmonic trap . . . 83 A.1 Series expansion in harmonic trap solutions...... 84
B. Effective spin hamiltonian from two-site perturbation theory ...... 90
C. The variational principle and variational Monte Carlo...... 94 C.1 Application of VMC to fermionized hamiltonians...... 97
viii List of Figures
Figure Page
1.1 Mean-field phase diagram of the Bose-Hubbard model...... 7 1.2 Schematic implementation of the Raman scheme for spin-orbit coupling . . 13 1.3 Spectrum of the hamiltonian for Raman-induced one-dimensional spin-orbit coupling...... 15
2.1 Energy spectrum of the Rashba hamiltonian...... 24 2.2 Energy spectrum of the Rashba hamiltonian in a harmonic trap ...... 27 2.3 Spatial variation of the eigenstates of the Rashba hamiltonian in a harmonic trap ...... 28
3.1 Energy spectrum of a single particle in a 2D square optical lattice with spin orbit coupling...... 38 3.2 Density of states and lower-band spin wavefunction for a Rashba coupled particle in a 2D square optical lattice...... 39 3.3 Weak-coupling energy expectation values of various ordered condensate states 41 3.4 Spin densities of various ordered condensate states...... 41 3.5 Illustration of frustration for compass model and Dzyaloshinski-Moriya in- teractions...... 46 3.6 Magnetic phase diagram in the deep Mott limit...... 47 3.7 Magnetic order in Mott insulating states of lattice bosons with spin-orbit coupling...... 48 3.8 Variational comparison of several classical spin states...... 50 3.9 Mean-field Mott lobes in the Rashba coupled Bose-Hubbard model . . . . . 55 3.10 Mean-field solutions to the Rashba coupled Bose-Hubbard model ...... 56 3.11 DMRG results for the Bose-Hubbard model with abelian spin-orbit coupling 62
4.1 The honeycomb lattice model...... 67 4.2 Comparison of fermion and boson ground state energies...... 71 4.3 Variational Monte Carlo energy and variance for the VMFT state ...... 73 4.4 Condensate fraction in the VMFT state...... 74
ix List of Tables
Table Page
3.1 Virtual state matrix elements for calculating the low-energy effective spin hamiltonian in the Mott insulator...... 43
x Chapter 1 Introduction, motivations, and outline
One of the primary motivations for studying ultracold atomic gases is the notion of quan- tum simulation. Remarkably rich phenomena, like magnetism and superconductivity, are encoded into the solutions of the familiar non-relativistic Schr¨odingerequation; however, analytic and even numerical estimates for physically meaningful quantities are often avail- able only approximately or else in very special limiting cases. Even minimalistic “effective” models are notoriously hard to solve when there are many interacting degrees of freedom.
On the other hand, with the unprecedented control that experimentalists have achieved over dilute gases of ultracold atoms, interesting model hamiltonians can now be engineered in the laboratory and their properties explored directly without recourse to approximations, across wide parameter regimes [1].
The typical starting point for most many-body theories is an effectively non-interacting
(i.e., mean-field) limit where the properties of a single quasiparticle are well understood.
p2 Deviations from the simplest single-particle behavior, H = 2m∗ , however, can have dramatic effects on the many-body physics. The origin of such deviations could be, for example, the presence of a periodic potential, which in turn gives rise to a band spectrum where certain energies are prohibited. In the cold atoms context, a static periodic potential can be created with lasers – the so-called “optical lattice.” In addition, substantial progress has recently been made in driving the occupation of higher energy Wannier states of the individual potential wells of an optical lattice [2,3], simulating the spatially anisotropic
1 p and d orbitals occupied by real electrons in solids. Other typical sources of nontrivial single-body physics are orbital magnetism, which gives rise to the Landau level spectrum, and various flavors of spin-orbit coupling (SOC) which generically give rise to Dirac points and spin-momentum locking within the single-particle states.
While optical lattices are now a standard tool in cold atom experiments, the implemen- tation of orbital magnetism and spin-orbit coupling are both very new and active areas of current research. Both of these phenomena arise in conventional solid-state experiments as a direct consequence of the electric charge carried by the electron and the coupling of that charge to electric and magnetic fields. Since cold atom experiments are performed using neutral atoms, one must find clever ways to artificially endow these atoms with the effects of minimal coupling to a gauge field, but obviously with a substantially different underlying cause. The key insight here will prove to be the use of Berry’s “geometric phase.” From this perspective, it will also turn out to be quite natural to view spin-orbit coupling in an identical framework, arising from the imprinting of spin-carrying particles with a nonabelian geometric phase [4].
In the following chapters, I narrow my focus to exploring the consequences of large spin-orbit coupling for interacting ultracold bosons, with and without an additional optical lattice potential. In the remainder of this introduction, however, I will describe why spin- orbit coupling and Bose gases are interesting, separately, and will also introduce the essential notions that lay the foundation for engineering synthetic gauge fields for neutral atoms.
Upon combining these ingredients, some of the usual effects of spin-orbit coupling survive
(such as the spin Hall effect), however, the underlying Bose statistics lead to states of matter that are quite foreign to the intuition developed either for solid-state spin-orbit coupled Fermi systems or for the conventional spinless Bose superfluids.
1.1 Spin-orbit coupling
Traditionally, spin-orbit coupling is understood as a consequence of special relativity. It appears in the first-order (in v/c) relativistic correction to the Schr¨odingerequation for an
2 electron [5], −|e| v Hrel = × E · S (1.1) 2mec c where, intuitively, the spin S of the electron (with charge e = −|e|) will precess as it moves in a static electric field E, because under a Lorentz transformation to the particle’s rest
0 v frame, that electric field becomes a magnetic field B = − c × E, which in turn yields a Zeeman coupling to the spin, ∝ B0 ·S. In the fully relativistic Dirac equation, this coupling between spin and spatial motion is made manifest
2 HDirac = cα · p + βmc (1.2) where (α, β) are a set of 4 × 4 matrices which act in the space of internal states for the electron (two spin projections, particle and antiparticle projections).
Na¨ıvely, one would expect relativistic corrections to be essentially negligible in the low- energy, nonrelativistic limit where condensed matter physics is typically studied. However, there are a variety of situations where the effects of spin-orbit coupling are simply unavoid- able. The familiar L · S coupling of atomic physics arises from the spherically symmetric
Coulomb field that binds electrons to positively charged nuclei. The relativistic nature of the atomic “fine-structure” is manifest in the scaling of the level splittings as ∆E ∼ (Zα)4
[5], with α ∼ 1/137 the fine structure constant. However, the presence of the atomic num- ber Z means that these splittings can easily be of the same order as the bandwidth in narrow-band materials that include heavy elements, for example the iridium (ZIr = 77) and osmium (ZOs = 76) oxides. Furthermore, in a bulk crystalline lattice, electrons can experience a nonzero average gradient of the electric potential. This can only happen in crystals that lack inversion symmetry (noncentrosymmetric lattices) and is called Dresselhaus spin-orbit coupling [6,7].
Here, it is the electron’s crystal momentum k, rather than orbital angular momentum, that couples to the spin. Similar to this is the Rashba effect, again related to an explicitly broken inversion symmetry, that in this case occurs at the surface of a crystal or at an interface between two different materials [8]. Rashba proposed this effect in the context of the two-
3 dimensional electron gas in semiconductor quantum wells, although it is also an important ingredient for the electronic structure at the interface between bulk LaAlO3 and SrTiO3 [9]. In recent years, interest in these materials with strong spin-orbit coupling has been on the rise. This has been driven both by the desire to use spin-orbit coupling as a resource in the blooming field of spintronics, where it is a crucial ingredient for new circuit devices such as spin transistors [10], as well as an appreciation of the fundamental role that spin-orbit coupling plays in stabilizing new exotic states of matter, for example, the quantum spin
Hall insulators and topological insulators [11].
1.2 Ultracold bosons
Condensed matter physics abounds with examples of macroscopic quantum phenomena.
The usual examples are the dramatic and closely-related phenomena of superconductivity and Bose-Einstein condensation (BEC), but even simple metals can be well understood by starting from an ideal gas of explicitly quantum mechanical fermions. These facts indeed hint at the notion that perhaps the most substantial effects imposed by quantum mechanics on condensed matter systems already appear at the level of exchange statistics, that is, the fundamental constraints satisfied by the many-body wavefunction of identical particles when those particles are exchanged.
Of course, the vast majority of matter that we might be interested in studying is made of fermions, whether we are considering the electron gases in metals and semiconductors or the protons and neutrons in nuclei or even the quarks that make up those nucleons.
One might naturally wonder what “boson matter” even means. The distinction between fermions and bosons first appears in the two possibilities for how the wavefunction of several indistinguishable particles changes when the coordinates of the particles are permuted. If we were to take two hydrogen atoms and swap their positions then the wavefunction necessarily accumulates a minus sign from swapping the two electrons, as well as a minus sign from swapping the two protons, meaning that in total there is no sign change, as expected for bosons. Each hydrogen atom is made up of two (distinguishable) fermions bound together
4 with an energy of 1 Ry, so unless we probe the atom at that energy scale (or on length scales shorter than the Bohr length), the electron and proton do not have separate identities.
Rather than working in terms of the electron and proton coordinates individually, then, it is much more useful to work with the center of mass coordinate and the specific set of quantum numbers that the electron possesses as it sits in the Coulomb potential of the proton. This remains the case even for much more complicated atoms.
So now, supposing that we do have some collection of noninteracting atoms, built in such a way that they are composite bosons, how does this collection behave? If their number is conserved, there is a finite temperature phase transition at which the atoms all pile into the lowest energy state available. This insight provides the first steps toward understanding the exotic properties of superfluid 4He, which was discovered in the 1930s and, for many decades, remained the paradigm of a bosonic condensed matter system. It was not, however, until 1995 that Bose-Einstein condensation was observed for the first time in a dilute (that is, essentially noninteracting) atomic gas after an experimental effort spanning decades, by
Cornell and Wieman, and then shortly thereafter reproduced independently by Ketterle, for which the trio shared the 2001 Nobel prize in physics [12, 13]. Since then, this field has blossomed in many directions. The study of bosonic matter is no longer restricted to the superfluidity of 4He, but instead encompasses a wide range of activity, such as systematically studying the effects of interactions on condensation, the role of internal degrees of freedom in “spinor” BECs, the effects of rotation, and the implementation and consequences of synthetic gauge fields such as the spin-orbit coupling I study in this thesis.
1.2.1 The Bose-Hubbard model
Well before the experimental realization of an ideal Bose gas in the cold atoms context, there were experiments where 4He was absorbed into porous media, as well as various studies of disordered “granular” superconductors, where the Cooper pairs could be imagined as approximately behaving like bosons, because their length scale is smaller than that of an individual grain. In either situation, Fisher et al. [14] posited that these experiments could be modeled by a network of localized regions that bosonic degrees of freedom (represented
5 † with creation and annihilation operators which satisfy [bi , bj] = δij) could occupy, weakly coupled together by quantum tunneling. The resulting model, called the “Bose-Hubbard model,” is characterized then by a hamiltonian
X U X H = (V − µ) b†b + b†b b†b − 1 + t b†b (1.3) BHM i i i 2 i i i i ij i j i ij
Where µ is a chemical potential that governs the density, Vi models a local (usually dis- order) potential, U is the potential energy penalty that every pair of bosons costs when occupying the same region i, while tij represents the amplitude for a boson in region j to tunnel into i. For a regular lattice network, and vanishing disorder, this is also an ef-
fective description of Josephson junction arrays (those being, in a sense, a special case of
granular superconductors). More relevant to this thesis, however, is that the above model
is a remarkably accurate description of a cold atomic gas in an optically induced periodic
potential, as demonstrated by Jaksch et al. [15].
Looking at this model on its own terms, there are two obvious limits. For U = 0, we have the lattice analog of an ideal Bose gas and, therefore, condensation in the lowest- lying eigenstate of HBHM. A small but nonzero value of U perturbs this result, however
it is insufficient to destroy the condensate [16]. On the other hand, taking all tij = 0,
the eigenstates of HBHM have a fixed, integer number of bosons at each lattice site and clearly do not describe a condensate. One might question if this state is stable against
small, nonzero tij and, indeed, it is. This is the celebrated incompressible “Mott insulator.” In Fig. 1.1, I also plot a zero-temperature mean-field phase diagram as a function of the
chemical potential and hopping strength, showing where these phases appear. The mean-
field method used here is expanded on in Chapter3.
1.3 Synthetic gauge fields
As the main thrust of this thesis involves cold atoms interacting with static gauge fields,
it will be useful to have some understanding of how these gauge fields are created as a
matter of principle. A charged particle couples to a static magnetic field, described by a
6 Figure 1.1: (a) Cartoon representation of bosons in an optical lattice. Each atom can tunnel between neighboring potential wells with an amplitude −t, while two bosons can occupy the same well with an energy penalty U. (b) Mean-field phase diagram of the Bose-Hubbard model. From this contour plot of the condensate order parameter, we observe large domes, the “Mott lobes,” where the condensate vanishes and an incompressible commensurate solid resides. For sufficiently large t, however, these states give way to a superfluid ground state.
vector potential A which satisfies B = ∇ × A through the “minimal coupling” substitution
(throughout, I will use natural units so that ~ = c = 1)
π = p − qA (1.4) representing the effect of the field on the canonical momentum π that enters the single particle hamiltonian. However, we are studying neutral atoms. Because q = 0, there is no coupling, and so the effects of orbital magnetism, such as the integer quantum Hall effect for fermions, or the creation of quantized vortices and vortex lattices in superfluids, seem irreparably out of reach. In the following sections, I will provide an elementary introduction to the methods by which the momenta of cold atoms can be modified, through the acquisition of a Berry’s phase or through optical processes that impart momentum to an atom in a way that depends on its internal state, to adopt the above form.
7 1.3.1 Berry’s connection and the adiabatic principle
We will build up the theoretical framework behind synthetic gauge fields (following some- what the discussions in [17, 18]) by considering a single atom with some internal degree of freedom (for concreteness, this is usually the hyperfine spin state). Now, imagine that we engineer a spatially varying, but time-independent1, coupling of these internal states with a general matrix U(r) , so that the atom is described by the Hamiltonian
1 H = − ∇2 + V (r) + U(r) (1.5) 2M I where the kinetic energy operator is, of course, agnostic to the internal state and V counts the part of the potential energy that is diagonal in the space of internal states. We can diagonalize U in this space, generating a spatially varying set of eigenvectors |χm(r)i (the “dressed states”), and then we can expand the state vector of the atom as
X |Ψ(r, t)i = ψm(r, t)|χm(r)i (1.6) m The time evolution of this vector is governed by the Schr¨odingerequation,
∂ X X 1 i ψ (r, t)|χ (r)i = − ∇2 + V (r) + U(r) ψ (r, t)|χ (r)i (1.7) ∂t m m 2M I m m m m
Acting on this from the left with hχ`| we obtain
∂ 1 X i ψ (r, t) = (V (r) + ) ψ (r, t) − hχ (r)|∇2ψ (r, t)|χ (r)i (1.8) ∂t ` ` ` 2M ` m m m where ` is the eigenvalue of U associated with |χ`i and the last term of the right-hand side
1Time dependence in the coupling is also easily included in this formalism and, indeed, some schemes for generating synthetic gauge fields rely on it, however, it is omitted here for a smoother presentation of the main ideas.
8 remains to be evaluated. Suppressing the dependence on r for clarity, we can write
2 2 2 hχ`|∇ ψm|χmi =hχ`| ∇ ψm |χmi + 2 (∇ψm) · |∇χmi + ψm|∇ χmi