studies on time-reversal invariant topological insulatorsfv691mt5830/maciejko_phd_the… · studies...
TRANSCRIPT
STUDIES ON TIME-REVERSAL INVARIANT TOPOLOGICAL
INSULATORS
A DISSERTATION
SUBMITTED TO THE DEPARTMENT OF PHYSICS
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
Joseph Maciejko
August 2011
http://creativecommons.org/licenses/by-nc/3.0/us/
This dissertation is online at: http://purl.stanford.edu/fv691mt5830
© 2011 by Joseph Maciejko. All Rights Reserved.
Re-distributed by Stanford University under license with the author.
This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.
ii
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Shoucheng Zhang, Primary Adviser
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
David Goldhaber-Gordon
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Xiaoliang Qi
Approved for the Stanford University Committee on Graduate Studies.
Patricia J. Gumport, Vice Provost Graduate Education
This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.
iii
Abstract
This dissertation brings together a number of topics in the theory of time-reversal (TR) in-
variant topological insulators. The first four chapters aredevoted to the transport properties
of the 2D quantum spin Hall (QSH) state. We explain nonlocal transport measurements in
HgTe quantum wells in terms of a Landauer-Buttiker theory of helical edge transport and
confirm the discovery of the QSH state in this material. We findthat decoherence can lead
to backscattering without breaking microscopic TR symmetry. As an example of incoherent
scattering, we study a Kondo impurity in an interacting helical edge liquid. A renormaliza-
tion group analysis shows the existence of an impurity quantum phase transition governed
by the Luttinger parameter of the edge liquid between a localhelical Fermi liquid withT 6
scaling of the low-temperature conductance, and an insulating strongly correlated phase
with fractionally charged emergent excitations. In the presence of a TR symmetry break-
ing magnetic field, it is known that even coherent scatteringcan lead to backscattering.
Through exact numerical diagonalization we find that nonmagnetic quenched disorder has
a strong localizing effect on the edge transport if the disorder strength is comparable to the
bulk gap. The predicted magnetoconductance agrees qualitatively with experiment.
The last two chapters are devoted to 3D topological insulators. We propose a com-
bined magnetooptical Kerr and Faraday rotation experimentas a universal measure of the
Z2 invariant. Finally, we propose a fractional generalization of 3D topological insulators
in strongly correlated systems, characterized by ground state degeneracy on topologically
nontrivial spatial 3-manifolds, a quantized fractional bulk magnetoelectric polarizability
without TR symmetry breaking, and a halved fractional quantum Hall effect on the sur-
face.
iv
Acknowledgments
First and foremost, I wish to thank my advisor Shou-Cheng Zhang. His remarkable
ability to distill beautiful and profound ideas from complex condensed matter problems,
his unwavering commitment to Dirac’s principle of simplicity and elegance in theoretical
physics, and his no-nonsense approach to the sometimes frustrating intricacies of day-to-
day problem-solving and overall project management alike have been a great source of
inspiration to me. I am most grateful for his constant reminder to trust in my own abilities
and judgment, as well as for his renewed encouragement to risk thinking outside the box
and choose challenging yet important research problems.
A special thanks goes to my other primary mentor and collaborator, Xiao-Liang Qi.
His generosity in sharing his amazing talent and enthusiasmfor physics with me as well as
his patient guidance have made working with him a truly formative experience, and a real
pleasure.
I would like to thank the faculty in the condensed matter group at Stanford, in particular
Steve Kivelson, Tom Devereaux, David Goldhaber-Gordon, Ian Fisher, and Hari Manoha-
ran, for many interesting discussions and for their help in various academic matters.
Wisdom is not found in books alone, and it is a pleasure to acknowledge my col-
laborators who did not skimp in sharing theirs with me: Laurens Molenkamp, Hartmut
Buhmann, Ewelina Hankiewicz, Eun-Ah Kim, Yuval Oreg, Andreas Karch, Congjun Wu,
Dennis Drew, Tadashi Takayanagi, Tom Devereaux, Rajiv Singh, Chao-Xing Liu, Cheng-
Chien Chen, Andreas Roth, Christoph Brune, Suk Bum Chung, Adam Sorini, and Brian
Moritz. I would also like to acknowledge financial support I have received over the past
five years from the Stanford Graduate Fellowship (SGF) Program, the National Science
and Engineering Research Council of Canada (NSERC), and theFonds Quebecois de la
v
Recherche sur la Nature et les Technologies (FQRNT), without which the work reported
in this dissertation would not have been possible. I also wish to thank Tom Devereaux for
kindly providing me access to the computing facilities of the Shared Hierarchical Academic
Research Computing Network (SHARCNET) and the Stanford Institute for Materials and
Energy Science (SIMES) at the SLAC National Accelerator Laboratory. Likewise, I thank
Laurens Molenkamp and Ewelina Hankiewicz for kindly providing me access to the com-
puting facilities of the Leibniz Rechenzentrum Munich.
A special thanks goes to Cheng-Chien Chen, my friend and fellow condensed matter
theorist at the physics department and CMITP. It has been an honor and a privilege to
share the highs and lows of graduate school with him. I also wish to thank my other
colleagues at the CMITP, Jiun-Haw Chu, Stathis Ilonidis, and George Karakonstantakis,
for their friendship and support.
I wish to thank my research group colleague Taylor Hughes forhis encouraging men-
torship. Benefitting from the guidance of such a fine physicist and friendly senior col-
league helped me through each of the new steps of my Ph.D. I also wish to thank my other
group colleagues, Rundong Li, Sri Raghu, Suk Bum Chung, Zhong Wang, Qin Liu, An-
drei Bernevig, Lukas Muchler, and Maissam Barkeshli for sharing their knowledge and
perspective on various physics topics.
I have benefitted a lot from various professional and personal interactions with my
fellow physics students and friends at Stanford, Hong Yao, Erez Berg, Andrei Garcia, John
Robertson, Wei-Feng Tsai, Katherine Luna, Reza Jamei, Ileana Rau, Alex Fried, Weejee
Cho, Yeming Shi, Sho Yaida, Li Liu, Francois Amet, Nachum Plonka, Daniel Harlow, and
Tzen Ong. You have all contributed to making the last five years truly enjoyable — thank
you!
I would like to express sincere thanks to Maria Frank, Roberta Edwards, Sybille Katz,
Corrina Peng, Ellie Lwin, and Catherine Meng for their invaluable help in dealing with
various administrative matters. Thank you for always finding time to help despite your
busy schedules.
I distinctly remember flipping through my dad’s Ph.D. dissertation on gauge theories of
particle physics as a kid, and being fascinated by the myriadof strange symbols and funny
little diagrams with wiggly lines that covered its pages. Little did I, or dad, know that this
vi
was the beginning of my career in theoretical physics, and that I would be writing my own
dissertation twenty or so years later, with its own collection of mysterious equations. Thank
you dad for introducing me to the amazing world of physics.
I am especially grateful to my parents Romain and Helene, my sister Eva, and all of
my family. Through your own example and your sense of humor, you have shown me that
belief in self and perseverance go a long way. Your unconditional love and support have
kept me plodding through during difficult times — thank you!
Finally, it is a joy to thank my wonderful fiancee, Michelle.Being engaged to a woman
who is both an incredibly smart Stanford Ph.D. student and anamazingly loving and sup-
portive person is a true blessing. I thank her for taking a genuine interest in all my endeav-
ors, academic and otherwise; for being the most balanced, optimistic, and joyful person
I know, especially during big research headaches; and most of all, for believing in me at
every step of the way. This dissertation is dedicated to her.
vii
Contents
Abstract iv
Acknowledgments v
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Dissertation overview . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 12
2 Nonlocal transport in the QSH state 14
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Ohm’s law versus nonlocal transport . . . . . . . . . . . . . . . . .. . . . 15
2.3 Device structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Transport on the edge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Dissipationless transport . . . . . . . . . . . . . . . . . . . . . . . .. . . 18
2.6 Helical versus chiral . . . . . . . . . . . . . . . . . . . . . . . . . . . . .19
2.7 Dephasing and nonlocal transport . . . . . . . . . . . . . . . . . . .. . . 23
2.7.1 Quantum kinetic formalism and Landauer-Buttiker equations . . . . 24
2.7.2 Dephasing in the QSH regime: numerical study . . . . . . . .. . . 27
2.7.3 Dephasing in the QH regime: numerical study . . . . . . . . .. . . 32
2.7.4 Self-energy and microscopic time-reversal symmetry. . . . . . . . 34
2.7.5 Nonlocal resistance in H-bar structure: numerical study . . . . . . . 38
2.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
viii
3 Kondo effect in the QSH state 41
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Theoretical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Weak coupling regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Strong coupling regime . . . . . . . . . . . . . . . . . . . . . . . . . . . .47
3.5 Experimental realization . . . . . . . . . . . . . . . . . . . . . . . . .. . 50
4 Magnetoconductance of the QSH state 52
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2 Theoretical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5 Spin AB effect and topological spin transistor 62
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.2 Phenomenological scattering matrix analysis . . . . . . . .. . . . . . . . . 65
5.3 Minimal model description . . . . . . . . . . . . . . . . . . . . . . . . .. 66
5.4 Experimental realization in HgTe QW . . . . . . . . . . . . . . . . .. . . 70
5.5 Conclusion and outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . .74
6 Topological quantization in units ofα 75
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.2 Magnetooptical Kerr and Faraday rotation . . . . . . . . . . . .. . . . . . 77
6.3 Reflectivity minima and total surface Hall conductance .. . . . . . . . . . 79
6.4 Reflectivity maxima and topological magnetoelectric effect . . . . . . . . . 80
6.5 Kerr-only measurements . . . . . . . . . . . . . . . . . . . . . . . . . . .82
6.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7 Fractional topological insulators in 3D 85
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.2 Projective construction and trial wave function . . . . . .. . . . . . . . . . 87
7.3 Effective gauge theory . . . . . . . . . . . . . . . . . . . . . . . . . . . .88
ix
7.4 Time-reversal symmetry and quantization of the axion angle . . . . . . . . 91
7.5 Bulk topology and surface “half” fractional quantum Hall effect . . . . . . 92
7.6 Three classes of parton models . . . . . . . . . . . . . . . . . . . . . .. . 93
7.6.1 Deconfined models . . . . . . . . . . . . . . . . . . . . . . . . . . 98
7.6.2 Higgs models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.7 Fractional quantized magnetoelectric effect . . . . . . . .. . . . . . . . . 107
7.7.1 Chiral anomaly . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.7.2 Topological quantization of the axion angle . . . . . . . .. . . . . 110
7.8 Ground state degeneracy . . . . . . . . . . . . . . . . . . . . . . . . . . .115
7.8.1 Ground state degeneracy on the 3-torus . . . . . . . . . . . . .. . 115
7.8.2 Ground state degeneracy on 3-manifolds with boundaries . . . . . . 123
7.9 Gapless surface states . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 126
7.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8 Conclusion 130
A Theoretical methods 133
A.1 Weak-coupling renormalization group equations . . . . . .. . . . . . . . . 133
A.1.1 Operator algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
A.1.2 Scaling dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . 138
A.1.3 Operator product expansions . . . . . . . . . . . . . . . . . . . . .140
A.2 Weak-coupling Kubo formula calculation of the edge conductance . . . . . 144
A.2.1 Analytic continuation in the time domain . . . . . . . . . . .. . . 158
A.2.2 Bosonic correlators . . . . . . . . . . . . . . . . . . . . . . . . . . 163
A.3 Coulomb gas description of the strong coupling regime . .. . . . . . . . . 165
A.3.1 Euclidean action . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
A.3.2 Semiclassical instanton calculation . . . . . . . . . . . . .. . . . . 168
A.3.3 Dilute instanton gas approximation . . . . . . . . . . . . . . .. . 171
A.3.4 Tree-level renormalization . . . . . . . . . . . . . . . . . . . . .. 174
A.3.5 Mapping to dual boundary sine-Gordon model . . . . . . . . .. . 176
A.3.6 Current operator in strong coupling regime . . . . . . . . .. . . . 179
A.4 Keldysh calculation of the shot noise . . . . . . . . . . . . . . . .. . . . . 180
x
A.4.1 Basics of noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
A.4.2 Schwinger-Keldysh action . . . . . . . . . . . . . . . . . . . . . . 185
A.4.3 Nonequilibrium current . . . . . . . . . . . . . . . . . . . . . . . . 186
A.4.4 Nonequilibrium noise . . . . . . . . . . . . . . . . . . . . . . . . 187
A.4.5 Schottky relation and fractional Fano factor . . . . . . .. . . . . . 188
A.5 Mean-field description of half-charge tunneling . . . . . .. . . . . . . . . 189
A.6 Phase space derivation of theT 6 behavior in the noninteracting case . . . . 192
A.7 Transport in a spin Aharonov-Bohm ring . . . . . . . . . . . . . . .. . . . 195
A.7.1 S-matrix analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
A.7.2 Scattering at the junction . . . . . . . . . . . . . . . . . . . . . . .197
A.7.3 Tight-binding model . . . . . . . . . . . . . . . . . . . . . . . . . 198
A.8 Magnetooptical Kerr/Faraday rotation in thick film geometry . . . . . . . . 199
A.9 Surface Chern-Simons theory for an Abelian fractional 3D TI . . . . . . . . 202
Bibliography 207
xi
List of Tables
7.1 Gauge charge assignments of parton fields: Abelian model. . . . . . . . . 98
7.2 Gauge charge assignments of parton fields: Higgs model . .. . . . . . . . 105
xii
List of Figures
1.1 Chiral versus helical: QH versus QSH effects . . . . . . . . . .. . . . . . 5
2.1 2- and 4-terminal resistances . . . . . . . . . . . . . . . . . . . . . .. . . 16
2.2 Devices D3 and D4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 2- and 4-terminal resistances for device D3 . . . . . . . . . . .. . . . . . 21
2.4 Nonlocal resistances for H-bar device D4 . . . . . . . . . . . . .. . . . . 22
2.5 2-terminal geometry in QSH regime with dephasing . . . . . .. . . . . . . 29
2.6 2-terminal conductance in QSH regime from TB model . . . . .. . . . . . 30
2.7 2- and 4-terminal conductances in QSH regime with dephasing . . . . . . . 31
2.8 Cartoon picture of edge transport in QSH regime with dephasing . . . . . . 32
2.9 2-terminal geometry in QH regime with dephasing . . . . . . .. . . . . . 33
2.10 2-terminal conductance in QH regime,Bz = 1 T . . . . . . . . . . . . . . 34
2.11 2-terminal conductance in QH regime,Bz = −1 T . . . . . . . . . . . . . 35
2.12 Cartoon picture of edge transport in QH regime with dephasing . . . . . . . 36
2.13 Numerical calculation of 4-terminal resistance in H-bar geometry . . . . . . 38
3.1 Edge conductance of the QSH state at finite temperature . .. . . . . . . . 42
3.2 Edge conductance in the strong coupling regime . . . . . . . .. . . . . . . 47
4.1 MagnetoconductanceG(B) of a QSH edge . . . . . . . . . . . . . . . . . 56
4.2 MagnetoconductanceG(B) for various disorder strengths . . . . . . . . . . 58
4.3 Effect onG(B) of BIA term,B-field orientation, and Dirac mass . . . . . . 60
5.1 Schematic picture of the spin AB effect . . . . . . . . . . . . . . .. . . . 64
5.2 Illustration of the minimal model describing a FM/QSH junction . . . . . . 67
xiii
5.3 Phenomenological analysis of the two-terminal conductance . . . . . . . . 69
5.4 Numerical study of the spin AB effect in HgTe QW . . . . . . . . .. . . . 71
6.1 Kerr and Faraday angles for a topological insulator thick film . . . . . . . . 77
6.2 Reflectivity and universal functionf(θ) . . . . . . . . . . . . . . . . . . . 79
6.3 Kerr-only measurement setup and universal functionfK(θ) . . . . . . . . . 82
7.1 Cartoon picture of fractional 3D topological insulator. . . . . . . . . . . . 89
7.2 Global center symmetry ofSU(Nc) effective gauge theory . . . . . . . . . 119
A.1 Integration contour for evaluating Matsubara sum . . . . .. . . . . . . . . 159
A.2 Analytic structure of the time-ordered boson propagator . . . . . . . . . . . 162
A.3 Incident, reflected, and transmitted light for Kerr/Faraday rotation . . . . . 200
xiv
Chapter 1
Introduction
Symmetry appears to be a profound organizational principleof nature. Beautifully sym-
metric natural patterns are ubiquitous, ranging from spiral galaxies and nearly spherical
planets to the five-fold symmetry of starfish and the helical symmetry of DNA. Symmetry
also turns out to be a remarkably useful theoretical principle by which we can structure our
understanding of complex physical systems. It is perhaps inthe field of condensed matter
physics — where one studies matter arranged in an endless variety of forms — that the
symmetry principle finds its most spectacular application.Paradoxically, one of the most
instructive ways to study stable phases of matter is to classify them according to the sym-
metries theybreak. Most classical and quantum gases and liquids do not break any sym-
metry at all; they enjoy the full translational and rotational symmetries of free space. Most
classical and quantum solids break these symmetries down toa discrete subgroup of trans-
lations and rotations (the space group). Magnetically ordered quantum spin systems such
as ferromagnets and antiferromagnets break spin rotation symmetry. Thisbroken symme-
try principle is at the heart of the phenomenological Ginzburg-Landau theory [1] of phase
transitions which, combined with microscopic many-body theories of condensed matter
systems, constitutes the cornerstone of pre-1980s condensed matter physics. In Ginzburg-
Landau theory, a stable phase of matter is characterized by alocal order parameter, which
is nonzero in an ordered phase but vanishes in a disordered phase. Phases with nonzero
order parameters are further distinguished by the way the order parameter transforms un-
der symmetry operations, i.e. by the representation of the symmetry group of the system
1
CHAPTER 1. INTRODUCTION 2
Hamiltonian to which it belongs.
The 1980s were marked by the discovery of the integer [2] and fractional [3] quantum
Hall (QH) effects. The QH effect occurs when a two-dimensional electron gas (2DEG), for
instance formed by electrons trapped in the inversion layerof a metal-oxide-semiconductor
structure or electrons in a semiconductor quantum well (QW), is subjected to a large mag-
netic field perpendicular to the plane of the 2DEG. A perpendicular magnetic field causes
the electrons to travel along circular cyclotron orbits, the radius of which becomes smaller
with increasing magnetic field. For large enough magnetic fields, electrons in the bulk of
the material form small, closed cyclotron orbits. On the other hand, electrons near the edge
of the sample can trace extended, open orbits that skip alongthe edge. At low tempera-
tures, quantum effects become important and two things happen. First, the area of closed
orbits in the bulk becomes quantized, bulk electrons becomelocalized (since they only
trace small, closed orbits) and the bulk turns into an insulator. Second, the skipping edge
orbits form extended one-dimensional channels with a quantized conductance ofe2/h per
channel. Furthermore, the transverse (Hall) conductanceσxy is quantized in integer (integer
QH) or rational (fractional QH) multiples ofe2/h.
It was soon realized that the bulk of a QH state is a featureless insulating state which
does not break any symmetries other than time-reversal (TR)symmetry, and thus cannot be
characterized by a local order parameter. Nevertheless, QHstates with different values of
the Hall conductance are truly distinct phases of matter, and correspond to quantum ground
states which cannot be adiabatically connected to each other without closing a spectral gap,
i.e. without going through a QH plateau transition. Even more surprising is the fact that
the quantization of the Hall conductance is found to be extremely accurate even in disor-
dered samples, where one would expect the randomizing effect of disorder to destroy any
quantization phenomenon. Indeed, if conduction proceeds only through one-dimensional
channels, one would naively expect these to be strongly affected by disorder due to Ander-
son localization [4].
The lack of a bulk local order parameter description a la Ginzburg-Landau and the
existence of boundary states robust to disorder can both be understood as defining charac-
teristics of a topological state of quantum matter. A usefulconcept in this context is that of
bulk-edge correspondence[5], of which the integer QH state provides a clear illustration.
CHAPTER 1. INTRODUCTION 3
A topological state of matter is insulating in the bulk, but supports gapless boundary states
which are perturbatively robust to disorder. Rather than being characterized by a local or-
der parameter, the bulk is characterized by atopological invariantwhich, in the case of the
integer QH state, is an integer denoted as the TKNN number [6]or Chern number [7]. The
bulk topological invariant is in turn related to the number of stable gapless boundary states.
In the integer QH state, the Chern number is simply equal to the number of stable gapless
edge states, and is also the value of the quantized Hall conductance in units ofe2/h. In that
sense, one says that the edge states areprotected by the bulk topology. But more concretely,
what is the mechanism for this “topological protection”? The answer is the following: the
bulk topology is responsible for some kind of fractionalization, in some sense, on the edge.
More precisely, the usual degrees of freedom of the electronare spatially separatedon
opposite edges. The usual degrees of freedom of an electron in a one-dimensional chan-
nel are twofold: right-moving and left-moving. However, ina QH sample, the top edge
has only right-moving electrons and the bottom edge has onlyleft-moving electrons (or
vice-versa, depending on the sign of the magnetic field). Backscattering on a given edge
is thus suppressed due to the inability of an electron to reverse its direction of motion, and
the QH edge channels completely evade Anderson localization. Since a single direction of
propagation is present on a given edge, the QH edge channels are termedchiral.
The TKNN integer relates the physical response of the Hall conductance to a topo-
logical invariant in momentum space. Whereas this work gives the first insight into the
topological nature of the QH state, it is limited to noninteracting systems. A more funda-
mental description of the QH effect is given by the topological field theory based on the
Chern-Simons term in2 + 1 dimensions [8, 9]. In this approach, the problem of electrons
in a 2DEG subject to an external perpendicular magnetic fieldB is exactly mapped to that
of bosons coupled to both the external magnetic field and an internal, emergent statistical
magnetic fieldb. This statistical magnetic field, the dynamics of which is described by
the Chern-Simons term, is responsible for the transmutation of the fermionic electrons into
bosons. At the magic filling fractionsν = 1/m (with m an odd integer) at which the QH
effect occurs, the external and statistical magnetic fieldsprecisely cancel each other, and
the bosons condense into a superfluid state. The effective field theory of a boson superfluid
is the(2 + 1)-dimensional Maxwell electrodynamics [10, 11]. In the long-wavelength and
CHAPTER 1. INTRODUCTION 4
low-energy limit, the Chern-Simons term dominates over theMaxwell term, and the ef-
fective theory of the QH state is just the topological Chern-Simons term. This topological
field theory is generally valid in the presence of disorder and interactions.
Until very recently, QH states were the only topological states the existence of which
had been firmly established by experimental observation. Compared to the rich variety
of “traditional” broken-symmetry states, one is led to the obvious question: should there
not be other topological states remaining to be discovered?In particular, the QH insulator
requires a large magnetic field for its existence. A natural question to ask is whether a
magnetic field, which breaks TR symmetry, is a necessary condition to obtain a topological
state. The first answer to this question was provided by the independent theoretical pre-
diction by Kane and Mele [12, 13] and Bernevig and Zhang [14] of a new state of matter,
the 2D time-reversal invariant topological insulator or quantum spin Hall (QSH) insulator.
This state displays robust quantized properties but does not require a TR symmetry break-
ing magnetic field for its observation. Roughly speaking, the QSH state can be viewed as
two copies of the QH state with opposite Hall conductances. The proposal of Kane and
Mele is based on the spin-orbit interaction of graphene, andis mathematically motivated
by the earlier work of Haldane [15] on the so-called quantum anomalous Hall (QAH) ef-
fect. The proposal of Bernevig and Zhang is based on the spin-orbit interaction induced
by strain in semiconductors. Neither proposals have yet been realized in actual condensed
matter systems, mostly because of the small spin-orbit interaction in the proposed systems.
However, they provided an important conceptual framework in which the stability of the
QSH state can be investigated.
In what sense is the QSH state a topological state of matter? This question is most
clearly answered by looking at whether this state supports stable gapless boundary modes,
robust to disorder. Let us proceed by comparison with the QH edge modes discussed pre-
viously (Fig. 1.1). As mentioned previously, the edge states of the QH state are such that
electrons can only propagate in a single direction on a givenedge. Compared with a one-
dimensional system of spinless electrons (Fig. 1.1, top left), the top edge of a QH system
contains only half the degrees of freedom (Fig. 1.1, bottom left). The QH system can thus
be compared to a “freeway” where electrons traveling in opposite directions have to be
“driving in different lanes”. This spatial separation resulting in chiral edge channels can be
CHAPTER 1. INTRODUCTION 5
Spinless 1D chain Spinful 1D chain
2 = 1 + 1 4 = 2 +22 1 + 1 4 2 +2
QH QSH
Figure 1.1: Chiral vs helical: spatial separation is at the heart of both the QH and QSH ef-fects. A spinless one-dimensional system (top left) has both right-moving and left-movingdegrees of freedom. Those two basic degrees of freedom are spatially separated in a QHsystem (bottom left), as illustrated by the symbolic equation “2 = 1 + 1”. The upper edgehas only a right-mover and the lower edge a left-mover. Thesechiral edge states are ro-bust to disorder: they can go around an impurity (green dot) without backscattering. Onthe other hand, a spinful one-dimensional system (top right) has twice as many degrees offreedom as the spinless system due to the twofold spin degeneracy. Those four degrees offreedom are separated in a TR invariant way in a QSH system (bottom right). The top edgehas a right-mover with spin up (denoted by a red dot) and a left-mover with spin down(denoted by a blue cross), and conversely for the lower edge.That separation is illustratedby the symbolic equation “4 = 2+ 2”. Thesehelicaledge states are robust to nonmagneticdisorder, i.e. impurities which preserve the TR symmetry ofthe QSH state.
illustrated by the symbolic equation “2 = 1+ 1” where each “1” corresponds to a different
chirality. This “chiral traffic rule” is particularly effective in suppressing electron scatter-
ing: since electrons travel always in the same direction, they are forced to avoid impurities
(Fig. 1.1, bottom left, green dot), and thus cannot backscatter.
On the other hand, as mentioned earlier, the QSH state can be roughly understood as
two copies of the QH state, with one copy for each spin. The edge state structure of the QSH
state (Fig. 1.1, bottom right) can thus be described pictorially by superposing two copies of
QH edge states (Fig. 1.1, bottom left), with opposite chirality for each spin. Compared to a
spinful one-dimensional system (Fig. 1.1, top right), the top edge of a QSH system contains
CHAPTER 1. INTRODUCTION 6
only half the degrees of freedom. The resulting edge states are termedhelical, because spin
is correlated with the direction of propagation. This new pattern of spatial separation can
be illustrated by the symbolic equation “4 = 2+2” where each2 corresponds to a different
helicity. Although electrons are now allowed to travel bothforward and backwards on
the same edge, there is a new “traffic rule” which suppresses backscattering: in order to
backscatter, an electron needs to flip its spin, which requires the breaking of TR symmetry.
If TR symmetry is preserved, as is the case for nonmagnetic impurities, no backscattering
is allowed.
What is the mechanism which allows this spatial separation?In the case of the QH
effect, the separation is achieved by an external magnetic field or, in the case of the QAH
effect, by some internal field which breaks TR symmetry. Thisinternal field takes the form
of a relativistic mass term for emergent Dirac fermions in2 + 1 dimensions, with the sign
of the internal field (and hence the chirality of the QAH edge states) dictated by the sign of
the mass. In the case of the QSH effect, the separation is achieved through the TR invariant
spin-orbit coupling.
Since the QSH state is characterized by a bulk insulating gapand gapless boundary
states robust to disorder (in the presence of TR symmetry), the QSH state is indeed a new
topological state of matter. However, because the Hall conductance of the QSH state van-
ishes, it is clear that the TKNN or Chern number discussed above, which corresponds to
the the value of the Hall conductance in units ofe2/h, cannot provide a useful classifica-
tion of the QSH state. This question has been addressed both within the topological band
theory [12] and the topological field theory [16], and it turns out that the proper topological
invariant is valued in theZ2 group containing only two elements,0 or 1, with 1 correspond-
ing to the topologically nontrivial QSH insulator and0 corresponding to a topologically
trivial insulator with no robust gapless edge states. Physically, thisZ2 invariant simply
counts the number of stable gapless edge states modulo2.
As mentioned previously, Kane and Mele [13] proposed graphene — a monolayer of
carbon atoms — as a possible candidate for the QSH effect. Unfortunately, this proposal
turned out to be unrealistic because the spin-orbit gap in graphene is extremely small [17,
18]. The QSH effect was also independently proposed in semiconductors in the presence of
strain gradients [14], but this proposal also turned out to be hard to realize experimentally.
CHAPTER 1. INTRODUCTION 7
Soon afterwards, Bernevig, Hughes, and Zhang [19] initiated the search for the QSH state in
semiconductors with an “inverted” bandstructure, and predicted a quantum phase transition
in type-III HgTe/CdTe QW between a trivial insulator phase and a QSH phase governed
by the thicknessd of the QW. The QSH state was observed experimentally soon after by
Konig et al. [20].
The QSH insulator in HgTe/CdTe QW is characterized by insulating gap in the bulk and
a single pair of helical edge states at each edge. A topological phase transition occurs due to
the band inversion at theΓ point driven by the spin-orbit interaction. The helical edge state
forms a single one-dimensional massless Dirac fermion withcounter-propagating states
forming a Kramers doublet under TR symmetry. In a strip geometry with two edges, each
edge supports one propagating mode of a given chirality withconductancee2/h, hence
the predicted longitudinal conductance in such a geometry is2e2/h [19]. Furthermore, the
helical state consisting of a single massless Dirac fermionis “holographic”, in the sense
that it cannot exist in a purely one-dimensional system, butcan only exist as the boundary
of a two-dimensional system [21].
The model Hamiltonian for the two-dimensional topologicalinsulator in HgTe/CdTe
QW [19] also gives a basic template for a generalization to three dimensions, leading to
a simple model Hamiltonian for a class of materials: Bi2Se3, Bi2Te3 and Sb2Te3 [22, 23].
Similar to their two-dimensional counterpart the HgTe/CdTe QW, these materials can be
described by a simple but realistic model, where the spin-orbit interaction drives a band
inversion transition at theΓ point. In the topologically nontrivial phase, the bulk states
are fully gapped, but there is a topologically protected surface state consisting of a single
two-dimensional massless Dirac fermion. This two-dimensional massless Dirac fermion is
“helical”, where the spin of the electron points perpendicularly to its momentum, forming
a left-handed helical texture in momentum space. Similarlyto the one-dimensional heli-
cal edge state, a single two-dimensional massless Dirac surface state is “holographic”, in
the sense that it cannot occur in a purely two-dimensional system with TR symmetry, but
can exist as the boundary of a three-dimensional insulator.A TR invariant single-particle
perturbation can not introduce a gap for the surface state. Agap can open up for the sur-
face state when a TR breaking perturbation is introduced on the surface. In this case, the
CHAPTER 1. INTRODUCTION 8
system becomes a full insulator, both in the bulk and on the surface. The topological prop-
erties of the fully gapped insulator are characterized by a novel topological magnetoelectric
effect [16].
Soon after the theoretical prediction of the three-dimensional topological insulator in
the Bi2Se3, Bi2Te3, and Sb2Te3 class of materials [22, 24], angle-resolved photoemission
(ARPES) experiments observed the surface state with a single Dirac cone [24, 25, 26].
Spin-resolved ARPES experiments subsequently observed the predicted left-handed helical
spin texture of the massless Dirac fermion [26]. These pioneering works inspired much of
the subsequent developments both in theory and experiment.
The general theory of the three-dimensional topological insulator has been developed
along two different routes. The topological band theory gives a general description of the
topological invariant in the single-particle momentum space [27, 28, 29]. In particular, a
method due to Fu and Kane [30] gives a simple algorithm to determine the topological
properties of any complex electronic structure with inversion symmetry. This method pre-
dicts that the semiconducting alloy BixSb1−x is a topological insulator for a certain range
of compositionx. ARPES experiments [31] have indeed observed topologically nontrivial
surface states in this system. However, the surface states in BixSb1−x are rather compli-
cated, and cannot be described by a simple model Hamiltonian.
The topological band theory is only valid for noninteracting systems in the absence of
disorder. The topological field theory is a more general theory which describes the electro-
magnetic response of the topological insulator [16]. In Ref. [16], Qi, Hughes, and Zhang
found that the electromagnetic response of three-dimensional topological insulators is de-
scribed by the Maxwell equations with an added topological term proportional toE·B. This
exact modification [16] had been proposed earlier in the context of high-energy physics [32]
as a modification to conventional electrodynamics due to thepresence of the Peccei-Quinn
axion field [33]. In this approach [16], theZ2 topological invariant from topological band
theory corresponds to a quantized emergent axion angleθ which is constrained by TR in-
variance to take only two values,0 (the trivial insulator) orπ (the topological insulator).
The equivalence between the two definitions has been proven recently [34]. Several unique
experiments based on axion electrodynamics in three-dimensional topological insulators
have been proposed: a topological Kerr and Faraday effect [16, 35, 36, 37], a topological
CHAPTER 1. INTRODUCTION 9
magnetoelectric effect [16], an image magnetic monopole effect [38, 39], and emergent dy-
namical axion particles [40, 41]. Efforts towards the discovery of these exotic phenomena,
as well as intensive searches for new three-dimensional topological insulator materials, are
ongoing.
1.1 Motivation
The work described in this dissertation can be divided in twoparts which roughly follow
the chronological evolution of the field of topological insulators from the years 2007 to
2010.
The first part of this dissertation is concerned with the transport properties of 2D QSH
insulators. Our work was mainly motivated by the experimental observation of a quantized
longitudinal conductance of2e2/h in HgTe QW [20]. Although the measurements reported
in Ref. [20] were strong evidence for the existence of robustedge channels, a number of
questions awaited explanation:
• Could the observed quantized conductance simply result from the usual subband
quantization in quasi-1D mesoscopic systems rather than from the existence of edge
channels?
• Why does the conductance deviate from2e2/h in long enough samples?
• Why does the conductance deviation increase rather than decrease with decreasing
temperature?
• Is there a way to directly measure theZ2 invariant in this material?
• Why does the conductance decrease so steeply with an appliedperpendicular mag-
netic field, and throughout the bulk gap, even if the edge gap opened by the magnetic
field is extremely small?
• Can we use the helical property of the edge channels to designa novel type of elec-
tronic device?
CHAPTER 1. INTRODUCTION 10
The first four chapters of this dissertation deal with answers to these questions.
The second part of this dissertation is concerned with 3D topological insulators. Here
the motivation was both experimental and theoretical. Our work was started soon after ex-
perimental observation of gapless surface states by ARPES in the Bi2Se3, Bi2Te3, Sb2Te3
class of materials [24, 25, 26]. Although this provided strong evidence for the existence of
the 3D topological insulator state in these materials, it was still unclear how to observe the
topological magnetoelectric effect predicted in Ref. [16], which is the true topologically
quantized physical observable in a 3D topological insulator (i.e. the analog of measuring
the quantized Hall conductance in a 2D QH insulator). Finally, the motivation for the last
part of our work came from the existence of two “competing” definitions of topological
order in condensed matter systems. Topological order as described earlier in this chapter
is fully characterized by band-theoretic invariants and arises in noninteracting or weakly
interacting systems. It can be called topological order of Thouless type, of which the in-
teger QH is a prime example. On the other hand, another notionof topological order was
pioneered by Wen [42] and is characterized by ground state degeneracies, fractionalization,
and emergent gauge theories, and arises in strongly correlated systems. FQH systems are
the canonical example of topologically ordered systems in the sense of Wen. Our work on
fractional 3D topological insulators was an effort to generalize the Thouless-type topolog-
ical order in ordinary 3D topological insulators to Wen-type topological order.
1.2 Contributions
This dissertation is built on the six publications listed and summarized below:
(1) Andreas Roth, Christoph Brune, Hartmut Buhmann, Laurens W. Molenkamp, Joseph
Maciejko, Xiao-Liang Qi, and Shou-Cheng Zhang,Nonlocal transport in the quan-
tum spin Hall state, Science325, 294 (2009).
This work confirms the existence of the QSH state in HgTe QW through an extensive
set of multi-terminal transport measurements. It is the result of a collaboration be-
tween the experimental group of L. W. Molenkamp at WurzburgUniversity and the
Stanford theory group. The latter developed a Landauer-Buttiker theory of transport
CHAPTER 1. INTRODUCTION 11
in the QSH state which produced excellent agreement with theexperimental data.
The theory also explains the different role played by voltage probes in the QH and
QSH states.
(2) Joseph Maciejko, Chao-Xing Liu, Yuval Oreg, Xiao-LiangQi, Congjun Wu, and
Shou-Cheng Zhang,Kondo effect in the helical edge liquid of the quantum spin Hall
state, Physical Review Letters102, 256803 (2009).
Using linear response and renormalization group methods, we calculate the low-
temperature edge conductanceG of a QSH insulator in the presence of a magnetic
impurity. We find a quantum phase transition between a weaklycoupled metallic
“local helical liquid” with unusual power lawsG ∼ T x, and a strongly coupled
insulating phase where transport proceeds by weak tunneling of quasiparticles with
half an electron charge.
(3) Joseph Maciejko, Xiao-Liang Qi, and Shou-Cheng Zhang,Magnetoconductance of
the quantum spin Hall state, Physical Review B82, 155310 (2010).
We study numerically the edge magnetoconductance of the QSHstate in the presence
of quenched nonmagnetic disorder. For a disorder strength on the order of the bulk
gap, the conductance decreases roughly linearly with the magnetic field, in qualitative
agreement with experiments on HgTe QW. We conjecture that for disorder small
compared to the gap the edge liquid suffers 1D Anderson localization, while for
disorder larger than the gap the electrons can hop to the 2D bulk and undergo 2D
diffusive motion and 2D antilocalization.
(4) Joseph Maciejko, Eun-Ah Kim, and Xiao-Liang Qi,Spin Aharonov-Bohm effect and
topological spin transistor, Physical Review B82, 195409 (2010).
Ever since its discovery, the electron spin has only been measured or manipulated
through the application of an electromagnetic force actingon the associated magnetic
moment. In this work, we use the helical property of the QSH edge states to propose
a spin Aharonov-Bohm effect in which the electron spin is controlled by a magnetic
flux while no electromagnetic field is acting on the electron.
CHAPTER 1. INTRODUCTION 12
(5) Joseph Maciejko, Xiao-Liang Qi, H. Dennis Drew, and Shou-Cheng Zhang,Topo-
logical quantization in units of the fine structure constant, Physical Review Letters
105, 166803 (2010).
Fundamental topological phenomena in condensed matter physics are associated
with a quantized electromagnetic response in units of fundamental constants. The
3D topological insulator is predicted to exhibit a topological magnetoelectric effect
quantized in units of the fine structure constantα = e2/~c. In this work, we propose
an optical experiment to directly measure this topologicalquantization phenomenon,
independent of material details.
(6) Joseph Maciejko, Xiao-Liang Qi, Andreas Karch, and Shou-Cheng Zhang,Frac-
tional topological insulators in three dimensions, Physical Review Letters105, 246809
(2010).
Topological insulators can be generally defined by a topological field theory with
an axion angleθ of 0 or π. In this work, we introduce the concept of fractional
topological insulator defined by a fractional axion angle and show that it can be
consistent with time-reversal invariance. The fractionalaxion angle can be measured
experimentally by a “halved” FQH effect on the surface with Hall conductance of the
form σH = pqe2
2hwith p, q odd integers.
1.3 Dissertation overview
This dissertation is structured as follows. Chapters 2-5 pertain to the transport properties
of the 2D QSH insulator. Chapter 2 covers publication (1) andincludes experimental work
by the group of L. W. Molenkamp at Wurzburg University, as well as supporting theoret-
ical material (Sec. 2.7). The latter section describes the effect of coherent and incoherent
scattering on the transport properties of the QSH state, andemphasizes the similarities
and differences with the QH state. Analytical and numericalcalculations allow us to re-
solve a paradox between the theoretically predicted absence of backscattering in the helical
edge state and the experimentally observed nonzero resistance induced by a voltage probe.
Chapter 3 covers publication (2). A key result of this chapter is the prediction of a fractional
CHAPTER 1. INTRODUCTION 13
Fano factor of1/2 at low temperatures, which is a direct measure of theZ2 invariant of the
QSH state. Details of the calculations involved in this chapter are given in Sections A.1-
A.6 of Appendix A. Chapter 4 covers publication (3). Chapter5 covers publication (4),
and details of theS-matrix transport calculations are given in Sec. A.7. Chapters 6 and 7
are concerned with 3D topological insulators and generalizations thereof. Chapter 6 covers
publication (5), and details of the Kerr/Faraday calculation are given in Sec. A.8. Sec-
tions 7.1-7.5 of Chapter 7 cover publication (6), while Sections 7.6-7.10 cover unpublished
material. An explicit derivation of the boundary Chern-Simons theory for an Abelian frac-
tional 3D topological insulator is given in Sec. A.9. Finally, we conclude in Chapter 8.
Chapter 2
Nonlocal transport in the quantum spin
Hall state
2.1 Introduction
The search for topological states of quantum matter has become an important goal in con-
densed matter physics. Inside a topological insulator, theconventional laws of electrody-
namics are dramatically altered [43], which may have applications in constructing novel
devices for processing of (quantum) information. The QSH state [13, 14] is a topologi-
cally nontrivial state of matter which exists in the absenceof any external magnetic field.
It has a bulk energy gap but gapless helical edge states protected by TR symmetry. In
the QSH regime, opposite spin states forming a Kramers doublet counter-propagate at the
edge [21, 44]. Recently, the QSH state has been theoretically predicted in HgTe QW [19].
There is a topological quantum phase transition at a critical thicknessdc of the QW, sepa-
rating the trivial insulator state ford < dc from the QSH insulator state ford > dc. Soon
after the theoretical prediction, evidence for the QSH state has been observed in transport
measurements [20]. In the QSH regime, experiments measure aconductanceG close to
twice the quantum unit of conductanceG = 2e2/h, which is consistent with quantum
transport due to helical edge states. However, such a conductance quantization in small
Hall-bar geometries does not allow us to distinguish experimentally between ballistic and
edge channel transport in a convincing manner. Thus it is important to be able to prove
14
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 15
experimentally in an unambiguous manner the existence of edge channels in HgTe QW.
2.2 Ohm’s law versus nonlocal transport
In conventional diffusive electronics, bulk transport satisfies Ohm’s law. The resistance is
proportional to the length and inversely proportional to the cross-sectional area, implying
the existence of a local resistivity or conductivity tensor. However, the existence of edge
states necessarily leads to nonlocal transport which invalidates the concept of local resis-
tivity. Such nonlocal transport has been experimentally observed in the QH regime in the
presence of a large magnetic field [45], and the nonlocal transport is well described by a
quantum transport theory based on the Landauer-Buttiker formalism [46]. These measure-
ments constitute definitive experimental evidence for the existence of edge states in the QH
regime.
In this chapter, we first describe nonlocal transport measurements in HgTe QW per-
formed by the group of Laurens W. Molenkamp at the Universityof Wurzburg which
demonstrate the existence of the predicted extended edge channels. Device structures that
are more complicated compared to a standard Hall bar allow a detailed investigation of
the transport mechanism. After describing the measurements, we present the theory of
quantum transport in the QSH regime, and uncover the effectsof macroscopic time irre-
versibility on the helical edge states.
2.3 Device structure
We present experimental results on four different devices.The behavior in these struc-
tures is exemplary for the around 50 devices we studied. The devices are fabricated from
HgTe/(Hg,Cd)Te QW structures with well thicknesses ofd = 7.5 nm (samples D1, D2 and
D3) and 9.0 nm (sample D4). Note that all wells have a thickness d > dc ≃ 6.3 nm, and
thus exhibit the topologically non-trivial inverted band structure. At zero gate voltage, the
samples aren-type and have a carrier density of aboutns = 3× 1011 cm−2 and a mobility
of 1.5× 105 cm2/(V·s), with small variations between the different wafers. Thedevices are
lithographically patterned using electron-beam lithography and subsequent Ar ion-beam
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 16
I1
2 3
4
56
V
I1
2 3
4
56
V
-0.5 0.0 0.5 1.0 1.5 2.00
5
10
15
20
25
30
35
40
R (
kΩ)
V* (V)
(1 x 0.5) µm2
(2 x 1) µm2
R14,14=3/2 h/e2
R14,23=1/2 h/e2
Figure 2.1: Two-terminal (R14,14) and four- terminal (R14,23) resistance versus (normalized)gate voltage for the Hall bar devices D1 and D2 with dimensions as shown in the insets.The dotted blue lines indicate the resistance values expected from the Landauer-Buttikerapproach.
etching. Devices D1 and D2 are micrometer-scale Hall bars with exact dimensions as indi-
cated in the insets of Fig. 2.1. D3 and D4 are dedicated structures for identifying non-local
transport, with schematic structure given in Fig. 2.2. All devices are fitted with a 110-nm-
thick Si3N4/SiO2 multilayer gate insulator and a 5/50 nm Ti/Au gate electrodestack. By
applying a voltageVg to the top gate the electron carrier density of the QW can be ad-
justed, going from ann-type behavior at positive gate voltages through the bulk insulator
state into ap-type regime at negative gate voltages. For reasons of comparison, the exper-
imental data in Fig. 2.1, 2.3, and 2.4 are plotted as a function of a normalized gate voltage
V ∗ = Vg−Vthr (Vthr is defined as the voltage for which the resistance is largest). Measure-
ments are performed at a lattice temperature of 10 mK for samples D1, D2 and D3 and at
1.8 K for sample D4 using low-frequency (13 Hz) lock-in techniques under voltage bias.
The four terminal resistance (Fig. 2.1) shows a maximum at about h/2e2, in agreement
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 17
with the results of Ref. [20]. The contact resistance shouldbe insensitive to the gate volt-
age, and can be measured from the resistance deep in the metallic region. By subtracting
the contact resistance we find that the two terminal resistance has its maximum of about
3h/2e2 (Fig. 2.1). This value is exactly what is expected from the theory of QSH edge
transport obtained from the Landauer-Buttiker formula.
2.4 Transport on the edge
Within the general Landauer-Buttiker formalism [47], thecurrent-voltage relationship is
expressed as
Ii =e2
h
∑
j
(TjiVi − TijVj), (2.1)
whereIi is the current flowing out of thei-th electrode into the sample region,Vi is the
voltage on thei-th electrode, andTji is the transmission probability from thei-th to thej-
th electrode. The total current is conserved in the sense that∑
i Ii = 0. A voltage leadj is
defined by the condition that it draws no net current,i.e. Ij = 0. The physical currents are
left invariant if the voltages on all electrodes are shiftedby a constant amountµ, implying
that∑
i Tij =∑
i Tji. In a TR invariant system, the transmission coefficients satisfy the
conditionTij = Tji.
For a general two-dimensional sample, the number of transmission channels scales
with the width of the sample, so that the transmission matrixTij is complicated and non-
universal. However, a tremendous simplification arises if the quantum transport is entirely
dominated by the edge states. In the QH regime, chiral edge states are responsible for the
transport. For a standard Hall bar withN current and voltage leads attached (see the insets
of Fig. 2.1 withN = 6), the transmission matrix elements for theν = 1 QH state are given
by T (QH)i+1,i = 1, for i = 1, . . . , N , and all other matrix elements vanish identically.
Here we periodically identify thei = N + 1 electrode withi = 1. Chiral edge states are
protected from backscattering, therefore, thei-th electrode transmits perfectly to the neigh-
boring (i+1)th electrode on one side only. In the example of current leads on the electrodes
1 and4, and voltage leads on the electrodes2, 3, 5 and6, one finds thatI1 = −I4 ≡ I14,
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 18
V2 − V3 = 0 andV1 − V4 = (h/e2)I14, giving a four-terminal resistance ofR14,23 = 0 and
a two-terminal resistance ofR14,14 = h/e2.
In the case of helical edge states in the QSH regime, oppositespin states form a Kramers
pair, counter-propagating on the same edge. The helical edge states are protected from
backscattering due to TR symmetry, and the transmission from one electrode to the next
is perfect. From this point of view, the helical edge states can be viewed as two copies of
chiral edge states related by TR symmetry. Therefore, the transmission matrix is given by
T (QSH) = T (QH) + T †(QH), implying that the only non-vanishing matrix elements are
given by
T (QSH)i+1,i = T (QSH)i,i+1 = 1. (2.2)
Considering again the example of current leads on the electrodes1 and4, and voltage leads
on the electrodes2, 3, 5 and6, one finds thatI1 = −I4 ≡ I14, V2 − V3 = (h/2e2)I14
andV1 − V4 = (3h/e2)I14, giving a four-terminal resistance ofR14,23 = h/2e2 and a two-
terminal resistance ofR14,14 = 3h/2e2. The experimental data in Fig. 2.1 confirm this
picture. For both micro Hall-bar structures D1 and D2 that differ only in the dimensions of
the area between the voltage contacts 3 and 4, we observe exactly the expected resistance
values forR14,23 = h/2e2 andR14,14 = 3h/2e2 for gate voltages where the samples are in
the QSH regime.
2.5 Dissipationless transport
Conceptually, one might sense a paradox between the dissipationless nature of the QSH
edge states and the finite four-terminal longitudinal resistanceR14,23, which vanishes for the
QH state. We can generally assume that the microscopic Hamiltonian governing the voltage
leads is invariant under TR symmetry, therefore, one would naturally ask how such leads
could cause the dissipation of the helical edge states, which are protected by TR symmetry?
In nature, the TR symmetry can be broken in two ways, either atthe level of the microscopic
Hamiltonian, or at the level of the macroscopic irreversibility in systems whose microscopic
Hamiltonian respects the TR symmetry. When the helical edgestates propagate without
dissipation inside the QSH insulator between the electrodes, neither forms of TR symmetry
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 19
breaking are present. As a result, the two counter-propagating channels can be maintained
at two different quasi chemical potentials, leading to a netcurrent flow. However, once
they enter the voltage leads, they interact with a reservoircontaining infinitely many low-
energy degrees of freedom, and the TR symmetry is effectively broken by the macroscopic
irreversibility. As a result, the two counter-propagatingchannels equilibrate at the same
chemical potential, determined by the voltage of the lead. Dissipation occurs with the
equilibration process. The transport equation (2.1) breaks the macroscopic TR symmetry,
even though the microscopic TR symmetry is ensured by the relationshipTij = Tji. In
contrast to the case of QH state, the absence of dissipation of the QSH helical edge states
is protected by Kramers’ theorem, which relies on the quantum phase coherence of wave
functions. Thus dissipation can occur once the phase coherence is destroyed in the metallic
leads. On the contrary, the robustness of QH chiral edge states does not require phase
coherence. A more rigorous and microscopic analysis on the different role played by a
metallic lead in QH and QSH states is provided in Sec. 2.7, theresult of which agrees
with the simple transport equations (2.1) and (2.2). These two equations correctly describe
the dissipationless quantum transport inside the QSH insulator, and the dissipation inside
the electrodes. One can subject these two equations to more stringent experimental tests
than the two- and four-terminal experiments of Fig. 2.1 by considering devices D3 and D4
(Fig. 2.2).
2.6 Helical versus chiral
A further difference between helical and chiral edge channels is evident from our experi-
ments on the six-terminal device D3 (Fig. 2.3). When the longitudinal resistance of device
D3 is measured by passing a current through contacts 1 and 4 and by detecting the voltage
between contacts 2 and 3 (R14,23), we find, similarly to the results of Fig. 2.1, the resistance
value ofh/2e2 when the bulk of the device is gated into the insulating regime (Fig. 2.3A).
However, the longitudinal resistance is significantly different in a slightly modified config-
uration, where the current is passed through contacts 1 and 3and the voltage is measured
between contacts 4 and 5 (R13,45) (Fig. 2.3B). We now findR13,45 ≈ 8.6 kΩ, which is
markedly different from what one would expect for either theQH transport, or the purely
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 20
1 μm 2 μm
1 2 3
6 5 4
1 μm
1 μm
5 μm
1 2
34
A B
Figure 2.2: Schematic layout of devices D3 (A) and D4 (B). Thegrey areas are the mesa’s,the yellow areas the gates, with dimensions as indicated in the figure. The numbers indicatethe coding of the leads.
diffusive transport, where this configuration would be equivalent to the previous. The ap-
plication of equations (2.1) and (2.2) actually predicts indeed that the observed behavior is
what one expects for helical edge channels. One finds that this resistance value can again be
expressed as an integer fraction of the inverse conductancequantae2/h: R13,45 = h/3e2.
This result shows that the current through the device is influenced by the number of ohmic
contacts in the current path. These ohmic contacts lead to the equilibration of the chemical
potentials between the two counter-propagating helical edge channels inside the contact.
There are also some devices for which the maximal resistancedoes not match the theo-
retical value obtained from Eq. (2.1) and (2.2), but still remains an integer fraction of the
quantumh/e2. This result can be naturally understood as due to inhomogeneities in the
gate action, e.g. due to interface trap states, inducing some metallic droplets close to the
edge channels while the bulk of the sample is insulating. A metallic droplet can cause
dephasing of the electronic wave function, leading to fluctuations in the device resistance.
For full dephasing, the droplet plays the role of an additional Ohmic contact, just as for
the chiral edge channels in the QH regime [45]. More details on the effects of additional
Ohmic contacts in the QSH state are given in Sec. 2.7.
Another measurement that directly confirms the nonlocal character of the helical edge
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 21
A B
-1 0 1 2 30
5
10
15
20
25
30
35
40
R(k
Ω)
V* (V)
I: 1-4
V: 2-3
R14,23=1/2 h/e2
R14,14=3/2 h/e2
I: 1-3
V: 5-6
R13,13=4/3 h/e2
R13,56=1/3 h/e2
-1 0 1 2 3 4
V* (V)
Figure 2.3: Four- and two-terminal resistance measured on device D3: (A)R14,23 (red line)andR14,14 (green line) and (B)R13,56 (red line) andR13,13 (green line). The dotted bluelines indicate the expected resistance value from a Landauer-Buttiker calculation.
channel transport in the QSH regime is shown in Fig. 2.4, which displays data obtained
from device D4, in the shape of the letter “H”. In this 4-terminal device the current is passed
through contacts 1 and 4 and the voltage is measured between contacts 2 and 3. In the
metallicn-type regime (low gate voltage) the voltage signal tends to zero. In the insulating
regime, however, the nonlocal resistance signal increasesto approximately6.5 kΩ, which
again fits perfectly to the result of Laudauer-Buttiker considerations:R14,23 = h/4e2 ≈6.45 kΩ. Classically, one would expect only a minimal signal in thisconfiguration (from
Poisson’s equation, assuming diffusive transport, one estimates a signal of about 40Ω), and
certainly not one that increases so strongly when the bulk ofthe sample is depleted. This
signal measured here is fully nonlocal, and can be taken (as was done twenty years ago for
the QH regime) as definite evidence of the existence of edge channel transport in the QSH
regime. A similar nonlocal voltage has been studied in a metallic spin Hall system with
the same H-bar geometry [48], in which case the nonlocal voltage can be understood as
a combination of the spin Hall effect and the inverse spin Hall effect [49]. The quantized
nonlocal resistanceh/4e2 we find here is the quantum counterpart of the metallic case.
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 22
Fig. 4
0.0 0.5 1.0 1.5 2.00
5
10
15
20
25
R (
kΩ)
V* (V)
I: 1-4
V: 2-3
1
3
2
4
R14,23=1/4 h/e2
R14,14=3/4 h/e2
Figure 2.4: Nonlocal four-terminal resistance and two-terminal resistance measured on theH-bar device D4:R14,23 (red line) andR14,14 (green line). Again, the dotted blue linerepresents the theoretically expected resistance value.
Assuming for example that the chemical potential in contact1 is higher than that in contact
4 (see the layout of D4 in Fig. 2.2B), more electrons will be injected into the upper edge
state in the horizontal segment of the H-bar than into the lower edge state. Because on
opposite edges, the right-propagating edge states have opposite spin, this implies that a
spin-polarized current is generated by an applied biasV1 − V4, comparable to a spin Hall
effect. When this spin-polarized current is injected into the right leg of the device, the
inverse effect occurs. Electrons in the upper edge flow to contact 2 while those in the lower
edge will flow to contact 3, establishing a voltage difference between those two contacts
due to the charge imbalance between the edges. The right leg of the device thus acts as a
detector for the injected spin-polarized current, which corresponds to the inverse spin Hall
effect.
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 23
2.7 Dephasing and nonlocal transport
In this section, we clarify the role played by contacts in edge transport in QSH systems by
answering the following questions:
• In the absence of microscopic TR symmetry breaking, why doesa contact contribute
an additional longitudinal resistance, if edge state backscattering is forbidden by TR
symmetry?
• How can we understand the different role played by contacts in QH and QSH sys-
tems?
• Besides a voltage probe, can other processes cause an additional resistance?
The answers are as follows:
• A contact isnot a TR symmetry breaking single-particle potential with matrix ele-
ments connecting counter-propagating edge channels with opposite spins. An contact
is, ultimately, a reservoir of electrons which populates both channels incoherently.
Furthermore, anideal contact populates both channelswith equal weight, i.e. it in-
jects spin up and spin down electrons with equal probability. A right-moving electron
entering a contact will disappear into the reservoir. Sincethe contact is assumed not
to draw any current, the reservoir has to inject a new electron into the device. This
new electron can be of either spin, and bears no phase relationship with the previous
electron. For an ideal contact, this electron has1/2 probability of being injected as a
left-mover, which is the origin of the resistance contributed by the contact. Note that
this incoherent momentum-relaxing event is different fromusual coherent momen-
tum relaxation caused by potential scattering in an ordinary (non-helical) 1D system.
• In the QH case, the channels on each edge are chiral such that the longitudinal two-
terminal conductance is always quantized tone2/h with n the number of channels
on each edge, regardless of the presence of additional lateral contacts. Quantization
of the conductance is maintained independently of whether transport is coherent or
incoherent.
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 24
• Based on analytical and numerical results, we show that a finite-sized region with
dephasing processes can cause additional longitudinal resistance in the same way as
a contact does.
The structure of this section is as follows. Starting from the quantum kinetic formalism
embodied in the Meir-Wingreen formula (Sec. 2.7.1), we firstshow how arbitrary dephas-
ing processes contributing a self-energy to the one-particle Green function can be formally
thought of as a contact. In particular, we show that the equilibrium two-terminal conduc-
tance calculated from the quantum kinetic formalism agreeswith the conductance calcu-
lated from thethree-terminal Landauer-Buttiker formula if the third terminal is identified
with the “dephasing bath”. Then, using a four-band tight-binding model on a square lattice
for HgTe/CdTe QW, we perform numerical calculations of the conductance in the QSH
regime (Sec. 2.7.2). We show how dephasing processes contribute the expected additional
resistance, which constitutes an explicit justification ofLandauer-Buttiker phenomenology.
By adding an orbital magnetic field to the model, we also perform conductance calculations
in the QH regime (Sec. 2.7.3) and compare to the previous QSH results. We then proceed
with a short discussion of the transformation properties ofthe self-energy under time re-
versal (Sec. 2.7.4) to emphasize that resistance in a QSH system is caused by macroscopic,
and not microscopic, TR symmetry breaking. Finally, we perform an explicit numerical
calculation of the nonlocal resistance in a H-bar structure(Sec. 2.7.5) similar to that of
sample D4 (Fig. 2.4) using the four-band tight-binding model. We find good agreement
with both the prediction of the Landauer-Buttiker formalism and the experimental results,
thus validating the picture of transport by edge states.
2.7.1 Quantum kinetic formalism and Landauer-Buttiker equations
The Meir-Wingreen formula [50] for the current through a terminal connected to a meso-
scopic region is a natural starting point since it is an exactresult for noninteracting leads
derived from the quantum kinetic formalism, valid out of equilibrium. It encompasses the
Kubo formula and the Landauer-Buttiker formula as specialcases. It expresses the current
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 25
flowing into the mesoscopic region through terminali as
Ii =ie
htr
∫
dE [Γi(E)G<(E)− Σ<
i (E)A(E)], (2.3)
whereA = i(G − G†) is the nonequilibrium spectral function,G,G< are nonequilibrium
Green functions, andΣ<i ,Γi = i(Σi − Σ†
i) are the lesser self-energy and broadening func-
tions of the semi-infinite leads, respectively.
Dephasing region and Landauer-Buttiker equations
The Green functionG contains the total self-energyΣ which comprises not only the self-
energies of the leadsΣj , but also the self-energyΣd of any processes occurring in the
mesoscopic region that cannot be described by a single-particle Hermitian Hamiltonian,
Σ = Σd +∑
j
Σj .
We will refer to the processes described byΣd asdephasingprocesses.G< andA are given
by [51, 52]
G< = G
(
Σ<d +
∑
j
Σ<j
)
G†, A = G
(
Γd +∑
j
Γj
)
G†.
One then makes the following ansatze,
Σ<j = ifjΓj, Σ<
d = ifdΓd, (2.4)
that is, the self-energies should obey the fluctuation-dissipation theorem [53]. Due to the
self-energies, the eigenstates of the isolated mesoscopicregion acquire a finite lifetime.
Equations (2.4) mean that electrons “leaking out” of the mesoscopic region end up in a
“reservoir” with distribution functionfj,d according to which they are equilibrated. It is
this precise requirement which enables us to derive the Landauer-Buttiker result. That is,
an arbitrary dephasing region and a contact can be treated onthe same footing if they both
obey the fluctuation-dissipation theorem.
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 26
Using the preceding equations, and restricting ourselves to linear response
fi,d ≃ f +
(
− df
dE
)
(µi,d − µ), (2.5)
wheref(E) = (eβ(E−µ) + 1)−1, we find easily
Ii =e
h
∫
dE
(
− df
dE
)
[
∑
j
Tij(µi − µj) + Tid(µi − µd)
]
,
where
Tij(E) ≡ tr ΓiGΓjG†, Tid(E) ≡ tr ΓiGΓdG
†. (2.6)
By virtue of the second equality in Eq. (2.4) it makes sense toconsider the current drawn
by the dephasing reservoir characterized by the distribution functionfd. After linearization
[Eq. (2.5)] we obtain
Id =e
h
∫
dE
(
− df
dE
)
∑
j
Tdj(µd − µj),
whereTdj(E) ≡ tr ΓdGΓjG†. We now require that the dephasing reservoir does not draw
any current,
Id = 0.
In other words, it acts as a voltage probe. DenotingGij ≡ e2
h
∫
dE(
− dfdE
)
Tij(E), we have
the set of equations
Ii =∑
j
Gij(Vi − Vj) +Gid(Vi − Vd),
Id =∑
j
Gdj(Vd − Vj) = 0,
which are just the multiterminal Landauer-Buttiker equations with a voltage probe.
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 27
Two-terminal case with dephasing
We study as a particular case the two-terminal geometry which we will use for the numer-
ical analysis. We consider two physical contacts1, 2 with I = I1 = −I2. Solving the
previous equations we obtain the two-terminal conductance,
G2t ≡I
V1 − V2= G12 +
G1dGd2
Gd1 +Gd2,
which is just the three-terminal Landauer-Buttiker result if we considerd as a terminal. We
write
G2t = Gcoh +Gincoh,
with Gcoh = G12 the coherent part of the conductance andGincoh = G1dGd2/(Gd1 + Gd2)
the incoherent part. This result is valid in the absence of TRsymmetry (hence holds for the
QH case to be considered later). If TR symmetry is preserved,thenG1d = Gd1 and we can
write
G2t = G12 +
(
1
G1d+
1
Gd2
)−1
,
which corresponds to two resistors in series, combined in parallel with a third.
2.7.2 Dephasing in the QSH regime: numerical study
To understand the effect of a dephasing region in the quantumspin Hall system, we now
perform an explicit numerical calculation of the two-terminal conductanceG2t in the HgTe
system. The geometry (Fig. 2.5) consists of a finite QSH region connected tosemi-infinite
terminals 1 and 2, and to afinitedephasing region 3. Corresponding to the continuumk ·pHamiltonian of Ref. [19], one can derive the following tight-binding (TB) model on a 2D
square lattice,
H =∑
i
c†iVici +∑
i
(
c†iTxci+x + c†iTyci+y + h.c.)
, (2.7)
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 28
where theci are four-component spin-3/2 spinors, and the4 × 4 matricesVi, Tx, Ty are
defined as
Tx =1
2(2D14×4 + 2B12×2 ⊗ σz + 2βσy ⊗ σy)
+1
2i
(
vσz ⊗ σx +∆e +∆h
2σx ⊗ 12×2 +
∆e −∆h
2σx ⊗ σz
)
,
Ty =1
2(2D14×4 + 2B12×2 ⊗ σz + 2βσy ⊗ σy)
+1
2i
(
−v12×2 ⊗ σy −∆e +∆h
2σy ⊗ σz −
∆e −∆h
2σy ⊗ 12×2
)
,
Vi = (C − 4D − µ+ Eg(i))14×4 + (M − 4B)12×2 ⊗ σz + (∆z − 4β)σy ⊗ σy,
whereµ is the chemical potential andEg is the gate potential used to tune the Fermi level
inside the bulk energy gap in the device region. The parameters v, B, C, D, M , β, ∆e,h,z
are obtained fromk · p theory [54]. The self-energies of the semi-infinite leadsΣj are
calculated by a transfer matrix method [55]. The dephasing region 3 is a metallic region
of finite sizeNy × L and differs from the other regions in that an imaginary term−iη is
added to the energy of each site, i.e. we have a local self-energy (Σd)iα,jβ = −iηδijδαβfor sitesi, j with spin indicesα, β in region 3. The Green functionG of the QSH region
in the presence of all three self-energies is calculated by arecursive method [56], and the
transmission functionsTij(µ) andTid(µ) at the Fermi level are subsequently obtained from
Eq. (2.6). We then study the dependence of the zero-temperature conductancesGcoh,Gincoh
andG2t onη andL.
First of all, one can understand the two limit cases intuitively. For a fixed small value of
η, the region 3 is a true terminal in the limit ofL → ∞, and the two-terminal conductance
G2t agrees with the prediction of the Landauer-Buttiker formalism. From Fig. 2.5 we can
just read offT12 = 1, T13 = 1 andT32 = 1 such that we expectG2t =32e2
h. On the other
hand, in the limit ofL→ 0 the top edge is unperturbed and we expectG2t = 2e2/h. To see
how a crossover between the two limits occurs, we study the two-terminal conductance for
genericL numerically.
Results of the numerical calculation are plotted in Fig. 2.6. For each fixed value ofη,
we see the crossover between the two limit values2e2/h (L → 0) and3e2/2h (L → ∞)
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 29
Figure 2.5: Two-terminal geometry in the QSH regime with dephasing. Regions 1 and 2are semi-infinite leads and region 3 has finite sizeNy×L and contains dephasing processesmodeled by a local self-energyΣiα,jβ = −iηδijδαβ . In the QSH region the Fermi level liesin the bulk gap, and it lies in the conduction band in regions1, 2, 3.
as expected from the Landauer-Buttiker analysis. The off-resonance conductanceG2t (see
Fig. 2.7 and the discussion of resonances in the next paragraph) scales withηL (Fig. 2.6B).
Upon increasingL, the convergence ofG2t to the asymptotic value of3e2/2h occurs at a
length scaleLc ∝ 1/η. Indeed, forη = 0, region 3 is just a single-particle TR invariant
Hamiltonian which cannot induce any backscattering. Transport is then entirely coherent
(Fig. 2.8A). Asη is increased from zero, there is a growing incoherent contributionGincoh
which increases at the expense of the coherent contributionGcoh in order to preserve cur-
rent conservation. The electron flux incoming from the left is partially decohered with
probability per unit time∼ η/~. The remaining flux contributes to the remaining coher-
ent contributionGcoh (Fig. 2.8B). AsηL increases, the coherent conductance eventually
levels off toe2/h which is just the contribution from the unperturbed edge. Transport on
the top edge then becomes totally incoherent (Fig. 2.8C) with Gcoh = 0.5 e2/h. This is
the signature of an ideal contact which populates the two counter-propagating edge states
equally. In this regime, the finite dephasing region completely mimicks a true semi-infinite
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 30
2040
6080
100
0
0.005
0.010
0.5
1
1.5
2
L (a)η (eV)
G (
e2 /h)
G2t
Gcoh
Gincoh
A
0 20 40 60 80 50 60 70 801.4
1.5
1.6
1.7
1.8
1.9
2
ηL (meV⋅a)
G2t
(e2 /h
)
0.0010.0680.130.200.270.330.400.470.672.0
Bη (meV)
Figure 2.6: (A) Two-terminal conductance in the QSH regime.Top curve (red): totalconductance, middle curve (green): coherent partGcoh, bottom curve (blue): incoherentpartGincoh. We useNy = 40, Nx = 120, and a TB lattice constanta = 60 A for allcalculations. (B) Scaling plot for the conductanceG2t ∼ G2t(ηL) for values ofL awayfrom transmission resonances (see Fig. 2.7).
lead (Fig. 2.8D).
There are however sharp resonant dips in the conductance forsmall but nonzero values
of η (Fig. 2.7). These resonances appear because forη smaller than the level spacing in the
dephasing region∆E ∼ 1/L, the self-energy contributed by the dephasing region to the
QSH region has sharp features. AsL is varied, these features cross the Fermi energy and
lead to a sharp decrease in the coherent conductance. From Fig. 2.7 we see that even for
small ηL, these resonances can be strong enough as to completely block coherent trans-
port on the top edge (G2t → 1.5e2/h). Consequently, even a small dephasing region may
have the same effect as a Buttiker probe. Experimentally, in some of the devices D3 we
have observed deviation of the four-terminal conductance from the value predicted by the
Landauer-Buttiker formalism. For example, in the data shown in Fig. 2.7B, the six-terminal
Landauer-Buttiker formula predicts a four-terminal resistance ofR45,12 = h/6e2, but the
experimental value of the maximal resistance is smaller andclose toh/7e2 instead. Such a
result is consistent with the existence of an additional dephasing region, as shown schemat-
ically by the red solid circle in the inset of Fig. 2.7B. Such dephasing regions can exist due
to the inhomogeneity of the sample. Experimentally, we havenoticed that changing the
gate voltage also influences the homogeneity of the sample bycharging and discharging
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 31
0 20 40 60 80 1001.5
1.6
1.7
1.8
1.9
2
Ny = 40, N
b = 40, E
g = 0.015 eV, η (meV) is varied
L (a)
G (
e2 /h)
0.0010.010.050.10.250.5
A
-1,5 -1,0 -0,5 0,0 0,5 1,00
1
2
3
4
5
6
R (
kΩ
)
V* (V)
B
I: 4-5
V: 1-2
R45,12=1/6 h/e2
R45,12=1/7 h/e2
Figure 2.7: (A) Total two-terminal conductanceG2t in the QSH regime for small valuesof dephasing strengthη. On resonance, coherent transport on the top edge can be entirelyblocked withG2t dropping to the completely incoherent limit 1.5e2/h. (B) Four-terminalresistanceR45,12 in a device D3. The blue dashed line shows the theoretical valueh/6e2
from the Landauer-Buttiker formalism. The green dashed line shows the theoretical valueh/7e2 when considering the effect of an additional probe, as shownschematically by thered solid circle in the inset (see text).
trap states at the semiconductor interface. This tends to aninhomogeneous potential profile
in the gated area so that metallic islands still exist when most of the gated area is insulating.
As shown by the theoretical analysis here, a metallic islandcan lead to a similar effect as
an additional probe.
The characteristic length for dephasingLc, which can be defined operationally as the
lengthL required for the conductance to reach the completely incoherent limit G2t = 1.5
e2/h, is thus seen to depend on the dephasing strength asLc ∼ 1/η whereτ ∼ ~/η
would correspond to the decoherence time associated with some physical dephasing pro-
cess, e.g. electron-phonon or electron-electron interactions. Note that the dephasing self-
energyΣ ∝ η does not have to break TR symmetry (see Sec. 2.7.4) to decrease the con-
ductance. Decoherence comes from the presence of the dephasing bath with a distribution
function included inΣ<, not fromΣ breaking TR symmetry.
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 32
Figure 2.8: Cartoon picture of edge transport in the QSH regime in the presence of de-phasing. Only right-moving states are shown. Bottom edge isunaffected by dephasing andcontributes a fixede2/h toG2t. Coherent transport is denoted by solid lines and incoherenttransport by dotted lines. (A) Forη = 0, top edge state propagates coherently without re-sistance. (B) Forη 6= 0, top edge state is phase-randomized (yellow cloud) with probabilityper unit timeη/~ and contributes to incoherent transport, or is not phase-randomized andcontributes to coherent transport. For increasingηL, incoherent transport is enhanced atthe expense of coherent transport. (C) ForηL large enough, transport is completely inco-herent. This case is equivalent to (D) which represents the true semi-infinite lead attachedto an electron reservoir.
2.7.3 Dephasing in the QH regime: numerical study
It is instructive to investigate the effect of dephasing in the QH regime and compare it to
the QSH case just studied. To reach the QH regime we add to the HgTe/CdTe TB model
an orbital magnetic fieldB = Bz perpendicular to the 2D electron gas. When the Fermi
level lies inside the Landau level gap, transport is carriedby chiral edge states (Fig. 2.9).
We chooseB andEF such that there is a single chiral state on each edge.
From the Landauer-Buttiker analysis we expect the two-terminal conductance to be
G2t = e2/h since there is only one right-moving channel, on the top or bottom edge de-
pending on the sign of the magnetic field. The numerical results for the conductance are
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 33
Figure 2.9: Two-terminal geometry in the QH regime with dephasing. The Fermi level liesbetween Landau levels and transport is carried by chiral edge states.
given in Fig. 2.10 forBz = 1 T and Fig. 2.11 forBz = −1 T.
In Fig. 2.10 withBz = 1 T, the right-mover is on the bottom edge and is therefore
unperturbed by the dephasing region (Fig. 2.12D). Transport is therefore entirely coherent,
with G2t = Gcoh = e2/h andGincoh = 0 (numerically, one obtains a very small but finite
Gincoh due to interedge tunneling across the finite device widthNy. Upon increasingNy,
Gincoh gets vanishingly small).
On the other hand, if we reverse the sign of the magnetic field,in Fig. 2.11 forBz = −1
T the right-mover is on the top edge and is therefore affectedby the dephasing region. For
weak enoughη, as in the QSH case transport is mostly coherent (Fig. 2.12A), except for
resonances. AsηL increases, the incoherent partGincoh increases (Fig. 2.12B). For large
enoughηL, transport is completely incoherent (Fig. 2.12C). However, the sum ofGcoh and
Gincoh (not shown in Fig. 2.11) remains quantized toe2/h. This is because of the chiral
nature of the edge states: unlike the QSH case, in the QH case decoherence does not lead
to momentum relaxation.
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 34
2040
6080
100
0
5
100
0.2
0.4
0.6
0.8
1
L (a)η (meV)
G (
e2 /h)
Gcoh
Gincoh
Figure 2.10: Two-terminal conductance in the QH regime withBz = 1 T, correspondingto Fig. 2.12D. Top curve (red): coherent partGcoh, bottom curve (blue): incoherent partGincoh.
2.7.4 Self-energy and microscopic time-reversal symmetry
In this section, we describe briefly the transformation properties of the self-energy under
TR and show that the dephasing self-energy used in the numerical calculation does not
break TR symmetry.
Consider a Hermitian single-particle Hamiltonian operator H = H† which is invariant
under the action of the TR operatorT , T H = HT . This impliesΘHΘ−1 = H∗ for the
single-particle Hamiltonian matrixH in a given basis, withT = ΘK the representation of
T in that basis,K is the complex conjugation operation which acts only onc-numbers and
Θ2 = −1. In the single-particle case, the retarded Green function in frequency space is
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 35
0
20
40
60
80
100
0
5
100
0.5
1
L (a)η (meV)
G (
e2 /h)
Gcoh
Gincoh
Figure 2.11: Two-terminal conductance in the QH regime withBz = −1 T, correspondingto Fig. 2.12A,B,C. Top curve (red): incoherent partGincoh, bottom curve (blue): coherentpartGcoh.
given by the matrix inverseG = (ω + iδ −H)−1. We then have
G∗ = (ω − iδ −H∗)−1
= (ω − iδ −ΘHΘ−1)−1
= Θ(ω − iδ −H)−1Θ−1
= ΘG†Θ−1,
or
ΘGΘ−1 = GT , (2.8)
usingΘ† = Θ−1.
We now show that Eq. (2.8) also holds in the case thatH is a general many-body
Hamiltonian as long as it is TR invariant. In other words, we want to show that Eq. (2.8)
holds even if the Green function contains a self-energy term. Consider the zero temperature
Green function〈0|ciα(t)c†jβ(t′)|0〉 whereiα denotes the single-particle basis,i being a
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 36
Figure 2.12: Cartoon picture of edge transport in the QH regime in the presence of de-phasing. Only right-moving states are shown. Bottom edge isunaffected by dephasing.Coherent transport is denoted by solid lines and incoherenttransport by dotted lines. (A)Forη = 0, top edge state propagates coherently without resistance,in exactly the same wayas in Fig. 2.8A. (B) Forη 6= 0, partial phase-randomization occurs as in Fig. 2.8B, exceptthat decohered electrons can only be injected as right-movers since there is no left-movingchannel on the top edge. (C) For large enoughηL, transport is totally incoherent. (D) Ifthe right-mover is on the bottom edge, transport is undisturbed by the dephasing region andthus completely coherent.
site index andα a spin index. (At finite temperature the result is the same since the density
matrix is TR invariant.) Since the ground state is TR invariant T |0〉 = |0〉, we have
〈0|ciα(t)c†jβ(t′)|0〉 =
(
c†iα(t)|0〉, c†jβ(t′)|0〉)
=
(
c†iα(t)T |0〉, c†jβ(t′)T |0〉)
=
(
T T−1c†iα(t)T |0〉, T T−1c†jβ(t′)T |0〉
)
.
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 37
Now the TR transformation of the creation operators is
T−1c†iα(t)T = θαα′c†iα′(−t),
with a sum over repeated indices, andt → −t since the complex conjugation operationK
acts on the factors ofi in the exponential of the Heisenberg operatorsc(t) = eiHtce−iHt (but
not on the Hamiltonian operatorH itself for which complex conjugation is not defined).
For fermions we haveθ2 = −1. We obtain
〈0|ciα(t)c†jβ(t′)|0〉 = θαα′θββ′
(
T c†iα′(−t)|0〉, T c†jβ′(−t′)|0〉)
.
Using the antiunitary property(
T |φ〉, T |χ〉)
= 〈χ|φ〉 of T , we obtain
〈0|ciα(t)c†jβ(t′)|0〉 = θαα′θββ′
(
c†jβ′(−t′)|0〉, c†iα′(−t)|0〉)
= θαα′θββ′〈0|cjβ′(−t′)c†iα′(−t)|0〉.
In equilibrium, the correlation functions are time-translationally invariantG(t, t′) = G(t−t′), hence
〈0|ciα(t)c†jβ(t′)|0〉 = θββ′〈0|cjβ′(t)c†iα′(t′)|0〉(θT )α′α,
or
ΘGΘ−1 = GT ,
whereΘ = θ⊕· · ·⊕θ, the transpose is with respect to the single-particle basisand we used
ΘT = Θ−1. We have carried out the derivation forG> but this property holds regardless of
the time ordering and is thus valid for the various Green functionsG<,>,R,A,T,T . It is also
valid in frequency space. The retarded self-energy is defined byΣ = G−10 − G−1, where
G0 andG are the unperturbed and perturbed retarded Green functions, respectively. From
Eq. (2.8) which is valid for an arbitrary interacting Hamiltonian, it immediately follows
that
ΘΣΘ−1 = ΣT .
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 38
Then a local self-energy, diagonal in spin space,
Σiα,jβ(ω) = −iηδijδαβ,
will satisfy this requirement. Hence this means that our additional self-energy does not
break TR symmetry.
2.7.5 Nonlocal resistance in H-bar structure: numerical study
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07−0.15
−0.1
−0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
Eg (eV)
R12
,34 (
h/e2 )
B
Figure 2.13: (A) Geometry of the H-bar structure. Semi-infinite metallic leads in then-typeregime are attached to contacts 1-4. Contacts 1 and 2 are current probes and contacts 3 and 4are voltage probes that draw no current. The calculation is performed forW = 0.24µm andL = 1.2µm, corresponding to the same aspect ratio as that of device D4(see Fig. 2.2). (B)Solid line: disordered-averaged nonlocal resistanceR12,34 as a function of gate potentialEg, dotted line: prediction of the Landauer-Buttiker formalism in the edge state picture,R12,34 = h/4e2.
Using the TB model [Eq. (2.7)], one can directly calculate the nonlocal resistanceR12,34
for the H-bar geometry illustrated in Fig. 2.13A. In the linear response regime, the nonlocal
resistance is simply obtained by solving the Landauer-Buttiker equations [57],
R12,34 ≡V3 − V4
I=
h
e2T31T42 − T32T41
D,
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 39
whereI ≡ I1 = −I2 and
D ≡ det
T12 + T13 + T14 −T12 −T13−T31 −T32 T31 + T32 + T34
−T41 −T42 −T43
,
where we consider for simplicity the case of zero temperature withTij ≡ Tij(µ), and the
transmission coefficientsTij are obtained from the Green function of the whole device as
in Eq. (2.6). For ad = 89.9 A device, the mass gap parameterM obtained fromk · pcalculations is|M | = 24.5 meV and the velocity is~v = 3.34 eV·A, such that the effective
mass of the carriers at the bottom of the conduction band is approximatelym∗ ≃ M/v2 =
0.0167m0 wherem0 is the bare electron mass. Considering the experimental mobility of
µ = 1.5 × 105 cm2/(V·s), we can estimate the inverse lifetime due to disorder as~/τ =e~
m∗µ= 0.46 meV.
The results of the calculation are plotted in Fig. 2.13B. We have averaged the resis-
tance over 50 configurations of Gaussian on-site disorder. The lattice constant of the tight-
binding model isa = 30 A. The resistance plateau in the nominally insulating regime
agrees very well with the predictionR12,34 = h4e2
of the Landauer-Buttiker formalism for
pure edge state transport. These numerical results thus confirm the edge state picture of
transport in the quantum spin Hall insulator regime.
2.8 Conclusion
The multi-terminal and nonlocal transport experiments on HgTe microstructures in the
QSH regime demonstrate that charge transport occurs through extended helical edge chan-
nels. We have extended the Landauer-Buttiker model for multi-terminal transport in the
QH regime to the case of helical QSH edge channels, and have shown that this model con-
vincingly explains the observations. Currently, logic devices based on the CMOS design
generate significant heating due to the ohmic dissipation within the channel. Our work
on the conductance quantization demonstrates that electrons can be transported coherently
CHAPTER 2. NONLOCAL TRANSPORT IN THE QSH STATE 40
within the edge channel without ohmic dissipation. Such an effect can be used to construct
a new generation of logic devices with improved performance.
Chapter 3
Kondo effect in the helical edge liquid of
the quantum spin Hall state
3.1 Introduction
The QSH insulator is a topologically nontrivial state of matter [54] that has recently been
observed in transport experiments carried out in HgTe QW [20] following its theoretical
prediction [19]. The two-dimensional QSH insulator has a charge excitation gap in the
bulk but supports one-dimensional gapless edge states forming a so-called “helical liquid”:
on each edge there exists a Kramers’ pair of counter-propagating states with opposite spin
polarization. The QSH insulator is robust against weak single-particle perturbations which
preserve TR symmetry such as weak potential scattering [13,21, 44].
This theoretical picture is consistent with experimental observations: the longitudinal
conductanceG in a Hall bar measurement is approximately quantized toG0 = 2e2/h,
independent of temperature, for samples of about a micron length [20, 58]. However,
larger samples exhibitG < G0 andG decreases with decreasing temperature [58]. De-
viations from the expected quantized value have been attributed to the presence of local
doped regions due to potential inhomogeneities within the sample arising from impurities
or roughness of the well/barrier interface [58]. Although pure potential scattering can-
not backscatter the edge states, the role of these potentialinhomogeneities is to trap bulk
electrons which may then interact with the edge electrons. These localized regions act as
41
CHAPTER 3. KONDO EFFECT IN THE QSH STATE 42
dephasing centers for the edge channels due to interaction effects and may cause backscat-
tering.
Figure 3.1: Temperature dependence of the conductance: thebehavior is logarithmic athigh temperatureT ≫ T ∗ and power-law at low temperatureT ≪ T ∗. At T = 0, ametal-insulator quantum phase transition is driven by Coulomb interactions in the helicalliquid: the system is a “Kondo metal” forK > 1/4 and a “Luttinger liquid insulator” forK < 1/4. The Fano factore∗ is defined as the ratio between shot noise and current, andreflects the charge of the current-carrying excitations.
In this work, we study theoretically the temperature dependence of the edge conduc-
tance of a QSH insulator. We consider the case where a local doped region in the vicinity
of the edge contains an odd number of electrons and acts as a magnetic impurity coupled
to the helical edge liquid. Our main results are as follows (Fig. 3.1):
1. At high temperatures, the conductance is logarithmic,−∆G ≡ −(G − G0) = η +
γ ln(D/T ) whereη, γ are interaction-dependent parameters andD is an energy scale
of order the bulk gap.
2. For weak Coulomb interactionsK > 1/4 whereK is the Luttinger parameter of the
edge liquid, the conductance is restored to unitarity atT = 0 due to the formation of
CHAPTER 3. KONDO EFFECT IN THE QSH STATE 43
a Kondo singlet. This is in stark contrast with the Kondo problem in a usual spinful
1D liquid where the conductance vanishes atT = 0 for all Kρ < 1 whereKρ is the
Luttinger parameter in the charge sector [59]. At low but finite T the conductance
decreases as an unusual power-law∆G ∝ −T 2(4K−1) due to correlated two-particle
backscattering. The edge liquid being helical, the decrease in conductance is a direct
measure of the spin-flip rate [60].
3. For strong Coulomb interactionsK < 1/4, two-particle backscattering processes
are relevant and the system becomes insulating atT = 0. At low but finite T , the
conductance is restored by tunneling of excitations with fractional chargee/2 and we
obtainG(T ) ∝ T 2(1/4K−1).
3.2 Theoretical model
We model the impurity by aS = 12
local spin coupled by exchange interaction to the 1D
helical liquid with Coulomb interactions. The helical liquid having the same number of
degrees of freedom as a spinless fermion, a single nonchiralbosonφ is sufficient for its
description in the bosonized language [21]. The system is described by the Hamiltonian
H = H0 +HK +H2 whereH0 is the usual Tomonaga-Luttinger Hamiltonian [61],
H0 =v
2
∫
dx
[
KΠ2 +1
K(∂xφ)
2
]
,
with K the Luttinger parameter andv the edge state velocity. The Kondo HamiltonianHK
has the form
HK =J‖a
2πξ
(
S− : e−i2
√πφ(0) : + H.c.
)
− Jza√πSzΠ(0), (3.1)
whereS± = Sx ± iSy andSz are the spin operators for the impurity localized atx = 0.
a is the lattice constant of the underlying 2D lattice and corresponds to the size of the
impurity (we assume that the impurity occupies a single lattice site). ξ, the penetration
length of the helical edge states into the bulk, acts as a short-distance cutoff for the 1D
continuum theory in the same way that the magnetic length, the penetration length of the
CHAPTER 3. KONDO EFFECT IN THE QSH STATE 44
chiral edge states, acts as a short-distance cutoff for the chiral Luttinger liquid theory of the
QH edge [62]. In addition to Kondo scattering, two-particlebackscattering is allowed by
TR symmetry [21, 44]. In HgTe QW the wavevector at which the edge dispersion enters the
bulk is usually much smaller thanπ/2a such that the uniform two-particle backscattering
(umklapp) term requiring4kF = 2π/a can be ignored. The impurity potential can however
provide a4kF momentum transfer and we must generally also consider a local impurity-
induced two-particle backscattering term [63]
H2 =λ2a
2
2π2ξ2: cos 4
√πφ(0) : ,
whereλ2 is the two-particle backscattering amplitude.
3.3 Weak coupling regime
We first consider the weak coupling regime whereJ‖, Jz andλ2 are small parameters.
The calculation of the conductance proceeds in two steps. Wefirst perform an explicit
perturbative calculation of the conductance to quadratic order in the bare couplingsJ‖ and
λ2 using the Kubo formula. This result is then extended to include all leading logarithmic
terms in the perturbation expansion by means of a weak coupling renormalization group
(RG) analysis of the scale-dependent couplingsJ‖(T ) andλ2(T ) where the scale is set by
the temperatureT .
The forward scattering termJz can be removed from the Hamiltonian by a unitary
transformation [64] of the formU = eiλSzφ(0) with λ = −Jza/vK√π, at the expense
of modifying the scaling dimension of the vertex operator: ei2√πφ : . The transformed
Hamiltonian reads
UHU † = H0 +H2 +J‖a
2πξ
(
S− : e−i2√πχφ(0) : + H.c.
)
,
whereχ = 1 − νJz/2K with ν = a/πv the density of states of the helical liquid. The
scaling dimension of: ei2√πχφ : is K ≡ Kχ2.
Using the transformed Hamiltonian we obtain the correctionto the conductance∆G(T ) =
CHAPTER 3. KONDO EFFECT IN THE QSH STATE 45
∆GK(T ) + ∆G2(T ) to quadratic order in the couplingsJ‖ andλ2 using the Kubo formula
(see Sec. A.2), where∆GK is the correction due to spin-flip Kondo scattering1,
∆GK
e2/h= −Γ(1
2)Γ(K)
Γ(12+ K)
π2
3S(S + 1)(νJ‖(T ))
2, (3.2)
whereνJ‖(T ) = νJ‖(T/D)K−1 to O(J‖) andD = ~v/πξ is a high-energy cutoff of order
the bulk gap.∆G2 is the correction due to two-particle backscattering (see Sec. A.2),
∆G2
e2/h= −Γ(1
2)Γ(4K)
Γ(12+ 4K)
a4
2π4ξ4
(
λ2(T )
D
)2
, (3.3)
whereλ2(T ) = λ2(T/D)4K−1. There are no crossed terms of the formO(J‖λ2) orO(Jzλ2)
for a S = 12
impurity since the two-particle backscattering operator flips two spins but
a S = 12
spin can be flipped only once. We however expect that such terms would be
generated for impurities with higher spin.
These results can be complemented by a RG analysis. The RG equation forλ2 follows
by dimensional analysisdλ2
dℓ= (1−4K)λ2 with ℓ = ln(D/T ), so thatλ2 is relevant forK <
1/4 and irrelevant forK > 1/4. The renormalized coupling isλ2(T ) = λ2(T/D)4K−1, and
second order renormalized perturbation theory∆G2(T ) ∝ −λ2(T )2 simply reproduces the
Kubo formula result (3.3). Perturbation theory fails forT . T ∗2 whereT ∗
2 ∝ (λ02)1/(1−4K)
is a scale for the crossover from weak to strong two-particlebackscattering withλ02 the bare
two-particle backscattering amplitude.
The one-loop RG equations [66, 21] for the Kondo couplingsJ‖, Jz read (see Sec. A.1)
dJ‖dℓ
= (1−K)J‖ + νJ‖Jz,dJzdℓ
= νJ2‖ . (3.4)
The family of RG trajectories is indexed by a single scaling invariantc = (νJ‖)2 − (νJz)
2
whereνJz ≡ νJz + 1 −K, which is fixed by the couplings at energy scaleD. In contrast
to the spinful case [59], the absence of spin-flip forward scattering in the helical liquid
preserves the stability of the ferromagnetic fixed line, as is the case in the usual Kondo
1This result is valid in the high-temperature regime~v/L ≪ T < D for Fermi liquid leads whereL isthe length of the QSH region (see Ref. [65]).
CHAPTER 3. KONDO EFFECT IN THE QSH STATE 46
problem. The renormalized spin-flip amplitudeJ‖(T ) is given in terms of the bare param-
etersJ0‖ , J
0z by
νJ‖(T ) =ανJ0
‖
sinh(
ανJ0‖ ln(T/T
∗K)) , (3.5)
such that∆GK is obtained to all orders in perturbation theory in the leading-log approxi-
mation by substituting Eq. (3.5) in Eq. (3.2). In Eq. (3.5),
α =
(
J0z
J0‖
)2
− 1
1/2
is an anisotropy parameter2 andT ∗K is the Kondo temperature,
T ∗K = D exp
(
− 1
νJ0‖
arcsinhα
α
)
.
In the isotropic caseα = 0, one recovers the usual formT ∗K = De−1/νJ0
‖ . In the limit 1 −K ≫ νJ0
‖ , νJ0z , Coulomb interactions dominate over Kondo physics and we obtainT ∗
K ≃
D
(
νJ0‖
1−K
)1/(1−K)
, a power-law dependence similar to results previously obtained [67] for
Kondo impurities in spinful Luttinger liquids. From the scaling exponent we see thatT ∗K
corresponds to the scale of the mass gap opened in a spinless Luttinger liquid by a point
potential scatterer of strengthνJ0‖ and the corresponding crossover is that of weak to strong
single-particle backscattering.
In the high temperature limitmaxT ∗2 , T
∗K ≪ T . D, both the Kondo and two-
particle scattering processes contribute logarithmically to the suppression of the conduc-
tance,−∆G(T ) ∼ η+γ ln(D/T ) whereη, γ are functions of the bare couplingsK, J0‖ , J
0z
andλ02.
2One can show that the Kondo model derived from the Anderson model for a single level coupled to the
helical liquid is isotropic due to time-reversal symmetry with J0‖ = J0
z = (|t|2 + |u|2)(
1ǫF−ǫd
+ 1ǫd+U−ǫF
)
whereǫF is the Fermi energy,ǫd is the impurity level with on-site Coulomb repulsionU , andu andt are thespin-flip and non-spin-flip hopping amplitudes, respectively. Coulomb interactions (K 6= 1) may howeverinduce an effective anisotropy (α 6= 0) even if the original Kondo model is isotropic.
CHAPTER 3. KONDO EFFECT IN THE QSH STATE 47
Figure 3.2: Strong coupling regime: the Kondo singlet effectively removes one site fromthe system. a) Luttinger liquid withKρ < 1: a punctured 1D lattice is disconnected, b)QSH edge liquid withK > 1/4: the edge liquid follows the boundary of the deformed 2Dlattice. c) Half-charge tunneling forK < 1/4 by flips of the Ising order parameterm.
3.4 Strong coupling regime
We now investigate the low temperature regime below the crossover temperaturesT ≪minT ∗
2 , T∗K. The topological nature of the QSH edge state as a “holographic liquid”
living on the boundary of a 2D system [21] results in a drasticchange of the low-energy
effective theory in the vicinity of the strong coupling fixedpoint as compared to that of a
usual 1D quantum wire. As suggested by the perturbative RG analysis, the nature of the
T = 0 fixed point depends on whetherK is greater or lesser than1/4.
We first consider the case whereK > 1/4. In this case, two-particle backscattering is
irrelevant and∆G2 flows to zero. On the other hand, for antiferromagneticJz the Kondo
strong coupling fixed pointJ‖, Jz → +∞ is reached atT = 0, with formation of a local
Kramers singlet and complete screening of the impurity spinby the edge electrons. As a
result, the formation of the Kondo singlet effectively removes the impurity site from the
underlying 2D lattice (Fig. 3.2b). In a strictly 1D spinful liquid, this has the effect of
cutting the system into two disconnected semi-infinite 1D liquids (Fig. 3.2a) and transport
CHAPTER 3. KONDO EFFECT IN THE QSH STATE 48
is blocked atT = 0 for all Kρ < 1 [59]. In contrast, due to its topological nature, the QSH
edge state simply follows the new shape of the edge and we expect the unitarity limitG =
2e2/h to be restored atT = 0. For finiteT ≪ T ∗K, the effective low-energy Hamiltonian
contains the leading irrelevant operators in the vicinity of the fixed point. In the case of
spinful conduction electrons, the lowest-dimensional operator causing a reduction of the
conductance is single-particle backscattering. However,the helical property of the QSH
edge states forbids such a term and it is natural to conjecture that the leading irrelevant
operator must be the two-particle backscattering operatorwith scaling dimension4K. We
thus expect a correction to the conductance at low temperaturesT ≪ T ∗K for K > 1/4 of
the form∆G ∝ − (T/T ∗K)
2(4K−1).
In particular, in the noninteracting caseK = 1 we predict aT 6 dependence in marked
contrast to both the usual Fermi liquid [68] and spinful 1D liquid [59] behaviors. This
dependence characteristic of a “local helical Fermi liquid” can be understood from a sim-
ple phase space argument (see Sec. A.6). The Pauli principlerequires the two-particle
backscattering operator to be defined through a point-splitting procedure [21] with the
short-distance cutoffξ, which translates into a derivative coupling in the limit ofsmall
ξ,
ψ†R(0)ψ
†R(ξ)ψL(0)ψL(ξ) → ξ2ψ†
R∂xψ†RψL∂xψL.
In the absence of derivatives, the four fermion term contributesT 2 to the inverse lifetime
τ−1k . The derivatives correspond to four powers of momenta closeto the Fermi points in
the scattering rateΓk,k′→p,p′ ∝ (k− k′)2(p− p′)2, which translates into an additional factor
of T 4. Furthermore, since at temperaturesT ≪ T ∗K suppression of the conductance is
entirely due to two-electron scattering, we expect the effective chargee∗ ≡ S/(2|〈IB〉|)obtained from a measurement of the shot noiseS in the backscattering current3 〈IB〉 to be
e∗ = 2e [63].
ForK < 1/4, the|λ2| → ∞ fixed point is reached atT = 0 and the system becomes
insulating. The fieldφ(x = 0, τ) is pinned at the minima of the cosine potentialH2 located
at±(2n + 1)√π/4 for λ2 > 0 and±2n
√π/4 for λ2 < 0, with n ∈ Z. The conductance
3In the weak coupling or high temperature regimeT ≫ T ∗2 , T
∗K, both the Kondo (e∗ = e) and 2P backscat-
tering (e∗ = 2e) contributions to the effective carrier charge are present, such that we expect a non-universalvalue for the Fano factor.
CHAPTER 3. KONDO EFFECT IN THE QSH STATE 49
is restored at finite temperatures by instanton processes corresponding to the tunneling
between nearby minima separated by∆φ =√π/2. From the relationje = e∂tφ/
√π
between electric currentje andφ field, the charge pumped by a single instanton (Fig. 3.2c)
is obtained as
∆Qinst =e√π∆φ =
e
2.
This fractionalized tunneling current can be understood asthe Goldstone-Wilczek cur-
rent [69, 70, 16] for 1D Dirac fermions with a mass termδL = gΨ(m1 + iγ5m2)Ψ where
g ∼ λ2, and the mass order parametersm1 = cos 2√πφ, m2 = sin 2
√πφ change sign
during an instanton process with∆φ =√π/2 (see Sec. A.5). The order is Ising-like be-
cause the two-particle backscattering term explicitly breaks the spinU(1) symmetry of the
helical liquid
H0 =πv
2
∫
dx
(
Kσ2z +
1
Kρ2)
down toZ2, whereσz = ρ+ − ρ− andρ = ρ+ + ρ− are the spin and charge densities, and
ρ± are the chiral densities for the two members of the Kramers pair.
Fractionalization of the tunneling current is confirmed by asaddle-point evaluation of
the path integral for largeλ2 in the dilute instanton gas approximation (see Sec. A.3), which
yields a Coulomb gas representation of the partition function that can be mapped exactly
to the boundary sine-Gordon theory
S[θ] =K
β
∑
iωn
|ωn||θ(iωn)|2 + t
∫ β
0
dτ cos√πθ(τ),
wheret is the instanton fugacity. The RG equation fort follows as
dt
dℓ=
(
1− 1
4K
)
t,
and the conductanceG(T ) ∝ t(T )2 is a power-lawG ∝ (T/T ∗2 )
2(1/4K−1) for T ≪T ∗2 , K < 1/4. In contrast to the strong coupling regime in a usual Luttinger liquid where
t corresponds to a single-particle hopping amplitude [71, 72], the unusual scaling dimen-
sion of the tunneling operator in the present case corresponds to half-charge tunneling. In
CHAPTER 3. KONDO EFFECT IN THE QSH STATE 50
particular, we calculate the shot noise in the strong coupling regime (see Sec. A.4) using
the Keldysh approach [73] and findS = 2e∗|〈I〉| where〈I〉 is the tunneling current and
e∗ = e/2.
3.5 Experimental realization
We find that the experimental results of Ref. [58] are consistent with our theoretical expres-
sions for the weak coupling regime with a weak Luttinger parameterK ≃ 1, but the small
number of available data points does not allow for a reliabledetermination of the model
parameters. The temperature dependence of the conductancebeing exponentially sensitive
to K, our predictions can be best verified in QW with stronger interaction effects. Due
to reduced screening of the Coulomb interaction4, we expect to see a steeper decrease of
conductance with decreasing temperature in HgTe samples with only a backgate.
Because of lower Fermi velocitiesvF , we expect even stronger interaction effects to
occur in InAs/GaSb/AlSb type-II QW [74] which have been recently predicted to exhibit
the QSH effect [75]. Interestingly, some recent experimental results [76, 77] suggest that
the QSH state is indeed realized in these materials. For QW widthswInAs = wGaSb = 10
nm in the inverted regime [75], and considering only screening from the front gate closest
to the QW layer, from ak ·p calculation of material parameters we obtainK ≃ 0.2 < 1/4,
making the insulating phase observable at low temperatures. Although the backgate will
cause additional screening,vF can be further decreased by adding a thin AlSb barrier layer
between the InAs and GaSb QW layers. The Fermi velocity is controlled by the overlap
between electron and hole subband wave functions [19] whichare localized in different
QW layers in the type-II configuration [75], and an additional barrier layer will decrease
this overlap. A lowervF also translates into higher Kondo temperatures sinceνJ ∝ 1/v2F ,
where one power ofvF comes from the matrix element of the localized impurity potential
between edge states, and one power comes from the density of statesν. SinceT ∗K depends
4K can be estimated [71] byK = [1 + α ln(d/ℓ)]−1/2 whereα = 2π2
e2/ǫ~vF
andǫ is the bulk dielectricconstant. The distanced from the QW layer to a nearby metallic gate acts as a screeninglength for theCoulomb potential, andℓ is a microscopic length scaleℓ = maxξ, w which acts as a short-distance cutofffor the Coulomb potential, wherew is the thickness of the QW layer.
CHAPTER 3. KONDO EFFECT IN THE QSH STATE 51
on νJ exponentially, we expect experimentally accessible Kondotemperatures in type-II
QW.
Chapter 4
Magnetoconductance of the quantum
spin Hall state
4.1 Introduction
A great deal of interest has been generated recently by the theoretical prediction [19] and
experimental observation [20, 78, 79] of the QSH insulator state [13, 14, 54]. The QSH
state is a novel topological state of quantum matter which does not break TR symmetry, but
has a bulk insulating gap and gapless edge states with a distinct helical liquid property [21].
The gaplessness of the edge states is protected against weakTR symmetry preserving per-
turbations by Kramers degeneracy [21, 44]. As a result, the QSH state exhibits robust
dissipationless edge transport [20, 78, 79] in the presenceof nonmagnetic disorder.
However, in the presence of an external magnetic field which explicitly breaks TR sym-
metry, the gaplessness of the edge states is not protected. This can be simply understood
by considering the generic form of the effective 1D HamiltonianH for the QSH edge [70]
to first order in the magnetic fieldB,H = H0 +H1(B), whereH0 = ~vkσ3 is the Hamil-
tonian of the unperturbed edge, andH1(B) =∑
a=1,2,3(ta · B)σa is the perturbation due
to the field. k is a 1D wave vector along the edge,v is the edge state velocity,σ1,2,3 are
the three Pauli spin matrices, andt1,2,3 are model-dependent coefficient vectors [70]. IfB
points along a special direction in spacet∗ ≡ t1 × t2, thenH1(B) ∝ σ3 commutes with
H0, the wave vectork is simply shifted, and the edge remains gapless, unless mesoscopic
52
CHAPTER 4. MAGNETOCONDUCTANCE OF THE QSH STATE 53
quantum confinement effects become important [80]. IfB ∦ t∗, then[H0, H1(B)] 6= 0 and
a gapEgap ∝ |B| opens in the edge state dispersion.
Experimentally [20, 58], one observes that the conductanceG(B) of a QSH device ex-
hibits a sharp cusp-like peak atB = 0, andG decreases for increasing|B|. Although the
explanation of a thermally activated behaviorG(B) ∝ e−Egap(|B|)/kBT with T the tempera-
ture accounts qualitatively for the observed cusp, it does so only if the chemical potential
µ lies inside the edge gap which, according to theoretical estimates [54], is rather small
(Egap ∼ 1 meV). Experimentally, a sharp peak is observed [58] throughout the bulk gap
(Eg ∼ 40 meV). Furthermore, this explanation ignores the effects ofdisorder. In the ab-
sence of TR symmetry, the QSH edge liquid is topologically equivalent to a spinless 1D
quantum wire, and is expected to be strongly affected by disorder due to Anderson localiza-
tion. Although the effect of disorder on transport in the QSHstate has been the subject of
several recent studies [21, 44, 81, 82, 83, 84, 85], except for studies addressing the effect of
magnetic impurities [21, 86] there have been no theoreticalinvestigations of the combined
effect of disorder and TR symmetry breaking on edge transport in the QSH state.
In this work, we study numerically the edge magnetoconductanceG of a QSH insu-
lator in the presence of quenched nonmagnetic disorder. Ourmain findings are: (1) For
a finite magnetic fieldB and disorder strengthW on the order of the bulk energy gap
Eg, G deviates from its quantized valueG(0) = 2e2/h at zero field [20] by an amount
∆G(B) ≡ G(B)−G(0) which seems roughly linear in|B| at smallB, at least in the range
of fields we study. We observe this behavior forµ across the bulk gap (Fig. 4.1c), which
agrees qualitatively with the cusp-like features reportedin Ref. [20]. (2) The slope∂G/∂B
of G(B) at smallB steepens rapidly whenW > Eg (Fig. 4.2b), which suggests that bulk
states play an important role in the backscattering of the edge states. (3)G is unaffected by
an orbital magnetic field in the absence of inversion symmetry breaking terms (Fig. 4.3a).
In the absence of such terms,t1 andt2 are entirely in thexy plane of the device [54], hence
t∗ ∝ z is out-of-plane and a perpendicular fieldB ‖ t∗ cannot lead to backscattering, as
discussed earlier. In the presence of inversion symmetry breaking terms, the effective edge
Hamiltonian becomesH ′ = ~vkσ′3+∑
a=1,2,3(t′a ·B)σ′
a, whereσ′3 has nonzero components
along the1 and2 directions. Thent′∗ = t′1×t′2 is not alongz anymore, and a perpendicular
field B = Bz can lead to backscattering.
CHAPTER 4. MAGNETOCONDUCTANCE OF THE QSH STATE 54
4.2 Theoretical model
We start from a simple 4-band continuum model Hamiltonian [19, 54] used to describe the
physics of the QSH state in HgTe QW,
H(k) =
(
H(k) ∆(k)
∆†(k) H∗(−k)
)
, (4.1)
written in the(E1+, H1+, E1−, H1−) basis whereE1, H1 are the relevant QW subbands
close to the Fermi energy and± denotes time-reversed partners. The diagonal blocks
H(k), H∗(−k) with H(k) = ǫ(k) + vk · σ +M(k)σz are related by TR symmetry and
correspond to decoupled 2D Dirac-like Hamiltonians, wherek = (kx, ky), σ = (σx, σy) is
a vector of Pauli matrices, and the velocityv is obtained fromk ·p theory. We also define a
quadratic kinetic energy termǫ(k) = C−Dk2 and the Dirac mass termM(k) =M−Bk2,
whereC,D,M,B arek · p parameters. The off-diagonal block∆(k) is given by [87]
∆(k) =
(
∆ek+ −∆z
∆z ∆hk−
)
, (4.2)
where∆e,∆h,∆z arek · p parameters andk± = kx ± iky. It originates from the bulk
inversion asymmetry (BIA) of the underlying microscopic zincblende structure of HgTe
and CdTe [88]. A nearest-neighbor TB model on the square lattice can be derived from
Eq. (4.1),
H =∑
i
c†iV ci +∑
i
(
c†iTxci+x + c†iTyci+y + h.c.)
, (4.3)
where the4× 4 matricesV, Tx, Ty depend solely on thek ·p parameters introduced above.
Equations (4.1) and (4.3) correspond to a translationally invariant system in the absence
of magnetic field or disorder. In the presence of disorder andan external magnetic field
B = (Bx, By, Bz), we perform the substitutions
V −→ V +HZ‖ +HZ⊥ +Wi,
Tx −→ Tx exp
(
2πi
φ0
∫ i+x
i
dℓ ·A)
= Txe−2πinzy/a,
CHAPTER 4. MAGNETOCONDUCTANCE OF THE QSH STATE 55
whereWi is a Gaussian random on-site potential with standard deviation W mimicking
quenched disorder,A = (−Bzy, 0) is the in-plane electromagnetic vector potential in the
Landau gauge,φ0 = h/e is the flux quantum, andnz = Bza2/φ0 is the number of flux
quanta per plaquette witha the lattice constant. We usea = 30 A which is a good approx-
imation to the continuum limit. The in-plane Zeeman termHZ‖ is given by [54]
HZ‖ = g‖µB
0 0 B− 0
0 0 0 0
B+ 0 0 0
0 0 0 0
, (4.4)
whereB± = Bx ± iBy, µB is the Bohr magneton, and the in-planeg-factorg‖ is obtained
from k · p calculations [87]. The out-of-plane Zeeman termHZ⊥ is given by [54]
HZ⊥ = µBBz diag(
gE⊥, gH⊥,−gE⊥,−gH⊥)
, (4.5)
and the out-of-planeg-factorsgE⊥, gH⊥ are also obtained fromk ·p calculations [87]. The
k ·p parameters used in the present work correspond to a HgTe QW thickness ofd = 80 A.
We calculate numerically theT = 0 disordered-averaged two-terminal conductanceG
and conductance fluctuationsδG of a finite QSH strip (Fig. 4.1a) using the standard TB
Green function approach [89, 90, 91]. We find thatNdis ∼ 100 disorder configurations are
enough to achieve good convergence forG andδG. For a strip of widthLy comparable
to the edge state penetration depthλ, interedge tunneling [92] backscatters the edge states
even atB = 0 and the system is analogous to a topologically trivial quasi-1D quantum
wire. To ensure that we are studying effects intrinsic to thetopologically nontrivial QSH
helical edge liquid, we first need to suppress interedge tunneling. The naive way to achieve
this is to use a very largeLy; however, this can be computationally rather costly. We usea
geometry (Fig. 4.1a) which allows us to effectively circumvent this problem while keeping
Ly reasonable. By adding a local Dirac mass term [54]δM < 0 on the first horizontal chain
of our TB model (Fig. 4.1a, red dots), the penetration depthλ2 at the top edge becomes
much smaller than that at the bottom edgeλ1 ≫ λ2. We then add disorder only to the last
Ldis/a chains of the central region withLdis ≫ λ1 andLy − Ldis ≫ λ2. The resulting top
CHAPTER 4. MAGNETOCONDUCTANCE OF THE QSH STATE 56
a) M
Ly
!2
1
Ldis
-1
-0.5
0
B [
T]
0 1 2 3G [e
2/h]c)
Lx
1 4
1.6
1.8
2
e2/h
]
0.12
0.18
0.24
Ly [ m]b) 0.5
1
B
-1
-0.5
]
Ee
Ed
Ec
Eb
Ead)
-1 -0.5 0 0.5 1
1
1.2
1.4
B [T]
G [
e
0
0.5
1-0.05 0 0.05
kx [ /a
]!, E(k
x) [eV]
Figure 4.1: MagnetoconductanceG of a QSH edge: a) TB model with asymmetric edgestatesλ2 ≪ λ1 to study a single disordered edge; b) dependence ofG on sample widthLy
for disorder strengthW = 55 meV larger than the bulk gap, lengthLx = 2.4µm, fixedclean widthLy − Ldis = 0.03µm, and local mass termδM = −70 meV, with error bars(plotted forLy = 0.12µm andB > 0 only) corresponding to conductance fluctuationsδG;c) dependence ofG on chemical potentialµ; d) quasi-1D spectrum of the device illustratedin a) for zeroW,B, showing bulk states (blue), top edge states (green) and bottom edgestates (red).
edge states are very narrow, contribute an uninteresting background quantized conductance
independent ofB andW , and are essentially decoupled from the bottom edge states (whose
magnetoconductance we wish to study) that are effectively propagating in a semi-infinite
disordered medium.
4.3 Numerical results
For µ inside the bulk gap, we expect edge transport to dominate thephysics. The typical
behavior of the magnetoconductanceG(B) for B = Bz and disorder strengthW larger
than the bulk gapEg ≃ 40meV is shown in Fig. 4.1b. The cusp-like feature atB = 0 agrees
CHAPTER 4. MAGNETOCONDUCTANCE OF THE QSH STATE 57
qualitatively with the results of Ref. [20].G(B) is independent ofLy, which suggests that
transport is indeed carried by the edge states.G(B = 0) is quantized toG0 ≡ 2e2/h
independent ofW up toW = 71 meV with extremely small conductance fluctuations
δG(B = 0)/G0 ∼ 10−5, which confirms that interedge tunneling is negligible evenfor
strong disorder. Furthermore,G tends toG0/2 for large |B| ∼ 1 T, which indicates that
the disordered bottom edge is completely localized for large W and |B|, and only the
unperturbed top edge conducts. ForW < Eg, G is approximately quadratic inB (not
shown), and|G(B) − G0|/G0 ≪ 1 even for large|B| ∼ 1 T. ForB 6= 0, we observe that
the amplitude of the fluctuationsδG does not decrease upon increasingNdis, and is roughly
independent ofW with δG/G0 ∼ O(10−1) for large enough disorderW & Eg. Since in
the absence of TR symmetry the QSH system is a trivial insulator and the edge becomes
analogous to an ordinary spinless 1D quantum wire with no topological protection, we
conclude thatδG corresponds to the well-known universal conductance fluctuations [89].
The dependence ofG(B) on µ is plotted in Fig. 4.1c. We considerW = 55 meV
slightly larger thanEg (Fig. 4.1d). This is not unreasonable as the bulk mobilityµ∗ of
the HgTe QW in Ref. [20] is estimated asµ∗ ≃ 105 cm2/(V·s), which corresponds to
a momentum relaxation timeτ = µ∗m∗/e ≃ 0.57 ps. The bulk carriers at the bottom
of the conduction subband have an effective massm∗ ≃ 0.01me whereme is the bare
electron mass.τ is given by~/τ ≃ 2πν(Wa)2, with ν the bulk continuum density of
states at the Fermi energy given byν ≃ m∗/π~2. This yieldsW ≃ 22 meV. However,
this estimate considers only bulk disorder and we expect edge roughness to yield a higher
effectiveW on the edge. Furthermore, this estimate is perturbative inW and neglects
interband effects which are expected to occur forW ∼ Eg. For the chosen value ofW
we observe that the bulk states (Fig. 4.1d, blue lines) are strongly localized withG ≪ G0
for µ < Ea andµ > Ee in the bulk bands, while the cusp-like feature atB = 0 with
G(B = 0) = G0 remains prominent forEb < µ < Ed in the bulk gap and even at the
bottom of the conduction bandEd < µ < Ee where the top edge states (Fig. 4.1d, red
lines) coexist with the bulk states. The sudden dip inG(B 6= 0) for µ ∼ Ec ≃ 15 meV
corresponds to the opening of the small edge gap discussed earlier. Finally, G ≃ G0 is
almost independent ofB for Ea < µ < Eb, where the disordered bottom edge and bulk
states are mostly localized while the clean top edge supports another channel (Fig. 4.1d,
CHAPTER 4. MAGNETOCONDUCTANCE OF THE QSH STATE 58
0 0.025 0.05
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
|B| [T]
G [e
2 /h]
0 20 40 600
5
10
15
20
25
30
W [meV]
−dG
/dB
[(e2 /h
).T
−1 ]
293337414549535761656971
a)
W [meV]
b)
Figure 4.2: a) Magnetoconductance for various disorder strengthsW ; b) small-B slope ofthe magnetoconductance (obtained by linear regression for0 < B < 15 mT). Device sizeis (Lx × Ly) = (2.4× 0.12)µm2.
dashed green line), with a total top edge conductance ofG = G0.
The magnetoconductance forB = Bz and various values ofW is plotted in Fig. 4.2.
Although not evident from the figure,G(B) is approximately quadratic inB for W < Eg,
and approximately linear in|B| at smallB for W > Eg (Fig. 4.2a). The slope ofG(B)
at small fields (obtained by linear regression for0 < B < 15 mT where the dependence
is approximately linear) is plotted in Fig. 4.2b, and is seento increase rapidly forW &
Eg ≃ 40 meV. ForB = 0, we have essentiallyG = G0 independent ofW (Fig. 4.2a).
This contrasts with the results of Ref. [81, 84] where deviations fromG = G0 atB = 0
occur forW larger than some critical valueWc > Eg. The reason for this difference is
that in Ref. [81, 84], disorder-induced collapse of the bulkgap is accompanied by the edge
states penetrating deeper into the bulk and eventually reaching the opposite edge, such
that interedge tunneling takes place and causes backscattering. Here, due to our special
geometry (Fig. 4.1a) the top edge state is unperturbed and always remains localized near
the edge, out of reach of the bottom edge state, even as the latter penetrates deeper into the
disordered bulk for increasingW .
CHAPTER 4. MAGNETOCONDUCTANCE OF THE QSH STATE 59
The BIA term∆k has an important effect onG for B = Bz (Fig. 4.3a). For simplicity,
we set∆e = ∆h = 0 and consider only the effect of∆z. For∆z = 0, the perturbation
H′ = ej · A due to an orbital field, withe the electron charge andj the current operator,
has no matrix element between the spin states of a counterpropagating Kramers pair on
a given edge [54], andG is unaffected. For an in-plane field,HZ‖ does have a nonzero
matrix element between these states, and there is a nontrivial magnetoconductance even in
the absence of BIA.
The dependence ofG(B) on the orientation ofB is plotted in Fig. 4.3b. Theg-
factors [87] used in the Zeeman terms are such that the Zeemanenergies for in-plane and
out-of-plane fields are of the same order [54]. The in-plane vs out-of-plane anisotropy
(Fig. 4.3b,x, y vs z) arises from the orbital effect of the out-of-plane fieldB = Bz, which
is absent for an in-plane field. In our model, the in-plane anisotropy is very weak (some-
what visible on Fig. 4.3b for|B| ∼ 1 T), and is due to the inequivalence between the
transportx and confinementy directions. Finally, theB = 0 peak inG is more pronounced
for a smaller mass termM [54] in the Dirac HamiltoniansHk, H∗−k
(Fig. 4.3c). Since
Eg ∝ |M | approximately, a smaller|M | results in a larger dimensionless disorder strength
W/Eg, which is equivalent to an increase inW (see Fig. 4.2b).
Although the mechanism behind the observed negative magnetoconductance∆G ∝−|B| (Fig. 4.1,4.2) for an orbital fieldB = Bz cannot be unambiguously inferred from our
numerical results, a dependence linear in|B| for smallB and the requirement of “strong”
disorderW & Eg for its observation seem to indicate that the effect has a nonperturbative
character. A treatment which is perturbative inW andB yields at most, to leading order, the
result−∆G ∝ ℓ−1 ∝ W 2eff(B) ∝ B2, whereℓ is the mean free path [93] andWeff(B) ≡
W |B|/B0 is some effective disorder strength, withB−10 ∝ ∆z if only the effect of∆z
is considered for simplicity. For “weak” disorderW < Eg, the 1D edge states which
enclose a negligible amount of flux are the only low-energy degrees of freedom, and the
magnetic field only has a perturbative effect on them. Indeed, if we choose the gaugeA =(
Bz(Ly − y), 0)
, for sufficiently smallBz we have thatA is small forLy − λ1 . y < Ly
with λ1 ≪ Ly where the bottom edge state wavefunction has finite support (Fig. 4.1a), and
the effect of an orbital fieldBz on a single edge can be treated perturbatively. In this case,
the amplitude∝ Weff(B) in perturbation theory for a leading order backscattering process
CHAPTER 4. MAGNETOCONDUCTANCE OF THE QSH STATE 60
1
1.2
1.4
1.6
1.8
2
G [
e2/h
]
-0.5 0 0.5
1.7
1.75
1.8
1.85
1.9
1.95
2
B [T]
G [
e2/h
]
x
y
z
-1 -0.5 0 0.5 1
1.4
1.5
1.6
1.7
1.8
1.9
2
B [T]
G [
e2/h
]
15
10
5
2
1
z = 0
z = -0.5
z = -1.5
z = -1
a) b)
c)
-M [meV]
Figure 4.3: Dependence of the magnetoconductanceG on a) strength of thek-independentBIA term∆z with ∆e = ∆h = 0; b) magnetic field orientation; c) Dirac mass termM < 0.Sample size is(Lx × Ly) = (2.4 × 0.12)µm2, disorder strength isW = 55 meV for a),b)andW = 30 meV for c).
on a single edge involves one power of∆z and one power ofBz to couple the spin states of
the counterpropagating Kramers partners [54] (with no momentum transfer as our choice of
gauge preserves translational symmetry in thex direction), and one power ofW to provide
the necessary momentum transfer for backscattering. Our observation that∆G ∝ −B2
for W < Eg corroborates this physical picture. On the other hand, the cusp-like feature
atB = 0 (Fig. 4.1b) occurs for “strong” disorderW & Eg, which seems to indicate that
the bulk states play an important role. This leads us to a different physical picture. For
W & Eg, the edge electrons easily undergo virtual transitions to the bulk. In other words,
the emergent low-energy excitations forW & Eg extend deeper into the bulk than the
“bare” edge electrons. The electrons spend a significant amount of time diffusing randomly
in the bulk away from the edge, with their trajectories enclosing finite amounts of flux
before returning to the edge, which endows the orbital field with a nonperturbative effect. In
CHAPTER 4. MAGNETOCONDUCTANCE OF THE QSH STATE 61
this way the conventional picture of 2D antilocalization [94] can apply, at least qualitatively,
to a single disordered QSH edge. We are thus led to the interesting picture, peculiar to the
QSH state, of a dimensional crossover between 1D antilocalization [95, 96, 97] in the weak
disorder regimeW < Eg with the orbital field having a perturbative effect, to an effect
analogous to 2D antilocalization in the strong disorder regimeW > Eg with the orbital
field having a nonperturbative effect.
4.4 Conclusion
We have shown that “strong” disorder effectsW/Eg ∼ 1 in a QSH insulator in the presence
of a magnetic fieldB and inversion symmetry breaking terms can give rise to a cusp-like
feature in the two-terminal edge magnetoconductance with an approximate linear depen-
dence∆G(B) ∝ −|B| for smallB. These results are in good qualitative agreement with
experiments. A possible physical intepretation of our results consists of a dimensional
crossover scenario where a weakly disordered, effectivelyspinless 1D edge liquid crosses
over, for strong enough disorder, to a state where disorder enables frequent excursions of
the edge electrons into the disordered flux-threaded 2D bulk, resulting in a behavior remi-
niscent of 2D antilocalization.
Chapter 5
Spin Aharonov-Bohm effect and
topological spin transistor
5.1 Introduction
The spin of the electron is one of the most fundamental quantum mechanical degrees of
freedom in Nature. Historically, the discovery of the electron spin helped to lay the foun-
dation of relativistic quantum mechanics. In recent years,the electron spin has been pro-
posed as a possible alternate state variable for the next generation of computers, which led
to extensive efforts towards achieving control and manipulation of the electron spin, a field
known as spintronics [98]. Despite the great variety of currently used or theoretically pro-
posed means of manipulating the electron spin, a feature common to all of them is that they
all make use of theclassicalelectromagnetic force or torque actinglocally on the magnetic
moment associated with the spin.
On the other hand, it is known that due to the Aharonov-Bohm (AB) effect [99], elec-
trons in a ring can be affected in a purelyquantum mechanicalandnonlocalway by the
flux enclosed by the ring even though no magnetic field — hence no classical force —
is acting on them. This effect could be termed “charge AB effect”, as it relies only on
the electron carrying an electric charge. In systems with spin-orbit coupling or magnetic
fields, a spin-dependent phase factor can be obtained and leads to modifications to the AB
62
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 63
effect [100, 101, 102, 103]. However, these effects usuallyinvolve classical forces act-
ing on the spin such as electromagnetic fields, and the pure gauge potential leading to the
charge AB effect does not directly couple to spin. This observation leads naturally to the
question of whether it is possible to observe a “spin AB effect” which would enable one
to manipulate the electron spin in a purely nonlocal and quantum mechanical way, without
any classical force or torque acting locally on the spin magnetic moment.
In this work, we show that the spin AB effect is indeed possible by making use of the
edge states of the recently discovered QSH insulators. In recent years, the QSH insulator
state has been proposed in several different materials [13,14, 104, 19, 75, 105]. In particu-
lar, this topologically nontrivial state of matter has beenrecently predicted [19] and realized
experimentally [20, 78, 79] in HgTe QW. The QSH insulator is invariant under TR, has a
charge excitation gap in the bulk, but has topologically protected gapless edge states that
lie inside the bulk insulating gap. These edge states have a distinct helical property: two
states with opposite spin polarization counterpropagate at a given edge [13, 21, 44]. The
edge states come in Kramers doublets, and TR symmetry ensures the crossing of their en-
ergy levels at TR invariant points in the Brillouin zone. Because of this level crossing, the
spectrum of a QSH insulator cannot be adiabatically deformed into that of a topologically
trivial insulator without closing the bulk gap. The helicity of the QSH edge states is the
decisive property which allows the spin AB effect to exist: the perfect correlation between
spin orientation and direction of propagation allows the transmutation of a usual charge AB
effect into a spin AB effect, as will be explained in detail below.
The mechanism we propose to realize the spin AB effect is illustrated in Fig. 5.1. Con-
sider a two-terminal device consisting of a bounded QSH insulator region pierced by a
hole which is threaded by a magnetic fluxφ. If the edge electrons propagating clock-
wise have their spin pointing out-of-plane alongz (spin up |↑〉 , red trajectory), due to
TR symmetry the electrons propagating counterclockwise must have opposite spin along
−z (spin down|↓〉 , blue trajectory). If we inject electrons spin-polarized along thex di-
rection |→〉 = 1√2(|↑〉 + |↓〉) from a ferromagnetic (FM) lead on the left, the electron
beam will be split coherently upon entering the QSH region atthe left junction into a|↑〉beam propagating along the top edge and a|↓〉 beam propagating along the bottom edge.
When the electron beams are recombined on the right side of the ring, the electrons along
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 64
Figure 5.1: Schematic picture of the spin AB effect. A ring ofQSH insulator threaded by amagnetic fluxφ is connected to two magnetic leads. Spin polarized electrons injected fromthe left lead enter the QSH region as a superposition of spin up and down states. The spinup (down) state can only propagate along the top (bottom) edge of the QSH ring, and thetwo spin states thus acquire an AB phase difference proportional toφ. Consequently, uponexiting the QSH region the two edge states recombine into a state with spin rotated withrespect to the injected direction. The magnetization direction of the right lead generallydiffers from that of the left lead by an angleθ. The two-terminal conductanceG = G(φ, θ)of the device depends on the relative angle between the spin polarization of the outgoingstate and that of the right lead.
top and bottom edges will acquire a phase difference ofϕ = 2πφ/φ0 due to the AB ef-
fect, whereφ0 = hc/e is the flux quantum. Consequently, the output state is given by1√2
(
|↑〉+e−iϕ |↓〉)
, such that the electron spin is rotated by an angleϕ in thexy plane. The
magnetic flux being confined to the hole in the device (Fig. 5.1), the electromagnetic fields
are zero in the region where the electrons propagate, and thespin is rotated by a purely
quantum mechanical Berry phase effect. In particular, for collinear FM leads (θ = 0 in
Fig. 5.1), one expects the conductance to be maximal forφ = 0 (modφ0) and minimal for
φ = 12φ0 (modφ0), thus realizing a “topological” spin transistor [Fig. 5.3(c)]. This effect is
topological in the sense that the spin is always rotated by one cycle for each period of flux
φ0, regardless of the details of the device, such as the size of the system or the shape of the
ring.
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 65
5.2 Phenomenological scattering matrix analysis
Before considering any microscopic model of transport in a QSH system, generic features
of two-terminal transport in the device of Fig. 5.1 that depend only on symmetry consid-
erations can be extracted from a simple phenomenological scattering matrix orS-matrix
analysis [106]. The left and right junctions are each described by a scattering matrixSL
andSR, respectively (e.g. Fig. 5.2(a) for the left junction). Considering the left junc-
tion first, SL consists of four submatricestL, t′L, rL, r′L which correspond respectively to
transmission from left to right, transmission from right toleft, reflection from the left, and
reflection from the right. One can define similar submatricesfor SR. We wish to obtain an
effectiveS-matrixS [see Eq. (A.63)] for the whole device, by combining theS-matrices of
the junctions together with theS-matrix for the central QSH region. Inside the QSH region,
the AB effect is described by the matrixΦ ≡ e−iϕσz/2 whereσx, σy, σz are the three Pauli
matrices. In addition to the geometric phaseϕ, the edge electrons also acquire a dynamical
phaseλ = 2kF ℓ identical for both spin polarizations, whereℓ is the distance travelled by
the edge electrons from left to right junction andkF is the edge state Fermi wave vector.
Details of the analysis are presented in Sec. A.7; here we discuss only the main results. We
obtain the effective2× 2 device scattering matrixS,
S(φ, θ) =(
1− eiλΦr′L(0)ΦrR(θ))−1
eiλ/2Φ, (5.1)
where the junction reflection matricesr′L(θL) andrR(θR) depend on the anglesθL, θR of
the magnetizationML,R in the left and right leads. For simplicity we considerθL = 0 and
defineθ ≡ θR (Fig. 5.1).
The two-terminal conductanceG of the device can be written as
G =e2
htr ρRSρLS
†, (5.2)
using Eq. (A.63) of Sec. A.7. HereρL, ρR are2 × 2 effective spin density matrices for the
FM leads, and have the form
ρα(θα) =12Tα(θα)
(
1 +Pα(θα) · σ)
, (5.3)
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 66
with α = L,R, whereTα = tr ρα is the transmission coefficient of the junction andPα
is a polarization vector. For simplicity, we can assume the device to have aπ-rotation
symmetry, which together with TR symmetry restricts the generic form of the reflection
matricesr′L andrR in Eq. (5.1) to be
r′L(θ) =
(
αθ βθ
γθ αθ+π
)
, rR(θ) =
(
αθ+π βθ
γθ αθ
)
. (5.4)
Physically,αθ is a non-spin-flip reflection amplitude whereasβθ, γθ are spin-flip reflection
amplitudes, withβθ corresponding to a|↓〉 → |↑〉 reflection andγθ to a |↑〉 → |↓〉 reflec-
tion. These amplitudes are generally different due to the breaking of TR symmetry at the
junctions by the nearby FM leads.
5.3 Minimal model description
These expressions being so far very general, to make furtherprogress it is useful to consider
a simple continuum Hamiltonian model for the FM/QSH junctions in which the reflection
matricesr′L andrR can be calculated explicitly. This model satisfies the symmetries in-
voked earlier and will be seen to be a good description of the realistic HgTe system in
spite of its simplicity. We model the FM leads as 1D spin-12
fermions with a term which
explicitly breaks theSU(2) spin rotation symmetry [107],
HFM =
∫
dx؆(
− 1
2m
∂2
∂x2−M(θ) · σ
)
Ψ,
whereM(θ) =M n, with n = x cos θ+ y sin θ, is an in-plane magnetization vector andΨ
is a two-component spinorΨ ≡(
ψ↑ ψ↓
)T
. In the absence of AB flux, the QSH edge
liquid consists of 1D massless helical fermions [21, 44]. When the spins of the edge states
are polarized along thez direction, the Hamiltonian is given by
HQSH = −iv∑
α=t,b
ηα
∫
dx(
ψ†α↑∂xψα↑ − ψ†
α↓∂xψα↓)
,
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 67
Figure 5.2: Illustration of the minimal model describing a FM/QSH junction. (a) Schematicpicture of the junction between the left FM lead and the QSH insulator. Incoming channelsal,1, . . . , al,pL from the left lead scatter at the junction into transmitted QSH edge channelsbl′,↑, bl′,↓ and reflected lead channelsbl,1, . . . , b1,pL. This scattering process is described bya scattering matrixSL. (b) Minimal model description of the junction. The FM lead isdescribed by 1D parabolic bands with a spin splitting2M , while the QSH edge states arelinearly dispersing and TR invariant, with opposite spin states counter-propagating.
wherev is the edge state velocity andα = t, b refers to the top and bottom edge, respec-
tively, with ηt = 1 andηb = −1.
In this simple model, the junction is described as a sharp interface between the FM
region and the QSH region, from which the reflection matrixr′L in Eq. (5.1) and the spin
density matrixρL in Eq. (5.2) can be obtained. The calculation yields the reflection matrices
precisely in the form of Eq. (5.4) withαθ = a andβθ = γ∗θ = be−iθ. In the limit of small
spin splittingM/εF ≪ 1 whereεF is the Fermi energy in the leads [Fig. 5.2(b)], the
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 68
reflection amplitudesa andb are given by
a ≃ v − vFv + vF
, b ≃ M
2εF
v3Fv(v + vF )2
, (5.5)
wherevF =√
2εF/m is the Fermi velocity in the FM leads. The off-diagonal spin-flip
reflection amplitudeb is proportional to the magnetizationM and along with its accompa-
nying scattering phase shifte±iθ is an explicit signature of TR symmetry breaking at the
junction. The diagonal non-spin-flip reflection amplitudea does not break TR symmetry
and is the same as would be obtained in the scattering from a nonmagnetic metal with
M = 0. The lead spin density matricesρL, ρR can also be calculated explicitly and are
found to follow the form of Eq. (5.3) as expected from the general S-matrix analysis. In
the limitM/εF ≪ 1, we obtainTL = TR = 8vvF/(v + vF )2 and
PL(θ) = PR(θ) ≡ P(θ) = −M(θ)
4εF
v2Fv(v + vF )
, (5.6)
i.e. the spin polarization vector is directly proportionalto the magnetizationM.
From the results obtained above, we can readily evaluate theconductanceG, which has
the following expression in the limitM/εF , P ≡ |P(θ)| ≪ 1 andλ = 0,
G(ϕ, θ;λ = 0) =e2
h
TLTR/2
1− 2a2 cosϕ+ a4
×[
1 +cos(θ − ϕ) + (1− t2)2 cos(θ + ϕ) + C(ϕ, θ)
1− 2a2 cosϕ+ a4P 2 +O(P 4)
]
,
(5.7)
wheret = 1 − a andC(ϕ, θ) ≡ γ cosϕ + δ cos θ with γ, δ some constants depending
only ona. The effect of a finiteλ will be addressed in the next section, where we study
numerically a more realistic model of the QSH state in HgTe QW. Physically,a and t
can be interpreted as reflection and transmission coefficients for theSz spin current. The
generic behavior of Eq. (5.7) is illustrated in Fig. 5.3. ThetermC(ϕ, θ) is an uninteresting
background term which manifests no correlation between AB phaseϕ and rotation angle
of the electron spinθ. The term∝ cos(θ−ϕ) corresponds to a rotation of the electron spin
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 69
Figure 5.3: Phenomenological analysis of the two-terminalconductance (top view ofFig. 5.1). (a) The two leading contributions to the spin AB rotation. The purple pathstands for the process with no spin flips, which leads to a spinrotation ofϕ ≡ 2πφ/φ0.The orange path stands for the process with spin-dependent reflections, which leads to aspin rotation of−ϕ. (b) Schematic intensity map of the two-terminal conductanceG(ϕ, θ).The conductance reaches its maximum along the linesϕ = θ (purple) andϕ = −θ (or-ange), which are contributed by the purple and orange paths in panel (a), respectively. (c)The on and off states of the topological spin transistor are defined forθ = 0 by ϕ = 0 andϕ = π, respectively, as also indicated in panel (c).
byϕ, and the term∝ cos(θ+ϕ) corresponds to a rotation by−ϕ. The conductance is thus
maximal forϕmax = ±θ [Fig. 5.3(b)], manifesting the desired flux-induced spin rotation
effect. Physically, theϕmax = θ term corresponds to a process in which electrons traverse
the device without undergoing spin flips (Fig. 5.3(a), purple trajectory) while theϕmax =
−θ term corresponds to a process involving at least one TR breaking spin-flip reflection
(Fig. 5.3(a), orange trajectory). As can be seen from Eq. (5.7), the relative intensity of the
two contributions to the conductance isI−θ/Iθ = (1 − t2)2 which can be close to unity
for strongly reflecting junctionst ≪ 1. As both contributions are minimal forϕ = π at
θ = 0, one can considerϕ = π, θ = 0 as the “off” state of a spin transistor (Fig. 5.3(c),
right) where the rotation of the spin is provided by a purely quantum mechanical Berry
phase effect. This is in contrast with the famous Datta-Das spin transistor [108] where
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 70
the rotation of the spin is achieved through the classical spin-orbit force. The “on” state
corresponds to the absence of spin rotation forϕ = 0 (Fig. 5.3(c), left).
5.4 Experimental realization in HgTe QW
We now show that this proposal can in principle be realized experimentally in HgTe QW.
We model the device of Fig. 5.1 as a rectangular QSH region threaded by a magnetic AB
flux through a single plaquette in the center, and connected to semi-infinite metallic leads
on both sides by rectangular QSH constrictions modeling quantum point contacts (QPC)
[Fig. 5.4(a)]. The QSH region is described by an effective4 × 4 tight-binding Hamil-
tonian [19, 54] with the chemical potential in the bulk gap, while the metallic leads are
described by the same model with the chemical potential in the conduction band. The de-
tailed form of the model is given in Sec. A.7.3. The injectionof spin-polarized carriers
by the FM layers of Fig. 5.1 is mimicked by the inclusion of an effective Zeeman term
in the Hamiltonian of the semi-infinite leads. We calculate numerically the two-terminal
conductance through the device of Fig. 5.4(a) for a QW thicknessd = 80 A. We use the
standard lattice Green function Landauer-Buttiker approach [89] in which the conductance
is obtained from the Green function of the whole device, the latter being calculated recur-
sively [90].
The results of the numerical calculation are plotted in Fig.5.4(b), (c), (d). In the ab-
sence of phase-breaking scattering processes, one distinguishes two temperatures regimes
T ≪ Tℓ andT ≫ Tℓ separated by a crossover temperatureTℓ = π~v/kBℓ with v the edge
state velocity, defined as the temperature for which a thermal spread∆µ ∼ kBT in the
energy distribution of injected electrons corresponds to aspread in the distribution of dy-
namical phasesλ = 2kF ℓ of ∆λ ∼ 2π. In the low temperature regimeT ≪ Tℓ, ∆λ ≪ 2π
and the dynamical phase is essentially fixed such thatG(T ≪ Tℓ) ≃ G(T = 0). In this
regime,G(T = 0, µ) is approximately periodic inµ for µ within the bulk gap, with period
∆µ ∼ kBTℓ. A crossing pattern (Fig. 5.4(b), top) occurs periodicallyand can be obtained
by tuning the chemical potential. It corresponds to the flux-induced spin rotation effect
(Fig. 5.3). In the high temperature regimeT ≫ Tℓ, one could expect that the crossing
pattern, and thus the spin rotation effect, would be washed out by thermal self-averaging of
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 71
Figure 5.4: Numerical study of the spin AB effect in HgTe QW. (a) Device geometry usedfor the numerical two-terminal conductance calculation:a = 30 A is the lattice constant ofthe tight-binding model,L = 18 nm, Lx = ℓ = 240 nm, Ly = 120 nm, φ is the AB flux,andW is the QPC width. (b) Intensity map of the conductanceG(φ, θ) for fixed chemicalpotentialµ = 0.06 eV (top panel) and averaged chemical potential over energy range∆µ =5meV corresponding to an average over∼ 2π dynamical phase (bottom panel). These twosituations correspond to low and high temperature, respectively (see text). (c) Logarithmicplot of on/off ratioGon/Goff of topological spin transistor as a function of spin polarizationP of injected carriers for fixed chemical potentialµ = 0.06 eV and different values of theQPC widthW . (d) Plot of on/off ratio as a function of QPC widthW for fixed chemicalpotentialµ = 0.06 eV and different values of the spin splitting∆s in the bulk leads.
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 72
the dynamical phase. Surprisingly, the pattern remains (Fig. 5.4(b), bottom), and actually
acquires a more symmetric structure through the self-averaging procedure. In both temper-
ature regimes, the conductance pattern agrees qualitatively with the result of the simple 1D
Hamiltonian model [Fig. 5.3(b)].
So far, our discussion has ignored the existence of phase-breaking processes. Such
processes introduce an additional characteristic temperatureTϕ, defined as the temperature
above which the phase coherence lengthℓϕ(T ) becomes smaller than the system sizeℓ,
that isℓϕ(Tϕ) = ℓ andℓϕ(T > Tϕ) < ℓ. As explained in the Introduction, the stability of
the QSH state is protected by Kramers’ theorem. However, Kramers’ theorem requires the
quantum phase coherence of electronic wave functions, hence for T > Tϕ the QSH state
can be destroyed [20, 78, 79]. Thus, the observation of the spin AB effect requiresT <
Tϕ, with the precise value ofTϕ depending on the particular nature of the phase-breaking
mechanisms. With this first requirement satisfied, two scenarios are possible depending on
the relative value of the two characteristic temperaturesTℓ andTϕ. If Tℓ < Tϕ, the scenario
described in the previous paragraph applies, with the existence of a low temperature regime
T ≪ Tℓ < Tϕ with well-defined dynamical phase and a high temperature regime Tℓ ≪T < Tϕ with completely randomized dynamical phase. On the other hand, if Tℓ > Tϕ, then
since we requireT < Tϕ for the observation of the spin AB effect the high temperature
regimeT ≫ Tℓ can be never be achieved. This could correspond for instanceto a very
small, fully phase-coherent deviceℓ ≪ ℓϕ, with no noticeable thermal fluctuation effects.
In transport measurements on HgTe QW [20, 78, 58] a robust QSHstate has been observed
in devices of sizeℓ ≃ 1 µm up to temperatures of4.2 K. This gives us a lower bound
estimate of a few Kelvins forTϕ, for a device of such size. For a typical edge state velocity
~v ∼ 3.5 eV·A one obtains a crossover temperatureTℓ ∼ 13 K & Tϕ, which indicates that
one would probably be in the low temperature regime with weakthermal fluctuation effects
and good tunability of the crossing pattern with chemical potentialµ. The other scenario
requiringTℓ < Tϕ can be realized if the lower bound estimate of4.2 K turns out to be too
conservative and we actually haveTϕ > 13 K, or if the edge state velocity is significantly
smaller than the value of3.5 eV·A used above. The latter possibility can occur in type-
II QW [75] where the edge state velocity is about one order of magnitude smaller, hence
Tℓ ∼ 1 K and the conditionTℓ < Tϕ would in principle be satisfied.
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 73
In our calculations, for simplicity we have assumed that electrons on both the top and
bottom edges acquire the same dynamical phaseλ. In a real system, the two arms of
the ring are not perfectly symmetric and the electrons propagating on different arms can
certainly acquire different dynamical phasesλbottom 6= λtop. However, the dynamical phase
differenceδ ≡ λbottom − λtop only leads to an additional flux-independent rotation of the
spin of the outgoing electrons, which leads to a shift of the conductance pattern in the angle
θ by an amountδ (see Eq. (A.64) of Sec. A.7). Thus the transistor remains effective if one
usesθ = δ instead ofθ = 0 in the right FM lead. If one prefers to useθ = 0, one can cancel
out the phase asymmetry by patterning an electrostatic gateon top of one given arm. By
tuning the potential of this gate, one can adjust the Fermi wave vector locally and introduce
a dynamical phase offset which cancels out the phase asymmetry δ.
In Fig. 5.4(c), (d) we plot the on/off ratioGon/Goff of the topological spin transistor,
which can be taken as the figure of merit of the device. We defineGon ≡ G(φ = 0, θ = 0)
andGoff ≡ G(φ = 12φ0, θ = 0) [Fig. 5.3(c)]. We use two parameters, the junction spin
polarizationP and the bulk spin splitting∆s to quantify the degree of spin polarization
of the injected carriers. An actual experimental implementation of the transistor concept
described here will require optimization of these or similar parameters. The junction spin
polarizationP is obtained for a given junction geometry, i.e. a given choice of QPC width
W and lengthL [Fig. 5.4(a)], by calculating the transfer matrix [55] of the junction di-
rectly from the TB model and using equation (5.3) withP ≡ |P|. The spin splitting∆s
is obtained from the continuumk · p HgTe QW Hamiltonian mentioned earlier, and is de-
fined as the energy difference between “spin up” (E1+) and “spin down” (E1−) energy
levels [54] at theΓ point. The on/off ratio increases rapidly for a polarization P of order
unity [Fig. 5.4(c)]. It is reasonable to expect that optimized junction designs, better that
the simplistic proof-of-concept geometry used here, wouldyield even higher on/off ratios.
There is also an optimal widthWopt ≃ 0.29Ly for the junction QPC [Fig. 5.4(d)]. For
W < Wopt, interedge tunneling [92] strongly backscatters the incoming electrons and re-
ducesGon, which suppresses the on/off ratio. ForW > Wopt, the edge states on opposite
edges are too far apart to recombine coherently and to produce the desired spin rotation
effect, which increasesGoff and also suppresses the on/off ratio. In our calculation, we
did not take into account the possible structural inversionasymmetry (SIA) which induces
CHAPTER 5. SPIN AB EFFECT AND TOPOLOGICAL SPIN TRANSISTOR 74
Rashba spin-orbit coupling in the QW [88]. However, it should be noticed that the usual
contributions of SIA to the AB effect, such as the Aharonov-Casher (AC) effect [109], are
absent because there are no 2D bulk carriers in the QSH state.Since the only conducting
channels in the QSH state are the 1D edge states, the only effect of SIA is some global
rotation of the edge state spin direction. The topological spin rotation induced by half of
a flux quantum is simply a consequence of the spatial separation of opposite spins on op-
posite edges, which is determined by the topological properties of the QSH state and thus
remains robust.
5.5 Conclusion and outlook
In this work, we have shown the possibility of using a topologically nontrivial state of
matter, the QSH insulator state, to manipulate the spin of the electron by purely nonlocal,
quantum mechanical means, without recourse to local interactions with classical electro-
magnetic fields. This spin AB effect, which is a spin analog ofthe usual charge AB effect,
relies on the helical and topological nature of the QSH edge states which is peculiar to that
state of matter, combined with a Berry phase effect. In addition, we have shown that the
spin AB effect can be used to design a new kind of spin transistor which is fundamentally
different from the previous proposals, in that there is no classical force or torque acting on
the spin of the electron. Furthermore, edge transport in theQSH regime being dissipation-
less [20, 78, 79], the proposed topological spin transistorwould have the advantage of a
lower power consumption in comparison to previous proposals for spin transistors. More
generally, such a quantum manipulation of the electron spin, if observed, could open new
directions in spintronics research and applications, and would at the same time demonstrate
the practical usefulness of topological states of quantum matter.
A recent paper by Usaj [110] discusses a similar effect in thespin-polarized edge states
of graphene ribbons. We expect our effect to be more robust toexternal perturbations due
to the topological protection of the QSH edge states. Indeed, the helical edge liquid of the
QSH state is a novel state of matter which is topologically distinct [21, 44] from the edge
states of graphene.
Chapter 6
Topological quantization in units of the
fine structure constant
6.1 Introduction
Topological phenomena in condensed matter physics are typically characterized by the
exact quantization of the electromagnetic response in units of fundamental constants. In
a superconductor, the magnetic flux is quantized in units of the flux quantumφ0 ≡ h2e
;
in the QH effect, the Hall conductance is quantized in units of the conductance quantum
G0 ≡ e2
h. Not only are these fundamental physical phenomena, they also provide the most
precise metrological definition of basic physical constants. For instance, the Josephson ef-
fect in superconductivity allows the most precise measurement of the flux quantum which,
combined with the measurement of the quantized Hall conductance, provides the most ac-
curate determination of Planck’s constanth to date [111]. The remarkable observation of
such precise quantization phenomena in these imprecise, macroscopic condensed matter
systems can be understood from the fact that they are described in the low-energy limit
by topological field theories with quantized coefficients. For instance, the QH effect is
described by the topological Chern-Simons theory [9] in2 + 1 dimensions, with coeffi-
cient given by the quantized Hall conductance. Superconductivity can be described by the
topologicalBF theory [112] with coefficient corresponding to the flux quantum.
More recently, a new topological state in condensed matter physics, the TR invariant
75
CHAPTER 6. TOPOLOGICAL QUANTIZATION IN UNITS OFα 76
topological insulator (TI), has been investigated extensively [113, 114, 115, 116]. The
concept of TI can be defined most generally in terms of the topological field theory [16]
with effective Lagrangian
L =1
8π
(
εE2 − 1
µB2
)
+θ
2π
α
2πE ·B, (6.1)
whereE andB are the electromagnetic fields,ε andµ are the dielectric constant and mag-
netic permeability, respectively, andθ is an angular variable known in particle physics as
the axion angle [32]. Under periodic boundary conditions, the partition function and all
physical quantities are invariant under shifts ofθ by any multiple of2π. SinceE ·B is odd
under TR symmetry, the only values ofθ allowed by TR symmetry are0 or π (modulo2π).
The second term of Eq. (6.1) thus defines a topological field theory with coefficient quan-
tized in units of the fine structure constantα ≡ e2
~c. The topological field theory is generally
valid for interacting systems, and describes a quantized magnetoelectric response denoted
topological magnetoelectric effect [16]. The quantization of the axion angleθ depends
only on the TR symmetry and the bulk topology; it is thereforeuniversal and independent
of any material details. More recently, it has been shown [34] that the topological field
theory [16] reduces to the topological band theory [12, 13, 27, 28] in the noninteracting
limit. Interestingly, the topological magnetoelectric effect is the first topological quantiza-
tion phenomenon in units ofα. It can therefore be combined with the two other known
topological phenomena in condensed matter, the QH effect and superconductivity, to pro-
vide a metrological definition of the three basic physical constants,e, h, andc.
The topological magnetoelectric effect has not yet been observed experimentally. An
insight into why this is so can be gained by comparing the3 + 1 dimensional topological
field theory (6.1) of TI to the2 + 1 dimensional Chern-Simons topological field theory of
the QH effect [9]. In2 + 1 dimensions, the topological Chern-Simons term is the only
term which dominates the long-wavelength behavior of the system, which leads to the
universal quantization of the Hall conductance. On the other hand, in3 + 1 dimensions
the topologicalθ-term in Eq. (6.1) and the Maxwell term are equally importantin the
long wavelength limit. Therefore, one has to be careful whendesigning an experiment
to observe the topological quantization of the topologicalmagnetoelectric effect, in which
CHAPTER 6. TOPOLOGICAL QUANTIZATION IN UNITS OFα 77
Figure 6.1: Measurement of Kerr and Faraday angles for a TI thick film of thicknessℓ andoptical constantsε2, µ2 on a topologically trivial insulating substrate with optical constantsε3, µ3, in a perpendicular magnetic fieldB. (We consider normal incidence in the actualproposal but draw light rays with a finite incidence angle in the figure for clarity.) Theexternal magnetic field can be replaced by a thin magnetic coating on both TI surfaces, assuggested in Ref. [16].
the dependence on the non-topological materials constantsε andµ are removed.
6.2 Magnetooptical Kerr and Faraday rotation
In this chapter, we propose an optical experiment to observethe topological quantization
of the topological magnetoelectric effect in units ofα, independent of material properties
of the TIsuch asε andµ. This experiment could be performed on any of the available TI
materials, such as the Bi2Se3, Bi2Te3, Sb2Te3 family or the recently discovered thallium-
based compounds [116]. Consider a TI thick film of thicknessℓ with optical constants
ε2, µ2 and axion angleθ deposited on a topologically trivial insulating substratewith optical
constantsε3, µ3 (Fig. 6.1). The vacuum outside the TI hasε = µ = 1 and trivial axion angle
CHAPTER 6. TOPOLOGICAL QUANTIZATION IN UNITS OFα 78
θvac = 0. The substrate being also topologically trivial, both interfaces atz = 0 andz = ℓ
support a domain wall ofθ giving rise to a surface QH effect with half-quantized surface
Hall conductanceσsH = (n + 1
2) e
2
hwith n ∈ Z [16]. The factor of1
2is a topological
property of the bulk and is protected by the TR symmetry. On the other hand, the value of
n depends on the details of the interface and may thus be different for the two interfaces. To
account for this general case we assignθsubs = 2pπ with p ∈ Z to the topologically trivial
substrate, corresponding toσs,0H = θ
2πe2
hon thez = 0 interface andσs,ℓ
H = (p − θ2π) e
2
hon
thez = ℓ interface. The experiment consists in shining normally incident monochromatic
light with frequencyω on the TI film, and measuring the Kerr angleθK of the reflected
light and Faraday angleθF of the transmitted light. However, the effective theory (6.1)
applies only in the regimeω ≪ Eg/~ whereEg is the surface gap [16]. Such a surface
gap can be opened by a thin magnetic coating on both surfaces of the TI, as first suggested
in Ref. [16], or by an applied perpendicular magnetic fieldB = Bz (Fig. 6.1) through the
surface Zeeman effect as well as the exchange coupling between surface electrons and the
paramagnetic bulk. We discuss the experimentally simpler case of the external magnetic
field. For incident light linearly polarized in thex directionEin = Einx, the Kerr and
Faraday angles are defined bytan θK = Eyr /E
xr andtan θF = Ey
t /Ext , respectively, with
Er = Exr (−x) + Ey
r y andEt = Ext x + Ey
t y the reflected and transmitted electric fields,
respectively (Fig. 6.1). Furthermore,θK andθF are to be measured as a function ofB. The
angles that we discuss in the following are defined as the linear extrapolation ofθK(B) and
θF (B) asB → 0+, in which limit the non-topological bulk contribution to optical rotation
is removed [16].
The problem of optical rotation at a TI/trivial insulator interface has been studied be-
fore [16, 35, 117]. In general,θK andθF depend on the optical constantsε2, µ2 of the TI. In
the thick film geometry considered here, they will also depend in a complicated manner on
the optical constantsε3, µ3 of the substrate, the film thicknessℓ, and the photon frequency
ω, due to multiple reflection effects at the two interfaces (see Sec. A.8). It seems therefore
dubious that one could extract the exact quantization of thetopological magnetoelectric
effect from such a measurement. However, we find that these multiple reflection effects
can be used for a universal measurement of the topological magnetoelectric effect, with no
explicit dependence onε2, µ2, ε3, µ3, ℓ, andω.
CHAPTER 6. TOPOLOGICAL QUANTIZATION IN UNITS OFα 79
Figure 6.2: (a) ReflectivityR as a function of photon frequencyω in units of the character-istic frequencyωℓ for a topological insulator Bi2Se3 thick film on a Si substrate; universalfunctionf(θ) for different values of (b) the substrate dielectric constant ε3, (c) p, the totalsurface Hall conductance in units ofe2
h, and (d) the TI dielectric constantε2. The position of
the zero crossing is universal and provides an experimentaldemonstration of the quantizedtopological magnetoelectric effect.
6.3 Reflectivity minima and total surface Hall conductance
In Fig. 6.2(a) we plot the reflectivityR ≡ |Er|2/|Ein|2 as a function of photon frequencyω
in units of a characteristic frequencyωℓ ≡ c√ε2µ2
πℓ, for ε2 = 100, ε3 = 13, andµ2 = µ3 = 1,
appropriate for a topological Bi2Se3 [22] thin film on a Si substrate [118, 119, 120]. We
observe that minima inR occur whenω/ωℓ is an integer, corresponding toℓ being an
integer multiple ofλ2
2with λ2 = 2πc
ω√ε2µ2
the photon wavelength inside the TI. For radiation
in the terahertz range this corresponds toℓ ∼ 100 µm. Whenω is tuned to any of these
CHAPTER 6. TOPOLOGICAL QUANTIZATION IN UNITS OFα 80
minima, we find
tan θ′K =4αp
Y 23 − 1 + 4α2p2
, tan θ′F =2αp
Y3 + 1, (6.2)
whereYi ≡√
εi/µi is the admittance of regioni, and the prime indicates rotation angles
measured at a reflectivity minimum, i.e. forω/ωℓ ∈ Z. We see thatθ′K andθ′F are indepen-
dent of the TI optical constantsε2, µ2. Equation (6.2) corresponds simply to the results of
Ref. [16, 35] for auniqueinterface with axion domain wall∆θ = 2pπ. Moreover, the two
angles can be combined [121] to obtain a universal result independent of both TIε2, µ2 and
substrateε3, µ3 properties,
cot θ′F + cot θ′K1 + cot2 θ′F
= αp, p ∈ Z. (6.3)
Since the rotation angles are measured at a reflectivity minimum, Eq. (6.3) has no explicit
dependence onℓ or ω either. Equation (6.3) clearly expresses the topological quantiza-
tion in units ofα solely in terms of experimentally measurable quantities, and is the first
important result of this work.
6.4 Reflectivity maxima and topological magnetoelectric
effect
However, neither Eq. (6.2) nor Eq. (6.3) depend explicitly on the TI axion angleθ, and
one may ask whether Eq. (6.3) is at all an indication of nontrivial bulk topology. In fact,
Eq. (6.3) describes the topological quantization of thetotal Hall conductance of both sur-
facesσs,totH = σs,0
H + σs,ℓH = p e2
h, which holds independently of possibleT breaking in the
bulk. In the special case that the two surfaces have the same surface Hall conductance, we
havep = 2σs,0H = θ
πand Eq. (6.3) is sufficient to determine the bulk axion angleθ. How-
ever, for a TI film on a substrate the two surfaces are generically different and can have
different Hall conductance. To obtain the axion angleθ in the more general case of dif-
ferent surfaces, we propose another optical measurement performed at reflectivitymaxima
ω = (n + 12)ωl, n ∈ Z [Fig. 6.2(a)]. We denote byθ′′K andθ′′F the Kerr and Faraday angles
CHAPTER 6. TOPOLOGICAL QUANTIZATION IN UNITS OFα 81
measured at an arbitrary reflectivity maximum. In contrast to θ′K andθ′F [Eq. (6.2)], these
depend onε2, µ2 as well as onε3, µ3,
tan θ′′K =4α[
Y 22
(
p− θ2π
)
− Y 23
θ2π
]
Y 23 − Y 4
2 + 4α2[
2Y 22
θ2π
(
p− θ2π
)
− Y 23
(
θ2π
)2] ,
tan θ′′F =2α(
p− θ2π
+ Y3θ2π
)
Y3 + Y 22 − 4α2 θ
2π
(
p− θ2π
) , (6.4)
where we defineY 23 = Y 2
3 +4α2(
p− θ2π
)2. More importantly,θ′′K andθ′′F depend explicitly
on the TI axion angleθ. It is readily checked that Eq. (6.4) reduces to Eq. (6.2) in the single-
interface limitθ = 2pπ, Y2 = Y3 or θ = 0, Y2 = 1. In general however, from the knowledge
of p [Eq. (6.3)] and eitherθ′K or θ′F we can extractY3 by using Eq. (6.2) without performing
any separate measurement. Moreover,θ′′K andθ′′F can be combined to cancel the explicit
dependence on the TI propertiesε2, µ2. We solve forY 22 in Eq. (6.4) in terms ofθ′′F , say,
and substitute the resulting expressionY 22 = Y 2
2 (θ) into the equation forθ′′K in Eq. (6.4).
The result can be expressed in the formf(θ′K , θ′F , θ
′′K , θ
′′F ; p, θ) = 0 wheref is “universal”
in the sense that it does not depend explicitly on any material parameterεi, µi. Substituting
the experimental values ofθ′K , θ′F , θ
′′K , θ
′′F andp into this expression, we obtain a function
of a single variablef(θ). If we plot f as a function ofθ, the zero crossingf(θ) = 0 gives
the value of the bulk axion angleθ with no2π ambiguity. Plots of the universal functionf
are given in Fig. 6.2(b), (c), and (d) for different values ofthe material parametersε2, ε3, p
(settingµ2 = µ3 = 1 without loss of generality) and for a bulk axion angleθ = π. The
zero crossing point is independent of material parameters and, together with Eq. (6.3),
can provide a universal experimental demonstration of the quantization of the topological
magnetoelectric effect in the TI bulk. In athin film geometryℓ ≪ λ2
2corresponding to
ω ≪ ωℓ, the optical response is always given by the sum of the Hall conductivities of the
two surfaces. Therefore, thick filmsℓ ≥ λ2
4to allow destructive interference and reflectivity
maxima are essential to the measurement of the bulk topological magnetoelectric effect.
CHAPTER 6. TOPOLOGICAL QUANTIZATION IN UNITS OFα 82
Figure 6.3: (a) Kerr-only measurement setup, with materialparameters the same as indi-cated in Fig. 6.1; (b), (c) and (d): universal functionfK(θ) for different material parameters[same as in Fig. 6.2(b), (c), (d)]. As in Fig. 6.2, the position of the zero crossing is universaland provides an experimental demonstration of the quantized topological magnetoelectriceffect.
6.5 Kerr-only measurements
Our proposal so far necessitates the measurement of both Kerr and Faraday angles. We now
show that it is possible to extractp andθ from Kerr measurements alone, if the Kerr angle
is measured in both directions [Fig. 6.3(a)]. Indeed, whilethe Faraday angle is generally
independent of the direction of propagation [122], the Kerrangle depends on it. Here we
exploit this asymmetry of the Kerr angle to extractp andθ. We denote byθ′13K andθ′′13K the
Kerr angles defined previously in Eq. (6.2) and (6.4), respectively. Conversely, we denote
by θ′31K andθ′′31K the Kerr angles for light traveling in the opposite direction, i.e. incident
from the substrate [Fig. 6.3(a)]. As before, the prime and double prime correspond to
CHAPTER 6. TOPOLOGICAL QUANTIZATION IN UNITS OFα 83
angles measured at reflectivity minima and maxima, respectively. We find
tan θ′31K = − 4αpY3Y 23 − 1 + 4α2p2
, (6.5)
tan θ′′31K =4αY3
[
Y 22
θ2π
− γ(
p− θ2π
)]
γY 23 + 4γα2
[
p2 −(
θ2π
)2]
− Y 42 − 8α2Y 2
2
(
θ2π
)2,
where we defineγ ≡ 1 + 4α2(
θ2π
)2. As previously,θ′13K and θ′31K can be combined to
eliminateY3 and provide a universal measure ofp ∈ Z,
cot θ′13K − sgn p√
1 + cot2 θ′13K (1− tan2 θ′31K ) = 2αp, (6.6)
providedY 23 ≡ ε3/µ3 > 1 + 4α2p2, which is satisfied in practice for lowp sinceα2 ∼
10−4. Furthermore, comparing Eq. (6.5) forθ′31K to Eq. (6.2) forθ′13K we see thatY3 is
easily obtained asY3 = − cot θ′13K tan θ′31K . Finally, to extract the bulk axion angleθ, we
need to solve forY 22 in Eq. (6.4) in terms ofθ′′13K , and substitute the resulting expression
Y 22 = Y 2
2 (θ) into the equation forθ′′31K in Eq. (6.5). The result of this analysis can once
again be expressed in the formfK(θ′13K , θ′31K , θ′′13K , θ′′31K ; p, θ) = 0, wherefK is a “universal”
function which only depends on the measured Kerr angles. As before, we substitute intofK
the experimental values ofθ′13K , θ′31K , θ′′13K , θ′′31K andp [obtained from Eq. (6.6)] and obtain
a function of a single variablefK(θ) which crosses zero at the value of the bulk axion
angle with no2π ambiguity. In Fig. 6.3(b), (c) and (d) we plot the universal functionfK
for different values of the material parametersε2, ε3, p and for a bulk axion angleθ = π.
The zero crossing point is independent of material parameters and, together with Eq. (6.6),
provides another means to demonstrate experimentally the universal quantization of the
topological magnetoelectric effect in the bulk of a TI.
6.6 Discussion
Recent work [36] has addressed the similar problem of optical rotation on a TI film, and
found interesting and novel results for the rotation angles. However, these results hold only
in certain limits which are less general than the ones discussed in this work. First, Ref. [36]
CHAPTER 6. TOPOLOGICAL QUANTIZATION IN UNITS OFα 84
considers a free-standing TI film in vacuum. Most films are grown on a substrate which can
affect the physics qualitatively. For instance, the giant Kerr rotationθK = tan−1(1/α) ≃π/2 found in Ref. [36] is a special case of our Eq. (6.2) withp = 1 andε3/µ3 = 1. It is
dramatically suppressed whenε3/µ3 − 1 is greater thanα2 ∼ 10−4, which is typically the
case in practice. Second, in Ref. [36] a correction proportional to∆/ǫc was introduced to
the surface Hall conductance, where∆ is the TR symmetry-breaking Dirac mass andǫc is
a non-universal high-energy cutoff. According to the general bulk topological field theory
of the TI [16], the surface Hall conductance is always quantized as long as the surface is
gapped and the bulk is TR invariant (in theB → 0 limit). Thus we conclude that such a
non-universal correction is absent and the requirement∆ ≪ ǫc is not necessary within the
topological field theory approach [16]. This difference clearly demonstrates the power of
the topological field theory approach [16] in predicting universally quantized topological
effects in condensed matter physics.
Chapter 7
Fractional topological insulators in three
dimensions
7.1 Introduction
Most states of quantum matter are classified by the symmetries they break. However, topo-
logical states of quantum matter [113] evade traditional symmetry-breaking classification
schemes, and are rather described by topological field theories in the low-energy limit. For
the quantum Hall effect, the topological field theory is the2+1 dimensional Chern-Simons
theory [9] with coefficient given by the quantized Hall conductance. In the noninteracting
limit, the integer QH [2] conductance in units ofe2
his given by the TKNN invariant [6] or
first Chern number. In the presence of strong correlations, one can also observe the frac-
tional QH effect [3, 123], where the Hall conductance is quantized in rational multiples ofe2
h. In both cases however, these topological states can exist only in a strong magnetic field
which breaks TR symmetry.
More recently, TR invariant TI have been studied extensively [113, 114, 115]. The
TI state was first predicted theoretically in HgTe QW, and observed experimentally [19,
20, 12, 14] soon after. The theory of TI has been developed along two independent routes.
Topological band theory identifiedZ2 topological invariants for noninteracting band insula-
tors [12, 27, 28]. The topological field theory of TR invariant insulators was first developed
in 4 + 1 dimensions, where the Chern-Simons term is naturally TR invariant [124, 125].
85
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 86
Dimensional reduction then gives the topological field theory for TI in 3 + 1 and2 + 1
dimensions [16]. The topological field theory is generally valid for interacting systems,
and describes the experimentally measurable quantized magnetoelectric response. The co-
efficient of the topological term, the axion angleθ, is constrained to be either0 or π by TR
invariance. The topological field theory has been further developed in Ref. [126, 35]. More
recently, it has been shown that it reduces to the topological band theory in the noninteract-
ing limit [34].
By analogy with the relation between the integer QH effect and fractional QH effect,
one is naturally led to the question whether there can exist a“fractional TI”. In 2+1 dimen-
sions, an explicit wave function for the fractional QSH state was first proposed in Ref. [14],
and the edge theory was investigated in Ref. [127]. The TR invariant fractional topological
state has also been constructed explicitly in4 + 1 dimensions [124]. Since TR invariant
TI form a dimensional ladder in4, 3 and2 dimensions [16, 128, 129], it is natural to in-
vestigate the TR invariant TI in3 + 1 dimensions. Fractional states generally arise from
strong interactions. Since topological band theory cannotdescribe such interactions, we
formulate the general theory in terms of the topological field theory. The TI is generally
described by the effective actionSθ = θ2π
e2
2π
∫
d3x dtE · B whereE andB are the elec-
tromagnetic fields [16]. Under periodic boundary conditions, the partition function and all
physical quantities are invariant under shifts ofθ by multiples of2π. SinceE · B is odd
under TR, it appears that the only values ofθ allowed by TR symmetry are0 or π mod2π.
In this paper, we show that there exist TR invariant insulating states in3+1 dimensions
with P3 ≡ θ2π
quantized in non-integer, rational multiples of12
of the formP3 = 12pq
with
p, q odd integers. The magnetoelectric polarizationP3 is defined by the response equation
P = − B
2π(P3 + const.), whereB is an applied magnetic field andP is the induced electric
polarization. Such a fractionalized bulk topological quantum number leads to a fractional
QH conductance ofpqe2
2hon the surface of the fractional TI. In contrast to the usual QH
effect in2 + 1 dimensions, the surface QH effect does not necessarily exhibit edge states
and thus cannot be directly probed by transport measurements. Alternatively, it can only
be experimentally observed through probes which couple to each surface separately, such
as magnetooptical Kerr and Faraday rotation [16, 35]. Generically, a slab of fractional TI
can have different fractional Hall conductance on the top and bottom surfaces, which can
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 87
be determined separately by combined Kerr and Faraday measurements, independent of
non-universal properties of the material [37].
7.2 Projective construction and trial wave function
Our approach is inspired by the composite particle, or projective construction of fractional
QH states [9, 130, 131, 132, 133, 134, 135, 136, 137]. The ideais to decompose the
electron with chargee into N fractionally charged, fermionic “partons”, which have a dy-
namics of their own. One considers the case that the partons form a known topological
state, say a topological band insulator. When the partons are recombined to form the phys-
ical electrons, a new topological state of electrons emerges. In the fractional QH case for
example, theν = 13
Laughlin state can be obtained by splitting each electron into N = 3
partons of chargee3. Each parton fills the lowest Landau level and forms a noninteracting
ν = 1 integer QH state. Ignoring the exponential factors, the parton wave function is the
Slater determinant integer QH wave functionΨ(zi) ∝ ∏
i<j (zi − zj), and the electron
wave function is obtained by gluing three partons together,which leads to the Laughlin
wave function [123]Ψ1/3(zi) ∝∏
i<j (zi − zj)3. Similarly, in 3+1 dimensions one can
construct an interacting many-body wave function by gluingpartons which are in aZ2
topological band insulator state. The parton ground state wave functionΨ1(rnsn) is a
Slater determinant describing the ground state of a noninteracting TI Hamiltonian such as
the lattice Dirac model [16], withrnsn, n = 1, . . . , N the position and spin coordinates
of the partons. The electron wave function is obtained by requiring the coordinates of all
Nc partons forming the same electron to be the same [130],
ΨNc(rnsn) = [Ψ1(rnsn)]Nc . (7.1)
Equation (7.1) is the(3+1)-dimensional generalization of the Laughlin wave function, and
serves as a trial wave function for the simplest fractional TI phases we propose.
More generally, we can considerNf different “flavors” of partons, withN (f)c partons
of each flavorf = 1, . . . ,Nf . This decomposition has to satisfy two basic rules. First, to
preserve the fermionic nature of the electron, the total number of partons per electron must
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 88
be odd (Fig. 7.1a),
N (1)c +N (2)
c + · · ·+N (Nf )c = odd. (7.2)
Second, ifqf < e is the (fractional) charge of partons of flavorf , the total charge of the
partons must add up to the electron chargee,
N (1)c q1 +N (2)
c q2 + · · ·+N (Nf )c qNf
= e. (7.3)
For instance, theν = 13
Laughlin state described above corresponds toNf = 1, N (1)c = 3,
andq1 = e3, which satisfies both conditions. Here we consider that partons of each flavor
f condense in a (generally different) noninteracting TR invariant TI state with axion angle
θf = π mod2π. This is the analog of having partons condense in various integer QH states
in the fractional QH construction. Finally, the partons have to be bound together to yield
physical electrons. As we will see, this can be done by coupling partons of flavorf to a
SU(N (f)c ) gauge field, which can be interpreted as a “color field” where partons of flavor
f come inN (f)c colors. Since the TI analog of theν = 1
3Laughlin state will involve three
partons coupled to aSU(3) gauge field in3 + 1 dimensions, we dub our partons “quarks”
by analogy with quantum chromodynamics (QCD).
7.3 Effective gauge theory
To obtain a more systematical understanding of the fractional TI, we now deduce its effec-
tive gauge theory by way of a gedanken experiment. We consider subjecting a noninteract-
ing TI to strong electron-electron interactions, and startwith the simplest case ofNf = 1
with Nc odd. The electron being split intoNc quarks of chargeeNc
, the electron operator
will be written as a product ofNc quark operatorsψiα, i = 1, . . . ,Nc. However, the quark
operators act in a Hilbert space which is larger than the physical electron Hilbert space.
We need to remove those states of the quark Hilbert space which are not invariant under
unitary transformations which leave the electron operatorunchanged, i.e.SU(Nc) trans-
formations with quarks in theNc representation. The projection onto the electron Hilbert
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 89
Figure 7.1: a) Quark picture of fractional TI with flavor and color degrees of freedom; b)surface fractional QH effect vs transport measurements [Eq. (7.7)]; c) nontrivial vs trivialfractional TI; d) Witten effect as a probe of bulk topology.
space can therefore be implemented by coupling the quarks toa SU(Nc) gauge fieldaµ
with a coupling constantg. Outside the fractional TI, we expect the system to be in the
confined phase, in analogy to quark confinement in QCD, which has onlySU(Nc) singlet
excitations in its low-energy spectrum, i.e. gauge-invariant “baryons”. Quarks of a given
flavor within the baryon are antisymmetric in theirN (f)c color indices; Fermi statistics then
implies that their spins are aligned. In a relativistic theory this would imply that in the
Nf = 1 theory the baryon has spinNc
2. In nonrelativistic lattice models this is not a con-
cern, but even within the context of relativistic continuumfield theories one can obtain
composite spin-12
electrons forNf > 1.
Inside the fractional TI, electron-electron interactionstranslate into complicated in-
teractions among quarks. We consider the case that these interactions lead the quarks to
condense at low energies into a noninteracting TR invariantTI state with axion angleθ, and
that the non-Abelian gauge fieldaµ enters a deconfined phase1. We now show that such a
phase is a fractional TI. A low-energy effective Lagrangianfor Nf = 1 can be conjectured
1This can be achieved either by adding additional colored butelectrically neutral matter or consideringthe special case with only Abelian groups,N (f)
c = 1 for all f .
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 90
in the form
L = ψ† (iD0 −Hθ[−iD])ψ + Lint(ψ†, ψ), (7.4)
whereDµ = (D0,−D) = ∂µ + i eNcAµ + igaµ is theU(1)em × SU(Nc) gauge covari-
ant derivative, andHθ = Hθ(p) is the single-particle Hamiltonian for a TR invariant TI
with axion angleθ. Lint represents weak TR invariant residual interactions which do not
destabilize the gapped TI phase, and can thus be safely ignored. The kinetic Yang-Mills
Lagrangian foraµ is generally present but not explicitly written.
Since the quarks are in a gapped TI phase, they can be integrated out to yield an effective
Lagrangian for the gauge fields [16],
Leff =θ
32π2ǫµνλρ tr
(
e
NcFµν + gfµν
)(
e
NcFλρ + gfλρ
)
=θeffe
2
32π2ǫµνλρFµνFλρ +
θg2
32π2ǫµνλρ tr fµνfλρ, (7.5)
wheretr is the trace in theNc representation ofSU(Nc), Fµν = ∂µAν − ∂νAµ andfµν =
∂µaν − ∂νaµ + ig[aµ, aν ] are theU(1)em andSU(Nc) field strengths, respectively, and the
electromagnetic response is governed by an effective axionangle
θeff =θ
Nc
= 0,± π
Nc
,±3π
Nc
,±5π
Nc
, . . . , Nc odd. (7.6)
Equation (7.5) is obtained by replacing theU(1)em “electron” field strengtheFµν in the
U(1)em topological term θ2π
e2
2πE · B = θe2
32π2 ǫµνλρFµνFλρ for noninteracting TI [16] by the
totalU(1)em×SU(Nc) “quark” field strength eNcFµν+gfµν . The crossed terms of the form
trFµνfλρ vanish due to the tracelessness of theSU(Nc) gauge field. In general,θeff can be
obtained from the Adler-Bell-Jackiw anomaly [138, 139], sinceθ corresponds to the phase
of the quark mass [16]. The effective theory can also be obtained for quarks in a trivial
insulator state withθ = 2nπ, n ∈ Z. However, such a state is adiabatically connected to
a θ = 0 vacuum and is a trivial insulator in the bulk, although a fractional θeff can still be
obtained due to pure surface effects. Since the focus of the present work is a fractional TI
state with nontrivial bulk, we always consider quarks withθ = π mod 2π in the following.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 91
7.4 Time-reversal symmetry and quantization of the ax-
ion angle
We are now faced with our initial question of whether the effective theory (7.5),(7.6) breaks
TR invariance. According to the first term in Eq. (7.5), TR invariance would requireθeff to
be quantized in integer multiples ofπ if the minimal electric charge wase [140]. However,
the minimal charge in our theory iseNc, i.e. that of the quarks. Therefore,θeff has to be
quantized in integer multiples ofπN 2c. On the other hand, the second term in Eq. (7.5)
requiresθ to be quantized2 in integer multiples ofπ, which means by Eq. (7.6) thatθeff has
to be quantized in units ofπNc. This latter constraint is consistent with, but stronger than,
the former3, and the values ofθeff allowed by TR invariance are thus correctly given by
Eq. (7.6).
Equations (7.5) and (7.6) constitute a topological field theory which, precisely because
it is topological, is insensitive to small TR invariant perturbations and defines a new stable
phase of matter, the TR invariant fractional TI in3 + 1 dimensions. The effective theory
can also be derived in the multi-flavor caseNf ≥ 1, with N (f)c satisfying rules (7.2) and
(7.3). Considering that quarks of flavorf form a noninteracting TI with axion angleθf =
π mod 2π and integrating them out yields an effective Lagrangian in the form of (7.5),
but with gauge groupU(1)em×∏Nf
f=1 U(N(f)c )/U(1)diag. HereU(1)diag is the overallU(1)
gauge transformation of the electron operator. The electromagnetic axion angleθeff is given
by θeff =(
∑Nf
f=1N (f)
c
θf
)−1
. When θfπ
is odd for each flavor, one can show thatθeff = πp/q
with p, q odd integers.
2For a recent discussion, see Ref. [141, 142].3Color neutral monopoles carryU(1)em magnetic chargeNc
e , and lead toθeff = nπN 2
c
, n ∈ Z. However,
monopoles with smallerU(1)em magnetic charge1e are allowed but also carry color magnetic charge [143],and lead toθeff = nπ
Nc
, n ∈ Z.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 92
7.5 Bulk topology and surface “half” fractional quantum
Hall effect
Important physical properties of the fractional TI can be read off from Eq. (7.5). The
surface of the fractional TI is an axion domain wall with theU(1)em axion angle jumping
from θeff in the fractional TI to0 in the vacuum. Such a domain wall has a surface QH effect
with surface Hall conductanceσH,s =θeff2π
e2
h[16]. Therefore, the surface Hall conductance
of the fractional TI has the general form
σH,s =p
q
e2
2h, p, q odd. (7.7)
For example, in the simplest single-flavor case withθ = π in Eq. (7.5), we haveσH,s =1Nc
e2
2hwith Nc an odd integer, corresponding to half of aν = 1
Ncfractional QH Laughlin
state. The more general result (7.7) corresponds to half of ageneric Abelian fractional QH
state [144, 145, 130].
The fractional axion angle and the associated surface Hall conductance (7.7) are prop-
erties of the bulk topology. It is important to distinguish them from a TI withθeff = ±πand where the surface Dirac fermions form a fractional QH state [146]. In a noninteracting
TI with θeff = −π for example, both the axion domain wall and the surface fractional QH
state contribute toσH,s,
σH,s =
(
−1
2+n
q
)
e2
h=
2n− q
q
e2
2h, (7.8)
with nq
an allowed filling fraction for a fractional QH state in2+1 dimensions. For Abelian
fractional QH statesq is odd, hence the surface Hall conductanceσH,s = 2n−qq
e2
2hhas the
same general form as for the fractional TI [Eq. (7.7)]. As thesimplest example, the Laugh-
lin state with nq= 1
3leads toσH,s = −1
6e2
h(Fig. 7.1c, right) which is the same as for a
genuine fractional TI with bulkP3 = −16
(Fig. 7.1c, left). However, the bulk topology is
very different in both cases. Therefore, surface measurements are not sufficient to deter-
mine the bulk topology and bulk measurements ofP3 are needed. One such measurement
would consist in embedding a monopole with magnetic chargeqm inside the fractional TI
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 93
(Fig. 7.1d) and measuring its electric chargeqe induced by the Witten effect [147, 38, 148].
Another possible “experiment” is to measure the ground state degeneracy on topolog-
ically nontrivial spatial 3-manifolds. Consider a fractional TI on a manifoldΣg × I with
Σg a Riemann surface of genusg andI = [0, L] a bounded interval, whereL is the sam-
ple “thickness” and the two copies ofΣg (at each end ofI) are the two bounding sur-
faces. We first discuss contributions to the ground state degeneracy arising solely from the
boundary, and comment on bulk contributions later on. A noninteracting TI with aν = 13
Laughlin state deposited on both surfaces is described by two independent Chern-Simons
theories [9, 149, 150, 151] and has a ground state degeneracyof 3g (mg for ν = 1m
) on
each surface for a total ground state degeneracy of(3g)2 = 32g. The situation is different
for a genuine fractional TI withP3 = ±16
(Fig. 7.1c, left). To study the ground state de-
generacy we set the external electromagnetic fields to zero in Eq. (7.5) and consider the
internalSU(3) θ-term. Assuming that the system stays gapped as we take the limit of
zero thicknessL → 0 where the gauge fieldsaµ on both surfaces become identified, the
system is described by a singleSU(3)k Chern-Simons theory onΣg where the levelk is
the sum of the contributions from both surfaces. If on both surfacesθ goes to the same
value outside the TI, thenk = 0 and the ground state is unique. Ifθ = 0 on one side and
θ = 2π on the other, we have aSU(3)1 Chern-Simons theory with ground state degeneracy
3g 6= 32g [150, 151]. In addition to the boundary contributions of theθ-term to the ground
state degeneracy, the gauge theory in the bulk can have a nontrivial ground state degeneracy
even in the absence of boundaries. For instance, the deconfined phase ofSU(Nc) gauge
theory has a ground state degeneracy ofN 3c onT 3 [152]. The total ground state degener-
acy has in general both bulk and boundary contributions, anddepends on the details of the
gauge group.
7.6 Three classes of parton models
Topological insulators [115, 116] are new states of quantummatter that cannot be adiabat-
ically connected to conventional insulators. They are fully gapped in the bulk but support
gapless boundary modes which are protected by discrete symmetries [128, 153, 129, 154].
Three-dimensional TR invariant TI [27, 29] have attracted agreat deal of attention in the
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 94
past few years, for the most part due to their theoretical prediction[30, 22] and subsequent
experimental detection [31, 24, 25] in Bi-based semiconductors, the latter through the ob-
servation of their protected helical surface states. The existence and stability of these sur-
face states is protected by a bulkZ2 topological invariant which corresponds physically to a
quantized magnetoelectric effect [16, 126]. The experimental observation of the quantized
magnetoelectric effect, for instance by way of magnetooptical measurements,[16, 36, 37]
is a key goal in the field which is being actively pursued [118,155, 156]. From a theo-
retical standpoint [16], the quantized magnetoelectric effect is described at energies much
smaller than the energy gap by the addition of a term proportional toE · B (with E and
B the electric and magnetic fields, respectively) to the usualLagrangian for Maxwell elec-
trodynamics, i.e. axion electrodynamics [32]. This term isin fact the Abelian version of
the topologicalθ-term in QCD [157, 158], and its coefficientθ is periodic (under certain
conditions [159]) with period2π. SinceE ·B is odd under TR symmetry, the only values
of θ allowed by TR symmetry are0 or π mod 2π, with θ = 0 for theZ2 trivial insula-
tor andθ = π for theZ2 nontrivial insulator [16]. TI can be described microscopically
by noninteracting, spin-orbit coupled electrons hopping on a lattice, and the axion angle
θ can be computed from a knowledge of the single-particle wavefunctions in momentum
space [16, 126].
The effect of electron-electron interactions on the surface modes of 3D TI has been a
topic of considerable current interest. Weak interactionscan either turn the noninteracting
helical surface state into a weakly interacting helical Fermi liquid with spin-charge coupled
collective modes [160], or drive a transition to a superconducting state [161, 162]. Strong
enough interactions can lead to spontaneous TR symmetry-breaking on the surface and fer-
romagnetic [163, 164, 165, 166] or helical spin density waveorder [167]. In contrast, the
bulk of a TI is fully gapped and thus expected to be perturbatively stable to interactions. On
the other hand, exotic states known as 3D topological Mott insulators [168, 169, 170] have
been theoretically proposed, whereas a topologically nontrivial bandstructure is dynami-
cally generated as a consequence of strong electron-electron interactions. Although these
are strongly interacting states, their mean-field description is still that of a topological band
insulator, and the axion angleθ remains quantized [171, 172] to0 or π mod2π. Another
type of topological Mott insulator has been theoretically proposed [173, 174] in which
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 95
spin-charge separation leads to a bulk insulator with a helical liquid of gapless spinons, but
an electromagneticθ-term is not generated because the spinons are electricallyneutral.
In many regards, the 3D TR invariant TI can be viewed as a generalization of the 2D
integer QH effect to 3D. The topologicalZ2 quantization of the bulk axion angleθ in 3D is
the direct analog of the topologicalZ quantization of the bulk Hall conductance in 2D. By
analogy with the relation between the integer QH effect and the fractional QH effect, one is
naturally led to the question whether there can exist a “fractional 3D TI” which preserves
TR symmetry but is characterized by afractional quantized axion angle, i.e. whereθ is
a non-integer, rational multiple ofπ. In 2D and in the case thatSz (the z component
of the electron spin) is conserved, recent theoretical work[14, 127, 175, 176] has shown
that a TR invariant fractional QSH insulator can in principle exist. In this special case of
conservedSz, the QSH insulator is equivalent to two decoupled integer QHsystems with
equal and opposite effective magnetic fields. Both integer QH systems can be driven into
fracitonal QH states by adding electron-electron interactions and appropriately tuning the
effective magnetic fields while keeping them equal and opposite, which yields a fractional
QSH insulator without breaking TR symmetry. On the other hand, this procedure is not
directly applicable to 3D case, where one cannot in general reduce a TR invariant TI to two
decoupled topological states which break TR symmetry in an equal and opposite way4.
In our previous work [180], we introduced a theory of fractional TI in 3D based on a
parton construction [130, 131, 132, 133, 134, 135]. Simply postulating that an electron, un-
der the influence of strong interactions in the underlying lattice Hamiltonian, fractionalizes
into Nc “colors” of partons gives a realization of a fractional TI, as long as each color of
partons forms a topological band insulator. In order to ensure that outside the fractional TI
the partons recombine into electrons we needed to add additional degrees of freedom, anal-
ogous to the statistical gauge fields in the theory of the fractional QH effect [9]. Outside
the TI these gauge fields are confining. As a consequence, the partons, which are charged
4Such a reduction is possible for weak TI [27] which essentially consist of stacked QSH states, inwhich case the two decoupled TR symmetry-breaking topological states would correspond to layered 3DQH phases [177, 178, 179]. We are interested in strong TI [27]which are genuinely 3D. Although in thiscase there is no obvious reduction to decoupled TR symmetry-breaking states, recent work [137] on novel“bulk” 3D fractional QH phases may provide a useful startingpoint.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 96
under the “statistical” gauge fields, can never be observed in isolation but are instead con-
fined into electrons, just like quarks are confined into mesons and baryons in QCD. The
theory of the partons together with the statistical gauge fields should be thought of as a
low-energy effective description of the system. The goal ofthis paper is to show that con-
sistent low-energy theories with the characteristic properties of a fractional TI exist and to
analyze their properties.
There are many choices and questions associated with the additional statistical gauge
fields. Are they Abelian or non-Abelian in nature? By construction they should confine out-
side the fractional TI, but what phase do they realize insidethe fractional TI? In Ref. [180]
we mostly focused on the case of a deconfined gauge field. It is easy to write down decon-
fined models and to demonstrate the theoretical consistencyof fractional TI in principle.
Such models can either be Abelian or non-Abelian. Examples of both types were presented
in our earlier work. One generic feature that all deconfined realizations of fractional TI
share is the presence of additional gapless degrees of freedom. One typically considers
TI phases that are fully gapped in the bulk. In the case of a deconfined realization of a
fractional TI we have to slightly generalize this understanding by demanding that all de-
grees of freedom charged under electromagnetism are gapped, while allowing for gapless
electrically neutral degrees of freedom. In that case, the system is indeed an insulator as
far as electrical transport is concerned. These additionalneutral gapless degrees of free-
dom play a role similar to the low-energy phonons in a band insulator, and do not spoil the
quantization of the electromagnetic response.
An alternative to the deconfined realization of the fractional TI is to put the statistical
gauge fields in a Higgs phase. We will present an explicit model of this kind in this work.
In this case the system is truly gapped with no gapless degrees of freedom. The statistical
gauge group is broken down to a discrete gauge group; the latter is sufficient to ensure that
in physical states the net electron number is integer.
As we will elaborate in more detail in this work, the basic topological feature of the
fractional TI, a fractionalθ angle, is completely robust in that it only depends on the total
number of partons and not on any of the details of the statistical gauge sector — whether
it is Abelian or non-Abelian, deconfined or Higgsed. This universalθ angle characterizes
the quantized magnetoelectric effect in the material together with its physically measurable
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 97
surface properties, such as a surface fractional QH effect with half the conductivity of a
typical Laughlin state per surface as well as the corresponding magnetooptical Kerr and
Faraday effects.
There is, however, a second topological feature of the fractional TI thatdoesdepend
on details of the gauge field sector: the ground state degeneracy. Some basic features of
the ground state degeneracy have already been discussed in Ref. [180], but here we will
elaborate on this. It has recently been proven [181] that fora gapped system,θ can only
be fractional if the ground state on the3-torusT 3 is degenerate. This observation makes
it clear that a confined phase is not an option for the fractional TI, as it would result in a
unique ground state in a completely gapped system. In the case of a deconfined realization
of the fractional TI, we will find that the ground state in the Abelian models is unique,
whereas in the non-Abelian models it is typically degenerate. The reason why this is not
in contradiction with the theorem of Ref. [181] can be tracedto the fact that we have
additional gapless degrees of freedom, therefore violating the assumption of Ref. [181]
that the system is gapped. We will demonstrate explicitly how this allows us to avoid the
arguments of Ref. [181]. In the Higgs models, the ground state is degenerate as the theorem
just mentioned requires, and we determine the ground state degeneracy.
We illustrate our findings with the study of three examples. All three models correspond
toNc = 3 partons. Model A is an Abelian, deconfined realization of a fractional TI based
on aU(1)× U(1) statistical gauge field. Model B is a non-Abelian, deconfinedrealization
based onSU(3). Model C is a Higgsed realization where the gauge group of model B is
broken down toZ3.
The structure of the paper is as follows. In Sec. 7.6, we introduce three general classes
of parton-gauge boson effective theories: the AbelianU(1) deconfined models, the non-
Abelian deconfined models, and the Higgs models. We illustrate the general ideas in each
case by a specific example of three colors of partons (Nc = 3), corresponding to three
classes of fractional TI but all with a fractional axion angle θ = π/3. In Sec. 7.7, we
prove the topological quantization of the axion angleθ in fractional multiples ofπ using
the chiral anomaly. This derivation allows us to derive the quantization of the axion angle
in both Abelian and non-Abelian models, whether gapless or gapped. In Sec. 7.8, we
discuss the issue of ground state degeneracy on spatial3-manifolds of nontrivial topology.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 98
field U(1)em U(1)A U(1)Bψ1 e/3 2g −gψ2 e/3 −g 2gψ3 e/3 −g −g
Table 7.1: Gauge charge assignments of parton fields under electromagneticU(1)em andstatisticalU(1)2 = U(1)A×U(1)B gauge groups in the simplest Abelian model forNc = 3.
This is the direct analog of the topological degeneracy in fractional QH states on Riemann
surfaces [149, 150, 151], and can be taken as an indication that fractional TI states exhibit
topological order in the many-body sense. In Sec. 7.9 we briefly speculate on the nature of
the gapless surface states, and summarize the paper in Sec. 7.10.
In this section, we introduce the three basic classes of possible statistical gauge sectors
for an fractional TI, and describe in more detail a specificNc = 3 model in each class.
7.6.1 Deconfined models
As described earlier, one way to realize a fractional TI is todrive the statistical gauge
fields into a deconfined phase. This can be realized either in an Abelian or a non-Abelian
setting with rather different properties. What is common toany deconfined realization of a
fractional TI is the appearance of extra gapless matter. This extra gapless matter is neutral
from the point of view of theU(1)em Maxwell gauge field5. Therefore, the system is still an
insulator, i.e. all degrees of freedom charged underU(1)em are gapped. These additional
gapless degrees of freedom can be considered as soft “phonons”, since they do not enter
the electrically charged sector.
Abelian models
Model A: In the simplest Abelian model of a fractional TI withNc = 3, the electron frac-
tionalizes into three fermionic partonsψi, i = 1, 2, 3 with a statisticalU(1)2 = U(1)A ×5The emergent gapless “photon” arising in the layered 3D fractional QH states considered in Ref. [137]
is a direct example of this.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 99
U(1)B gauge group in addition to theU(1)em Maxwell gauge group. The charge assign-
ments are given in Table 7.1, wheree is the electromagnetic gauge charge andg is the
U(1)2 “statistical” gauge charge. These gauge groups ensure thatthe only gauge-invariant
operator that carries Maxwell charge is the product of the three parton operators, hence
the gauge-invariant electron operator isψ1ψ2ψ3. Indeed, the generators of the weight lat-
tice (see Sec. 7.7.2) corresponding to the representation of Table 7.1 aree1 = (2g,−g),e2 = (−g, 2g), e3 = (−g,−g), and the equation
∑
i niei = 0 requiring an operator
ψn11 ψ
n22 ψ
n33 to be gauge-invariant has the one-parameter family of solutionsn1 = n2 = n3,
i.e. operators of the form(ψ1ψ2ψ3)n of which onlyn = 1 corresponds to an operator with
electromagnetic chargee. For a TR invariant system, all three partons have a real mass. As
we will review in the next section, in the topologically nontrivial phase all three partons
have a real but negative mass.
This model is the simplest of theU(1)Nc/U(1)diag Abelian models described in
Ref. [180]. The generic model starts out with a statisticalU(1)Nc gauge group andNc
partons. Each parton carries chargee/Nc under the MaxwellU(1)em gauge group. Theith
parton carries chargeg under theithU(1) factor of the statistical gauge group, and is neutral
under the remainingU(1) factors. One then takes the quotient of the statistical gauge group
by its diagonalU(1)diag subgroup whose generator is simply the sum of the generatorsof
the individualU(1)i factors. For the particular case ofNc = 3, the gauge group presented
above represents the two remaining gauge group factorsU(1)A = 2U(1)1−U(1)2−U(1)3andU(1)B = −U(1)1 + 2U(1)2 − U(1)3. These are two linear combinations that are
orthogonal toU(1)diag, in the sense that their generators satisfytr tAtem = tr tBtem = 0,
with tA = diag(2g,−g,−g), tB = diag(−g, 2g,−g), andtem = diag(e/3, e/3, e/3), and
tr is the trace in the representation of Table 7.1. For this choice of generators, one ob-
tains the convenient feature that the statistical parton charges are integer multiples ofg.
These generators are however neither orthogonal to each other (tr tAtB 6= 0) nor properly
normalized (tr tA,BtA,B 6= g2). Therefore, if one really starts out with aU(1)3 gauge the-
ory, the Maxwell gauge kinetic termtrFµνFµν would contain a mixedFA
µνFµνB term. The
generators can however be easily orthonormalized, and upondoing so we find precisely
the generatorsH1 andH2 [Eq. (7.14)] of the maximal diagonal subgroupU(1)2 of SU(3)
(see Sec. 7.7.2). The same construction works for general number of “colors”Nc, yielding
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 100
once again parton charges which can be chosen to only take values2g and−g under the
various Abelian factors, at the price of non-diagonal kinetic terms. As the gauge kinetic
terms do not impact the topological properties this is of no importance. This allows one to
study a simplified, topologically equivalent version of themodel where one takes the above
charge assignments with standard Maxwell terms (no mixing)and forgets about the fact
that theU(1)Nc−1 gauge group originated fromU(1)Nc in the first place. If desired, a set
of orthonormal generators is provided by the Cartan generatorsH1, . . . , HNc−1 of SU(Nc),
i.e. the generators of its maximal diagonal subgroupU(1)Nc−1. In fact, as far as the elec-
trically charged degrees of freedom are concerned, theU(1)Nc/U(1)diag model discussed
above is equivalent to a HiggsSU(Nc) model whereSU(Nc) is spontaneously broken to
its maximal diagonal subgroupU(1)Nc−1.
Upon integrating out the massive partons and setting the electromagnetic gauge poten-
tial to zero, in the fractional TI phase we are left with a pureU(1)Nc−1 dynamical gauge
theory with θ = π mod 2π. For a continuum theory, there is a single phase, the de-
confined Coulomb phase. This phase corresponds to a free fieldinfrared fixed point at
which the renormalized couplingg vanishes. In other words, in the infrared we have free
Dirac fermions (the partons) and photons (the statistical gauge fields). For a theory de-
fined on the lattice, we obtain two phases, the deconfined Coulomb phase and the confined
phase [182, 183, 184]. The fractional TI phase corresponds to the Coulomb phase, which
is equivalent to the continuum theory in the infrared exceptfor a doubling of the number of
fermion species. The topological properties are the same inboth the continuum and lattice
cases. However, the value (0 orπ mod2π) of theθ angle for the (free) partons is dependent
upon the choice of regularization procedure in the continuum theory, while it is fixed by the
lattice Hamiltonian in the lattice theory [16]. In analogy with the results of Cardy and Ra-
binovici [185] and Cardy [186] forZN gauge theory in3+1 dimensions, one may however
wonder whether the statisticalθ-term with nonzeroθ = π could affect the phase diagram of
the statistical gauge sector, since the gauge theory is now dynamical. This is not so because
aU(1) gauge theory can be viewed as theN → ∞ limit of a ZN gauge theory, and in this
limit the “electric” charges ofZN gauge theory disappear from the spectrum [187] and the
θ-term has no effect on the bulk free energy. Therefore, the Abelian models are simplest to
analyze. For example, the free fields do not allow for a nontrivial ground state degeneracy.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 101
Non-Abelian models
One may question whether a deconfined phase is actually realizable for non-Abelian mod-
els. Although pureSU(Nc) Yang-Mills theory is generally believed to be confining at zero
temperature, adding gapless matter can result in a deconfined phase. For example,N = 4
supersymmetric Yang-Mills (SYM) theory with any gauge group, in particular with gauge
groupSU(Nc), is actually conformally invariant and hence in a deconfinedphase [188]. In
fact, this is the generic case even in non-supersymmetric Yang-Mills theories with enough
gapless matter: for matter in the fundamental representation as well as for matter in two-
index tensor representations it is believed thatSU(m), SO(m) andSp(m) gauge theories
all exhibit a “conformal window”, that is, the gauge theory is in a conformally invariant
and deconfined phase as long as the number of flavorsNf is within a certain finite range.
A conjecture for the exact values ofNf that bound the conformal windows has been put
forward for example in Ref. [189]. The upper end of the conformal window is theoretically
well established: forNf larger than this maximal value the gauge theory loses asymptotic
freedom. It is weakly coupled in the infrared but ill-definedin the ultraviolet. Just below
the upper end of the conformal window it can be established, using perturbation theory,
that at least for largeNc the theory indeed flows to a conformal fixed point [190]. The
lower bound of the conformal window is mainly conjectural based on partial resumma-
tions. There has been a lot of recent activity on numerical studies of the conformal window
using lattice gauge theory [191]. While the precise lower bound of the conformal window
is still up to debate, the existence of a conformal window hasbeen firmly established. In
gauge theories withN = 1 supersymmetry the full conformal window has been mapped
out using the power of holomorphy [192]. TheN = 1 supersymmetricSU(Nc) gauge
theory is for example conformal for32Nc < Nf < 3Nc.
As for the Abelian models, the deconfined non-Abelian modelshave additional gapless
degrees of freedom. For example, in the case ofN = 4 SYM theory the extra matter
consists of adjoint fermions and scalars, in addition to thegauge fields. One important
difference compared to the Abelian case is that in the case ofa non-Abelian gauge theory
at a nontrivial fixed point the extra matter is not free, but remains interacting with a fixed
renormalized couplingg∗, whereg∗ is fixed by the requirement that the renormalization
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 102
groupβ-function vanishes. In the special case ofN = 4 SYM theory, g∗ is in fact a
free parameter and theβ-function vanishes identically for all values of the coupling [188].
Not only do the statistical gauge fields remain strongly coupled to each other, the partons
remain strongly coupled to the statistical gauge fields. Below we will argue that the value
of the fractionalθ angle only depends on the number of partons the electron fractionalizes
into, and is completely robust even against these strong interactions that remain in the
low-energy effective theory of the partons. In fact, in caseof N = 4 SYM theory this
has been demonstrated explicitly in the extreme limit of very large coupling, employing a
holographic realization of this particular model [193].
Model B: The canonical example in this class, which we will refer to asan example
in various sections, is a model withNc = 3 partons of electric chargee/3 coupled in the
fundamental representation to aSU(3) gauge field with some additional adjoint matter to
drive it into the deconfined phase. Details of this extra matter are largely irrelevant from
a topological point of view, except for the question of the ground state degeneracy onT 3.
N = 4 SYM theory can serve as the canonical example.
One consequence of the residual interactions in the infrared is that sometimes TR sym-
metry is spontaneously broken in the topologically nontrivial phase, that is atθ = π.
Whether this happens or not depends on many details of the statistical sector [194]. For the
purpose of constructing parton models of fractional TI, theimportant lesson to remember is
that, even though spontaneous TR symmetry-breaking does occur sometimes, non-Abelian
gauge theories with unbroken TR symmetry do exist. In the largeNc limit the situation
is slightly better understood [195]: while pureSU(Nc) Yang-Mills theory is believed to
spontaneously break TR symmetry atθ = π, additional gapless matter can prevent this. In
particular,N = 4 SU(Nc) SYM theory is TR invariant atθ = π in the largeNc limit.
One other important difference between the Abelian and the non-Abelian models con-
cerns spin. In model A, each of the three partons belongs to a different one-dimensional
irreducible representation of the statistical gauge groupU(1)2. The requirement that for a
fermionic state the many-body wave function has to be antisymmetric under interchange
of both color and spin indices does not constrain the symmetrization properties of the spin
quantum numbers, and hence the total spin, because gauge invariance does not require a
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 103
separate, complete antisymmetrization of the color indices in this case. A spin-1/2 elec-
tron is always possible. In model B however, gauge invariance under theSU(3) group
requires that the many-body wave function be completely antisymmetric in color indices.
Correspondingly the spin indices have to be completely symmetrized, and in model B the
electron would have spin3/2. Because spin rotation invariance is already broken in a
topological band insulator by spin-orbit coupling, this does not create a problem, but it is
certainly an aspect of our non-Abelian models to keep in mind.
General deconfined model
The general deconfined model will have some Abelian (free) factors and some non-Abelian
(interacting) factors. A large class of models of this type was introduced in our earlier
work [180]. The statistical gauge group in this general model is
Nf∏
a=1
U(Nac )/U(1)diag,
and the total number of partons is
Nc =
Nf∑
a=1
Nac .
For every flavora = 1, . . . , Nf we haveNac partons transforming in the fundamental rep-
resentation of the non-AbelianSU(Nac ) factor and carryingU(1)a chargeqa = g/Na
c as
well as electromagnetic chargeqema , while being neutral under all the other gauge groups.
The diagonal subgroupU(1)diag, whose generator is the sum of all theU(1)a generators, is
modded out. The only gauge-invariant operator that carriesnonvanishing electromagnetic
charge is the product of all the partons. To be gauge-invariant under each of the individual
non-Abelian factors, one needs to form baryonic operators out of the partons in the fun-
damental representation of that factor. These individual baryons however will carryU(1)a
chargeg. Only the product of all the individual baryons is gauge-invariant, as this is the
only way to get a gauge-invariant operator whose only statisticalU(1) charge is the charge
under the diagonal subgroupU(1)em, which is removed from the statistical gauge group.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 104
The electromagnetic charge of this gauge invariant operator is
Q =
Nf∑
a=1
Nac q
ema , (7.9)
hence theqema have to be chosen in such a way thatQ = e.
Another alternative one might want to consider is to use models based on orthogonal
or symplectic gauge groups. The gauge-invariant operatorsin such theories are the mesons
qaqbδab andqaqbωab, respectively, whereδab is the Kronecker delta andωab the correspond-
ing antisymmetric invariant tensor for the symplectic group. Both theories have baryons
qa1 · · · qaNcǫa1···aNc . However, in the symplectic theorySp(Nc) whereNc is even, these
baryons are not independent operators but are equivalent toa product of mesons, because
the symplectic tensorωab is antisymmetric. By analogy with the[U(M)×Sp(2k)]1 Chern-
Simons theory of theZk parafermion fractional QH states obtained from a parton construc-
tion [136], these theories have the promise of generating more exotic surface states. How-
ever, if we take the partons to have electric chargeq, not only we obtain baryons of electric
chargeNcq which would suggestq = e/Nc as before, but the mesons are also charged as
we can make mesons from two fundamental quarks. In theSU(Nc) case, mesons are made
from a fundamental quark and an antifundamental antiquark.Therefore, this time we are
forced to identify the mesons with the electrons [136]. Furthermore, we have to assign
the partons chargeq/2, and at the same time take the number of partonsNc to be even so
that the baryons carry an integer multiple of the electron charge, which is consistent with
the fact that symplectic groups are only defined for evenNc. To ensure that this mesonic
electron is a fermion, we need the electron to split into two different partons, a fermion
and a boson, both in the fundamental representation of the gauge group. We would also
get gauge-invariant scalars with the same chargee as the electron, but presumably these
can be made gapped. In this case, the axion angleθ has to be an integer multiple ofNcπ/4
from theNc fermions of chargee/2. This structure is somewhat reminiscent of theZ2 spin
liquid model of a fractional TI put forward in Ref. [181].
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 105
field U(1)em Z3
ψ1 e/3 1ψ2 e/3 1ψ3 e/3 1
Table 7.2: Gauge charge assignments of parton fields under electromagneticU(1)em andunbroken statisticalZ3 gauge groups in the simplest Higgs model forNc = 3. The thirdcolumn corresponds to the triality of the representation.
7.6.2 Higgs models
In order to get a completely gapped system that realizes a TI,it is sufficient to add elec-
trically neutral Higgs fields to the deconfined models of Sec.7.6.1, and consider a specific
pattern of spontaneous symmetry breakingG → H with G the original gauge group, such
that the unbroken gauge groupH in the Higgs phase is discrete. One can modify model B
above to realize this possibility:
Model C: A simple Higgs model can be obtained by augmenting the simplest non-
Abelian three-parton model, model B above, by inclusion of two complex scalar Higgs
fields in the adjoint representation ofSU(3).
The center ofSU(3) is Z3 and the adjoint Higgs fields are neutral under this center
symmetry. If one of the scalars acquires a vacuum expectation value, the continuous part
of the gauge group is broken to its maximal diagonal subgroupU(1)2 and one recovers
the fermions and gauge fields of model A together with severalcharged scalars. Giv-
ing generic noncommuting expectation values to both adjoint Higgs scalars completely
breaks the continuous part of the gauge group and generically only the discreteZ3 cen-
ter symmetry is unbroken. Under this discrete gauge group the parton charges are given
in Table 7.6.2. The third column corresponds to the nonzero triality k = 1 mod 3 of
the fundamental representation with charactere2πi/3. In general, for aSU(Nc) → ZNc
Higgs mechanism we want the partons to be in a representationof SU(Nc) with nonzero
Nc-ality k = 1, 2, . . . , Nc − 1 modNc with charactere2πik/Nc 6= 1. This discrete sub-
group of the original continuous statistical gauge group iscompletely sufficient to ensure
that all gauge-invariant operators have charges that are aninteger multiple of the electron
charge. Indeed, the centerZ3 of SU(3) is generated by a single element of the Cartan
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 106
subalgebra,H2 ∝ diag(1, 1,−2) in Eq. (7.14). The statistical weight lattice is generated
by the (unnormalized) weight vectorse1 = 1, e2 = 1, e3 = −2. A statistically neutral
operatorψn11 ψ
n22 ψ
n33 must satisfy
∑
i niei = 0 which leads ton1 + n2 = 2n3, hence the
total Maxwell charge ise(n1 + n2 + n3)/3 = n3e ∈ Ze. ψ1ψ2ψ3 is still the simplest
gauge-invariant operator, but as long as the fermions have internal spin states that can be
antisymmetrized,ψ3i or ψ2
i ψj with i, j = 1, 2, 3 and i 6= j would give rise to additional
gauge-invariant operators. In the relativistic continuumtheory the fermions carry spin1/2
andψ3i vanishes identically. As the fundamental degrees of freedom of the theory are the
partons, the presence of these extra bound states carrying integer electron chargee does not
affect the physics. The important requirement is that Gauss’ law, which enforces overall
Z3 neutrality, ensures that the net charge of the whole sample is an integer multiple of the
electron charge.
For Higgs fields that transform trivially under the centerZNcof SU(Nc), as is the case
here, it is known that the confined and Higgs phases are distinct [196]. In the Higgs phase,
the system is completely gapped even in the charge neutral sector because the unbroken
gauge groupZ3 is discrete. However, in the limit of infinite Higgs stiffness κ → ∞ the
system behaves like aZNcgauge theory which has an additional gapless Coulomb phase
separating the (gapped) deconfined and confined phases for large enoughNc [196, 187]. It
it possible that this phase persists for finite but large enough κ [187]. In this case, there
would be two distinct fractional TI phases with the same value of θ, one gapless and the
other fully gapped. We also note that the presence of aθ = π term in the statisticalZNc
gauge theory gives rise to the presence of oblique confined phases with dyon condensation
in addition to the usual confined phase [185, 186], but does not remove the gapped de-
confined and gapless Coulomb phases corresponding to the fractional TI. Finally, one can
also construct Abelian Higgs models in addition to the non-Abelian Higgs model discussed
here, withU(1)Nc−1 broken to a discrete subgroup.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 107
7.7 Fractional quantized magnetoelectric effect
7.7.1 Chiral anomaly
The calculation of the effective axion angleθ via the chiral or Adler-Bell-Jackiw (ABJ)
anomaly [138, 139] was mentioned in our original work [180] following earlier discus-
sions [16], and has been spelled out in more detail elsewhere[197, 154, 193, 176, 198, 199].
This argument can be used to obtain TI in any even spacetime dimension [154, 176, 199],
but for here let us specialize to the case of the 3D TR invariant TI. The goal is to calculate
the contribution to the effectiveθ angle of aU(1) gauge group that arises from integrating
out a fermion of chargeq. As (at least in model A) our partons are charged under more
than one Abelian group, we want to calculate all terms of the form
Sθ =i∑
a,b θabe2
32π2
∫
Md4x ǫµνλρF
aµνF
bλρ, (7.10)
whereM is the (here, Euclidean) spacetime manifold and the labela runs over all the
Abelian groups in the problem, i.e. the statistical gauge groups as well as the Maxwell
gauge group. For example, in model A we havea, b ∈ em, A, B. The non-Abelian
case works similarly as will be discussed below. Let us denote by qai the charge of the
ith parton under theath gauge group in units of the corresponding gauge coupling (e for
the electromagnetic sector,g for the statistical sector). In the Dirac kinetic term for the
parton, one can write a complex mass term with massM as the complex bilinear operator
ψ(ReM + iγ5 ImM)ψ. The TR operator takesM into its complex conjugateM∗, so the
system is only TR invariant ifM is real. In other even spacetime dimensions it is a different
discrete symmetry that takes the role of enforcing a real mass term [154]. OnceM is real
we see that there is aZ2 choice of mass terms:M can be real and positive or real and
negative. The two can not be smoothly deformed into each other in a TR invariant fashion
without crossingM = 0, that is, without closing the gap.
As all that matters in terms of physics of interfaces is the difference inθ, we can choose
the θ-term to be zero in the case thatM is real and positive, which one identifies as the
topologically trivial case. At the classical level, we can always perform a chiral rotation
ψ → eiαγ5ψ to absorb the phase ofM . This chiral rotation is a symmetry of the massless
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 108
theory. One can think of the mass as a spurion, i.e. as arisingfrom the expectation value of
a non-dynamical background field, to restore the symmetry inthe massive case by letting
M transform asM → e2iαM . If the phase ofM is originally θ0, with θ0 = π for the
topologically nontrivial TR invariant insulator,M can be made real and positive by a chiral
rotation with angleα = −θ0/2. However, this chiral rotation is not a symmetry of the
quantum effective action, as the path integral measure is not invariant. Performing such
a rotation generates aθ-term of the form given in Eq. (7.10), with a coefficient that is
determined by a triangle diagram with one axial current and twoU(1) currents,
θab =∑
i
qai qbi θ0, (7.11)
i.e. it is determined entirely by the gauge group representations to which the integrated
fermions belong. For the non-Abelian case, the mixedθ-terms (mixed between two gauge
groups) vanish identically due to the tracelessness of the representation matrices. The
diagonalθ-terms are given by a similar formula with the charges replaced by the trace over
the generators of the group in the representation of the partons. A single Dirac fermion
in the fundamental representation contributesθ = θ0. This calculation is robust against
inclusion of interactions [200] as recently discussed in Ref. [199].
From Eq. (7.11) it follows immediately thatNc partons of electric chargee/Nc generate
a θ-term for the Maxwell field withθ = θ0/Nc, that isθ = π/Nc if the partons realize a
TR invariant TI. No reference to the statistical gauge groupis necessary. In the Abelian
case it is advantageous to ensure that no mixedθ-terms involving the Maxwell field and
a statistical gauge field are generated. Such mixed terms give rise to extra contributions
to the effectiveθ-term for the Maxwell field (denoted by “em”) once the statistical gauge
fields are integrated out. The latter is a delicate thing to doin the deconfined phase, but
these mixed terms would alter the topological properties ofthe Maxwell field. Vanishing
of the mixed terms requires that for anya 6= em,
∑
i
qai qemi = 0. (7.12)
If all partons have the same electromagnetic chargeqemi = 1/Nc, this reads∑
i qai = 0,
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 109
which is simply the requirement that the electron is gauge-invariant under the statistical
gauge fields. This, by construction, is automatically satisfied in our Abelian models pre-
sented above. Alternatively, Eq. (7.12) follows from the orthogonality of the generators
tr tatem = 0, a = A,B (Sec. 7.6.1). On the other hand, the general non-Abelian models
introduced in Sec. 7.6.1 have, for every gauge group factor labeled bya, a total ofNac
partons with statistical chargeqa = 1/Nac and a Maxwell electric charge that is more or
less unconstrained up to the overall condition Eq. (7.9). Inthis case mixed terms will be
generated. The final value quoted forθ in Ref. [180] is only obtained after integrating out
the statistical gauge fields in this case.
As an explicit example, take the AbelianNc = 3 parton model, model A (Sec. 7.6.1).
Assume that all partons have a topologically nontrivial mass, i.e.θ0 = π. Using the chiral
symmetry to rotate the phase of all three mass terms to a real and positive mass, we generate
via the ABJ anomaly an electromagneticθ-term withθ = Cπ, where
C =∑
i
(qemi )2 =1
3.
No off-diagonalFem ∧ FA or Fem ∧ FB terms are generated, because the corresponding
anomalies∑
i qemi qAi and
∑
i qemi qBi vanish. In the statistical sector, nonzeroθ-terms will
be generated,
θAA =∑
i
(qAi )2π = 6π, θBB =
∑
(qBi )2π = 6π,
θAB = θBA = qAi qBi π = −3π. (7.13)
These extra terms play an important role in the charge quantization considerations below.
We note that the above calculations only depend on the fermion content of the theory, and
therefore can not distinguish between the deconfined (modelA or B) or Higgs (model C)
realizations of this particular fermion content.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 110
7.7.2 Topological quantization of the axion angle
In a topological band insulator, the low-energy description of the electromagnetic response
at energy scales below the gap is in terms of a topological field theory with aSθ ∼ θE ·Bterm [16]. From the point of view of the low-energy theory, the partition function is TR
invariant only ifeiSθ (or e−Sθ in Euclidean signature) is TR invariant which, together with
the Dirac quantization of fluxes appearing inSθ, constraintsθ to take discrete values. In
the fractional TI, after integrating out the massive partons, the low-energy electromagnetic
response still takes the formθE ·B, but with possibly fractional values forθ. The fractional
values ofθ allowed in this case follow from modified flux quantization conditions due to
the presence of the statistical gauge fields.
Quantization in the Abelian models
Consider theθ-term for the Maxwell field,θeffe2
32π2 ǫµνλρFµνFλρ. The quantization condition
on θeff follows directly from the Dirac quantization condition formagnetic charges, and
the argument applies either in Euclidean [140] or Minkowskispacetime [159] (with cer-
tain conditions on the electromagnetic fields in the latter case). We consider the Euclidean
case for simplicity. With periodic boundary conditions, spacetime is topologically equiv-
alent to the4-torusT 4. The minimal value of the spacetime integralSθ of the θ-termiθeffe
2
32π2 ǫµνλρFµνFλρ is obtained when the smallest allowed magnetic monopole is inserted
inside both2-tori T 212 andT 2
34 whereT 4 ∼= T 212 × T 2
34, andT 2µν is the2-torus generated by
directionsµ andν. In Euclidean spacetime, all directions can be taken to be spacelike and
thus all fluxesFµν are magnetic. We obtain
Sθ ≡iθeffe
2
32π2
∫
T 4
d4x ǫµνλρFµνFλρ =iθeffe
2
4π2
∫
T 212
dx1dx2 F12
∫
T 234
dx3dx4 F34,
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 111
where the factor of8 comes from the permutations of theǫ-tensor6. If the fundamental
charge ise, the smallest allowed magnetic monopole has magnetic flux
∫
T 2
F =2π
e≡ B0,
andSθ = iθeff , i.e. θeff is periodic with period2π. In the parton model however, the
fundamental charge is nowe/Nc, and the above argument would yield∫
T 2 F = 2πNc
eand
Sθ = iN2c θeff , which means thatθeff would have period2π
N2c. Also, the requirement that
the minimal allowed magnetic Maxwell flux beNcB0 and no longerB0 seems to be in
stark contrast to the real world, whereB0 fluxes have certainly been realized. Both of
those puzzles get resolved by taking into account that in models with fractional charges
interacting with statistical gauge fields, Maxwell magnetic charges can be accompanied by
“color” magnetic charges, i.e. magnetic charges of the statistical gauge fields [201, 202,
149, 150, 151]. We now explain how this increases the periodicity of θeff from 2πN2
cto 2π
Nc. In
order to do that, we must first review how the Dirac quantization condition is modified in
the presence of multiple Abelian gauge fields.
For multiple Abelian gauge fields, the Dirac quantization condition qeqm ∈ 2πZ with
qe, qm the electric and magnetic fluxes, respectively (in the first example, we hadqe = e
andqm = 2πe
), through a closed2-manifold such asT 2, is replaced by the more generic
condition∑
a qae q
am ∈ 2πZ, where again the superscripta labels the various Abelian gauge
group factors, including the Maxwell gauge group. For any given gauge groupa, qae qam
does not have to be an integer multiple of2π. Therefore a minimalB0 flux is allowed,
even though the productqae qam of the Maxwell magnetic charge producing this flux and
the electric charge of a parton would be2π/Nc, as long as there are also color magnetic
fluxes present. This is possible becauseqe for the parton is nonzero for the statistical
gauge groups. Outside the TI, the statistical fields are confined and their magnetic flux
has no physical consequence. Indeed, confinement of the color electric fluxes corresponds
to condensation of the magnetic fluxes [203], so the latter fluctuate wildly and there is no
energy cost associated with them. Therefore, we recover thefact that in a topologically
6The same value applies for all manifolds which are spin, suchasT 4. For a manifold that is not spin theminimal value of the integral ofF ∧ F is half as large [140]. As our parton model contains fermionsandhence can only be formulated on manifolds that are spin, we will restrict ourselves to that case.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 112
trivial insulator aB0 flux is possible. Inside a fractional TI, the statisticalU(1) gauge field
is deconfined and the color magnetic flux does have an effect.
To determine the periodicity ofθ, we need to consider the contribution of all gauge
fields to the actionSθ. We consider the general topological term Eq. (7.10) withM = T 4,
together with the result of the anomaly calculation Eq. (7.11). Recall thatqai denotes the
charge of theith parton under theath gauge group in units of the gauge coupling, so that
qae = eqai for the parton. Since∫
T 2 Fa = qam, one finds
Sθ =iθ04π2
∑
a,b,i
[
(eqai )(eqib)] [
qamqbm
]
=iθ04π2
∑
i
[
∑
a
(qe)ai (qm)
ai
]2
.
The Dirac quantization condition∑
a qae q
am ∈ 2πZ ensures that this isθ0 times a sum of
integers squared, soθ0 has the standard2π periodicity. The Maxwellθ angle isθ0/Nc,
according to Eq. (7.11), and so has periodicity2π/Nc as announced earlier.
We now demonstrate this explicitly on the example of model A.The Dirac quantization
condition, i.e. the condition that the parton wave functionshould be single-valued, allows a
B0 flux for the Maxwell field together with a color magnetic fluxB0/3 for the (say)U(1)B
magnetic field. To see this, we note that the phaseαi by which the wave function of thei
parton changes when taken around a loop enclosing this flux is
αi = 2π
(
eqBiB0
3+ eqemi B0
)
,
which yieldsα1 = 0, α2 = 2π, α3 = 0 and the wave function is indeed single-valued. A
calculation of theθab angles in model A, similar to Eq. (7.13) but with a generalθ0, yields
θem = θ0/3 and θBB = 6θ0, and we obtain witha, b ∈ em, A, B and in Lorentzian
signature
eiSθ = exp
(
i∑
a,b
θab2π
e2
2π
∫
d4xEa ·Bb
)
= exp
[
i
(
6θ0 ×1
32+θ03× 12
)]
= eiθ0,
as announced earlier.θ0 has periodicity2π, and hence the Maxwellθ angle has periodicity
2π/3.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 113
Quantization in the non-Abelian models
In the non-Abelian case, the discussion of charge quantization closely follows the Abelian
case discussed above. The proper quantization condition for magnetic fluxes in that case
is ea · mb ∈ 2πZδab, whereea is a “electric flux vector”,mb is a “magnetic flux vector”,
and the indicesa, b run over the generators of the gauge group [149, 150, 151, 204, 205].
Mathematically,ea is a vector in the weight lattice of the Lie algebra of the gauge group,
and the quantization condition definesmb as a vector in the dual weight lattice, i.e. the
“reciprocal” weight lattice [206].
Let us consider the example of a deconfined non-Abelian gaugetheory model based on
Nc = 3 partons and anSU(3) gauge group (model B). Including electromagnetism, the
total gauge group isSU(3)× U(1)em which has the following Cartan generators,
H1 =g√2λ3 =
g√2
1 0 0
0 −1 0
0 0 0
, H2 =g√2λ8 =
g√6
1 0 0
0 1 0
0 0 −2
,
H3 =
e3
0 0
0 e3
0
0 0 e3
, (7.14)
whereλ3, λ8 are Gell-Mann matrices, and we have explicitly written theSU(3) andU(1)em
gauge couplingsg and e, respectively. H1 andH2 are the two Cartan generators of
SU(3) [206], normalized totr(HaHb) = g2δab, a, b = 1, 2, andH3 is the generator of
U(1)em with all three quarks having the same electric chargee/3. The weight lattice is
generated by the fundamental weights which are
e1 =
(
g√2,g√6,e
3
)
, e2 =
(
− g√2,g√6,e
3
)
, e3 =
(
0,− 2g√6,e
3
)
. (7.15)
We recall thateia is the eigenvalue of the Cartan generatorHi associated with theath com-
mon eigenvector|ea〉 of all three Cartan generators, i.e.Hi|ea〉 = eia|ea〉 [206]. The first
two entries ofea correspond to non-AbelianSU(3) “color” charges, and the last entry cor-
responds to the usualU(1)em electric charge. The dual weight lattice is generated by its
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 114
own fundamental weights, which are defined as the “reciprocal lattice basis vectors”mb,
ea ·mb = 2πδab, a, b = 1, 2, 3. (7.16)
The linear system Eq. (7.15), (7.16) is easily solved to yield7
m1 = 2π
(
1√2g,
1√6g,1
e
)
, m2 = 2π
(
− 1√2g,
1√6g,1
e
)
, m3 = 2π
(
0,− 2√6g,1
e
)
.
(7.17)
The allowed magnetic monopoles, i.e. the allowed magnetic flux configurations, are given
by linear combinations of the fundamental dual weightsmb with integer coefficients,
m = n1m1 + n2m2 + n3m3, n1, n2, n3 ∈ Z. (7.18)
We are now in a position to discuss the periodicity ofθeff . The “colorless” magnetic
monopole configuration discussed earlier, which led to a periodicity of 2πN2
c= 2π
9, corre-
sponds to the dual weight vector
m = m1 +m2 +m3 =
(
0, 0,6π
e
)
,
i.e. n1 = n2 = n3 = 1 in Eq. (7.18). However, we now see that this is not the “smallest”
magnetic monopole: we can choose a smaller monopole for which some of theni, i =
1, 2, 3 are zero. In particular, the smallest monopoles have only one ni equal to1. But as
seen in Eq. (7.17), these monopoles will necessarily be “colored”, i.e. they will carry some
amount of non-Abelian magnetic charge. Let us now evaluateSθ for a colored magnetic
monopole, saym = m1, and see how it affects the periodicity ofθeff . Since the monopole
is colored, we cannot simply discard theSU(3) θ-term in the effective action [Eq. (6) in
Ref. [180]]. Denoting byF andf theU(1)em andSU(3) field strengths, respectively, we
7Were it not for the thirdU(1)em entry, the weight lattice would be self-dual, i.e.ea = ma
2π . This wouldbe the case forU(3), or more generally forU(Nc) [141, 142]. We are concerned with the Lie algebra ofSU(Nc)×U(1)em, which is different from that ofU(Nc) because we have chosen a different normalizationfor theSU(Nc) generators than for theU(1)em generator. This difference is important, and is the mathemat-ical reason for the appearance ofθ/Nc rather thanθ/N2
c in theθ-term for theU(1)em gauge field.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 115
have
Sθ =iθeffe
2
32π2
∫
T 4
d4x ǫµνλρFµνFλρ +iθg2
32π2
∫
T 4
d4x ǫµνλρfaµνf
aλρ
=iθ
3
e2
4π2
(
2π
e
)2
+iθg2
4π2
[
(
2π√2g
)2
+
(
2π√6g
)2]
=iθ
3+iθ
2+iθ
6= iθ,
i.e. θ has periodicity2π, which means thatθeff = θ/3 has periodicity2π3
, as claimed
in Ref. [180]. It is readily checked that the other “minimal”monopoles,m = m2 and
m = m3, give the same quantization condition.
Quantization in the Higgs models
The analysis of charge quantization in the Higgs models is very similar to the Abelian case.
The basic statement is as before. A MaxwellB0 flux is possible only if it is accompanied
by color magnetic flux of the unbroken gauge groupH, otherwise onlyNcB0 is possible.
While color magnetic flux before was quantized in integer multiples of a basic unit, it only
now takes a finite number of discrete values. For example, in our model C, magnetic flux
for H = Z3 can only take three different values: three units of color magnetic flux are
equivalent to no color magnetic flux.
7.8 Ground state degeneracy
7.8.1 Ground state degeneracy on the 3-torus
It has recently been shown [181] that a fractionalθ angle in a TR invariant, gapped system
necessarily implies multiple ground states onT 3. However, a deconfined gauge theory may
or may not be gapped, and a unique ground state onT 3 is possible if the system is gap-
less. In general, deconfined gauge theories with continuousgauge groups will be gapless
and deconfined gauge theories with discrete gauge groups will be gapped. In particular,
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 116
we will show that the Abelian models in the deconfined phase contain gapless but electro-
magnetically neutral gauge bosons, and have a unique groundstate onT 3. On the other
hand, the non-Abelian models do exhibit a nontrivial groundstate degeneracy onT 3, even
though they avoid the theorem of Ref. [181] due to the extra massless degrees of freedom.
In the Higgs models, the theory obeys all the assumptions of the theorem and there has to
be multiple ground states onT 3. We will see that this is indeed the case. Therefore, various
realizations of a fractional TI with the same value of the Maxwell θ angle can be distin-
guished by their topological ground state degeneracy onT 3. This is similar to the fact that
the Hall conductance, although a topological invariant, isnot sufficient to fully characterize
the topological order in a fractional QH system [149, 150, 151].
Before we discuss our three different models in detail, we must first define precisely
what we mean by ground state degeneracy. As discussed in detail in the previous sections,
in at least two of our three models (the models in the deconfined phase including models
A and B), the theory contains gapless degrees of freedom. In order to arrive at a mean-
ingful definition of topological ground state degeneracy ina gapless system, we need to
study the theory for a finite-size system. Denoting byL the linear system size, the mass-
less degrees of freedom will develop a finite-size gap of order 1/L. In order to focus on
the ground state manifold, we want to study the theory at energies below that finite-size
gap. For a finite-size system, the ground state degeneracy will be generally lifted by non-
perturbative effects [149, 150, 151] corresponding to the tunneling of fractionally charged
partons around noncontractible loops inT 3, which leads to an energy splitting of order
∼ e−mgapL/L wheremgap is the dynamically generated parton mass gap already present at
infinite volume. For finiteL, these states are truly degenerate only in the strictmgap → ∞limit (i.e. for infinitely massive partons), but even at finitemgap we can identify the ground
states by a finite-size scaling analysis of the many-body spectrum.
Abelian models
We first investigate the question of ground state degeneracyonT 3 in the simplest Abelian
three-parton model, model A. When studying the model on aT 3, the ground state is unique.
As the statisticalU(1) fields are free in the infrared, this is simply a question of quantizing
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 117
Maxwell electrodynamics. As is well known,U(1)Nc−1 gauge theory onT 3 in the decon-
fined phase is simply equivalent to a set of2(Nc − 1) decoupled three-dimensional (in the
x, y, z directions onT 3) harmonic oscillators, i.e. a harmonic oscillator for eachof theU(1)
gauge fields with two polarizations. For a finite size system,the ground state is unique and
corresponds to each of these harmonic oscillators being in their ground state. The gap to
the first excited state is of order∼ 1/LwhereL is the linear system size, but vanishes in the
L → ∞ limit. According to Ref. [181], a gapped system can not have afractionalθ angle
if it has a unique ground state onT 3. As mentioned, the way our simple Abelian model
avoids this constraint is because it is not gapped (for infinite size). To see how the extra
free photons can circumvent the no-go theorem of Ref. [181],we observe that the argument
of Ref. [181] is essentially a Lorentzian version of the usual Dirac quantization argument.
Consider a gapped scenario where a2-torusT 2 of the spatialT 3 is pierced by a magnetic
flux Bz. In the presence of a unit fluxBz = B0 = 2πe
, the ground-state to ground-state
(G2G) amplitude when inserting the same minimal flux throughthe noncontractible loop
in thez direction picks up a phase given by
eiSθ = exp
(
iθ
2π
e2
2π
∫
d4xEem ·Bem
)
= eiθ.
The time-reversed process picks up a phasee−iθ, therefore for a TR invariant theory we
requireeiθ = e−iθ andθ has to be an integer multiple ofπ.
This argument relies on the fact that the minimal magnetic flux isB0. In a theory of a
Maxwell gauge field alone with partons of chargee/3 a magnetic fluxB0 is not consistent
with single-valuedness of the parton wave function; rather, the minimal flux allowed is3B0.
In this case, the above G2G amplitude picks up a phase factor of ei9θ, hence apparently any
multiple ofπ/9 would be an allowed TR invariant value forθ. Clearly, this situation can not
correspond to nature as we know it since aB0 flux can exist. At this stage, the discussion
is completely parallel to that of Dirac quantization in Sec.7.7.2. As we have seen, a basic
B0 flux is allowed as long as it is accompanied by a magnetic flux ofthe statistical gauge
fields. The Dirac quantization condition implies that the allowed combinations of fluxes
are exactly the ones for whicheiS evaluates toeinπ for integern. The G2G amplitude is TR
invariant despite the fact that the Maxwellθ angle takes the fractional valueπ/3. In fact,
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 118
Dirac quantization will ensure that the same happens in all the Abelian models proposed in
Ref. [180]. The no-go theorem is avoided even without groundstate degeneracy onT 3 due
to the presence of extra massless gauge fields.
Non-Abelian models
A deconfinedSU(Nc) gauge theory onT 3 hasN3c degenerate ground states [152]. There
are two complementary ways of establishing this result, oneby keeping the fundamental
partons in the spectrum, and the other by sending the parton mass to infinity,mgap → ∞,
and considering the pureSU(Nc) gauge theory. For concreteness we will discuss the case
Nc = 3. In the presence of fundamental quarks, one can construct a topological symmetry
algebra consisting of operatorsUa, a = 1, 2, 3 which insert a2π flux of theU(1)em gauge
field through theath noncontractible loop ofT 3, and operatorsTa which move a parton
around theath noncontractible loop. In the TR invariant case, we have[Ta, Tb] = [Ua, Ub] =
0, a, b = 1, 2, 3, but [152]
TaUb = e−2πi/3δabUbTa, (7.19)
which simply means that partons can pick up a nontrivialU(1)em Aharonov-Bohm phase
because they are fractionally charged. The operatorTa moves a parton around a closed
loop and is clearly a symmetry of the partition function. TheoperatorUa is also a sym-
metry of the partition function because theU(1)em phasee2πi/3 acquired by the partons is
also an element ofZ3 = 1, e2πi/3, e4πi/3, the center ofSU(3), and can thus be gauged
away. The existence of an algebra of operators which commutewith the Hamiltonian
but not among themselves implies the degeneracy of the energy eigenstates. The sym-
metry algebra Eq. (7.19) is somewhat reminiscent of the topological symmetry algebra
in the fractional QH states [149, 150, 151]. A representation of the algebra Eq. (7.19)
can be constructed by first constructing a representation ofthe Abelian subalgebra gen-
erated by theTa, i.e. by taking a set of states which diagonalizes theTa simultaneously,
Ta|η〉 = eiηa |η〉 with η = (η1, η2, η3). Pick a particular noncontractible loopa. By ap-
plyingUa to |η〉 repeatedly and using the commutation relations (7.19), onecan show that
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 119
Figure 7.2:SU(Nc) gauge transformationΩ on a noncontractible loopCx in the spatialmanifold [Eq. (7.21)], periodic up to an elemente2πik/Nc of the centerZNc
of SU(Nc),and under which the global Wilson loopW (Cx) transforms nontrivially asW (Cx) →e−2πik/NcW (Cx).
|η〉 = U3a |η〉, Ua|η〉, U2
a |η〉 is a 3-dimensional representation of the subalgebra gener-
ated byT1, T2, T3 and theUa corresponding to this specific loop. For generalNc, we would
obtain theNc-dimensional representation|η〉 = UNca |η〉, Ua|η〉, U2
a |η〉, . . . , UNc−1a |η〉.
The analysis can then be repeated by starting from this enlarged set of states and applying
theUa corresponding to the remaining noncontractible loops. At the end of this process one
finds that the dimension of the representation of the full symmetry algebra (7.19), which
is the same as the ground state degeneracy, isNb1(M)c whereb1(M) = dimH1(M,R), the
first Betti number [207] of the spatial manifoldM , corresponds to the number of noncon-
tractible loops inM . It is the dimension of the first homology groupH1(M,R) of M with
real coefficients. ForT 3 we haveb1(M) = 3.
Alternatively, this result can be obtained by studying the pureSU(Nc) gauge theory
which is the low-energy effective description at energies much less than the parton mass
gapmgap. In this language, the deconfined phase onT 3 corresponds to the condensation of
spatial Wilson loops, i.e. a spatial version of the condensation of the Polyakov loop [208]
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 120
(temporal Wilson loop). If all fields are in the adjoint (suchas is the case once the partons
have been integrated out),SU(Nc) gauge theory formulated on a spacetime 4-manifoldMwith nontrivial first homology groupH1(M) 6= 0 develops a globalZNc
symmetry [209,
210] originating from the center ofSU(Nc). This global symmetry is generated by large
gauge transformations, where the gauge parameter is not periodic onT 3 but only periodic
up to an element of the center. An order parameter for spontaneous breaking of this center
symmetry is a spatial Wilson loop along a noncontractible loopCa,
W (Ca) = trP exp
(
ig
∮
Ca
a
)
, (7.20)
whereg is the Yang-Mills gauge coupling,a is the statisticalSU(Nc) gauge potential,P
indicates path-ordering along the loop, and the trace is in the fundamental representation.
The fact thatW (Ca) transforms nontrivially under the center ofSU(Nc) is easily seen
by regularizing the theory on a lattice. Denote lattice sites by a triplet of integersn =
(nx, ny, nz), with na = 1, . . . , Na, a = x, y, z, and periodic boundary conditions on the
link variablesUn,µ = Un+Naa,µ, µ = t, x, y, z, in all spatial directionsa. Consider the
following family of ZNc⊂ SU(Nc) local gauge transformationsΩ(mx)
n parameterized by
an integermx,
Ω(mx)n
=
1, nx = 1, . . . , mx, . . . , Nx,
e2πik/Nc , nx = mx +Nx,(7.21)
with k = 1, . . . , Nc. Although the gauge transformation Eq. (7.21) is not periodic Ω(mx)n 6=
Ω(mx)n+Nxx
, it still is a valid gauge transformation because the usual gauge-invariant operators∑
trU are invariant underany local gauge transformation, including multivalued ones
(recall that the plaquette variableU ≡ Un,µUn+µ,νU†n+ν,µU
†n,ν transforms in the adjoint
asUΩ−→ ΩnUΩ
†n
for an arbitrary local gauge transformationΩn). Another way to say
this is that althoughΩ(mx)n is not periodic as aSU(Nc) gauge transformation, it is periodic
as aSU(Nc)/ZNcgauge transformation. Since the gauge fields transform trivially under
ZNcbecause they are in the adjoint, such a gauge transformationpreserves the periodic
boundary conditions on the gauge fields.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 121
We can now show that Eq. (7.20) transforms nontrivially under Ω(mx)n . The global
Wilson loop Eq. (7.20) has a natural lattice regularization,
W (Ca) = tr∏
n∈Ca
Un,a.
Denotingn⊥ ≡ (ny, nz), we see thatW (Cx) transforms underΩ(mx)n as
W (Cx) = trU(1,n⊥),xU(2,n⊥),x · · ·U(Nx,n⊥),x
= trU(mx,n⊥),xU(mx+1,n⊥),x · · ·U(mx+Nx−1,n⊥)
Ω−→ tr(1 · U(mx,n⊥),x · 1)(1 · U(mx+1,n⊥),x · 1) · · · (1 · U(mx+Nx−1,n⊥) · e−2πik/Nc)
= e−2πik/NcW (Cx), (7.22)
using the periodicity of the trace to shift the base point of the loop from(1,n⊥) to (mx,n⊥)
and the fact that the link variablesUn,µ transform asUn,µΩ−→ ΩnUn,µΩ
†n+µ. When the theory
is quantized, expectation values of operators such asW (Ca) are computed by averaging
over all SU(Nc) gauge transformations that are periodicΩn = Ωn+Naa. Note that the
transformation law Eq. (7.22) is independent of the local data mx specifying where the
discontinuity of the gauge functionΩ(mx)n occurs. Therefore, although such local data can
be lost by performing theSU(Nc) gauge averaging, the global data (i.e. the parameter
k) is not averaged out andW (Ca) can develop a nonzero expectation value. This is why
the transformation law Eq. (7.22) is in fact a global symmetry. As a result, spontaneous
breaking of this symmetry〈W (Ca)〉 6= 0 does not violate Elitzur’s theorem [211].
The Wilson loop Eq. (7.20) can be interpreted as the semiclassical process of creating
a heavy parton-antiparton pair atx = 0 and annihilating them again atx = L/2 along the
noncontractible loopCa, or in other words the worldline of a virtual parton taken around
the loop once. Therefore, we recover essentially the same physics as when we explicitly
considered the fundamental partons in the previous approach of establishing the ground
state degeneracy. The Polyakov loop is the finite-temperature version of Eq. (7.20), where
the spatial manifold need not have a nontrivial first homology but the relevant noncon-
tractible loop in that case is the periodic imaginary time directionτ ∈ [0, β], with β = 1/T
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 122
the inverse temperature. In this case, the logarithm of the Polyakov loop is the negative
of the free energy associated with a single parton. In a confined phase, this free energy is
infinite as it takes an infinite amount of energy to separate a single color charge from its
color charge conjugate, and the Polyakov loop has zero expectation value. Center symme-
try is unbroken. On the other hand, in a deconfined phase the Polyakov loop typically has
a nonzero expectation value and center symmetry is spontaneously broken. In our case we
are considering a zero-temperature theory and the spatial Wilson loop Eq. (7.20) character-
izes G2G amplitudes at zero temperature. As eachW (Ca) carries charge underZNc, there
areNc different values thatW (Ca) can take for each noncontractible loopCa. Therefore,
there will beN3c degenerate ground states onT 3 andN b1(M)
c ground states for a general
spatial manifoldM without boundary.
We remark that this ground state degeneracy was not requiredby the theorem of
Ref. [181]. Whether thisN b1(M)c degeneracy actually occurs depends on details of the sys-
tem beyond the number of partons and the gauge group. In particular, if the extra gapless
neutral matter added in order to drive the non-Abelian statistical gauge field into a decon-
fined phase is in thefundamentalrepresentation of the gauge group, there is no global
center symmetry and we expect that the ground state would be unique. Another example
where spontaneous breaking of the globalZNcsymmetry (and hence presumably the cor-
responding ground state degeneracy) does not necessarily occur isN = 4 SYM theory.
As long as one imposes periodic boundary conditions for the adjoint fermions along all
noncontractible loops, the potential for the corresponding Wilson loop is flat due to super-
symmetry. The ground state expectation value of the Wilson loop is one more modulus in
the theory. One can tune its expectation value arbitrarily,hence we can obtain a deconfined
phase withZNceither broken or unbroken.
Higgs models
Because the Higgs models of Sec. 7.6.2 are completely gapped, they must have multiple
ground states onT 3 to be consistent with the theorem of Ref. [181]. As we necessarily
have unbroken discrete gauge groups, this is indeed ensured. Discrete gauge groups give
rise to degenerate ground states that differ by the value of the Wilson loop of the discrete
gauge field around noncontractible loops, i.e. discrete global fluxes. For example, in our
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 123
model C with itsZ3 unbroken gauge group, there are three different discrete fluxes per
noncontractible loop of the spatial manifoldT 3, which label the various ground states for a
total ground state degeneracy of33 = 27. A genericSU(Nc) Higgs model on a boundary-
less spatial manifoldM with all fields in the adjoint, and whereSU(Nc) is spontaneously
broken to its centerZNcwill have a ground state degeneracy ofN
b1(M)c . This is the three-
dimensional version of theN2 ground state degeneracy onT 2 in the deconfined phase of
ZN gauge theory in2 + 1 dimensions [212]. A similar mechanism is also responsible for
the ground state degeneracy in theZ2 spin liquid model put forward in Ref. [181] as an
explicit realization of a fully gapped fractional TI.
7.8.2 Ground state degeneracy on 3-manifolds with boundaries
On a3-manifold with boundaries, such as the cartesian productM = Σ× I of a Riemann
surfaceΣ and an intervalI discussed in our previous work [180], we need to also consider
the effect of the Chern-Simons terms induced on the boundary. To be specific, let us focus
on theM = Σ × I case with boundary∂M = Σ ∪ Σ. The nontrivialθ-term of the
statistical gauge fields in the bulk induces a Chern-Simons term on the boundary due to
the axion domain wall [16] betweenθ 6= 0 inside the fractional TI andθ = 0 outside the
fractional TI. For the case ofθ = π, the corresponding Chern-Simons term has level1/2.
The fact that this term is not gauge-invariant as a purely(2+1)-dimensional theory does not
matter, because the gauge noninvariance of the boundary theory is compensated by the bulk.
This is made clear by writing theθ-term as a manifestly gauge-invarianttr ǫµνλρfµνfλρ
term [213, 214], wheref is the statistical field strength. The statistical gauge field also has
a kinetic (Yang-Mills) term∼ − 1g2tr fµνf
µν . The details of the ground state degeneracy
will depend on the low-energy dynamics of the gauge fields. Wehave already shown that
the gauge fields need to be in a deconfined phase to have a fractional TI. In the following,
we will further assume that the deconfined phase occurs at weak couplingg ≪ 1, so that
we can consider the gauge fields as being essentially free. The situation is less clear in the
case of a deconfined but strongly coupled theory such asN = 4 SYM theory. With this
assumption, spatially-varying gauge field configurationsaµ(x) will cost a nonzero energy
∼ 1/L for a system of linear sizeL. Therefore, in the ground state the gauge field has to be
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 124
independent of the spatial coordinates,aµ(x, t) = aµ(t). In particular, this means that the
gauge fields on both copies ofΣ at each end of the intervalI = [−L/2, L/2] are identified,
aµ(z = −L/2) = aµ(z = L/2), where we denote byz the coordinates alongI and byx, y
the coordinates onΣ. Denoting byωCS = a ∧ da + ig 23a ∧ a ∧ a the Chern-Simons form
and using the fact [207] thattr f ∧ f = dωCS with f = da + iga ∧ a the statistical field
strength, we have
g2
8π2
∫
M×R
θ(z) tr f ∧ f = − g2
8π2
(
∆θz=−L2
∫
Σ×R
ωCS +∆θz=L2
∫
Σ×R
ωCS
)
,
with ∆θz=±L2≡ θ(z = ±L/2 + ǫ) − θ(z = ±L/2 − ǫ), for ǫ a positive infinitesimal,
andR denotes time. Sincez = ±L/2 correspond to the boundary between a fractional TI
and the vacuum which is a trivial insulator, we generically have∆θz=±L2= (2k± + 1)π
wherek+, k− ∈ Z. The resulting theory is that of a single Chern-Simons gaugefield on
Σ × R with integer levelk ≡ k+ + k− + 1. This Chern-Simons theory gives a additional
contribution to the ground state degeneracy when∂M 6= 0.
In the Abelian case, we obtain a sum ofU(1)k Chern-Simons terms on the boundary
(see Sec. A.9). However, the ground state degeneracy is not simply the product of the
ground state degeneracy for eachU(1)k Chern-Simons term, because large gauge transfor-
mations which mix severalU(1) factors give additional constraints on the ground state
Hilbert space [150, 151]. For the AbelianU(1)2 theory (Sec. 7.6.1) corresponding to
Nc = 3, the ground state degeneracy is12(k + 1)(k + 2) for Σ = T 2 [150, 151], which
reproduces the familiar threefold degeneracy of theν = 1/3 fractional QH state on the
torus fork = 1. For theZNcHiggs models, the Maxwell term of theZNc
gauge theory
itself defines a topological field theory and the ground statedegeneracy isN2gc whereg is
the genus ofΣ [131, 132, 133, 134, 135].
In the non-Abelian case, we obtain a non-Abelian Chern-Simons term onΣ [180]. The
ground state degeneracy for aSU(Nc)k Chern-Simons theory onΣ is equal to the number
of conformal blocks of the level-k SU(Nc) Wess-Zumino-Witten (WZW) conformal field
theory [215]. This number has been determined for any gauge groupG, levelk and genusg
of Σ [216, 217], but the answer is particularly easy for the special case of the torus,g = 1.
In this case the ground state degeneracy at levelk is given by1/(Nc − 1)!∏Nc−1
j=1 (k + j).
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 125
For the special case ofk = 1 we are mostly interested in this simply evaluates toNc. For
Nc = 3, the ground state degeneracy is12(k + 1)(k + 2) in complete agreement with the
answer found in Ref. [150, 151]. This is the same as the AbelianU(1)2 theory.
The boundary Chern-Simons theory involves only the gauge field componentsax, ay
with x, y the coordinates onΣ. In addition to the ground state degeneracy associated with
these boundary degrees of freedom, there is also a contribution to the ground state degen-
eracy coming from the bulk, i.e. a “remnant” of the physics considered in Sec. 7.8.1. For
the Abelian models, the bulk is in the Coulomb phase and does not contribute any addi-
tional factor to the ground state degeneracy (Sec. 7.8.1). We therefore only discuss the
non-Abelian and Higgs models. In the non-AbelianSU(Nc) case, there is an additional
contribution due to center symmetry breaking (as in Sec. 7.8.1) associated withaz. If the
ZNccenter symmetry is spontaneously broken due to a nonzero expectation value of the
Wilson lineW (I) ≡ trPeig∫Idz az on the intervalI, this contributes an additional factor of
Nc to the ground state degeneracy. A similar factor ofNc arises in the case of the Higgs
model with a discrete unbrokenZNcgauge group. To obtain the full ground state degen-
eracy, this factor multiplies the boundary ground state degeneracy we obtained from the
Chern-Simons theory onΣ. To demonstrate that the Wilson lineW (I) is a gauge invariant
observable even in the case thatI is an open interval and not a closed loop, let us focus
on the Abelian case to simplify the discussion. The non-Abelian case is similar. Under
a gauge transformationaz → az + g−1∂zλ, the Wilson lineW (I) transforms by a factor
ei∆λ where∆λ ≡ λ(z = L/2) − λ(z = −L/2) is the difference between the two val-
uesλ takes on the two endsz = ±L/2 of the intervalI. The only gauge transformations
that are allowed in this theory are the ones such that∆λ vanishes modulo2π. Therefore,
W (I) is a legitimate gauge invariant observable. The reason why∆λ has to vanish modulo
2π on Σ × I is very similar to the reason why the Wilson loopW (Ca) [Eq. (7.20)] is a
gauge invariant observable for noncontractible loopsCa. Let us briefly review that case.
Even for a noncontractible loopCz, one could imagine performing a gauge transformation
which sets∮
Czdz az = 0, e.g. for a constantaz = a this can be achieved by choosing
λ = −ga(z + L/2). This would imply thatW (Cz) is not a gauge invariant observable,
even for noncontractible loops. However, this gauge transformation is not allowed because
of its action on the wave functions of charged particles. Fora unit charge particle, the wave
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 126
functionψ atz = −L/2 would transform asψ(z = −L/2) → ψ(z = −L/2), but the wave
function atz = L/2 (identified withz = −L/2 for the closed loopCz) would transform as
ψ(z = L/2) → ψ(z = L/2)e−igaL. For generica this gauge transformation changes the
boundary conditions imposed on the matter particles and hence changes the theory. Ifa is
an integer multiple of2π/gL however, the boundary conditions on the charged fields are
unchanged, we stay within the same theory and so this gauge transformation is allowed.
This is the reason why the variable∮
Czdz az is periodic and only its complex exponen-
tial W (Cz) is single valued. Applying the same logic on the interval, one reaches a very
similar conclusion. The gauge transformationλ = −ga(z + L/2) that would set∫
Idz az
to zero changes the boundary conditions imposed on charged matter fields. The boundary
conditions are part of the definition of the theory onΣ× I, even ifI is not a closed loop. A
gauge transformation that changes boundary conditions isnot a redundancy of the theory
and so should not be allowed. The set of gauge transformations which leaves the bound-
ary conditions on the matter fields unchanged impose a periodic identification on∫
Idz az.
HenceW (I) is a gauge invariant observable for the theory onΣ× I, and thus can acquire
a nonzero expectation value.
7.9 Gapless surface states
So far we have considered effective gauge theories for systems with periodic boundary con-
ditions in all spatial directions, i.e. the3-torusT 3, or for systems with boundaries but where
the boundary is gapped since it is described by a Chern-Simons term. In these two cases,
the fermionic partons are gapped everywhere including on the boundary, which allows us
to integrate them out. The Chern-Simons terms break TR symmetry on the boundary and
are absent if TR symmetry is preserved everywhere. In the latter case, we expect that the
fractional TI should support gapless surface states since each color of partons condenses
(at the mean-field level) into a topological band insulator state, which does support gapless
surface states. The question therefore arises: what is the nature of the gapless surface state
of a fractional TI?
Since we do not at present have a microscopic model of fractional TI, i.e. a model of
interacting electrons, it is at present difficult to answer this question. Since as we have seen,
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 127
several different effective gauge theories can give rise tothe same quantized fractionalθ
angle, we expect a variety of gapless surface states with properties largely dependent on
the details of the microscopic model.
From the point of view of the effective gauge theories discussed in this work, the gap-
less surface states consist of a helical liquid of partons interacting with a three-dimensional
gauge field. An effective theory for the(2 + 1)-dimensional surface can be obtained
by integrating out the bulk gauge fluctuations. A similar calculation was performed re-
cently [174] for the surface helical spinon liquid in spin-charge separated topological Mott
insulators [173], using a perturbative approach. In this case, the bulk consists of a decon-
fined U(1) gauge field. Therefore, the results of Ref. [174] should apply qualitatively
for the models of Sec. 7.6.1, i.e. the deconfinedU(1)Nc−1 models. Indeed, since the
U(1)Nc−1 gauge theories are deconfined at weak couplingg ≪ 1 (g is the parton-gauge
boson coupling), we expect that perturbation theory ing should be reliable. Furthermore,
Ref. [174] shows that due to the three-dimensional nature ofthe gauge fluctuations, pertur-
bation theory is better controlled than in two dimensions. However, since the microscopic
degrees of freedom in a fractional TI are electrons which correspond to gauge-invariant
“baryons” in the parton gauge theory, one should only calculate correlation functions of
gauge-invariant operators. Ref. [174] finds that perturbation theory at 1-loop gives only a
logarithmic modification of the tree-level result for the2kF surface spin-spin correlation
function, i.e. 〈S+(r)S−(0)〉 ∼ 1/r2 → 1/(r2 ln kF r). We expect that the2kF surface
current-current correlation function of the Abelian deconfined fractional TI, i.e. its surface
electromagnetic response, should also only exhibit logarithmic modification compared to
the noninteracting helical Fermi liquid. In the fully gapped Higgs models,g is an irrelevant
coupling because the bulk gauge fluctuations are massive. Weexpect that the electromag-
netic response of the gapless surface state should be the same as that of the noninteracting
helical Fermi liquid, up to corrections that are irrelevantat low energies. For the deconfined
non-Abelian models, since most known examples of these (such asN = 4 SYM theory)
occur at nonzero couplingg = g∗, we expect that the gapless surface state will be a strongly
correlated version of the helical Fermi liquid and it is difficult to guess what its properties
will be. We conjecture that such a state is a “helical non-Fermi liquid”, and holographic
realizations of fractional TI [193] may be a useful tool to compute its properties.
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 128
7.10 Conclusion
In this work, we considered a variety of gauge theories in3 + 1 dimensions and discussed
the conditions they must fulfill to be consistent low-energydescriptions of a fractional TI. A
fractional TI was defined phenomenologically in previous work [180, 181] as a TR invariant
state of interacting electrons which exhibits a quantized fractional axion angleθ in its low-
energy electromagnetic response. We considered AbelianU(1) models and non-Abelian
models. In both cases, the confined phase is not an option for afractional TI because
there would be no fractionally charged excitations in the spectrum, and the existence of
fractionally charged states is necessary for a fractionalθ angle to be consistent with TR
symmetry. This leaves us with two options: a deconfined (or Coulomb) phase and a Higgs
phase. The deconfined phase of AbelianU(1) models is a theory of noninteracting, gapless
gauge bosons. The gaplessness of the gauge bosons does not affect the quantization ofθ
because they are electrically neutral. We showed the fractional quantization ofθ explicitly
using the Adler-Bell-Jackiw chiral anomaly, which did not require the assumption that the
gauge bosons should be gapped. Achieving a deconfined phase in non-Abelian models is
more difficult, but non-Abelian models with sufficient electrically neutral gapless matter,
such asN = 4 SYM theory, are known to realize deconfined phases. These arehowever
usually strongly coupled phases. However, the chiral anomaly still holds in the case of
non-Abelian gauge groups, and we could again show explicitly the fractional quantization
of θ. Higgs models in which the AbelianU(1) or non-AbelianSU(N) groups were broken
down to a discreteZN group were shown to lead to fractional TI as well, with a fullygapped
spectrum in this case.
We investigated the ground state degeneracy of these effective gauge theories on spa-
tial 3-manifolds of nontrivial topology. On the three-torusT 3, the deconfined Abelian
U(1) models have a unique ground state. Indeed, this correspondssimply to quantizing
several independent flavors of Maxwell electrodynamics in abox with periodic bound-
ary conditions in all three directions. The deconfined non-Abelian models can have a
nontrivial ground state degeneracy onT 3 due to the fact that the first homology group
H1(T3,Z) = Z× Z× Z is nontrivial, corresponding to the existence of three inequivalent
CHAPTER 7. FRACTIONAL TOPOLOGICAL INSULATORS IN 3D 129
noncontractible loops inT 3. Whether there are multiple ground states or not in the decon-
fined phase of aSU(N) non-Abelian model depends on whether the centerZN of SU(N)
is spontaneously broken in the ground state or not. This is a question of dynamics which
depends on the details of the model. Higgs models with residual ZN gauge group can be
viewed as a subset of the previous case. Indeed,SU(N) gauge theories with all fields in
the adjoint representation develop the centerZN as a global symmetry which can be spon-
taneously broken in the ground state. Our Higgs models consist of adding adjoint Higgs
fields to pureSU(N) gauge theory such thatSU(N) is spontaneously broken to its cen-
ter. Wilson loops around noncontractible loops will still transform nontrivially underZN ,
and their acquiring nonzero expectation values means spontaneous breaking of this global
ZN symmetry and multiple degenerate ground states. We restricted our consideration of
3-manifoldsM with boundary∂M to the caseM = Σ× I with Σ a Riemann surface (say,
in thex, y directions) andI an interval (in thez direction). In this case, a Chern-Simons
term was induced on the boundary∂M = Σ∪Σ, and in the ground state the Chern-Simons
gauge fields on both copies ofΣ were identified. The resulting Chern-Simons theory had
integer level and its contribution to the total ground statedegeneracy (bulk and surface)
could be computed using standard methods.
Finally, we briefly commented on what one should expect for the electromagnetic re-
sponse properties of the gapless surface states of a fractional 3D TI based on the general
characteristics of the effective gauge theories discussedhere. We expect that deconfined
AbelianU(1) models should give at most a logarithmically modified version of the nonin-
teracting helical Fermi liquid, while the fully gapped Higgs models should only give cor-
rections that are irrelevant at low energies. The deconfined, strongly coupled non-Abelian
models should give rise to the most interesting case. We conjecture that the gapless surface
states of non-Abelian fractional TI are “helical non-Fermiliquid” states and suggest that
holographic methods [193] should be a promising way to studytheir properties.
Chapter 8
Conclusion
This dissertation reported some theoretical studies on 2D and 3D time-reversal invariant
topological insulators which were largely motivated by theexperimental realization of the
former in inverted HgTe quantum wells in 2007 and of the latter in Bi-based compounds in
2008.
The first four chapters provided answers to the questions raised in the introductory
Chapter 1. The nonlocal transport measurements reported and theoretically analyzed in
Chapter 2 constituted strong evidence that transport in inverted HgTe quantum wells does
indeed proceed by robust edge channels, as predicted by Bernevig, Hughes, and Zhang. The
Landauer-Buttiker theory of edge transport reported in this work was able to explain the
paradoxical observation of a finite resistance caused by voltage probes, as well as provide
evidence for the helical nature of the edge channels. The explanation of the different role
played by incoherent scattering in the quantum Hall and quantum spin Hall states paved
the way for the study reported in Chapter 3 of an impurity spinon the edge of the quantum
spin Hall insulator. Kondo scattering from the fluctuating impurity spin provided a possible
explanation for the experimentally observed increased suppression of the conductance with
decreasing temperature. In addition, we found that strong Coulomb interactions can lead
to the formation of an unconventional insulating phase withfractionally charged emergent
excitations. The associated quantized fractional Fano factor of 1/2 was found to be a direct
measure of theZ2 invariant of the quantum spin Hall state. Chapter 4 showed that the
cusplike magnetoconductance peak observed in experimentson HgTe quantum wells could
130
CHAPTER 8. CONCLUSION 131
be explained by a disorder effect, provided that the disorder strength was on the order of the
bulk insulating gap. Chapter 5 proposed a novel way of manipulating spins in condensed
matter systems in a purely quantum-mechanical way, by making use of the helical property
of the quantum spin Hall edge states.
The last two chapters focused on 3D topological insulators.Chapter 6 proposed an op-
tical rotation experiment to measure the topological magnetoelectric effect in a universal
way, independent of material properties such as dielectricconstant and magnetic perme-
ability. Although the topological magnetoelectric effecthas not yet been observed at the
time of this writing, we believe that its observation remains the prime goal of experimen-
tal studies on 3D topological insulators. Several experimental groups around the world
are currently involved in this pursuit at the experimental frontier of the field of topologi-
cal insulators. Chapter 7 introduced the concept of fractional 3D topological insulator and
showed that it is consistent with time-reversal invariance. This was an effort to reconcile
two notions of topological order — that of Thouless and that of Wen — in time-reversal in-
variant systems, in the same way that the quantum Hall effectwith its integer and fractional
realizations does so for time-reversal breaking systems. Although we did not write down a
microscopic Hamiltonian which realizes the fractional 3D topological insulator phase, re-
cent theoretical work has discussed 2D lattice models with nearly flat topologically nontriv-
ial bands [218, 219] as well as numerical evidence for fractional 2D topological insulators,
both time-reversal breaking [220, 221, 222] and time-reversal invariant [223, 224]. Fur-
thermore, a recently proposed microscopic model for a 3D topological insulator with flat
bands [225] paves the way for the construction of a microscopic model of a fractional 3D
topological insulator. Such studies lie at the theoreticalfrontier of the field of topological
insulators.
The experimental discovery of the integer quantum Hall effect in semiconductor het-
erostructures led to the theoretical discovery of topologyin condensed matter physics.
Given the elegant simplicity of the theory of this effect, itis surprising to note that fifty years
of solid state physics passed before its discovery. Nevertheless, it tremendously deepened
our understanding of quantum matter, revealing unexpectedconnections between it and
topics as diverse as elementary particle physics, conformal field theory, noncommutative
CHAPTER 8. CONCLUSION 132
geometry, and characteristic classes. It seems as though time-reversal invariant topolog-
ical insulators will add a new chapter to this story. Likewise, given that the partition of
time-reversal invariant band insulators into topologically distinct classes can be illustrated
using a simple4 × 4 Hamiltonian matrix, it is striking that this result escapednotice for
decades, despite the development of increasingly sophisticated band structure calculation
techniques. Although conceptually similar to the quantum Hall effect, topological insula-
tors build bridges of their own between condensed matter physics and topics in high-energy
physics and mathematics as diverse as the theta-vacuum, axions,K-theory, and Clifford al-
gebras. It is our hope that the study of topological insulators will encourage theorists and
experimentalists alike to revisit classic physics with newquestions, so that we might use
the past to imagine the future.
Appendix A
Theoretical methods
A.1 Weak-coupling renormalization group equations
We model the impurity on the edge of the QSH system by a singleS = 1/2 local spin
coupled by exchange interaction to a one-dimensional helical liquid with electron-electron
interactions. We also consider that the impurity leads to a local two-particle backscattering
term. The system is then described by the Hamiltonian
H = H0 +HK +H2.
H0 is the Hamiltonian of the helical liquid in the absence of theimpurity and reads in the
continuum limit
H0 =− ivF
∫
dx(
ψ†R+∂xψR+ − ψ†
L−∂xψL−)
+ vF
∫
dx[
g4(
(ρR+)2 + (ρL−)
2)
+ g2(
ρR+ρL− +H.c.)
]
,
whereψ†λσ(x) with λ = R,L creates, respectively, a right and left moving electron with
helicity σ = ± at positionx, andρλσ =: ψ†λσψλσ : are the corresponding normal-ordered
densities.vF is the Fermi velocity of the edge electrons, andg4 andg2 are the (dimension-
less) forward and dispersion scattering amplitudes [61]. In HgTe QW the wave vector at
which the edge dispersion enters the bulk is usually much smaller thanπ/2 such that the
133
APPENDIX A. THEORETICAL METHODS 134
umklapp term can be ignored.HK is the Kondo Hamiltonian,
HK = J‖a(
S− : ψ†R+(0)ψL−(0) : + S+ : ψ
†L−(0)ψR+(0) :
)
+ JzaSz
(
ρR+(0)− ρL−(0))
,
(A.1)
whereS± = Sx ± iSy andSz are the spin operators for the impurity located atx = 0, and
a is the size of the impurity. Note thatJ‖,z have units[E] and the continuum field operators
ψR,L(x) have units[L]−1/2 so that a factor of length is necessary to have the correct units for
the Hamiltonian. Essentially, if we start from a lattice model for the Kondo problem [226],
H =∑
k
ǫkc†kσckσ +
J
NS ·∑
kk′
c†kασαβck′β ,
then the continuum limit is
H → L
∫
dk
2πǫkc
†kσckσ + L2 J
NS ·∫
dk
2π
∫
dk′
2πc†kασ
αβck′β
≡∫
dk
2πǫkψ
†kσψkσ + JaS ·
∫
dk
2π
∫
dk′
2πψ†kασ
αβψk′β,
wherea = L/N is the lattice constant andψk =√Lck are the continuum field operators.
Henceψk has units[L]1/2, which means that the Fourier transformψ(x) =∫
(dk/2π)eikxψk
has units[L]−1/2.
Finally,H2 is the two-particle backscattering term,
H2 = λ2a2ψ†
R+(0)ψ†R+(ξ)ψL−(0)ψL−(ξ) + H.c.,
where point splitting with a short-distance cutoffξ is required by the Pauli exclusion prin-
ciple, andλ2 is the two-particle backscattering amplitude and has units[E]. Here the short-
distance cutoff is given by the penetration of the edge stateinto the bulkξ, in the same
way that the magnetic lengthℓB provides a short-distance cutoff for chiral edge theories
in the QH regime [62]. The HamiltonianH2 is manifestly nonlocal, and upon expanding
APPENDIX A. THEORETICAL METHODS 135
ψ(ξ) ≃ ψ(0) + ξ∂xψ(0), taking theξ → 0 limit gives the local interaction term
H2 = λ2a2ξ2ψ†
R+(0)∂xψ†R+(0)ψL−(0)∂xψL−(0) + H.c.
The Hamiltonian can be bosonized in the standard way [227, 228] using the chiral
vertex operatorsψ†R+,L−(x) = 1√
2πξ: e∓i2
√πφR+,L−(x) : . A single nonchiral boson field
φ = φR+ + φL− is sufficient to describe the helical liquid, reflecting the fact that it has the
same number of degrees of freedom as a spinless fermion. The Hamiltonian in bosonized
form reads
H = H0 +J‖a
2πξ
(
S− : e−i2
√πφ(0) : + H.c.
)
− Jza√πSzΠ(0)
+λ22π2
(
a
ξ
)2
: cos 4√πφ(0) : ,
whereH0 is the usual Tomonaga-Luttinger Hamiltonian [61]
H0 =v
2
∫
dx
[
KΠ2 +1
K(∂xφ)
2
]
, (A.2)
with K =√
π+g4−g2π+g4+g2
the Luttinger parameter andv = vF
√
(
1 + g4π
)2 −(
g2π
)2the renor-
malized edge state velocity.
We now derive the renormalization group equations (RGE) Eq.(3.4) for the flow of the
coupling constantsJ‖ andJz. We start from the Kondo Hamiltonian (A.1)
HK = J‖(S−ψ†RψL + S+ψ
†LψR) + JzSz(ψ
†RψR − ψ†
LψL). (A.3)
whereψR,L ≡ ψR+,L−(0), and we have rescaled the couplingsJ‖, Jz by 1/a. In the
bosonization treatment, all the fermion bilinears in this Hamiltonian are considered to be
normal ordered as far as their boson representation is concerned. In this section, we will
also neglect the factors of1/√ξ in the bosonization formulae for the chiral vertex operators
(ξ is the short-distance cutoff) since the scaling dimensionsof the operators including their
anomalous dimensions appear explicitly in the operator product expansions (OPE), and it
is those dimensions which control the RG flow of the couplings. In other words, when
APPENDIX A. THEORETICAL METHODS 136
writing the(0 + 1)-dimensional OPE
Oi(τ)Oj(τ′) ∼
∑
k
ckij|τ − τ ′|∆i+∆j−∆k
Ok(τ′),
where∆ is the scaling dimension (and the sum is over the members of the operator algebra),
we actually mean
Oi(τ)Oj(τ′) ∼
(
1
ξ
)∆engi +∆eng
j −∆engk ∑
k
ckij
(
ξ
|τ − τ ′|
)∆i+∆j−∆k
Ok(τ′),
which explicitly respects dimensional analysis, where∆eng is the engineering dimension.
We use boson fields on the infinite line [227] such that[φR(τ), φL(τ′)] = i
4, hence the
fermions properly anticommute and we do not need Klein factors. Since the number of
degrees of freedom of the helical liquid is the same as that ofa spinless fermion, we only
need a single species of bosons. The normal-ordered fermionbilinears are [227]
: ψ†RψL : =
1
2π: e−i2
√πφ : ,
: ψ†LψR : =
1
2π: ei2
√πφ : ,
: ψ†RψR − ψ†
LψL : = : ρR : − : ρL : =: jx : = − Π√π,
whereφ ≡ φR + φL is the nonchiral boson, and we work in imaginary timeτ = it. As
mentioned previously, the Hamiltonian (A.3) becomes in bosonized form
HK(Π, φ) = J‖1
2π
(
S− : e−i2
√πφ : + S+ : e
i2√πφ :
)
− Jz1√πSzΠ,
where we have omitted the free part. The canonical momentum is given in this case by
Π =∂L∂∂tφ
=1
vK
(
∂tφ+2√πJzSz
)
,
whereL is the Lagrangian density andK is the Luttinger parameter. FromH = Π∂tφ−L
APPENDIX A. THEORETICAL METHODS 137
it can be shown that the corresponding Euclidean action is
SK[φ] =
∫ β
0
dτ
[
J‖1
2π
(
S− : e−i2
√πφ : + S+ : e
i2√πφ :
)
− Jzi
vK√πSz∂τφ
]
,
with β the inverse temperature.
A.1.1 Operator algebra
In view of performing the OPE, we first define the members of theoperator algebra to be
O1(τ) =1
2π
(
S− : e−i2
√πφ : + S+ : e
i2√πφ :
)
,
O2(τ) = − i
vK√πSz∂τφ,
such that the total action reads
S[φ] = S0[φ] +∑
i
gi
∫ β
0
dτ Oi(τ),
where
S0[φ] =1
βK
∑
iωn
|ωn||φ(iωn)|2
is the conformally invariant Tomonaga-Luttinger action and g1 ≡ J‖ andg2 ≡ Jz are the
couplings. The coefficients of the OPE
Oi(τ)Oj(τ′) ∼
∑
k
ckij|τ − τ ′|∆i+∆j−∆k
Ok(τ′) (A.4)
enter the RGE [229, 230]
dgkdℓ
= (1−∆k)gk − π∑
ij
ckijgigj +O(g3), (A.5)
where the first term corresponds to tree level and the second term is the one-loop contribu-
tion. We will first calculate the scaling dimensions∆i of the operators, and then perform
APPENDIX A. THEORETICAL METHODS 138
the OPE to calculate the coefficientsckij . While RGE in the weak coupling limit are often
derived using the Wilson momentum-shell RG [228], for field theories in1 + 1 and0 + 1
dimensions conformal invariance makes it more convenient to work in real space and use
the OPE [229, 230].
A.1.2 Scaling dimensions
The scaling dimension∆i of an operatorOi(τ) is defined as [230]
〈Oi(τ)O†i (τ
′)〉 ∼ 1
|τ − τ ′|2∆i,
where for the scaling dimension, the numerical prefactors in the operators are not important.
We have
〈O1(τ)O†1(τ
′)〉 ∼〈S−(τ)S+(τ′)〉〈 : e−i2
√πφ(τ) : : ei2
√πφ(τ ′) : 〉
+ 〈S+(τ)S−(τ′)〉〈 : ei2
√πφ(τ) : : e−i2
√πφ(τ ′) : 〉.
For aSz eigenstate of the impurity spin withSz componentm, we have
〈S±(τ)S∓(τ′)〉 = S(S + 1)−m2 ±m sgn(τ − τ ′).
The spin-spin correlators have no dynamics and contribute nothing to the scaling dimen-
sions. We have
〈 : e±i2√πφ(τ) : : e∓i2
√πφ(τ ′) : 〉 = e4π〈φ(τ)φ(τ
′)〉 =1
|τ − τ ′|2K ,
since the boson correlator is
〈φ(τ)φ(τ ′)〉 = K
2πln
1
|τ − τ ′|
APPENDIX A. THEORETICAL METHODS 139
as we will show now. We have
〈φ(τ)φ(0)〉 = 1
β
∑
iωn
e−iωnτG0(iωn), (A.6)
whereG0(iωn) = K/2|ωn| is obtained from the unperturbed action
S0[φ] =1
βK
∑
iωn
|ωn||φ(iωn)|2
≡ 1
2
1
β
∑
iωn
φ(iωn)∗G−1
0 (iωn)φ(iωn). (A.7)
Note that theωn = 0 term does not appear in the action because it is zero, therefore we
have to omit then = 0 term in the Fourier transform Eq. (A.6). We use the sum
∞∑
n=1
e−αn
n= − ln(1− e−α),
so we get
〈φ(τ)φ(0)〉 = K
4πln
1
2− 2 cos(2πτ/β)=K
4πln
1
4 sin2(πτ/β),
where we can drop the factor of 4 in front of the sinus squared,which is just a constant
(the propagator is a logarithm, so this does not matter for the purposes of the OPE). We
therefore obtain
〈φ(τ)φ(0)〉 = K
4πln
1
sin2(πτ/β).
At zero temperatureβ → ∞, sin2 x→ x2 and we obtain
〈φ(τ)φ(τ ′)〉 = K
4πln
1
(τ − τ ′)2=K
2πln
1
|τ − τ ′| .
Once more, we have dropped a constantπ which we could have absorbed into some short-
time cutoffτc in the numerator of the argument of the logarithm, i.e.ln(τc/|τ |). Again, this
APPENDIX A. THEORETICAL METHODS 140
does not matter for the purposes of the OPE. Thus
〈O1(τ)O†1(τ
′)〉 ∼ 1
|τ − τ ′|2K ,
and we read off∆1 = K.
ForO2, we have
〈O2(τ)O†2(τ
′)〉 ∼ 〈Sz(τ)Sz(τ′)〉〈∂τφ(τ)∂τ ′φ(τ ′)〉.
Again,〈Sz(τ)Sz(τ′)〉 = m2 has no dynamics. We have
〈O2(τ)O†2(τ
′)〉 ∼ 〈∂τφ(τ)∂τ ′φ(τ ′)〉 = ∂τ∂τ ′〈φ(τ)φ(τ ′)〉 ∼1
|τ − τ ′|2 ,
hence∆2 = 1.
A.1.3 Operator product expansions
We now calculate the OPE. In this case, the numerical prefactors are important but we need
only retain the terms which are most singular in the limit|τ − τ ′| → 0. The impurity spin
operators have a nonsingular product and can be taken at equal timesSα(τ)Sβ(τ′) → SαSβ.
The first OPE is
O1(τ)O1(τ′) ∼ 1
4π2
(
S− : e−i2
√πφ(τ) : + S+ : e
i2√πφ(τ) :
)
×(
S− : e−i2√πφ(τ ′) : + S+ : e
i2√πφ(τ ′) :
)
.
First consider the product
: e±i2√πφ(τ) : : e±i2
√πφ(τ ′) : =: e±i2
√π[φ(τ)+φ(τ ′)] : e−4π〈φ(τ)φ(τ ′)〉 ∼ |τ − τ ′|2K ,
APPENDIX A. THEORETICAL METHODS 141
which is nonsingular sinceK > 0, hence we drop this term. Then we have
: e±i2√πφ(τ) : : e∓i2
√πφ(τ ′) : =: e±i2
√π[φ(τ)−φ(τ ′)] : e4π〈φ(τ)φ(τ
′)〉
=1
|τ − τ ′|2K : e±i2√π[φ(τ)−φ(τ ′)] : .
We Taylor expand
: e±i2√π[φ(τ)−φ(τ ′)] : =: e±i2
√π[(τ−τ ′)∂τ ′φ(τ
′)+...] : ,
hence we have
O1(τ)O1(τ′) ∼ 1
4π2
1
|τ − τ ′|2K(
S+S− : 1 + i2√π(τ − τ ′)∂τ ′φ(τ
′) + . . . :
+S−S+ : 1− i2√π(τ − τ ′)∂τ ′φ(τ
′) + . . . :)
.
The termS+, S− = 2(S2x + S2
y) is a constant since the HamiltonianHK = J‖S‖ ·σ‖ + JzSzσz is invariant underO(2) rotations in thexy plane. In the second term, us-
ing [S+, S−] = 2Sz we have
O1(τ)O1(τ′) ∼ const. +
i
π√π
1
|τ − τ ′|2K−1Sz∂τ ′φ(τ
′),
hence the OPE ofO1 with itself is
O1(τ)O1(τ′) ∼ const.− vK
π
1
|τ − τ ′|2K−1O2(τ
′). (A.8)
The OPE ofO2 with itself is
O2(τ)O2(τ′) ∼ − 1
π(vK)2S2z : ∂τφ(τ) : : ∂τ ′φ(τ
′) : .
APPENDIX A. THEORETICAL METHODS 142
By Taylor expanding, we obtain
: ∂τφ(τ) : : ∂τ ′φ(τ′) : =: ∂τ ′φ(τ
′) + (τ − τ ′)∂2τ ′φ(τ′) + . . . : : ∂τ ′φ(τ
′) :
=: (∂φ)2 : + (τ − τ ′) : (∂2φ)∂φ : + . . . ,
where it is clear that there are no singular terms. Thereforewe have
O2(τ)O2(τ′) ∼ const. (A.9)
Finally, we have the OPE ofO2 with O1,
O2(τ)O1(τ′) ∼ − i
2πvK√π
(
SzS−∂τφ(τ) : e−i2
√πφ(τ ′) : + SzS+∂τφ(τ) : e
i2√πφ(τ ′) :
)
.
We first need to prove the following result,
∂τφ(τ) : eiαφ(τ ′) : ∼ −iαK
2π
1
τ − τ ′: eiαφ(τ
′) : . (A.10)
First expand the exponential
∂τφ(τ) : eiαφ(τ ′) : = ∂τφ(τ) : 1 +
∞∑
n=1
(iα)n
n!φ(τ ′)n : .
We note that
: φ(τ ′)n : =: φ(τ ′) : n,
where we neglect infinite constants, since all fields are at the same point. Then consider
applying Wick’s theorem,
: ∂τφ(τ) : : φ(τ′) : n = : ∂τφ(τ)φ(τ
′)φ(τ ′)n−1 : +: ∂τφ(τ)φ(τ′)φ(τ ′)φ(τ ′)n−2 : + . . .
= n〈∂τφ(τ)φ(τ ′)〉 : φ(τ ′)n−1 :
= −nK2π
1
τ − τ ′: φ(τ ′)n−1 : ,
APPENDIX A. THEORETICAL METHODS 143
where we have used
〈∂τφ(τ)φ(τ ′)〉 = ∂τ 〈φ(τ)φ(τ ′)〉 = ∂τ
[
K
4πln
1
(τ − τ ′)2
]
= −K
2π
1
τ − τ ′,
hence we obtain
∂τφ(τ) : eiαφ(τ ′) : ∼ −K
2π
1
τ − τ ′
∞∑
n=1
(iα)n
(n− 1)!: φ(τ ′)n−1 : ,
where we have neglected the first term∂τφ(τ) = ∂τ ′φ(τ′) + (τ − τ ′)∂2τ ′φ(τ
′) + . . . which
is nonsingular. Pulling out a factor ofiα, this is just the exponential again and we obtain
Eq. (A.10). Using this result, and the identitySzS± = ±12S±, we obtain
O2(τ)O1(τ′) ∼ − 1
(2π)2v
1
τ − τ ′
(
S− : e−i2
√πφ(τ ′) : + S+ : e
i2√πφ(τ ′) :
)
,
hence the desired OPE is
O2(τ)O1(τ′) ∼ − 1
2πv
1
τ − τ ′O1(τ
′). (A.11)
One can also show thatO1(τ)O2(τ′) has exactly the same OPE. Note that Eq. (A.8), (A.9)
and (A.11) are consistent with Eq. (A.4) and the scaling dimensions∆1 = K, ∆2 = 1.
From these equations we can extract the coefficients
c211 = −vKπ, c112 = c121 = − 1
2πv.
The RGE (A.5) thus take the form
dJ‖dℓ
= (1−K)J‖ +1
vJ‖Jz,
dJzdℓ
= vKJ2‖ .
Upon rescaling the couplingsJ‖ → K−1/2J‖ andJz → vJz, we obtain the RGE in the
form of Eq. (3.4),dJ‖dℓ
= (1−K)J‖ + JzJ‖,dJzdℓ
= J2‖ . (A.12)
APPENDIX A. THEORETICAL METHODS 144
In this section we worked in units such that the density of statesν = 1, but the factor ofν
in Eq. (3.4) can be restored by a further rescalingJz → νJz andJ‖ → νJ‖.
A.2 Weak-coupling Kubo formula calculation of the edge
conductance
The finite-temperature Euclidean action corresponding to the Tomonaga-Luttinger Hamil-
tonian (A.2) is
S0 =
∫ β
0
dτ
∫
dxv
2K
[
1
v2(∂τφ)
2 + (∂xφ)2
]
, (A.13)
where1∫
dx ≡∫∞−∞ dx. The nonlocal conductivityσxx(x, x′;ω) defined as
jex(x, ω) =
∫ L
0
dx′ σxx(x, x′;ω)Ex(x
′, ω), (A.14)
wherejex = −ejx is the electric charge current, is given in linear response by the Kubo
formula [226],
σxx(x, x′;ω) =
ie2
ω + iδΠR
xx(x, x′;ω),
where the retarded current-current correlation function is
ΠRxx(x, x
′; t, t′) = −iθ(t− t′)〈[jx(x, t), jx(x′, t′)]〉.
We consider a uniform electric fieldEx(x′, ω) = Ex(ω). In 1D it is given byEx = V/L
whereV is the applied bias andL is the length of the region to which we apply the elec-
tric field. To establish the formalism, we first compute the conductance of the unperturbed
helical liquid without the Kondo impurity, and afterwards compute the conductance in the
presence of the impurity. In the absence of the impurity, theproblem is translationally
1Here we use the “field-theoretic” formulation [227] in whichφ(x) is defined on a infinite regionR withthe boundary conditionsφ(x → ±∞) = 0. This does not contradict the fact that the electric field is nonzeroonly in a region of finite sizeL because the correlation functions are calculated at equilibrium, in the absenceof the electric field. Note however that the ensuing commutation relations between the chiral fieldsφR,L and,most notably, the mode expansions, will in general differ from the finite-size formulationx ∈ [0, L] withperiodic boundary conditionsφ(x+ L) = φ(x).
APPENDIX A. THEORETICAL METHODS 145
invariant. If the current-current correlation function istranslationally invariant, we have∫
dx′ ΠRxx(x, x
′;ω) =∫
dxΠRxx(x, ω) andjx(x, ω) = jx(ω) is independent ofx. Further-
more, in 1D the current densityjex = I is the current itself. Hence we obtain
I(ω) = G(ω)V (ω),
where
G(ω) =ie2
(ω + iδ)L
∫ L
0
dxΠRxx(x, ω)
is the ac conductance. We are interested in the dc conductance,
G = limω→0
ie2
(ω + iδ)L
∫ L
0
dxΠRxx(x, ω), (A.15)
where
ΠRxx(x, ω) =
∫ ∞
−∞dt eiωtΠR
xx(x, t),
andΠRxx(x, t) = −iθ(t)〈[jx(x, t), jx(0, 0)]〉. This retarded function is obtained by ana-
lytic continuation of the Euclidean correlatorΠxx(x, τ) = −〈Tτ jx(x, τ)jx(0, 0)〉. At finite
temperature we have
ΠRxx(x, ω) = lim
iωn→ω+iδΠxx(x, iωn),
where
Πxx(x, iωn) =
∫ β
0
dτ eiωnτΠxx(x, τ),
andωn = 2πn/β, n ∈ Z are discrete bosonic Matsubara frequencies. At zero temperature
β → ∞, the frequencyωn → ω becomes a continuous variable and we have
Πxx(x, iω) =
∫ ∞
0
dτ eiωτΠxx(x, τ),
and all following expressions can be converted to zero temperature by settingβ → ∞ and
ωn → ω.
Taking theω → 0 limit is equivalent to takingiωn → 0, and sinceω+iδ → iωn appears
APPENDIX A. THEORETICAL METHODS 146
already in Eq. (A.15), we might as well work directly with theMatsubara frequencies,
G = − limωn→0
ie2
L
1
iωn
∫ L
0
dx
∫ β
0
dτ eiωnτ 〈Tτjx(x, τ)jx(0, 0)〉
= − limωn→0
e2
ωnL
∫ L
0
dx
∫ β
0
dτ eiωnτ 〈Tτjx(x, τ)jx(0, 0)〉.
According to the bosonization prescription in imaginary timeτ = it, the fermion number
current isjx(x, τ) = −i∂τφ(x, τ)/√π, hence
G = limωn→0
e2
πωnL
∫ L
0
dx
∫ β
0
dτ eiωnτ 〈Tτ∂τφ(x, τ)∂τφ(0, 0)〉.
It is easily found from the action (A.13) that theφ propagatorD(x, τ ; x′, τ ′) =
〈Tτφ(x, τ)φ(x′, τ ′)〉 satisfies the differential equation
(
1
vK∂2τ +
v
K∂2x
)
D(x, τ ; x′, τ ′) = −δ(x− x′)δ(τ − τ ′).
By translational invariance the propagator depends onx−x′ andτ−τ ′ only. For an infinite
system (see Eq. (A.13)), we impose the boundary conditionD(x − x′ → ±∞) = 0. In
imaginary time, the boundary condition is periodic with period β such that
D(x− x′, τ − τ ′) =1
β
∑
iωn
e−iωn(τ−τ ′)D(x− x′, iωn),
and we get(
ω2n
vK− v
K∂2x
)
D(x− x′, iωn) = δ(x− x′),
which shows thatD has a jump of magnitude−K/v at x = x′. From these boundary
conditions we obtain
D(x− x′, iωn) =K
2|ωn|e−|ωn||x−x′|/v.
Note the clear nonanalytic dependence on the frequency|ωn|. We can obtain the current-
current correlator〈Tτ∂τφ(x, τ)∂τφ(0, 0)〉 by differentiatingD, where the contact terms
APPENDIX A. THEORETICAL METHODS 147
coming from taking the derivative of theθ-function in theτ -ordered product do not con-
tribute to the conductance and are ignored2. Hence we obtain
〈Tτ∂τφ(x, τ)∂τ ′φ(x′, τ ′)〉 = ∂τ∂τ ′D(x− x′, τ − τ ′).
Denoting
G(x, iωn) =
∫ β
0
dτ eiωnτ 〈Tτ∂τφ(x, τ)∂τφ(0, 0)〉,
then from the double derivative∂τ∂τ ′ we obtainG(x, iωn) = ω2nD(x, iωn). Since we
take the limit ofωn → 0 from above (we consider physical, positive driving frequencies),
|ωn| = ωn and
G = limωn→0
e2ωn
πL
∫ L
0
dxD(x, iωn).
For small frequencies, we have
∫ L
0
dxD(x, iωn) =Kv
2ω2n
(
1− e−|ωn|L/v) ≃ KL
2|ωn|,
Hence theωn andL cancel, and we are left with
G =Ke2
2π=Ke2
h, (A.16)
upon restoring~ = h/2π. Note that this is valid for an infinite helical Luttinger liquid in
the absence of Fermi liquid leads [71, 72]. The presence of Fermi liquid leads contacting a
finite segment of helical Luttinger liquid of lengthL introduces a infrared cutoff frequency
ωL = ~vF/L in the theory. The present calculation is thus only valid forfrequencies above
this cutoff, in which case the helical Luttinger liquid dominates the physics (L is large). For
frequencies below this cutoff, e.g. for the true dc conductance of a real system contacted
to Fermi liquid leads, the transport is dominated by the Fermi liquid leads and there is a
crossover to the noninteracting valueG = e2/h atω . ωL [71, 72, 232].
In the presence of the impurity, the current-current correlation function is not trans-
lationally invariant, but the dc current is still independent of x. Indeed, the continuity
2They actually cancel the diamagnetic part of the response [231] which we have not included explicity.
APPENDIX A. THEORETICAL METHODS 148
equation
−iωρ(x, ω) + ∂xjx(x, ω) = 0,
implies that∂xjx(x, ω = 0) = 0 so thatjx(x, ω = 0) = jx independent ofx. Hence
generally we can calculate the current at an arbitrary point,
jex(ω) =
∫ L
0
dx′ σxx(0, x′;ω)Ex(x
′, ω),
so that the ac conductance is given by
G(ω) =ie2
(ω + iδ)L
∫ L
0
dx′ ΠRxx(0, x
′;ω).
Finally, for impurity problems where translation invariance is broken only at a single point
x = 0, we can choose to apply the electric field only at this point [Eq. (A.22)]. The
conductance involves then only the local current-current correlation function [Eq. (A.23)].
Effective action for x = 0
Since only the local current-current correlation functionatx = 0 is needed for the computa-
tion of the conductance in the presence of the impurity, we will integrate out the Tomonaga
wavesφ away fromx = 0 and obtain an effective action forx = 0. This can be done
exactly since the perturbation due to the impurity is localized atx = 0, and therefore does
not couple to the Tomonaga waves away fromx = 0. We will present two derivations of
this (0 + 1)-dimensional effective action.
Heuristic derivation [233].—Because of periodic boundary conditions in imaginary
timeφ(x, τ + β) = φ(x, τ), we have the expansion
φ(x, τ) =1
β
∑
iωn
e−iωnτφ(x, iωn),
APPENDIX A. THEORETICAL METHODS 149
with ωn = 2πn/β, n ∈ Z. The Tomonaga-Luttinger action (A.13) can then be rewritten as
S0 =1
β
∑
iωn
∫
dxv
2K
(
ω2n
v2φ(x,−iωn)φ(x, iωn) + ∂xφ(x,−iωn)∂xφ(x, iωn)
)
(A.17)
Since the perturbation acts only atx = 0, to simplify the problem we treat the degrees of
freedom forx 6= 0 as classical and consider quantum fluctuations only forx = 0. In other
words, we can say that forx 6= 0, we expand the field asφ(x) = φcl(x) + δφ(x) where
the classical fieldφcl(x) satisfies the classical equation of motion, and integrate out the
fluctuationsδφ(x) which contribute only a constant to the effective action forφ(0) since
the actionS0 is quadratic. Hence we can neglect these fluctuations altogether and treat
φ(x 6= 0) as classical.
The field equation is easily obtained from Eq. (A.17) andδS/δφ = 0,
(
− ω2n
vK+
v
K∂2x
)
φ(x, iωn) = 0, (A.18)
which is obviously the equation for the propagator without the contact term. Under the
boundary conditionsφ(x→ ±∞, iωn) = 0 we obtain
φ(x, iωn) = φ(iωn)e−|ωn||x|/v, (A.19)
which is taken to be our quantum field, withφ(iωn) ≡ φ(0, iωn) the fluctuating variable.
Since this solution is symmetric underx ↔ −x, we can replace∫
dx → 2∫∞0dx in
Eq. (A.17). Integrating by parts∂xφ∂xφ, we obtain
S =1
β
∑
iωn
v
K
∫ ∞
0
dx φ(x,−iωn)
(
ω2n
v2− ∂2x
)
φ(x, iωn)
− 1
β
∑
iωn
v
Kφ(x,−iωn)∂xφ(x, iωn)
∣
∣
∣
∞
0.
The first term vanishes becauseφ(x, iωn) satisfies the field equation (A.18). The last term is
the boundary term coming from the integration by parts. Substituting the solution (A.19),
only φ(iωn) remains (the field vanishes at the upper boundaryx → ∞ because of the
APPENDIX A. THEORETICAL METHODS 150
boundary conditions) and the effective action becomes [71,72]
Seff =1
βK
∑
iωn
|ωn||φ(iωn)|2.
where we have used the fact that sinceφ(τ) ≡ φ(0, τ) is real, thenφ(−iωn) = φ(iωn)∗.
Derivation through auxiliary fields.—The same result can be obtained somewhat more
rigorously by using auxiliary fields [234]. Consider the following representation of the
delta “functional”,
δ[φ(0, τ)− φ0(τ)] =
∫
Dλ(τ) ei∫ β0 dτ λ(τ)[φ(0,τ)−φ0(τ)],
whereλ(τ) is an auxiliary field. We insert a resolution of unity
1 =
∫
Dφ0 δ[φ(0, τ)− φ0(τ)],
in the path integral for the partition function,
Z =
∫
Dφ0DλDφ e−
∫ β0 dτ(
L0(φ,∂φ)+iλ(τ)[φ(0,τ)−φ0(τ)])
,
where
L0(φ, ∂φ) =
∫
dxv
2K
[
1
v2(∂τφ)
2 + (∂xφ)2
]
,
is the Lagrangian. Going to momentum space
D(x, τ) =1
β
∑
iωn
∫
dq
2πe−iωnτeiqxD(q, iωn),
the propagator is
D(q, iωn) =vK
ω2n + v2q2
,
and the action is
S0[φ] =1
2
1
β
∑
iωn
∫
dq
2πφ(−q,−iωn)D(q, iωn)
−1φ(q, iωn).
APPENDIX A. THEORETICAL METHODS 151
We have
∫ β
0
dτ λ(τ)φ(0, τ) =1
β
∑
iωn
∫
dq
2πλ(−iωn)φ(q, iωn),
∫ β
0
dτ λ(τ)φ0(τ) =1
β
∑
iωn
λ(−iωn)φ0(iωn),
therefore
Z =
∫
Dφ0DλDφ
×e− 12
1β
∑iωn
∫dq2π
φ(−q,−iωn)D(q,iωn)−1φ(q,iωn)+i 1β
∑iωn
λ(−iωn)[∫
dq2π
φ(q,iωn)−φ0(iωn)].
The integral being Gaussian, we can integrate outφ,
Z =
∫
Dφ0Dλ e− 1
21β
∑iωn
∫dq2π
λ(−iωn)D(q,iωn)λ(iωn) e−i 1β
∑iωn
λ(−iωn)φ0(iωn),
but∫
dq
2πD(q, iωn) =
∫ ∞
−∞
dq
2π
vK
ω2n + v2q2
=K
2|ωn|,
using contour integration. We can finally integrate out the auxiliary field λ,
Z =
∫
Dφ0 e− 1
βK
∑iωn
|ωn||φ0(iωn)|2 =
∫
Dφ0 e−S0[φ0],
where
S0[φ0] =1
βK
∑
iωn
|ωn||φ0(iωn)|2,
is our effective(0+1)-dimensional action for the impurity site. As will be shown later, this
effective action is nonlocal in the time domain (we drop the subscript onφ0 for simplicity),
1
βK
∑
iωn
|ωn||φ(iωn)|2 =π
2β2K
∫ β
0
dτ
∫ β
0
dτ ′(
φ(τ)− φ(τ ′))2
sin2 π(τ − τ ′)/β, (A.20)
and therefore cannot be written as a simple quantum mechanics problem with a Hermitian
single-particle Hamiltonian. In fact, Eq. (A.20) corresponds to a finite temperature version
APPENDIX A. THEORETICAL METHODS 152
of the Caldeira-Leggett action for quantum dissipative tunnelling, with φ ≡ φ(x = 0)
playing the role of the coordinateq of a quantum mechanical particle,φ(x 6= 0) playing
the role of the bath degrees of freedom, and1/K playing the role of the friction coefficient
η [235].
Potential barrier at x = 0
As a first exercise, we will compute the conductance in the presence of a classical magnetic
impurity, corresponding to the local perturbation
δH =
∫
dxU(x)ψ†(x)ψ(x),
with a magnetic potentialU(x) localized aroundx = 0. We neglect forward scattering (i.e.
theψ†RψR, ψ†
LψL terms) and consider only2kF backscattering (i.e. theψ†RψL,ψ†
LψR terms).
We will then be able to essentially read off the results for the Kondo and two-particle
backscattering terms from the calculation for a magnetic potential. Upon bosonization we
obtain the following contribution to the effective action,
δSeff = u1
∫ β
0
dτ cos 2√πφ(τ),
whereu1 is the2kF component of the Fourier transform ofU(x). Therefore our effective
actionS = S0 + δSeff becomes
S =1
βK
∑
iωn
|ωn||φ(iωn)|2 + u1
∫ β
0
dτ cos 2√πφ(τ), (A.21)
which is known as the boundary sine-Gordon model. This modelappears in various con-
texts, such as quantum diffusion in the presence of a periodic potential [236, 237], resis-
tively shunted Josephson junctions [238, 239], and a potential impurity in an interacting
one-dimensional electron gas [71, 72].
APPENDIX A. THEORETICAL METHODS 153
Conductance from the Kubo formula
In this section we will calculate the equilibrium conductance toO(u21) using the Kubo
formula. Instead of considering a constant electric field ona segment of lengthL, we
consider an abrupt potential which has a step precisely at the impurity site (but the overall
potential difference between left and right “leads” is the same). The electric field is thus
localized on the impurity,
Ex(x′, ω) = V (ω)δ(x′), (A.22)
hence the conductance is given by
G(ω) = σxx(0, 0;ω) =ie2
ω + iδΠR(0, ω), (A.23)
whereΠR ≡ ΠRxx. Since what we callG is actually only the real part of the dc conductance,
we have
G = − limω→0
e2
ωImΠR(0, ω). (A.24)
For 1D systems, analytic continuation in the frequency domain is more difficult than in the
time domain. We will therefore rather calculate the Matsubara correlatorΠ(0, τ) in the
time domain, and obtain the frequency-dependent retarded functionImΠR(0, ω) directly
from the Matsubara function (see Sec. A.2.1). The current-current correlator is
Π(0, τ) =1
π〈Tτ∂φ(τ)∂φ(0)〉,
where we use the short-hand notation∂φ(τ) ≡ ∂τφ(τ). ToO(u21), we have
〈Tτ∂φ(τ)∂φ(0)〉 = 〈∂φ(τ)∂φ(0)〉0 +u212
∫ β
0
dτ1
∫ β
0
dτ2
×〈∂φ(τ)∂φ(0) cos 2√πφ(τ1) cos 2√πφ(τ2)〉0,conn
+O(u41), (A.25)
APPENDIX A. THEORETICAL METHODS 154
where the connected 4-point correlator is
〈∂φ(τ)∂φ(0) cos 2√πφ(τ1) cos 2√πφ(τ2)〉0,conn
= 〈∂φ(τ)∂φ(0) cos 2√πφ(τ1) cos 2√πφ(τ2)〉0
−〈∂φ(τ)∂φ(0)〉0〈cos 2√πφ(τ1) cos 2
√πφ(τ2)〉0,
and we denote the (time-ordered) path integral average withS0[φ] as〈· · · 〉0 for simplicity.
In Sec. A.2.2 we show that the linear term〈∂φ(τ)∂φ(0) cos 2√πφ(τ1)〉0 vanishes due to the
neutrality condition [Eq. (A.35)], and derive the second-order term, i.e. the 4-point function
Eq. (A.36). The first term in Eq. (A.25) gives the unperturbedconductanceKe2/h which
we have already obtained, hence we evaluate only the second term.
Using the notation
P (τ1 − τ2) = e4π〈φ(τ1)φ(τ2)〉0 , f(τ − τ1) = 〈∂φ(τ)φ(τ1)〉,
and noting that we neglect theδ-function terms which arise when the derivative is pulled
out of a time-ordered product (as explained previously), weobtain
Π(0, τ) = −K2u21
4
1
β
∑
iωn
e−iωnτ [2P (0)− P (iωn)− P (−iωn)],
whereP (0) = P (iωn = 0). Theφ propagator was derived previously from Eq. (A.6) and
(A.7). Keeping the short-time cutoffτc explicitly, we have
〈φ(τ)φ(0)〉0 =K
2πln
πτc/β
sin(π|τ |/β) ,
at finite temperature, while at zero temperatureβ → ∞, we have
〈φ(τ)φ(0)〉0 β→∞−→ K
2πlnτc|τ | . (A.26)
APPENDIX A. THEORETICAL METHODS 155
We therefore have
〈Tτ : ei2√πφ(τ1) : : e−i2
√πφ(τ2) : 〉0 = e4π〈φ(τ1)φ(τ2)〉0
=
(
πτc/β
sin(π|τ1 − τ2|/β)
)2K
= P (τ1 − τ2). (A.27)
The functionP (τ) is thus given by
P (τ) =
(
πτc/β
sin π|τ |/β
)2K
= θ(τ)P+(τ) + θ(−τ)P−(τ),
hence we can extract
P±(τ) =
(
πτc/β
sin π(±τ)/β
)2K
,
which will be necessary for the analytic continuation. SinceP (τ) is even and is periodic
with periodβ, it is easy to show thatP (−iωn) = P (iωn) and we thus have
Π(0, iωn) = −K2u21
2[P (0)− P (iωn)],
with the corresponding (imaginary part of the) retarded function
ImΠR(0, ω) = −K2u21
2[ImPR(0)− ImPR(ω)].
Since we do not know whatP (iωn) is, we need to evaluateImPR(ω) using the analytic
continuation formula in the time domain, Eq. (A.30) (see Sec. A.2.1). We have
ImPR(ω) = 12
(
πτcβ
)2K ∫ ∞
−∞dt eiωt
[
1
sin2K(πit/β)− 1
sin2K(π(−it)/β)
]
.
We remark that the term in the square brackets is odd int. Since we are interested in the dc
limit, we expandeiωt ≃ 1+ iωt, and thus the first term in the expansion vanishes. Thus we
APPENDIX A. THEORETICAL METHODS 156
see thatImPR(0) = 0 and
ImΠR(0, ω) = 12K2u21 ImPR(ω). (A.28)
We obtain
ImPR(ω → 0) = 2ω sin πK
(
πτcβ
)2K (β
π
)2 ∫ ∞
0
dx x sinh−2K x,
were we used1
i2K− 1
(−i)2K = e−iπK − eiπK = −2i sin πK.
The integral is given by
∫ ∞
0
dx x sinh−2K x =π
4
Γ(12)Γ(K)
Γ(12+K)
csc πK,
hence
ImPR(ω → 0) =ω
2π
Γ(12)Γ(K)
Γ(12+K)
(πτc)2KT 2(K−1).
From Eq. (A.28) and (A.24), we thus obtain
∆G ≡ G−G0 = −(
Ke2
h
)
12K
Γ(12)Γ(K)
Γ(12+K)
(u1Λ
)2(
T
Λ
)2(K−1)
, (A.29)
whereG0 = K e2
h, we have restored~ andΛ = ~/πτc is a high-energy cutoff.
In the presence of Fermi liquid leads, Eq. (A.29) would rather read
∆G = −(
KLe2
h
)
12KL
Γ(12)Γ(KW )
Γ(12+KW )
(u1Λ
)2(
T
Λ
)2(KW−1)
,
for ~v/L ≪ T < Λ, whereKL is the Luttinger parameter of the leads andKW is that
of the wire [65]. This is because if we treated the boundary conditions correctly for ax-
dependentK(x) parameter, Eq. (A.28) would readImΠR(0, ω) = 12K2
Lu21 ImPR(ω), i.e.
theff prefactor becomes long-ranged in the dc limit and probes thephysics of the leads,
whereas theP correlator corresponds to2kF backscattering and as such depends only on
APPENDIX A. THEORETICAL METHODS 157
short-distance physics and probes only the physics of the wire.
This was for a classical magnetic impurity. For the Kondo problem, the calculation is
very similar, and we only need to perform the substitutions
u21 →2
3π2
(
a
ξ
)2
S(S + 1)J2‖ , KL → 1 (for Fermi liquid leads), KW → K,
where the factorS(S + 1) comes from the evaluation of the impurity spin correlators as
we now discuss. The thermodynamic average of a functionO(Si) of the impurity spin
operatorsSi is given by
〈O(Si)〉 =
S∑
m=−S
〈m|O(Si)|m〉
S∑
m=−S
1
,
where theSz eigenstates|m〉 such thatSz|m〉 = m|m〉 form a complete set of states for the
impurity spin and the impurity spin has no dynamics (HS = 0). It is not hard to show by
using theSU(2) commutation relations[Si, Sj] = iǫijkSk that
〈m|TτS±(τ1)S∓(τ2)|m〉 = S(S + 1)−m2 ±m sgn(τ1 − τ2),
〈m|TτSz(τ1)Sz(τ2)|m〉 = m2.
Using the following sums,
S∑
m=−S
1 = 2S + 1,
S∑
m=−S
m = 0,
S∑
m=−S
m2 =1
3S(S + 1)(2S + 1),
we find
〈S±(τ1)S∓(τ2)〉 =2
3S(S + 1), 〈Sz(τ1)Sz(τ2)〉 =
1
3S(S + 1)(2S + 1),
where we see explicitly that the spin correlators have no dynamics. In this calculation,
we neglect the forward scattering contribution proportional to 〈SzSz〉, as it must vanish by
unitarity (the conductance cannot exceed the unitarity limit G = e2/h per edge).
APPENDIX A. THEORETICAL METHODS 158
The Kubo formula result for Kondo scattering in the weak coupling limit is thus
∆GK
e2/h= −Γ(1
2)Γ(K)
Γ(12+ K)
1
3π2S(S + 1)
(
J‖Λ
)2(a
ξ
)2(T
Λ
)2(K−1) [
1 + 2νJz ln
(
Λ
T
)]
,
whereΛ = ~/πτc = ~v/πξ with vτc = ξ, andK = K(1 − νJz/2K)2 as explained in
Sec. 3.3. However we have
(
J‖Λ
)2(a
ξ
)2
= π4(νJ‖)2,
whereν = a/πv is the (renormalized) density of states. Hence we obtain
∆GK
e2/h= −Γ(1
2)Γ(K)
Γ(12+ K)
π2
3S(S + 1)(νJ‖)
2
(
T
Λ
)2(K−1) [
1 + 2νJz ln
(
Λ
T
)]
.
The calculation for the correction due to two-particle backscattering is very similar to the
case of a magnetic potential; the only difference is the scaling dimension of the cosine
operator.
A.2.1 Analytic continuation in the time domain
The Matsubara Green function for theφ boson is
D(τ) = −〈Tτφ(τ)φ(0)〉 = θ(τ)D+(τ) + θ(−τ)D−(τ),
which definesD±(τ). With the spectral functionA(ω) = −2 ImDR(ω), we have the
Lehmann representation
D(iωn) =
∫
dω′
2π
A(ω′)
iωn − ω′ = −∫
dω′
π
ImDR(ω′)
iωn − ω′ ,
and
D(τ) = −∫
dω′
πImDR(ω′)
1
β
∑
iωn
e−iωnτ
iωn − ω′ .
We need to calculate the Matsubara sum. We need to pick a function which has poles at
APPENDIX A. THEORETICAL METHODS 159
Figure A.1: Integration contour for evaluating Matsubara sum.
the (bosonic) Matsubara frequenciesiωn. For τ > 0, we picknB(−z) which has residues
−1/β and forτ < 0, we picknB(z) which residues1/β. (nB(z) ≡ (eβz − 1)−1 is the
Bose function.) We choose the contour as in Fig. A.1a. The contour encloses the poles
clockwise, so we have
∫
dznB(−z)e−zτ
z − ω′ = −2πi∑
iωn
(
− 1
β
)
e−iωnτ
iωn − ω′ , τ > 0,
∫
dznB(z)e
−zτ
z − ω′ = −2πi∑
iωn
(
1
β
)
e−iωnτ
iωn − ω′ , τ < 0,
so we can write
1
β
∑
iωn
e−iωnτ
iωn − ω′ =
1
2πi
∫
dznB(−z)e−zτ
z − ω′ , τ > 0,
− 1
2πi
∫
dznB(z)e
−zτ
z − ω′ , τ < 0.
APPENDIX A. THEORETICAL METHODS 160
Now we deform the contour as in Fig. A.1b, pushing the vertical lines to the left and to the
right. Our choice of different functionsnB(±z) ensures that the integral on the right and
left infinite arcs vanishes. The only contribution is the small circle which encloses the pole
at z = ω′ counterclockwise. We thus have
1
β
∑
iωn
e−iωnτ
iωn − ω′ =
e−ω′τ
e−βω′ − 1, τ > 0,
− e−ω′τ
eβω′ − 1, τ < 0,
so we obtain
D+(τ) =
∫
dω′
πImDR(ω′)
e−ωτ ′
1− e−βω′ ,
D−(τ) =
∫
dω′
πImDR(ω′)
e−ωτ ′
eβω′ − 1.
UsingnB(−ω) + nB(ω) = −1, we have
D+(τ)−D−(τ) =
∫
dω
πImDR(ω)e−ωτ ,
such that settingτ = it, we have
12[D+(it)−D−(it)] =
∫
dω
2πImDR(ω)e−iωt.
Taking the inverse Fourier transform, we have
ImDR(ω) = 12
∫ ∞
−∞dt[D+(it)−D−(it)]e
iωt. (A.30)
Equation (A.30) allows us to perform analytic continuationfrom the Matsubara Green func-
tions directly in the time domain.
More generally, the time-ordered functionDT (t) defined as
iDT (t) = 〈Ttφ(t)φ(0)〉 =tr e−βKTtφ(t)φ(0)
tr e−βK, (A.31)
APPENDIX A. THEORETICAL METHODS 161
with K ≡ H − µN and
DT (t) = θ(t)D>(t) + θ(−t)D<(t), (A.32)
has a branch cut on the real axis (we consider the complext-plane now), with a disconti-
nuity across the cut given
DT (t + iǫ)−DT (t− iǫ) = D<(t)−D>(t), (A.33)
for t real andǫ > 0 (see Fig. A.2). The sums in the trace in Eq. (A.31) are assumedto be
absolutely convergent if the terms in each sum are of the forme−A with ReA > 0 [240].
Then
iD>(t) =e−βKeitHφ(0)e−itHφ(0)
tr e−βK,
converges absolutely forRe(β−it) > 0 andRe(it) > 0 (there are two sums in the Lehmann
representation because we need to insert two complete sets of states), and
iD<(t) =e−βKe−itHφ(0)eitHφ(0)
tr e−βK,
converges absolutely forRe(β + it) > 0 andRe(−it) > 0. Hence
D>(t) is analytic for− β < Im t < 0,
D<(t) is analytic for0 < Im t < β.
From its definition on the real axis [Eq. (A.32)], we extend the definition ofDT (t) to the
complext-plane as shown in Fig. A.2, withDT (t) = D<(t) in the upper half-plane and
DT (t) = D>(t) in the lower half-plane. So in particular we have
DT (t+ iǫ) = D<(t) andDT (t− iǫ) = D>(t),
so Eq. (A.33) follows. From the definition of the Matsubara Green function, we have
D±(τ) = −tr e−βKe±τKφ(0)e∓τKφ(0)
tr e−βK,
APPENDIX A. THEORETICAL METHODS 162
Figure A.2: Analytic structure of the time-ordered functionDT (t) in the complext-plane.The function has a cut on the real axis and is analytic elsewhere for−β < Im t < β.
from which we immediately see that
iD>(t) = −D+(it) andiD<(t) = −D−(it).
As a result, the retarded and advanced functions can be written as
DR,A(t) = ±θ(±t)[D>(t)−D<(t)]
= ±iθ(∓t)[D+(it)−D−(it)],
with the upper sign (lower sign) corresponding to the retarded (advanced) function. The
spectral function follows,
A(t) = i[DR(t)−DA(t)] = D−(it)−D+(it),
APPENDIX A. THEORETICAL METHODS 163
so we obtain
ImDR(ω) = −12
∫
dt eiωtA(t)
= 12
∫
dt eiωt[D+(it)−D−(it)],
and we recover Eq. (A.30). The imaginary part of the retardedfunction can thus be directly
obtained from the knowledge of the Matsubara in the (imaginary) time domain. This is
useful for one-dimensional systems where it is more efficient to work in the time domain
than in the frequency domain.
A.2.2 Bosonic correlators
In this section we calculate the correlators of the masslessfree boson with vertex operators.
Here〈· · · 〉 ≡ 〈· · · 〉0 means the correlators evaluated in the free theoryS0[φ], Eq. (A.7).
The correlators calculated here are not the connected ones.The disconnected parts have to
be subtracted to get〈· · · 〉conn.
Three-point function
We want to calculate〈φ(τ)φ(τ ′)eiαφ(τ1)〉. We have
〈φ(τ)φ(τ ′)eiαφ(τ1)〉 = 〈φ(τ)φ(τ ′)〉+∞∑
n=1
(iα)n
n!〈φ(τ)φ(τ ′)φ(τ1)n〉. (A.34)
We can now use Wick’s theorem. The(n + 2)-point function vanishes unlessn = 2p + 2
is even. We have
〈φ(τ)φ(τ ′)φ(τ1)2p+2〉 = 〈φ(τ)φ(τ ′)〉(2p+ 1)!!〈φ(τ1)φ(τ1)〉p+1
+(2p+ 2)〈φ(τ)φ(τ1)〉(2p+ 1)〈φ(τ ′)φ(τ1)〉×(2p− 1)!!〈φ(τ1)φ(τ1)〉p
= (2p+ 1)!!〈φ(τ1)φ(τ1)〉p[
〈φ(τ)φ(τ ′)〉〈φ(τ1)φ(τ1)〉+(2p+ 2)〈φ(τ)φ(τ1)〉〈φ(τ ′)φ(τ1)〉
]
.
APPENDIX A. THEORETICAL METHODS 164
Substituting back in the sum in Eq. (A.34) and using
(2p+ 1)!!
(2p+ 2)!=
1
2p+1(p+ 1)!,
the sum gives exponentials back, and we have
〈φ(τ)φ(τ ′)eiαφ(τ1)〉 =[
〈φ(τ)φ(τ ′)〉 − α2〈φ(τ)φ(τ1)〉〈φ(τ ′)φ(τ1)〉]
e−12α2〈φ(τ1)φ(τ1)〉.
However, the bosonic correlator is [241]
〈φ(z, z)φ(0, 0)〉 = − 1
4πln
(
zz + ξ2
R2
)
,
with z = τ + ix andz = τ − ix, whereξ → 0 is a short-distance (ultraviolet) cutoff and
R → ∞ is a large-distance (infrared) cutoff. The exponential factor in the correlator then
becomes
e−2πα2 ln(R2/ξ2) → 0,
in the thermodynamic limitR2/ξ2 → ∞, unlessα = 0, which is just a statement of the
neutrality condition [230]. Hereα = 2√π so the correlator vanishes,
〈φ(τ)φ(τ ′) cos 2√π(τ1)〉 = 0. (A.35)
Four-point function
We now calculate the 4-point function,
〈φ(τ)φ(τ ′)eiαφ(τ1)eiβφ(τ2)〉 = 〈φ(τ)φ(τ ′)eiα[φ(τ1)+βφ(τ2)]〉,
where the equality is valid because in the path integral average, the operators are justc-
numbers (we have a boson theory) and we can move them at will. (Wick’s theorem takes
care of the product of normal-ordered vertex operators.) Wecan again use Wick’s theorem
as for the 3-point function withαφ(τ1) + βφ(τ2) as our “third” field (since it is just a linear
combination of fields, Wick’s theorem applies also to this composite object). In this case
APPENDIX A. THEORETICAL METHODS 165
we get
〈φ(τ)φ(τ ′)eiαφ(τ1)eiβφ(τ2)〉 =[
〈φ(τ)φ(τ ′)〉
−〈φ(τ)[αφ(τ1) + βφ(τ2)]〉〈φ(τ ′)[αφ(τ1) + βφ(τ2)]〉]
×e−12〈[αφ(τ1)+βφ(τ2)][αφ(τ1)+βφ(τ2)]〉,
and the exponential factor becomes
e−12〈[αφ(τ1)+βφ(τ2)][αφ(τ1)+βφ(τ2)]〉 =
(
ξ2
zz + ξ2
)−αβ/4π
e−18π
(α+β)2 ln(R2/ξ2).
ForR2/ξ2 → ∞, the correlator survives only ifα + β = 0 (neutrality condition). Hence
we takeβ = −α and get
〈φφ′eiα[φ1−φ2]〉 = eα2〈φ1φ2〉
[
〈φφ′〉 − α2(
〈φφ1〉 − 〈φφ2〉)(
〈φ′φ1〉 − 〈φ′φ2〉)
]
, (A.36)
in obvious notation. The〈φφ′〉 disappears in the connected correlator.
A.3 Coulomb gas description of the strong coupling regime
In this section, we derive the low-energy effective theory around the strong coupling fixed
point corresponding to the insulating phaseK < Kc = 1/4. Using a semiclassical instan-
ton calculation, we show that the effective theory is equivalent to a classical 1D Coulomb
gas. The coordinate of a particle in the Coulomb gas corresponds to the (imaginary) time
at which an instanton pumps chargee/2 across the cut between the two semi-infinite he-
lical liquids. We also derive the RG equation for the half-charge tunneling amplitudet in
the strong coupling limit, as well as the correct scaling exponent for the low temperature
behavior of the conductance.
APPENDIX A. THEORETICAL METHODS 166
A.3.1 Euclidean action
Let us recall the action we are working with. The free part is
S0[φ] =1
βK
∑
iωn
|ωn||φ(iωn)|2.
Sinceφ(τ) is a real scalar field, we haveφ∗(ωn) = φ(−ωn) so we get
S0[φ] =2
βK
∞∑
n=1
ωn|φ(iωn)|2,
since we have bosonic frequenciesωn = 2nπ/β. By Fourier transformation we have
ωnφ(iωn) =∫ β
0dτ eiωnτ i∂τφ(τ), hence we can write
S0[φ] =2
βK
∞∑
n=1
∫ β
0
dτ
∫ β
0
dτ ′ e−iωn(τ−τ ′)φ(τ)i∂τ ′φ(τ′).
From the geometric sum
∞∑
n=1
e−iωn(τ−τ ′) =e−iπ(τ−τ ′)/β
2i sin π(τ − τ ′)/β,
we can write
S0[φ] =1
βK
∫ β
0
dτ
∫ β
0
dτ ′ φ(τ)e−iπ(τ−τ ′)/β
sin π(τ − τ ′)/β∂τ ′φ(τ
′).
However the action should be realS0 = ReS0, hence we obtain
S0[φ] =1
βK
∫ β
0
dτ
∫ β
0
dτ ′ φ(τ)[cot π(τ − τ ′)/β]∂τ ′φ(τ′).
We now integrate by parts onτ ′. Sincecot πτ/β is periodic inτ with periodβ and the bo-
son field obeys periodic boundary conditionsφ(τ+β) = φ(τ), the boundary term vanishes.
APPENDIX A. THEORETICAL METHODS 167
Usingcot′ x = −1/ sin2 x, we obtain
S0[φ] = − π
β2K
∫ β
0
dτ
∫ β
0
dτ ′φ(τ)φ(τ ′)
sin2 π(τ − τ ′)/β.
However, we have
∫ β
0
dτ
∫ β
0
dτ ′φ2(τ)
sin2 π(τ − τ ′)/β= −
∫ β
0
dτ φ2(τ)
∫ τ−β
τ
ds
sin2 πs/β
=β
π
∫ β
0
dτ φ2(τ)
[
cotπ(τ − β)
β− cot
πτ
β
]
= 0,
because of the periodicity of the cotangent. By symmetry thesame is true forφ2(τ ′), hence
we can write
S0[φ] =π
2β2K
∫ β
0
dτ
∫ β
0
dτ ′(φ(τ)− φ(τ ′))2
sin2 π(τ − τ ′)/β. (A.37)
This was for the free part. Now, we also have the cosine potential, i.e. the two-particle
backscattering term,
S2[φ] = λ2
∫ β
0
dτ cos 4√πφ(τ).
To the action we add an ultraviolet regulator [61]
S ′0[φ] ≡
M
2
∫ β
0
dτ(∂τφ)2,
with a large massM , which will suppress the high-energy/short-time contribution to the
path integral with a factore−M2
∫dτ(∂φ)2 ∼ e−M
∑ω ω2φ2
ω/2. Indeed, the free action has the
same short-time behavior,
∫ β
0
dτ
∫ β
0
dτ ′(φ(τ)− φ(τ ′))2
sin2 π(τ − τ ′)/β
|τ−τ ′|≪β−→∫ β
0
dτ
∫ β
0
dτ ′(∂τ ′φ)
2(τ − τ ′)2(
πβ
)2
(τ − τ ′)2∝∫ β
0
dτ(∂τφ)2,
APPENDIX A. THEORETICAL METHODS 168
hence the full action reads
S[φ] =π
2β2K
∫ β
0
dτ
∫ β
0
dτ ′(φ(τ)− φ(τ ′))2
sin2 π(τ − τ ′)/β
+M
2
∫ β
0
dτ(∂τφ)2 + λ2
∫ β
0
dτ cos 4√πφ(τ).
A.3.2 Semiclassical instanton calculation
In order to perform a saddle-point approximation to the pathintegral for largeλ2, we first
rescale timeτ → τ =√λ2τ , φ(τ) → φ(τ), sin π(τ − τ ′)/β → sin π(τ − τ ′)/β
√λ2. We
obtain
S0[φ] → S0[φ] = (√
λ2)−2
∫ β√λ2
0
dτ
∫ β√λ2
0
dτ ′(
φ(τ )− φ(τ ′))2
sin2 π(τ − τ ′)/β√λ2.
Since we are concerned with the low-temperature regime, we will always consider thatβ
is very large. Therefore the sine function is in the linear region,
sin π(τ − τ ′)/β√
λ2 ≃1√λ2
sin π(τ − τ ′)/β,
hence we obtain
S0[φ] ≃∫ β
√λ2
0
dτ
∫ β√λ2
0
dτ ′(
φ(τ)− φ(τ ′))2
sin2 π(τ − τ ′)/β.
This means that the free action is unchanged (forβ → ∞, β√λ2 = β = ∞ anyway). We
will omit the integration bounds with the understanding that we always consider largeβ (in
the vicinity of zero temperature). However, the regulator and the cosine term are affected
by the rescaling,
S2[φ] → S2[φ] =√
λ2
∫
dτ cos 4√πφ(τ),
and
S ′0[φ] → S ′
0[φ]M
2√λ2
∫
dτ
(
∂φ(τ )
∂(τ /√λ2)
)2
=√
λ2M
2
∫
dτ(∂τ φ)2,
APPENDIX A. THEORETICAL METHODS 169
hence the rescaled action reads (we now writeφ asφ for simplicity)
S[φ] = S0[φ] +√
λ2
(∫
dτM
2(∂τφ)
2 +
∫
dτ cos 4√πφ
)
.
Sinceλ2 is very large, it dominates over the kinetic partS0, hence we neglectS0 and
perform a saddle-point approximation to the path integral∫
Dφ e−√λ2F [φ] by minimizing
the functionalF [φ] =∫
dτ L(φ, ∂τφ). We have
δ
δφ
∫
dτ L(φ, ∂τφ) = 0 ⇒ ∂τ
(
∂L
∂∂τφ
)
=∂L
∂φ,
and we obtain the Euler-Lagrange equation
Md2φ
dτ 2=dV (φ)
dφ, (A.38)
which is obviously the sine-Gordon equation, whereV (φ) = cos 4√πφ is the “inverted”
potential. We wish to consider an instanton solution corresponding to a soliton solution
of the sine-Gordon equation. We want to tunnel between nearby minima of the cosine
potential, say betweenφ = ±√π/4. We thus search for a “kink” solution withφ(τ =
±∞) = ±√π/4. The anti-kink corresponds toφ(τ = ±∞) = ∓√
π/4.
We multiply the sine-Gordon equation (A.38) by∂τφ and integrate from−∞ to τ . We
obtainM
2
(
dφ
dτ
)2
= V (φ)− V (−√π/4),
assuming thatdφ(−∞)/dτ = 0 which is required if the kink is to have a finite action. The
cosine potential givesdφ
dτ= ±
√
2
M
√
1 + cos 4√πφ.
We pick the positive slope which corresponds to the kink (theanti-kink has negative slope).
This equation can be integrated. We fix the origin of the kink (whereφ = 0) at τ = τ0, and
APPENDIX A. THEORETICAL METHODS 170
integrate fromτ0 to τ , meaning fromφ = 0 to φ = φ(τ):
∫ φ(τ)
0
dφ√
1 + cos 4√πφ
=
√
2
M
∫ τ
τ0
dτ =
√
2
M(τ − τ0).
Performing the integral, we obtain the kink solution,
φ(τ) =1√πarctan
(
tanh 2
√
π
M(τ − τ0)
)
.
One can check that it satisfies the sine-Gordon equation, theboundary condition on the
slopedφ(±∞)/dτ ∝ sech(x = ±∞) = 0 and on the endpointsφ(±∞) = ±√π/4. The
latter is easy to check sincetanh(±∞) = ±1 andarctan(±1) = ±π/4.
The action for this instanton can be calculated by substituting back the solution into the
action functional,
Sinst ≡ S[φ(τ)] =√
λ2
∫ ∞
−∞dτ
[
M
2(∂τ φ)
2 + cos 4√πφ
]
=√
λ2
∫ ∞
−∞dτ M(∂τ φ)
2 + infinite negative const.,
where we have extended the range of integration from0 < τ < β to −∞ < τ < ∞ since
the length of the intervalβ√λ2 is assumed to be much larger than the width of the instanton
∼√M . The infinite constant comes from the finite negative value ofthe potential minima
V (±√π/4) = −1 (it shows up from usingcos 4
√πφ = M
2(∂τφ)
2−1) but can be gotten rid
of by shifting the potential up by a constant, so it is ignored. The integral is easily evaluated
and gives
Sinst = 2
√
λ2M
π, (A.39)
which is manifestly finite. By symmetry the anti-kink has thesame action. The tunneling
ratet between nearby minima of the cosine is thus given by
t = e−Sinst,
APPENDIX A. THEORETICAL METHODS 171
which is the fugacity of the instanton gas. This semiclassical tunneling amplitude is ac-
tually renormalized by small fluctuation effects containedin the determinant arising from
integrating out the Gaussian fluctuationsδφ(τ) about the classical solutionφ(τ) = φ(τ) +
δφ(τ), but we will not consider these effects here (we are interested only in the leading or-
der temperature dependence of the tunneling amplitude which is a tree-level property while
these effects are one-loop effects).
A.3.3 Dilute instanton gas approximation
Sinceλ2 is large, we consider a rarefied instanton gas (few instantonevents sincet ≪ 1).
Direct tunneling across two valleys (e.g. from−√π/4 to 3
√π/4) is possible, but the
action for this event is much larger so the corresponding tunnel amplitude is exponentially
suppressed with respect tot. The leading order effect comes from several “sequential”
tunneling events.
We thus consider a multi-instanton configuration, i.e. an instanton gas configuration
where the boson fieldφ(τ) is of the form
φ(τ) =∑
i
φqi(τ − τi), (A.40)
where the “charge” indexqi = ±1 indicates a kink (qi = +1) or anti-kink (qi = −1). There
is aneutrality requirement∑
i
qi = 0,
arising from the periodic boundary condition to be satisfiedby the boson fieldφ(0) = φ(β)
in the partition function (the number of anti-kinks has to equal the number of kinks if the
field is to return to its initial value). Hencei runs from 1 to2p wherep > 0 is the number
of kink/anti-kink pairs.
Before we substitute the field configuration (A.40) into the action, we make the further
approximation ofneglecting the width of the kink, i.e. we take the limit√
M/4π → 0. The
derivative of the field isdφ
dτ=∑
i
dφqi
dτ.
APPENDIX A. THEORETICAL METHODS 172
The slopedφ/dτ of the kink is strongly peaked at its originτ = τ0. In the limit√M → 0,
it becomes a delta function
lim√M→0
dφ
dτ= aδ(τ − τ0),
where the value of the constanta is easily found by integration,
a =
∫ ∞
−∞dτ aδ(τ − τ0) =
∫ ∞
−∞dτ
dφ
dτ= φ(∞)− φ(−∞) =
√π
2,
from the boundary conditionφ(±∞) = ±√π/4. Hence the derivative of the field is
dφ
dτ=
√π
2
∑
i
qiδ(τ − τi), (A.41)
since the anti-kink (qi = −1) is simply−φ. So we see thatdφ/dτ is none but the charge
density of the instanton gas (which makes sense since in bosonizationjx ∝ ∂τφ is the
current density which corresponds to the density of tunneling events).
From the knowledge of the derivative∂τφ we can compute the action of the instanton
gas. We first know that
S ′0[φ] + S2[φ] =
∑
i
Sinst = 2pSinst,
corresponding to the action ofp noninteracting kink/anti-kink pairs. The nonlocal kinetic
(free) partS0[φ] gives rise to a long-range interaction between the instantons (i.e.|τ−τ ′| ≫width of the instanton∼
√M ). We substitute Eq. (A.41) in Eq. (A.37) and consider the
APPENDIX A. THEORETICAL METHODS 173
low-temperatureβ → ∞ limit. We obtain
S0[φ] =1
2πK
∫
dτ
∫
dτ ′(φ(τ)− φ(τ ′))2
(τ − τ ′)2
= − 1
πK
∫
dτ
∫
dτ ′φ(τ)φ(τ ′)
(τ − τ ′)2
= − 1
πK
∫
dτ
∫
dτ ′ ∂τφ(τ)∂τ ′φ(τ′) ln |τ − τ ′|
= − 1
4K
∑
ij
qiqj ln |τi − τj |
= − 1
2K
∑
i<j
qiqj ln |τi − τj |,
where we used the identity [242]
∫
dx dx′f(x)f(x′)
(x− x′)2=
∫
dx dx′ f ′(x)f ′(x′) ln |x− x′|,
which is easily proved by integrating by parts. Putting everything together, the partition
function becomes
Z =
∫
Dφ(τ) e−S0[φ]−S′0[φ]−S2[φ]
=
∞∑
p=1
t2p
(
2p∏
i=1
′∑
qi=±
)
∫ β
0
dτ2p
∫ τ2p−δ
0
dτ2p−1 · · ·∫ τ2−δ
0
dτ1
× exp
(
1
2K
∑
i<j
qiqj ln|τi − τj |
δ
)
, (A.42)
where we have added a short-time cutoffδ, i.e. the instanton gas is a gas of ‘hard rods’ with
diameterδ (the long-range logarithmic interaction is valid only for distances larger than the
cutoff |τ − τ ′| > δ). The prime on the sum indicates that the neutrality condition∑
i qi = 0
should be obeyed, and the instanton fugacityt = e−Sinst is also the tunneling amplitude.
APPENDIX A. THEORETICAL METHODS 174
A.3.4 Tree-level renormalization
We now perform renormalization of the partition functionat tree levelwhich is the leading
nonvanishing order in the beta functionβ(t) for the tunneling amplitudet. Consider an
action with reduced cutoff,
Zbδ(t) =∞∑
p=1
t2p
(
2p∏
i=1
′∑
qi=±
)
∫ bβ
0
dτ2p
∫ τ2p−bδ
0
dτ2p−1 · · ·∫ τ2−bδ
0
dτ1
× exp
(
1
2K
∑
i<j
qiqj ln|τi − τj |
δ
)
,
with b = 1 + dℓ. We now perform a scale transformationτi = τi/b to restore the cutoff to
δ. The integration measure transforms as
dτ2p · · · dτ1 =∂(τ2p, . . . , τ1)
∂(τ2p, . . . , τ1)dτ2p . . . dτ1 = b2p dτ2p . . . τ1,
hence
∫ bβ
0
dτ2p
∫ τ2p−bδ
0
dτ2p−1 · · ·∫ τ2−bδ
0
dτ1 = b2p∫ β
0
dτ2p
∫ τ2p−δ
0
dτ2p−1 · · ·∫ τ2−δ
0
dτ1.
The exponential becomes
exp
(
1
2K
∑
i<j
qiqj ln|τi − τj |
δ
)
exp
(
1
2K
∑
i<j
qiqj ln b
)
,
where the last exponential factor can be written as
exp
(
1
2K
∑
i<j
qiqj ln b
)
= b1
2K
∑i<j qiqj .
However, by neutrality of the Coulomb plasma we have
0 =
(
∑
i
qi
)2
=∑
ij
qiqj =∑
i 6=j
qiqj +∑
i
q2i = 2∑
i<j
qiqj +∑
i
1,
APPENDIX A. THEORETICAL METHODS 175
hence∑
i<j qiqj = −12
∑
i 1 = −p, so that we haveb1
2K
∑i<j qiqj = b−p/2K . Therefore, the
renormalized action is
Zδ =
∞∑
p=1
t2pb2p(1−1/4K)
(
2p∏
i=1
′∑
qi=±
)
∫ β
0
dτ2p
∫ τ2p−δ
0
dτ2p−1 · · ·∫ τ2−δ
0
dτ1
× exp
(
1
2K
∑
i<j
qiqj ln|τi − τj |
δ
)
≡ Zδ(t),
where we can now identity the renormalized tunneling amplitude by comparison with the
original partition function,
t = tb1−1/4K .
Expanding(1+ dℓ)1−1/4K = 1+ (1− 1/4K)dℓ andt = t+ (dt/dℓ)dℓ, we obtain the RGE
for the tunneling amplitude,dt
dℓ=
(
1− 1
4K
)
t.
The square of the tunneling amplitude obviously followsdt2
dℓ= 2
(
1− 14K
)
t2. Cutting off
the RG flow at a scale set by the temperatureT , we obtain the renormalized tunneling
strengtht2 as a function of temperature,
t2(T ) = t2(Λ)
(
T
Λ
)2(1/4K−1)
,
whereΛ ∼ EF is some high-energy scale of the order of the bandwidth, andt2(Λ) is the
“bare” tunneling amplitude at this energy, i.e.t2 = e−2Sinst with Sinst given by Eq. (A.39)
andλ2 taken to be the bare two-particle backscattering amplitude. Since the conductance
proceeds by tunneling, we have
G(T ) ∝ t2(T ) ∝ T 2(1/4K−1).
APPENDIX A. THEORETICAL METHODS 176
A.3.5 Mapping to dual boundary sine-Gordon model
To get an intuitive understanding of the low-energy physics, we want to show the equiv-
alence of the Coulomb gas partition function with the partition function of the boundary
sine-Gordon model for the dual bosonθ. By rephrasing the cosine term as a tunneling term
in the fermion language, we can identify what physical processes restore the conductance
at finite temperature.
We conjecture that the Coulomb gas partition function just derived can be written as a
functional integral over the dual bosonθ(τ) ≡ θ(x = 0, τ),
Z =
∫
Dθ e−Sdual[θ],
where
Sdual[θ] =K
β
∑
iωn
|ωn||θ(iωn)|2 + t
∫ β
0
dτ cosα√πθ(τ), (A.43)
and we have to findα andt in terms of the fugacityt = e−Sinst of the Coulomb gas model
Eq. (A.42). From the unperturbed action (the first term), theθ propagator is
〈θ(τ)θ(τ ′)〉 = − 1
2πln |τ − τ ′|1/K . (A.44)
We expand the partition function in powers oft. We have
Z
Z0= 1 +
∞∑
n=1
(−t)nn!
∫ β
0
dτn · · ·∫ β
0
dτ1〈T cosα√πθ(τ1) · · · cosα
√πθ(τn)〉,
where〈· · · 〉 =∫
Dθ e−S0[θ] · · · /Z0, with Z0 =∫
Dθ e−S0[θ] andS0[θ] the unperturbed
action. Only terms withn even survive the average, so that we have
Z
Z0
= 1 +∞∑
n=1
(−t)2n(2n)!
∫ β
0
dτ2n · · ·∫ β
0
dτ1〈T cosα√πθ(τ1) · · · cosα
√πθ(τ2n)〉.
We have explicitly written the time-ordering operatorT to stress the fact that the2n-point
correlator is automatically time-ordered because it was defined as a path integral average
APPENDIX A. THEORETICAL METHODS 177
(Feynman-Kac formula). We rewrite the integral over[0, β]2n as an integral over a2n-
simplex in the standard way3 [226],
∫ β
0
dτ2n · · ·∫ β
0
dτ1〈TA(τ1) · · ·A(τ2n)〉
= (2n)!
∫ β
0
dτ2n
∫ τ2n
0
dτ2n−1 · · ·∫ τ2
0
dτ1 〈A · · ·A〉,
where the last operator product is not time-ordered. The factorial is cancelled and the
partition function becomes
Z
Z0= 1 +
∞∑
n=1
t2n∫ β
0
dτ2n
∫ τ2n
0
dτ2n−1 · · ·∫ τ2
0
dτ1〈cosα√πθ(τ1) · · · cosα
√πθ(τ2n)〉.
Writing the cosine as
cosα√πθ(τi) =
12
∑
qi=±eiαqi
√πθ(τi),
we have
〈cosα√πθ(τ1) · · · cosα√πθ(τ2n)〉
=1
22n
(
2n∏
i=1
∑
qi=±
)
〈 : eiαq1√πθ(τ1) : · · · : eiαq2n
√πθ(τ2n) : 〉.
We now use the result〈 : eA1 : · · · : eAN : 〉 = exp∑N
i<j〈AiAj〉 for expectation values of
products of vertex operators [230] and theθ propagator Eq. (A.44) to obtain
〈cosα√πθ(τ1) · · · cosα√πθ(τ2n)〉 =
1
22n
(
2n∏
i=1
′∑
qi=±
)
exp
(
α2
2K
2n∑
i<j
qiqj ln |τi − τj |)
,
3Special thanks to T. T. Ong for pointing this out.
APPENDIX A. THEORETICAL METHODS 178
where the primed sum indicates the neutrality condition∑
i qi = 0, which is verified be-
cause the2n-point correlator of vertex operators: eiαqi√πθ(τi) : vanishes otherwise. Sub-
stituting back into the expression for the partition function, we find
Z
Z0= 1 +
∞∑
n=1
(
t
2
)2n(
2n∏
i=1
′∑
qi=±
)
∫ β
0
dτ2n
∫ τ2n
0
dτ2n−1 · · ·∫ τ2
0
dτ1
× exp
(
α2
2K
2n∑
i<j
qiqj ln |τi − τj |)
,
which is the same partition function as that of the Coulomb gas Eq. (A.42), apart from (1)
a multiplicative factorZ0 which is just a shift of free energy, (2) the first termZ/Z0 = 1
which corresponds to a configuration with no instantons (we could have included it in
Eq. (A.42)), and (3) the presence of a short-time cutoffδ which we should incorporate in
the present expression. We see that the identification of thetwo partition functions requires
t = 2t = 2e−Sinst andα2 = 1,
which implies that our “tunneling” action should read
Stunnel = t
∫ β
0
dτ cos√πθ(τ).
The dimension of this operator is easily obtained,
〈 : cos√πθ(τ) : : cos
√πθ(τ ′) : 〉 ∼ eπ〈θ(τ)θ(τ
′)〉 =1
|τ − τ ′|1/2K =1
|τ − τ ′|2∆ ,
meaning that∆ = 1/4K, which leads to the expected RGE
dt
dℓ=
(
1− 1
4K
)
t.
From the scaling dimension of the tunneling operator, we cansee that it corresponds to
tunneling the “square root” (or half) of an electron (ordinary single-particle tunneling cor-
responds to the operatorcos 2√πθ [71, 72]). In the next section we derive the form of
APPENDIX A. THEORETICAL METHODS 179
the current operator in the strong coupling regime and confirm the half-charge tunneling
picture.
A.3.6 Current operator in strong coupling regime
Suppose we would like to calculate the expectation value of the electric currentje = −ejat equilibrium,
〈je(τ)〉eq =ie√π〈∂τφ〉 =
ie
2
⟨
∑
i
qiδ(τ − τi)
⟩
=
∞∑
p=1
t2p
(
2p∏
i=1
′∑
qi=±
)
∫ β
0
dτ2p
∫ τ2p−δ
0
dτ2p−1 · · ·∫ τ2−δ
0
dτ1
×ie2
∑
i
qiδ(τ − τi) exp
(
1
2K
∑
i<j
qiqj ln|τi − τj |
δ
)
,
where we have used Eq. (A.41). Consider now a modified form of the action Eq. (A.43),
S[θ, a] =K
β
∑
iωn
|ωn||θ(iωn)|2 + 2t
∫ β
0
dτ cos[√πθ(τ) + 1
2ea(τ)]. (A.45)
The Coulomb gas representation of this action now reads
Z[a]
Z0
= 1 +∞∑
p=1
t2p
(
2p∏
i=1
′∑
qi=±
)
∫ β
0
dτ2p
∫ τ2p−δ
0
dτ2p−1 · · ·∫ τ2−δ
0
dτ1
× exp
(
1
2K
∑
i<j
qiqj ln|τi − τj |
δ+ie
2
∑
i
qia(τi)
)
.
We observe that
1
Z0
δZ[a]
δa(τ)=
∞∑
p=1
t2p
(
2p∏
i=1
′∑
qi=±
)
∫ β
0
dτ2p
∫ τ2p−δ
0
dτ2p−1 · · ·∫ τ2−δ
0
dτ1
× exp
(
1
2K
∑
i<j
qiqj ln|τi − τj |
δ+ie
2
∑
i
qia(τi)
)
ie
2
∑
i
qiδ(τ − τi),
APPENDIX A. THEORETICAL METHODS 180
hence we see that
〈je(τ)〉eq = lima→0
1
Z0
δZ[a]
δa(τ). (A.46)
Note that the factor of12
is required in the boson action Eq. (A.45) to correctly matchthe
prefactor in the instanton representation of the current Eq. (A.41). If we calculate the right-
hand side directly from the action Eq. (A.45), as we take the limit a(τ) → 0 we obtain
〈je(τ)〉 = te〈sin√πθ(τ)〉, hence we can identify
j = −t sin√πθ, je = te sin√πθ, (A.47)
as the current operators in the strong coupling regime, whereas the usual single-particle
tunneling operator is
iet(ψ†1ψ2 − h.c.) ∝ et sin 2
√πθ,
where the indices1, 2 refer to two disconnected semi-infinite Luttinger liquids [71, 72].
Hence we can already expect from Eq. (A.47) that in our case the excitations which tunnel
through the impurity are the “square root” of an electron, orhalf-charged quasiparticles.
A.4 Keldysh calculation of the shot noise
As the dual bosonθ is expected to transform in some way under a (local) gauge transforma-
tion, the currents Eq. (A.47) are clearly not gauge invariant. It is meaningful to calculate the
linear response of these non gauge invariant currents to an appliedelectromagnetic field,
because the piece of the current which restores gauge invariance gives a contribution to the
response which is nonlinear in the applied electromagneticfield, and therefore neglected in
a linear response calculation. However, if one is to calculate the fully nonlinear, nonequi-
librium current, then the gauge invariant (kinematical) current must be used. To obtain the
shot noise which is the noise in the presence of a finite current, we need to use the gauge
invariant current. The gauge invariant current is obtainedsimply bynot letting a → 0 in
Eq. (A.46),
〈je(τ)〉 =1
Z0
δZ[a]
δa(τ)= te sin[
√πθ(τ) + 1
2ea(τ)], (A.48)
APPENDIX A. THEORETICAL METHODS 181
which can be interpreted as “taking the square root” of the current operator considered by
Kane and Fisher [71, 72] for single-particle tunneling:
〈j1P〉 ∝ sin[2√πθ + ea(τ)].
In this section we want to derive an expression for the nonequilibrium noise spectrum in
the strong coupling regime at low temperatures and finite bias.
A.4.1 Basics of noise
We first review the basic ideas of noise. The noise spectrum measures the amount of
fluctuations in the current operator,
S(ω) ≡∫
dt eiωt〈∆I(t),∆I(t′)〉,
whereI(t) is the current operator in the Heisenberg picture and∆I(t) ≡ I(t)− 〈I〉 where
we assume steady-state such that the average current〈I〉 does not depend on time. We have
〈∆I(t),∆I(t′)〉 = 〈I(t), I(t′)〉 − 2〈I〉2 = 〈I(t), I(t′)〉conn,
and the zero-frequency noise is thus given by
S(ω = 0) =
∫
dt 〈I(t), I(t′)〉conn.
Classical shot noise: Schottky formula
Consider a mesoscopic device connected to external leads. There are two sources of noise.
Thermalnoise occurs due to the thermal fluctuations of the occupation numbers in the
leads, exists at zero bias, and vanishes at zero temperature. Shotnoise is the excess noise
in the presence of a finite current and is a nonequilibrium process. It is due to the fact that
current is carried by discrete charges.
The expression for the noise in a classical conductor is simply derived in the following
way. Consider that the current proceeds by random “tunneling” events in which a particle
APPENDIX A. THEORETICAL METHODS 182
with chargee is transported from one lead to the other. If we assume that the tunneling
events are uncorrelated, the probability for havingN tunneling events during timeτ is
given by the Poisson distribution,
PN(τ) =〈N〉NN !
e−〈N〉.
It is well-known that for the Poisson distribution the variance and mean are equal,
〈N2〉 − 〈N〉2 = 〈N〉. (A.49)
If we define the current as the random variableI ≡ eN/τ , the average current is given by
〈I〉 = e〈N〉/τ. (A.50)
The zero-frequency noise is then given by
S(ω = 0) =
∫ τ
0
dt(
〈I(t), I(t′)〉 − 2〈I〉2)
= 2
∫ τ
0
dt(
〈I2〉 − 〈I2〉)
= 2τe2(〈N2〉 − 〈N〉2)/τ 2.
Using Eq. (A.49) and (A.50), this is just
S(ω = 0) = 2e〈I〉, (A.51)
which is the Schottky formula. In the presence of a finite current, we thus see that the noise
contains information about the charge of the tunneling particles.
Equilibrium/thermal/Johnson-Nyquist noise
The zero-frequency thermal noise is
S(ω = 0) = 4GkBΘ,
APPENDIX A. THEORETICAL METHODS 183
whereG is the equilibrium conductance andΘ is the temperature. A simple derivation
follows below [243]. The current through a noninteracting two-terminal device with trans-
mission coefficientT (E) and twofold spin degeneracy is given by
I =2e
h
∫
dE T (E)[fL(E)− fR(E)],
whereE is the energy of a tunneling electron andfL, fR are the electron distribution
functions in the left and right leads, respectively. For simplicity, consider the classical limit
(the conclusions are unchanged in the Fermi-Dirac case [243]) where(E − µ)/kBΘ ≫ 1,
f(E) = (eβ(E−µ) + 1)−1 → e−β(E−µ).
Then one has
fL(E)− fR(E) = e−β(E−µ−eVL) − e−β(E−µ−eVR) = (eβeVL − eβeVR)e−β(E−µ).
Now assume the limit of small biasVL,R ≪ kBΘ,
fL(E)− fR(E) → βe(VL − VR)e−β(E−µ),
hence the equilibrium conductance is given by
G = limVL−VR→0
I
VL − VR=
2e2
kBΘh
∫
dE T (E)e−β(E−µ).
The idea is that although the net current in equilibrium is zero I = IL + IR = 0 since
IL = −IR, the noise produced by each stream|IL| = |IR| adds:
IS = |IL|+ |IR| = 2|IL| 6= 0.
At equilibriumVL = VR = 0, the current injected by the left lead is
I =2e
h
∫
dE T (E)fL(E) →2e
h
∫
dE T (E)e−β(E−µ),
APPENDIX A. THEORETICAL METHODS 184
Hence the total “noise current”IS is
IS = 2|IL| =4e
h
∫
dE T (E)e−β(E−µ) =2GkBΘ
e.
Thermal noise is the “shot noise” produced by this current. Using the Schottky relation
Eq. (A.51) withI = IS, we have
S(ω = 0) = 4GkBΘ,
hence we see that the thermal noise contains the same information as the conductance, i.e.
it does not contain any information about the charge of the tunneling particle.
Quantum shot noise: thermal/shot noise crossover
A quantum mechanical treatment yields [73]
S(ω = 0) =4e2
h
[
2kBΘT 2 + eV coth
(
eV
2kBΘ
)
T (1− T )
]
,
whereT is the transmission coefficient. We consider the limit of weak tunnelingT ≪ 1
and thus neglect the termsO(T 2). We obtain
S(ω = 0) = 2e〈I〉 coth(
eV
2kBΘ
)
.
The thermal noise is recovered at low biaseV ≪ kBΘ,
S(ω = 0) → 4GkBΘ, eV ≪ kBΘ,
whereG = limV→0〈I〉/V is the equilibrium conductance, and we usedcoth x ≃ 1/x for
x ≪ 1. In the high bias regimeeV ≫ kBΘ, the Schottky relation for the shot noise is
recovered,
S(ω = 0) → 2e〈I〉, eV ≫ kBΘ,
usingcothx→ 1 for x≫ 1.
APPENDIX A. THEORETICAL METHODS 185
A.4.2 Schwinger-Keldysh action
We now go back to our initial problem. Since we are considering a nonequilibrium situ-
ation, we need to consider a real-time action defined on the Schwinger-Keldysh contour
C [244, 245],
S[θ, A] = S0[θ] + 2t
∫
Cdt cos[
√πθ(t) + 1
2eA(t)],
where we denote the tunneling amplitude byt ≡ t to avoid confusion with the real time
variablet, andA(t) is the vector potential in real time,A(t) = −V t. In this case the current
is given by〈I(t)〉 = 1Z
δZδiA(t)
. We can make a consistency check of this formula by noticing
that in the presence of the usual electromagnetic coupling term
δS[θ, A] =
∫
dt je(t)A(t),
the above-mentioned functional derivative does indeed give rise to the expectation value of
the current,
〈I(t)〉 = 1
Z
δ
δiA(t)
∫
Dθ eiS0[θ]+i∫dt je(t)A(t) = 〈je(t)〉.
We want to derive a relation [73] between the zero-frequencynoise spectrum and the
nonequilibrium current,
S(ω = 0) = 2e∗|〈I〉| coth(
e∗V
2kBT
)
, (A.52)
wheree∗ is the charge of the tunneling quasiparticles. This relation holds for the interedge
tunneling of quasiparticles in the fractional QH regime (inthe weak tunneling regime) and
has been used experimentally to confirm the fractional charge e∗ = νe of the quasiparti-
cles [246, 247] whereν is the filling factor. The goal is to show that it also holds in our case
and thate∗ = e/2 in the strong coupling regime. The calculation proceeds in two steps:
a) calculating the nonequilibrium current〈I〉, b) calculating the zero-frequency noise spec-
trum S(ω = 0). By comparing the obtained results with Eq. (A.52), we can read out the
quasiparticle chargee∗.
APPENDIX A. THEORETICAL METHODS 186
A.4.3 Nonequilibrium current
In the Hamiltonian representation we have [73]
〈I(t)〉 = 12
∑
η
〈TCI(tη)e−i∫C dt1 H′(t1)〉, (A.53)
where the current operator is in the interaction representation with respect to the perturba-
tionH ′ defined as
H ′(t1) = −2t : cos[√πθ(t1) +
12eA(t1)] : .
In Eq. (A.53), TC is the contour-ordering operator on the Schwinger-KeldyshcontourC,
andη = ±1 denotes the forward (+) or backward (−) part of the contour. Since the current
〈I(t)〉 is a one-point correlation function, it does not matter on which branch of the contour
we choose to evaluate it, and we choose the symmetric combination 〈I(t)〉 = 12
∑
η〈I(tη)〉.At low temperaturesT ≪ T ∗
2 , the tunneling is weak and we calculate the current to
O(t2). Since the current operatorI is alreadyO(t), we need only expand the evolution
operatore−i∫C dt1 H′(t1) to first order inH ′. The integration over the two branches of the
Schwinger-Keldysh contour is written as
∫
Cdt1 F (t1) =
(∫ ∞
−∞+
∫ −∞
∞
)
dt1 F (t1) =∑
η1
η1
∫ ∞
−∞dt1 F (t
η11 ),
wheret±1 indicates a time argument on the± branch of the contour. Using Eq. (A.48) for
the current, we have
〈I(t)〉 =iet2
2
∑
ηη1
η1
∫ ∞
−∞dt1 sin
e
2[A(tη)−A(tη11 )]〈TC : ei
√πθ(tη) : : e−i
√πθ(t
η11 ) : 〉
= −iet2
2
∑
ηη1
η1
∫ ∞
−∞dt1 sin
eV
2(t− t1)e
πDηη1 (t−t1),
whereDηη1(t − t1) = 〈TCθ(tη)θ(tη11 )〉 is the Keldysh Green function of the dual boson
θ. We see that the current does not depend on the timet since we can perform a change of
APPENDIX A. THEORETICAL METHODS 187
variablest− t1 → t1. We subsequently relabelt1 ≡ t for simplicity and write
〈I〉 = −iet2
2
∑
ηη1
η1
∫ ∞
−∞dt sin
eV t
2eπDηη1 (t).
The causal and anti-causal Green functionsD++,D−− are functions of|t| only,
D±±(t) =1
2πKln
(
πτc/β
sin π(±i|t|)/β
)
,
and thus even undert → −t, hence they do not contribute to the integral which becomes
(with η1 → −η)
〈I〉 = iet2
2
∑
η
η
∫ ∞
−∞dt sin
eV t
2eπDη,−η(t).
The Green functionsD±∓ are given by
D+−(t) = D<(t) =1
2πKln
(
πτc/β
sin π(−it)/β
)
,
D−+(t) = D>(t) =1
2πKln
(
πτc/β
sin πit/β
)
.
The integral can then be computed explicitly and one obtains
〈I〉 = −2et2(
πτcβ
)1/2Kβ
π
21/2K−2
Γ(1/2K)
∣
∣
∣
∣
Γ
(
1
4K+iβeV
4π
)∣
∣
∣
∣
2
sinh
(
βeV
4
)
, (A.54)
whereΓ is the usual Gamma function.
A.4.4 Nonequilibrium noise
The nonequilibrium noise spectrumS(ω) is given by the Fourier transform of the nonequi-
librium current-current correlator
S(t, t′) = 〈I(t), I(t′)〉,
APPENDIX A. THEORETICAL METHODS 188
where, is the anticommutator. Again, in the Keldysh formalism we have
S(t, t′) =∑
η
〈TCI(tη)I(t′−η)e−i∫Cdt1 H′(t1)〉.
Since we want to calculate the noise spectrum toO(t2) as well, and each current operator
isO(t), we have to expand the evolution operator toO(1) only,
e−i∫Cdt1 H′(t1) → 1.
We thus obtain
S(t, t′) = e2t2∑
η
〈TC : sin[√πθ(tη) + 1
2eA(tη)] : : sin[
√πθ(t′−η) + 1
2eA(t′η)] : 〉
+O(t4)
=e2t2
2
∑
η
coseV (t− t′)
2eπDη,−η(t−t′) = S(t− t′).
The zero-frequency noise spectrum is then
S(ω = 0) =
∫ ∞
−∞dt S(t) =
e2t2
2
∑
η
∫ ∞
−∞cos
eV t
2eπDη,−η(t).
The integral is very similar to the one done previously, and we obtain
S(ω = 0) = 2e2t2(
πτcβ
)1/2Kβ
π
21/2K−2
Γ(1/2K)
∣
∣
∣
∣
Γ
(
1
4K+iβeV
4π
)∣
∣
∣
∣
2
cosh
(
βeV
4
)
. (A.55)
A.4.5 Schottky relation and fractional Fano factor
Taking the ratio of Eq. (A.54) and (A.55), we obtain
S(ω = 0) = e|〈I〉| coth(
eV
4kBT
)
, (A.56)
APPENDIX A. THEORETICAL METHODS 189
which has the same form as Eq. (A.52) with a quasiparticle charge
e∗ =e
2,
hence a noise measurement should be able to confirm the fractional charge of the tunneling
quasiparticles.
Equation (A.56) contains the full thermal/shot noise crossover. The Johnson-Nyquist
thermal (equilibrium) noise is recovered at low biaseV ≪ kBT ,
S(ω = 0) → 4GkBT, eV ≪ kBT,
whereG = limV→0〈I〉/V is the equilibrium conductance, and we usedcoth x ≃ 1/x for
x ≪ 1. The quasiparticle chargee∗ cancels out in this expression, so we need to consider
a large enough biaseV > kBT . In the high bias regimeeV ≫ kBT , the Schottky relation
for the shot noise is recovered,
S(ω = 0) → 2e∗|〈I〉|, eV ≫ kBT,
usingcothx→ 1 for x≫ 1.
A.5 Mean-field description of half-charge tunneling
In this section we use a mean-field description to show that the fractionalized tunneling cur-
rent in the insulating regimeK < 1/4 corresponds to the Goldstone-Wilczek current [69]
for (1 + 1)D Dirac fermions in an instanton background in the time domain. We consider
the vicinity of the zero temperature insulating fixed point forK < 1/4. At the fixed point
the boson density fieldφ is classical and pinned at some odd multiple of√π/4 since the
two-particle backscattering coupling constantλ2 has flown to infinity.
The discreteφ → φ+√π/2 symmetry and TR symmetry are spontaneously broken at
T = 0. This can be understood from the fact that the(0 + 1)D two-particle backscattering
problem can be mapped to a classical 1D Coulomb gas problem with long-range interac-
tions (Sec. A.3.3). In the presence of short-range interactions, due to the Mermin-Wagner
APPENDIX A. THEORETICAL METHODS 190
theorem there can only be spontaneous breaking of these (discrete) symmetries atTeff = 0
whereTeff is the effective temperature of the classical problem. Indeed,d = 1 is the lower
critical dimension for spontaneous breaking of a discrete symmetry for systems with short-
range forces. In the presence of long-range forces however,one can have spontaneous
symmetry breaking below a finiteTeff = Tc, which corresponds to a finiteK = Kc = 1/4
in the original problem. In the present case, the RGE tell us that any nonzeroλ2 will lead
to spontaneous symmetry breaking forK < Kc.
This means that the TR symmetry-breaking mass order parametersO1 ∼ cos 2√πφ and
O2 ∼ sin 2√πφ are also classical at zero temperature [21]. At finite temperatures,λ2 is
finite and TR symmetry is restored by fluctuations of the orderparameters, but at very low
temperatures these fluctuations are small and it is reasonable to perform a mean field de-
composition of the point-split two-particle backscattering operatorψ†R(0)ψ
†R(a)ψL(a)ψL(0).
The two-particle backscattering operatorψ†R(0)ψ
†R(a)ψL(a)ψL(0) + H.c. with a some
small distance of the order of the lattice constant is decomposed as
−〈ψ†R(0)ψL(a)〉ψ†
R(a)ψL(0)− 〈ψ†R(a)ψL(0)〉ψ†
R(0)ψL(a)
+〈ψ†R(0)ψL(0)〉ψ†
R(a)ψL(a) + 〈ψ†R(a)ψL(a)〉ψ†
R(0)ψL(0) + H.c.,
where the pairing terms〈ψ†R(0)ψ
†R(a)〉 and 〈ψL(a)ψL(0)〉 vanish. The point splitting of
the four-fermion operator is important: the mean field decomposition of a naive operator
product vanishes (since the naive operator product is itself zero by the Pauli principle). At
this stage, the point splitting can be formally removed fromthe two-fermionoperators, al-
though we need to keep it to evaluate the two-pointcorrelators. By translational invariance
we have
ψ†R(0)ψL(0)
(
2〈ψ†R(0)ψL(0)〉 − 〈ψ†
R(0)ψL(a)〉 − 〈ψ†R(a)ψL(0)〉
)
+H.c.,
where the last two terms are equal as we will see. We evaluate the correlators at zero tem-
perature where the density fieldφ is classical (pinned) but the phase fieldθ is fluctuating.
APPENDIX A. THEORETICAL METHODS 191
The correlator thus contains a free boson average overθ while keepingφ fixed,
〈ψ†R(0)ψL(a)〉 =
1
2πξ
⟨
e−i√π[φ(0)−θ(0)]ei
√π[−φ(a)−θ(a)]
⟩
=1
2πξ
⟨
ei√π[θ(0)−θ(a)]
⟩
e−i√π[φ(0)+φ(a)],
where the point splitting can be removed in the last factor,e−i√π[φ(0)+φ(a)] ≃ e−i2
√πφ(0).
Theθ correlator is
⟨
ei√π[θ(0)−θ(a)]
⟩
= exp(
−π2〈[θ(0)− θ(a)]2〉
)
= eπ〈θ(a)θ(0)〉 = exp
[
− 1
4Kln
(
a2 + ξ2
ξ2
)]
.
The mean field decomposition thus reads
1
πξψ†RψLe
−i2√πφ
1− exp
[
− 1
4Kln
(
a2 + ξ2
ξ2
)]
+H.c.
For the point splitting witha to make sense in a theory which is already regularized in the
ultraviolet by a short-distance cutoffξ, we have to consider the limitξ/a ≪ 1 in which
ln(a2+ξ2
ξ2) ≃ ln(a2/ξ2) ≫ 1. As a result, the last term vanishes and we are left simply with
1
πξψ†RψLe
−i2√πφ +H.c.
The mean field backscattering operator is thus
H2 = gψ†R
(
cos 2√πφ(0, t)− i sin 2
√πφ(0, t)
)
ψL +H.c. = gΨ†ma(0, t)σaΨ,
whereg = λ2/πξ,m1 = cos 2√πφ,m2 = sin 2
√πφ andΨ is the Dirac spinor [Eq. (A.57)].
Since the superconducting order parameters〈ψ†Rψ
†R〉 and〈ψLψL〉 are destroyed by the fluc-
tuations of the Josephson phase fieldθ, which unlikeφ is not pinned, the mean-field La-
grangian can be written as a mass term for Dirac fermions in1 + 1 dimensions,
δL = gΨ(m1 + iγ5m2)Ψ,
wherem1(0, τ) = cos 2√πφ(τ) ∼ O1 the normal mass andm2(0, τ) = sin 2
√πφ(τ) ∼ O2
APPENDIX A. THEORETICAL METHODS 192
the axial mass. The Dirac field is
Ψ =
(
ψR↑
ψL↓
)
, (A.57)
andγ0 = σ1, γ1 = iσ2 andγ5 = iγ0γ1 = −iσ3 are the Dirac matrices. The mass terms
are localized in space atx = 0 but “slowly-varying” in time (classical instanton gas). Asis
well-known, the instanton background induces a topological Goldstone-Wilczek fermion
current [69] given by
〈jµ〉 = 1
2πǫµνǫab
ma∂νmb
|m|2 ,
with µ, ν = t, x anda, b = 1, 2. A single instanton event pumps a quantizede/2 charge
along the edge, which is just the effect of rotating the “θ” mass domain wall parameter by
π in the Qi-Hughes-Zhang fractional charge proposal [70]. The electric current is just
j = −e〈jx〉 = 1
2π(m2∂tm1 −m1∂tm2) =
e√π∂tφ(0, t),
which is the standard bosonization formula. As a result, thecharge pumped by a single
instanton configuration in whichφ jumps, say, from−√π/4 to
√π/4, is
∆Qinst =
∫ t2
t1
dt j =e√π[φ(t2)− φ(t1)] =
e
2.
The instanton gas representation at low temperatures thus corresponds to the tunneling
of half-charged particles. The effect studied here can alsobe seen as the time-domain
counterpart to the statice/2 charge induced by a spatial magnetic domain wall on the QSH
edge [70].
A.6 Phase space derivation of theT 6 behavior in the non-
interacting case
The purpose of this section is to explain the∝ −T 6 behavior of the conductance in the
noninteracting regimeK = 1. In this case, we do not need to rely on bosonization and we
APPENDIX A. THEORETICAL METHODS 193
can use the fermion language. Consider the two-particle backscattering process
R, k;R, k′ −→ L, p;L, p′, (A.58)
whereR,L indicates a right/left mover, andk, k′, p, p′ are the momenta. The two-particle
backscattering Hamiltonian is
H2 = λ2ψ†R(0)ψ
†R(a)ψL(0)ψL(a) + H.c.,
where a point-splitting with a short-distance cutoffa is required by the exclusion principle.
We thus see that the interaction is nonlocal in space. In the continuum limit, we can expand
ψ(a) ≃ ψ(0) + a∂xψ(0) and we obtain
H2 = λ2ψ†R(0)∂xψ
†R(0)ψL(0)∂xψL(0) + H.c.,
where we have absorbed the factor ofa2 in the coupling constantλ2. Fourier transforming,
we obtain
H2 = λ2∑
k1k2k3k4
k2k4(c†Rk1
c†Rk2cLk3cLk4 +H.c.).
The scattering rate for the process of Eq. (A.58) is given by the Fermi Golden rule,
Γk,k′→p,p′ =2π
~|〈L, p;L, p′|H2|R, k;R, k′〉|2δ(εk + εk′ − εp − εp′).
The matrix element between the two-particle states|R, k;R, k′〉 = c†Rkc†Rk′|0〉 and
|L, p;L, p′〉 = c†Lpc†Lp′|0〉 is easily obtained,
〈L, p;L, p′|H2|R, k;R, k′〉 = −λ2(k − k′)(p− p′),
and the contribution to the inverse lifetime of the single-particle state|R, k〉 due to this
process is given by
1
τk∝∑
k′pp′
Γk,k′→p,p′nF (εk′)[1− nF (εp)][1− nF (εp′)],
APPENDIX A. THEORETICAL METHODS 194
where we have include the Fermi occupation numbersnF (ε) = (eβ(ε−µ) + 1)−1 to account
for the exclusion principle. At this stage we can already understand where theT 6 behavior
comes from. We sum over three variablesk′, p, p′. There is no constraint from momentum
conservation because the backscattering occurs at a singlepoint in space. There is one
constraint coming from energy conservation (theδ function in the scattering rateΓ) hence
we really sum over only two independent variables, for example εp andεp′. Each of these
variables contributes a phase space factorT/EF , hence so far we haveT 2. However the
matrix element squared contains(k−k′)2(p−p′)2 and since all momenta are confined to the
vicinity of the Fermi points, this contributes(k−kF )4 which translates into(ε−εF )4 ∼ T 4
at finite temperature, because the spectrum is linearε ∝ k (more generally, it can always
be linearized in the vicinity of the Fermi points). Hence we see that the helical nature of
the system suppresses the phase space for backscattering (which is another manifestation
of its robustness to disorder) because of the nonlocal nature of the backscattering.
We can do the calculation directly to verify our conjecture.We consider the linear
dispersionεk = ±~vFk. The right-movers haveεk,k′ = ~vF (k, k′) and the left-movers
haveεp,p′ = −~vF (p, p′), hence the energy-conservingδ function is
δ(εk + εk′ − εp − εp′) =1
~vFδ(k + k′ + p + p′),
which imposes the constraintk′ = −(k + p+ p′), and we have
1
τk∝∑
pp′
(2k + p+ p′)2(p− p′)2nF (ε−(k+p+p′))[1− nF (εp)][1− nF (εp′)].
Instead of doing the calculation at finiteT , we considerT = 0 but withk−kF > 0 slightly
away from the Fermi surface. Since the dispersion is linear,k − kF ∝ ε − εF and we can
then make the substitutionk − kF → T to estimate the finiteT result. Converting sums to
integrals and consideringT = 0, we have
1
τk∝∫
−∞dp
∫
−∞dp′ (2k + p+ p′)2(p− p′)2θ(kF + k + p+ p′)θ(−p− kF )θ(−p′ − kF ),
sincenF (εk) = nF (~vFk) = θ(kF − k) at T = 0 with εF = ~vFkF . The domain of
APPENDIX A. THEORETICAL METHODS 195
integration is easily drawn and we have
1
τk∝∫ −kF
−k
dp
∫ −kF
−(kF+k+p)
dp′ (2k + p+ p′)2(p− p′)2 =11
90(k − kF )
6 ∼ T 6,
which confirms the estimate.
A.7 Transport in a spin Aharonov-Bohm ring
A.7.1 S-matrix analysis
We wish to obtain an expression for theS-matrix S relating outgoingb to incominga
current amplitudes,
(
bl
br
)
= S(
al
ar
)
with S =
(
r t′
t r′
)
, (A.59)
whereal andbl (ar andbr) arepL × 1 (pR × 1) column vectors of the current amplitudes
outside the QSH region in the left (right) lead (see Fig. 5.1), andpL (pR) is the number of
propagating channels at the Fermi energy in the left (right)lead. The matrixS therefore
has dimensions(pL + pR) × (pL + pR) and the submatricesr, r′ and t, t′ are reflection
and transmission matrices, respectively. The two-terminal conductanceG from left to right
is given by the Landauer formula [89]G = e2
htr tt†. We assume that phase coherence
is preserved throughout the sample so thatS can be obtained by combiningS-matrices
for different portions of the device coherently [89]. We define the(pL,R + 2) × (pL,R +
2) scattering matricesSL, SR for the left (L) and right (R) FM/QSH junctions (e.g. see
Fig. 5.2(a) for the left junction),
(
bl
bl′
)
= SL
(
al
al′
)
,
(
br′
br
)
= SR
(
ar′
ar
)
, (A.60)
wherel′ (r′) is the QSH region immediately to the right (left) of the left(right) junction,
such thatal′ , ar′ andbl′ , br′ are the 2-component spinors of edge state current amplitudes.
APPENDIX A. THEORETICAL METHODS 196
They are related through the geometric AB phaseϕ (different for each spin polarization)
and the dynamical phaseλ = 2kF ℓ (identical for both spin polarizations) whereℓ is the
distance travelled by the edge electrons from left to right junction andkF is the edge state
Fermi wave vector,
ar′↑,↓
al′↑,↓
= eiλ/2e∓iϕ/2
bl′↑,↓
br′↑,↓
, (A.61)
where the upper sign forϕ corresponds to spin up. Using Eq. (A.60) and (A.61), we can
write(
e−iλ/2Φ†al′
br
)
= SR
(
eiλ/2Φbl′
ar
)
, (A.62)
where we defineΦ ≡ e−iϕσz/2. Using the first equality in Eq. (A.60) together with
Eq. (A.62), we can eliminate the intermediate amplitudesal′ , bl′ and obtain relations be-
tween the left lead amplitudesal, bl and the right lead amplitudesar, br, which gives usS[Eq. (A.59)]. The2×2 transmission matrixt, i.e. the lower left block ofS, is then obtained
in the form
t = tRStL, (A.63)
wheretL andtR are the2× pL andpR × 2 transmission matrices for the left and right junc-
tions, respectively (i.e. the lower left blocks ofSL, SR following the notation of Eq. (A.59)),
andS is a2 × 2 matrix defined in Eq. (5.1). The effective spin density matricesρL, ρR of
the FM leads used in Eq. (5.2) are defined asρL = tLt†L andρR = t†RtR.
If the arms of the ring are asymmetric, the dynamical phaseλ is generally different for
each arm and we haveλbottom − λtop ≡ δ 6= 0. In this case, one can show that Eq. (5.2)
still holds, but with the substitutions
ρL(θL) → RδρL(θL)R−1δ = ρL(θL + δ),
r′L(θL) → Rδr′L(θL)R
−1δ = r′L(θL + δ),
whereRδ ≡ e−iσzδ/2 rotates the spin about thez axis by an angleδ. In other words, a
phase asymmetry is equivalent to a rigid flux-independent rotation of the electron spin, and
APPENDIX A. THEORETICAL METHODS 197
simply shifts the conductance pattern by a constant angleδ:
G(φ, θ ≡ θR − θL) → G(φ, θR − (θL + δ))
= G(φ, θ − δ). (A.64)
A.7.2 Scattering at the junction
In order to solve the 1D scattering problem at the FM/QSH interface, we first observe that
the number of degrees of freedom is equal on either side of thejunction. If the Fermi level
εF is chosen such that both spin subbands in the FM leads are occupied, there are four
propagating modes on each side of the junction (two spins andtwo chiralities). The QSH
spin statesφQSH(±) areσz eigenstates while the FM spin statesφFM(±)(θ) are eigenstates of
n · σ and depend explicitly onθ. The Schrodinger equation for the junction is then solved
by the following scattering ansatz,
ψ(+)σ (x) =
φ<(+)σ√v<σeik
<σ x +
∑
σ′ rσ′σφ<(−)
σ′√v<σ′
e−ik<σ′x, x < 0,
∑
σ′ tσ′σφ>(+)
σ′√v>σ′
eik>
σ′x, x > 0,
for a right-moving scattering state, and with similar expressions for a left-moving scattering
stateψ(−)σ . Spin is denoted byσ, chirality by ± and side of the junction by<,>. The
propagating modes are explicitly normalized to unit flux such that rσ′σ and tσ′σ are the
desired reflection and transmission matrices. Requiring the continuity ofψ(±)σ andvxψ
(±)σ
at the interfacex = 0 (with vx ≡ ∂H/∂kx the velocity operator), we obtain a system of
linear equations for the sixteen matrix elementsrL, tL, r′L, t
′L constitutingSL. As illustrated
in Fig. 5.1, the magnetization angle is set to zero in the leftlead and toθ in the right lead
and we obtainr′L(0) andrR(θ) in Eq. (5.1).
APPENDIX A. THEORETICAL METHODS 198
A.7.3 Tight-binding model
The effective tight-binding model describing HgTe QW is defined as [19, 54]
H =∑
i
c†iVici +∑
ij
(
c†iTijeiAijcj + h.c.
)
, (A.65)
whereTij = Txδj,i+x + Tyδj,i+y is the nearest-neighbor hopping matrix,Aij =e~c
∫ j
idr ·A
is the Peierls phase withA the electromagnetic vector potential, andVi, Tx andTy are4×4
matrices containing thek ·p parameters and the effective Zeeman term. The4×4 matrices
Tx, Ty andVi used in the tight-binding Hamiltonian (A.65) are given by
Tx =
D+ − iA2
− i∆e
20
− iA2
D− 0 − i∆h
2
− i∆e
20 D+
iA2
0 − i∆h
2iA2
D−
,
Ty =
D+A2
∆e
20
−A2
D− 0 −∆h
2
−∆e
20 D+
A2
0 ∆h
2−A
2D−
, (A.66)
and
Vi = (C − 4D − εF + Eg(i))14×4 + (M − 4B)12×2 ⊗ σz +HeffZ‖ +Heff
Z⊥,
whereD± ≡ D ± B andA, B, C, D, M , ∆e, ∆h arek · p parameters [54], and1n×n
denotes then× n unit matrix. The Fermi energyεF is uniform throughout the device. The
gate potentialEg(i) is different in the QSH and lead regions [Fig. 5.4(a)], and isused to
tune the central region into the QSH insulating regime. The in-planeHeffZ‖ and out-of-plane
HeffZ⊥ effective Zeeman terms, which are used to mimick the injection of spin-polarized
APPENDIX A. THEORETICAL METHODS 199
carriers from a FM layer (Fig. 5.1), are given by [54]
HeffZ‖ = g‖µB
0 0 Beff− 0
0 0 0 0
Beff+ 0 0 0
0 0 0 0
,
HeffZ⊥ = µBB
effz
gE⊥ 0 0 0
0 gH⊥ 0 0
0 0 −gE⊥ 0
0 0 0 −gH⊥
, (A.67)
whereBeff± = Beff
x ± iBeffy , Beff = (Beff
x , Beffy , B
effz ) is some effective magnetic field the role
of which is to induce a spin polarization in the leads,µB is the Bohr magneton, andg‖ and
gE⊥, gH⊥ are the in-plane and out-of-planeg-factors, respectively.
A.8 Magnetooptical Kerr/Faraday rotation in thick film
geometry
In this section, we give some details of the calculation of the Kerr and Faraday angles in
the thick film geometry of Fig. 6.1 (repeated in Fig. A.3), fornormal incidence. Defining
E as in Ref. [35],
E =
(
2αE
H
)
, D =
(
D
2αB
)
, (A.68)
we have
E(z, t) =
E+1 e
i(k1z−ωt) + E−1 e
i(−k1z−ωt), z < 0,
E+2 e
i(k2z−ωt) + E−2 e
i(−k2z−ωt), 0 < z < L,
E+3 e
i(k3z−ωt), z > L.
APPENDIX A. THEORETICAL METHODS 200
Figure A.3: Incident, reflected, and transmitted light for Kerr/Faraday rotation experimenton thick topological insulator film of thicknessL.
Since we are at normal incidence,E is entirely in thexy plane and the only boundary
condition atz = 0 andz = L is the continuity ofE . This gives us two relations,
E+1 + E
−1 = E
+2 + E
−2 ,
E+2 e
ik2L + E−2 e
−ik2L = E+3 e
ik3L.
Using the fact that for a plane waveE = ck× (−iσy)D wherec is the speed of light in the
medium in which a given plane wave propagates as well as the constitutive relations [35]
D = ME, we obtain two more equations,
T12(E+1 − E
−1 ) = E
+2 − E
−2 ,
T23(eik2LE
+2 − e−ik2LE
−2 ) = eik3LE+
3 ,
APPENDIX A. THEORETICAL METHODS 201
where we define a “transfer matrix”Tij ≡ cicjM−1
j Mi. After some algebra, we obtain
eik3LE+3 = SFE
+1 ,
with
SF =(
I+ T23Q∗12[P∗
12]−1)−1 T23
(
Q12 +Q∗12[P∗
12]−1P12
)
, (A.69)
where we define
P12 ≡ cos k2L · I+ i sin k2L · T12,
Q12 ≡ cos k2L · T12 + i sin k2L · I.
Note that Eq. (A.69) contains all the multiple reflection effects. Using Eq. (A.68), we can
extract the relation between the transmittedEt ≡ E+3 and incidentEin ≡ E+
1 electric fields,
Et = S11F Ein +
c1ε12α
S12F z × Ein,
where we have removed the global phase factoreik3L which is irrelevant, because we care
only about therelativephase between thex andy components (i.e. the polarization). Con-
sider a wave initially polarized in thex direction,Eyin = 0. Then the Faraday rotation will
be given by the ratio betweeny andx components of the transmitted electric fields,
tan θF =Ey
t
Ext
=c1ε12α
S12F
S11F
=1
2α
S12F
S11F
, (A.70)
where we have used the fact thatc1ε1 = 1 in vacuum. The Kerr angle can be extracted in a
similar way,
tan θK =Ey
r
Exr
=c1ε12α
S12K
S11K
=1
2α
S12K
S11K
, (A.71)
whereSK is defined as
SK = [P∗12]
−1 (SF − P12) .
APPENDIX A. THEORETICAL METHODS 202
In general there can be additional “trivial” integer QH layers at the interfaces. We would
like to be able to measure the bulk axion angle independentlyof these trivial layers. To
account for them, we consider the general case that the substrate has a trivial axion an-
gle θ = 2pπ with p integer (see Fig. A.3). Generally, Eq. (A.70) and (A.71) depend in a
complicated way on the materials constants of both the topological insulator and the sub-
strate,c2, ε2 andc3, ε3, respectively, as well as on the topological insulator film thickness
L and the photon frequencyω. However, the results simplify at the “magic frequencies”
ω = 2πnc/√ε2µ2L andω = 2π(n + 1
2)c/
√ε2µ2L, n ∈ Z, corresponding to reflectivity
minima and maxima, respectively, allowing us to form combination of measured angles
that are independent of the material propertiesε2, c2 (or µ2), ε3, andc3 (or µ3).
A.9 Surface Chern-Simons theory for an Abelian fractio-
nal 3D topological insulator
In this section, we consider the AbelianU(1) × U(1) fractional topological insulator on
the spatial 3-manifoldM = T 2 × I with I = [−L/2, L/2] in thez direction. Because the
bulk 3D statistical gauge fields are free, we can integrate them out explicitly to obtain an
effective action for the gauge fields on the 2D boundary∂M = T 2 ∪ T 2. It is a Maxwell-
Chern-Simons theory with two coupled gauge fieldsa+µ anda−µ corresponding to the two
copies ofT 2. In the long-wavelength limitq ≪ 1/L the two gauge fields become identified
a+µ = a−µ ≡ αµ and the level of the Chern-Simons term forαµ is the sum of that for the two
surfaces, i.e. it is integer.
The spatial manifold isM = T 2 × I with I = [−L/2, L/2]. We can always choose the
generators ofU(1)×U(1) to satisfytr tatb = δab, a, b = 1, 2. The action in imaginary time
is
S3D[aµ] =
∫
d4x
(
1
4g2faµνf
aµν −
iθ(z)
32π2ǫµνλρf
aµνf
aλρ
)
,
with faµν = ∂µa
aν − ∂νa
aµ, a = 1, 2 theU(1)× U(1) “statistical” field strength and
∂zθ =∑
η=±1
(2kη + 1)πδ(z − ηL/2),
APPENDIX A. THEORETICAL METHODS 203
with kη ∈ Z. To derive an effective 2D action on∂M = T 2 ∪ T 2, we introduce a Lagrange
multiplier which constrains the gauge field to live on∂M . Then we integrate out the bulk
gauge fieldaaµ. (The approach is essentially the same as that used in Sec. A.2 to derive an
effective action for thex = 0 degrees of freedom in the helical liquid.) In other words, we
introduce a “resolution of unity”
1 =
∫
D a+µ D a−µ δ[aaµ(x, L/2)− a+,a
µ (x)]δ[aaµ(x,−L/2)− a−,aµ (x)],
in the partition function, whereaη,aµ , η = ± are two auxiliary gauge fields which are defined
only on the 2D surfacex = (x0 = t, x1, x2), with a+,aµ living on the 2-torus atz = L/2 and
a−,aµ living on the 2-torus atz = −L/2. We represent the functional delta function as
∏
η
δ[aaµ(x, ηL/2)−aη,aµ (x)] =
∫
D j+µ D j−µ exp
(
i
∫
d3x jη,aµ (x)[aaµ(x, ηL/2)− aη,aµ (x)]
)
where jη,aµ is a Lagrange multiplier which implements the constraint that aη,aµ (x) =
aaµ(x, ηL/2). The idea is to integrate out firstaµ, and then the Lagrange multiplierjηµ,
to get an effective actionSeff2D in terms of the 2D gauge fieldsaηµ alone. In other words, the
partition function is
Z =
∫
D a+µ D a−µ D j+µ D j−µ Daµ
× exp
(
−S3D[aµ] + i
∫
d3x jη,aµ (x)[aaµ(x, ηL/2)− aη,aµ (x)]
)
≡∫
D a+µD a−µ e−Seff
2D[a+µ ,a−µ ].
We want to see whether these gauge fields will be identified or not, i.e. do we get two
Chern-Simons theories or only one?
First of all, since theθ-term is a total derivative it contributes only to the boundary
piece. Therefore if we writeSeff2D = SMax
2D + Sθ2D we immediately know that
Sθ2D = −i
∑
η
kη +12
4π
∫
d3x ǫµνλaη,aµ ∂ν a
η,aλ ,
APPENDIX A. THEORETICAL METHODS 204
i.e. we obtain two decoupled Chern-Simons terms of half-odd-integer levelkη + 12. To
computeSMax we need to integrate out the 3D bulk gauge fluctuations, whichwe can do
exactly in the Abelian case because the gauge bosons are noninteracting. We can gauge-fix
the Maxwell term in the usual way [248] by adding a12g2ξ
(∂µaaµ)
2 term to the Lagrangian.
In the Feynman gaugeξ = 1 the 3D gauge boson propagator is
Dabµν(k, kz) =
g2δabδµν
k2 + k2z, (A.72)
with k = (k0 = ω, k1, k2). Because the Lagrange multipliersjη,aµ (k) =∫
d3x e−ikν xν jη,aµ (x)
are independent ofkz, we have to sum over allkz to obtain an effective 2D propagator. The
effective 2D Maxwell propagator for the 2D gauge fieldsaη,aµ is therefore
Dab,ηη′
µν (k) = g2δabδµν1
L
∑
kz
e−ikz(η−η′)L/2
k2 + k2z,
with |k| =√
ω2 + k2. Because the fieldsaη,aµ must satisfy some sort of boundary condition
at z = ±L/2 (either Dirichlet or Neumann),kz is a discrete variable,kz = nπ/L, n ∈ Z.
Performing the discrete sum overkz, we obtain
Dabµν(q) =
g2δabδµν|q| sinh |q|L
(
cosh |q|L 1
1 cosh |q|L
)
,
where we denoteq ≡ k for simplicity. The inverse propagator is
[D−1]abµν(q) =1
g2Lδabδµν
|q|L|q| sinh |q|L
(
cosh |q|L −1
−1 cosh |q|L
)
.
To obtain the final form ofSeff2D, we need to integrate out the Lagrange multipliersjη,aµ which
simply amounts to inverting the2 × 2 matrix propagator (A.72). Since now all quantities
APPENDIX A. THEORETICAL METHODS 205
are 2D, we can drop all the tildes for simplicity, and obtain
Seff2D[a
+µ , a
−µ ] =
1
2g2Lz
∫
d3q
(2π)3|q|Lz
sinh |q|Lzδµν
×(
a+,aµ a−,a
µ
)
−q
(
cosh |q|Lz −1
−1 cosh |q|Lz
)(
a+,aν
a−,aν
)
q
− i∑
η
kη +12
4π
∫
d3x ǫµνλaη,aµ ∂νa
η,aλ , (A.73)
with q ≡ (ω,q) and|q| =√
ω2 + q2. We denotedLz ≡ L for clarity.
Consider fixing the scaling dimension of the gauge fields by the Chern-Simons term.
The latter is therefore marginal and contains one power ofq. The effective Maxwell term
in Eq. (A.73) contains higher powers of|q|. Let us expand it to quadratic order in|q|Lz,
Seff2D[a
+µ , a
−µ ] =
1
2g2Lz
∫
d3q
(2π)3δµν
×(
a+,aµ a−,a
µ
)
−q
(
1 + 13q2L2
z −1 + 16q2L2
z
−1 + 16q2L2
z 1 + 13q2L2
z
)(
a+,aν
a−,aν
)
q
− i∑
η
kη +12
4π
∫
d3x ǫµνλaη,aµ ∂νa
η,aλ .
We now consider the long-wavelength, low-energy limit|q| ≪ 1/Lz. In this limit, the
quadratic Maxwell termsq2L2z are irrelevant and the leading term is
Seff2D[a
+µ , a
−µ ] =
1
2g2Lz
∫
d3q
(2π)3δµν
(
a+,aµ a−,a
µ
)
−q
(
1 −1
−1 1
)(
a+,aν
a−,aν
)
q
=1
2g2Lz
∫
d3x (a+,aµ − a−,a
µ )2.
This term contains no derivatives ofaη,aµ but simply implements a constraint. The equations
APPENDIX A. THEORETICAL METHODS 206
of motion read
0 =δSeff
2D
δaa,+µ
=∂L
∂aa,+µ
= 2(aa,+µ − aa,−µ ),
0 =δSeff
2D
δaa,−µ
=∂L
∂aa,−µ
= −2(aa,+µ − aa,−µ ),
which imply thataa,+µ = aa,−µ and the gauge fields on the two 2-tori are identified. There-
fore, in the limit|q| ≪ 1/Lz the Chern-Simons term in Eq. (A.73) becomes
Sθ2D = −i
∑
η
kη +12
4π
∫
d3x ǫµνλaη,aµ ∂νa
η,aλ = −i k
4π
∫
d3x ǫµνλαaµ∂να
aλ,
with αaµ ≡ aa,+µ = aa,−µ andk =
∑
η(kη +12) = k+ + k− + 1 is the effective Chern-Simons
level, which is integer.
Bibliography
[1] L. D. Landau and E. M. Lifshitz.Statistical Physics. Pergamon Press, Oxford, 1980.
[2] K. von Klitzing, G. Dorda, and M. Pepper. New method for high-accuracy determi-
nation of the fine-structure constant based on quantized Hall resistance.Phys. Rev.
Lett., 45:494, 1980.
[3] D. C. Tsui, H. L. Stormer, and A. C. Gossard. Two-dimensional magnetotransport
in the extreme quantum limit.Phys. Rev. Lett., 48:1559, 1982.
[4] E. Abrahams, P. W. Anderson, D. C. Licciardello, and T. V.Ramakrishnan. Scaling
theory of localization: Absence of quantum diffusion in twodimensions.Phys. Rev.
Lett., 42:673, 1979.
[5] X. L. Qi, Y.-S. Wu, and S. C. Zhang. General theorem relating the bulk topological
number to edge states in two-dimensional insulators.Phys. Rev. B, 74:045125, 2006.
[6] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs. Quantized Hall
conductance in a two-dimensional periodic potential.Phys. Rev. Lett., 49:405, 1982.
[7] B. Simon. Holonomy, the quantum adiabatic theorem, and Berry’s phase.Phys. Rev.
Lett., 51:2167, 1983.
[8] S. C. Zhang, T. H. Hansson, and S. Kivelson. Effective-field-theory model for the
fractional quantum Hall effect.Phys. Rev. Lett., 62:82, 1989.
[9] S. C. Zhang. The Chern-Simons-Landau-Ginzburg theory of the fractional quantum
Hall effect. Int. J. Mod. Phys. B, 6:25, 1992.
207
BIBLIOGRAPHY 208
[10] V. Ambegaokar, B. I. Halperin, D. R. Nelson, and E. D. Siggia. Dynamics of super-
fluid films. Phys. Rev. B, 21:1806, 1980.
[11] M. P. A. Fisher and D.-H. Lee. Correspondence between two-dimensional bosons
and a bulk superconductor in a magnetic field.Phys. Rev. B, 39:2756, 1989.
[12] C. L. Kane and E. J. Mele.Z2 topological order and the quantum spin Hall effect.
Phys. Rev. Lett., 95:146802, 2005.
[13] C. L. Kane and E. J. Mele. Quantum spin Hall effect in graphene.Phys. Rev. Lett.,
95:226801, 2005.
[14] B. A. Bernevig and S. C. Zhang. Quantum spin Hall effect.Phys. Rev. Lett.,
96:106802, 2006.
[15] F. D. M. Haldane. Model for a quantum Hall effect withoutLandau levels:
Condensed-matter realization of the “parity anomaly”.Phys. Rev. Lett., 61:2015,
1988.
[16] X. L. Qi, T. L. Hughes, and S. C. Zhang. Topological field theory of time-reversal
invariant insulators.Phys. Rev. B, 78:195424, 2008.
[17] H. Min, J. E. Hill, N. A. Sinitsyn, B. R. Sahu, L. Kleinman, and A. H. MacDon-
ald. Intrinsic and Rashba spin-orbit interactions in graphene sheets.Phys. Rev. B,
74:165310, 2006.
[18] Y. Yao, F. Ye, X. L. Qi, S. C. Zhang, and Z. Fang. Spin-orbit gap of graphene:
First-principles calculations.Phys. Rev. B, 75:041401(R), 2007.
[19] B. A. Bernevig, T. L. Hughes, and S.C. Zhang. Quantum spin Hall effect and topo-
logical phase transition in HgTe quantum wells.Science, 314:1757, 2006.
[20] M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buhmann, L. W. Molenkamp, X. L.
Qi, and S. C. Zhang. Quantum spin Hall insulator state in HgTequantum wells.
Science, 318:766, 2007.
BIBLIOGRAPHY 209
[21] C. Wu, B. A. Bernevig, and S. C. Zhang. Helical liquid andthe edge of quantum
spin Hall systems.Phys. Rev. Lett., 96:106401, 2006.
[22] H. J. Zhang, C. X. Liu, X. L. Qi, X. Dai, Z. Fang, and S. C. Zhang. Topological
insulators in Bi2Se3, Bi2Te3 and Sb2Te3 with a single Dirac cone on the surface.
Nature Phys., 5:438, 2009.
[23] C. X. Liu, X. L. Qi, H. J. Zhang, X Dai, Z. Fang, and S. C. Zhang. Model Hamilto-
nian for topological insulators.Phys. Rev. B, 82:045122, 2010.
[24] Y. Xia, D. Qian, D. Hsieh, L. Wray, A. Pal, H. Lin, A. Bansil, D. Grauer, Y. S. Hor,
R. J. Cava, and M. Z. Hasan. Observation of a large-gap topological-insulator class
with a single Dirac cone on the surface.Nature Phys., 5:398, 2009.
[25] Y. L. Chen, J. G. Analytis, J.-H. Chu, Z. K. Liu, S.-K. Mo,X. L. Qi, H. J. Zhang,
D. H. Lu, X. Dai, Z. Fang, S. C. Zhang, I. R. Fisher, Z. Hussain,and Z.-X. Shen. Ex-
perimental realization of a three-dimensional topological insulator, Bi2Te3. Science,
325:178, 2009.
[26] D. Hsieh, Y. Xia, L. Wray, D. Qian, A. Pal, J. H. Dil, J. Osterwalder, F. Meier,
G. Bihlmayer, C. L. Kane, Y. S. Hor, R. J. Cava, and M. Z. Hasan.Observation of
unconventional quantum spin textures in topological insulators. Science, 323:919,
2009.
[27] L. Fu, C. L. Kane, and E. J. Mele. Topological insulatorsin three dimensions.Phys.
Rev. Lett., 98:106803, 2007.
[28] J. E. Moore and L. Balents. Topological invariants of time-reversal-invariant band
structures.Phys. Rev. B, 75:121306(R), 2007.
[29] R. Roy. Topological phases and the quantum spin Hall effect in three dimensions.
Phys. Rev. B, 79:195322, 2009.
[30] L. Fu and C. L. Kane. Topological insulators with inversion symmetry.Phys. Rev.
B, 76:045302, 2007.
BIBLIOGRAPHY 210
[31] D. Hsieh, D. Qian, L. Wray, Y. Xia, Y. S. Hor, R. J. Cava, and M. Z. Hasan. A
topological Dirac insulator in a quantum spin Hall phase.Nature, 452:970, 2008.
[32] F. Wilczek. Two applications of axion electrodynamics. Phys. Rev. Lett., 58:1799,
1987.
[33] R. D. Peccei and H. R. Quinn.CP conservation in the presence of pseudoparticles.
Phys. Rev. Lett., 38:1440, 1977.
[34] Z. Wang, X. L. Qi, and S. C. Zhang. Equivalent topological invariants of topological
insulators.New J. Phys., 12:065007, 2010.
[35] A. Karch. Electric-magnetic duality and topological insulators. Phys. Rev. Lett.,
103:171601, 2009.
[36] W.-K. Tse and A. H. MacDonald. Giant magneto-optical Kerr effect and univer-
sal Faraday effect in thin-film topological insulators.Phys. Rev. Lett., 105:057401,
2010.
[37] J. Maciejko, X. L. Qi, H. D. Drew, and S. C. Zhang. Topological quantization in
units of the fine structure constant.Phys. Rev. Lett., 105:166803, 2010.
[38] X. L. Qi, R. Li, J. Zang, and S. C. Zhang. Inducing a magnetic monopole with
topological surface states.Science, 323:1184, 2009.
[39] F. D. M. Haldane and L. Chen. Magnetic flux of “vortices” on the two-dimensional
Hall surface.Phys. Rev. Lett., 53:2591, 1984.
[40] R. Li, J. Wang, X. L. Qi, and S. C. Zhang. Dynamical axion field in topological
magnetic insulators.Nature Phys., 6:284, 2010.
[41] J. Wang, R. Li, S. C. Zhang, and X. L. Qi. Topological magnetic insulators with
corundum structure.e-print arXiv:1008.2666, 2010.
[42] X. G. Wen. Quantum Field Theory of Many-Body Systems: From the Origin of
Sound to an Origin of Light and Electrons. Oxford University Press, Oxford, 2004.
BIBLIOGRAPHY 211
[43] S. C. Zhang. Topological states of quantum matter.Physics, 1:6, 2008.
[44] C. Xu and J. Moore. Stability of the quantum spin Hall effect: effects of interactions,
disorder, andZ2 topology.Phys. Rev. B, 73:045322, 2006.
[45] C. W. J. Beenakker and H. van Houten. Quantum transport in semiconductor nanos-
tructures.Solid State Phys., 44:1, 1991.
[46] M. Buttiker. Absence of backscattering in the quantumHall effect in multiprobe
conductors.Phys. Rev. B, 38:9375, 1988.
[47] M. Buttiker. Four-terminal phase-coherent conductance. Phys. Rev. Lett., 57:1761,
1986.
[48] C. Brune, A. Roth, E. G. Novik, M. Konig, H. Buhmann, E.M. Hankiewicz,
W. Hanke, J. Sinova, and L. W. Molenkamp. Evidence for the ballistic intrinsic
spin Hall effect in HgTe nanostructures.Nature Phys., 6:448, 2010.
[49] E. M. Hankiewicz, L. W. Molenkamp, T. Jungwirth, and J. Sinova. Manifestation of
the spin Hall effect through charge-transport in the mesoscopic regime.Phys. Rev.
B, 70:241301(R), 2004.
[50] Y. Meir and N. S. Wingreen. Landauer formula for the current through an interacting
electron region.Phys. Rev. Lett., 68:2512, 1992.
[51] H. M. Pastawski. Classical and quantum transport from generalized Landauer-
Buttiker equations.Phys. Rev. B, 44:6329, 1991.
[52] H. M. Pastawski. Classical and quantum transport from generalized Landauer-
Buttiker equations. II. Time-dependent resonant tunneling. Phys. Rev. B, 46:4053,
1992.
[53] L. P. Kadanoff and G. Baym.Quantum Statistical Mechanics. Benjamin, Menlo
Park, 1962.
BIBLIOGRAPHY 212
[54] M. Konig, H. Buhmann, L. W. Molenkamp, T. L. Hughes, C. X. Liu, X. L. Qi, and
S. C. Zhang. The quantum spin Hall effect: theory and experiment. J. Phys. Soc.
Jpn, 77:031007, 2008.
[55] S. Sanvito, C. J. Lambert, J. H. Jefferson, and A. M. Bratkovsky. General Green’s-
function formalism for transport calculation with spd Hamiltonians and giant mag-
netoresistance in Co-and Ni-based magnetic multilayers.Phys. Rev. B, 59:11936,
1999.
[56] K. Kazymyrenko and X. Waintal. Knitting algorithm for calculating Green functions
in quantum systems.Phys. Rev. B, 77:115119, 2008.
[57] S. Datta.Electronic Transport in Mesoscopic Systems. Cambridge University Press,
Cambridge, 1995.
[58] M. Konig. Spin-related transport phenomena in HgTe-based quantum well struc-
tures. PhD thesis, University of Wurzburg, 2007.
[59] A. Furusaki and N. Nagaosa. Kondo effect in a Tomonaga-Luttinger liquid. Phys.
Rev. Lett., 72:892, 1994.
[60] M. Garst, P. Wolfle, L. Borda, J. von Delft, and L. Glazman. Energy-resolved in-
elastic electron scattering off a magnetic impurity.Phys. Rev. B, 72:205125, 2005.
[61] T. Giamarchi.Quantum Physics in One Dimension. Clarendon Press, Oxford, 2003.
[62] X. G. Wen. Edge transport properties of the fractional quantum Hall states and
weak-impurity scattering of a one-dimensional charge-density wave. Phys. Rev. B,
44:5708, 1991.
[63] D. Meidan and Y. Oreg. Multiple-particle scattering inquantum point contacts.Phys.
Rev. B, 72:121312(R), 2005.
[64] V. J. Emery and S. Kivelson. Mapping of the two-channel Kondo problem to a
resonant-level model.Phys. Rev. B, 46:10812, 1992.
BIBLIOGRAPHY 213
[65] D. L. Maslov. Transport through dirty Luttinger liquids connected to reservoirs.
Phys. Rev. B, 52:R14368, 1995.
[66] A. Schiller and K. Ingersent. Exact results for the Kondo effect in a Luttinger liquid.
Phys. Rev. B, 51:4676, 1995.
[67] D.-H. Lee and J. Toner. Kondo effect in a Luttinger liquid. Phys. Rev. Lett., 69:3378,
1992.
[68] P. Nozieres. A “Fermi-liquid” description of the Kondo problem at low temperatures.
J. Low Temp. Phys., 17:31, 1974.
[69] J. Goldstone and F. Wilczek. Fractional quantum numbers on solitons.Phys. Rev.
Lett., 47:986, 1981.
[70] X. L. Qi, T. L. Hughes, and S. C. Zhang. Fractional chargeand quantized current in
the quantum spin Hall state.Nature Phys., 4:273, 2008.
[71] C. L. Kane and M. P. A. Fisher. Transmission through barriers and resonant tunneling
in an interacting one-dimensional electron gas.Phys. Rev. B, 46:15233, 1992.
[72] C. L. Kane and M. P. A. Fisher. Transport in a one-channelLuttinger liquid. Phys.
Rev. Lett., 68:1220, 1992.
[73] T. Martin. Noise in mesoscopic physics.e-print arXiv:cond-mat/0501208, 2005.
[74] L. J. Cooper, N. K. Patel, V. Drouot, E. H. Linfield, D. A. Ritchie, and M. Pepper.
Resistance resonance induced by electron-hole hybridization in a strongly coupled
InAs/GaSb/AlSb heterostructure.Phys. Rev. B, 57:11915, 1998.
[75] C. X. Liu, T. L. Hughes, X. L. Qi, K. Wang, and S. C. Zhang. Quantum spin Hall
effect in inverted type-II semiconductors.Phys. Rev. Lett., 100:236601, 2008.
[76] I. Knez, R. R. Du, and G. Sullivan. Finite conductivity in mesoscopic Hall bars of
inverted InAs/GaSb quantum wells.Phys. Rev. B, 81:201301, 2010.
BIBLIOGRAPHY 214
[77] I. Knez, R. R. Du, and G. Sullivan. Evidence for helical edge modes in inverted
InAs/GaSb quantum wells.e-print arXiv:1105.0137, 2011.
[78] A. Roth, C. Brune, H. Buhmann, L. W. Molenkamp, J. Maciejko, X. L. Qi, and S. C.
Zhang. Nonlocal transport in the quantum spin Hall state.Science, 325:294, 2009.
[79] M. Buttiker. Edge-state physics without magnetic fields. Science, 325:278, 2009.
[80] G. Tkachov and E. M. Hankiewicz. Ballistic quantum spinHall state and enhanced
edge backscattering in strong magnetic fields.Phys. Rev. Lett., 104:166803, 2010.
[81] D. N. Sheng, Z. Y. Weng, L. Sheng, and F. D. M. Haldane. Quantum spin Hall effect
and topologically invariant Chern numbers.Phys. Rev. Lett., 97:036808, 2006.
[82] M. Onoda, Y. Avishai, and N. Nagaosa. Localization in a quantum spin Hall system.
Phys. Rev. Lett., 98:076802, 2007.
[83] H. Obuse, A. Furusaki, S. Ryu, and C. Mudry. Two-dimensional spin-filtered chiral
network model for theZ2 quantum spin Hall effect.Phys. Rev. B, 76:075301, 2007.
[84] J. Li, R.-L. Chu, J. K. Jain, and S.-Q. Shen. TopologicalAnderson insulator.Phys.
Rev. Lett., 102:136806, 2009.
[85] D. Li and J. Shi. Dorokhov-Mello-Pereyra-Kumar equation for the edge transport of
a quantum spin Hall insulator.Phys. Rev. B, 79:241303(R), 2009.
[86] J. Maciejko, C. X. Liu, Y. Oreg, X. L. Qi, C. Wu, and S. C. Zhang. Kondo effect in
the helical edge liquid of the quantum spin Hall state.Phys. Rev. Lett., 102:256803,
2009.
[87] D. G. Rothe, R. W. Reinthaler, C. X. Liu, L. W. Molenkamp,S. C. Zhang, and E. M.
Hankiewicz. Fingerprint of different spin-orbit terms forspin transport in HgTe
quantum wells.New J. Phys., 12:065012, 2010.
[88] R. Winkler. Spin-Orbit Coupling Effects in Two-Dimensional Electron and Hole
Systems. Springer-Verlag, Berlin, 2003.
BIBLIOGRAPHY 215
[89] D. K. Ferry and S. M. Goodnick.Transport in Nanostructures. Cambridge Univer-
sity Press, Cambridge, 1997.
[90] S. Y. Wu, J. Cocks, and C. S. Jayanthi. General recursiverelation for the calcula-
tion of the local Green’s function in the resolvent-matrix approach. Phys. Rev. B,
49:7957, 1994.
[91] M. P. Lopez Sancho, J. M. Lopez Sancho, and J. Rubio. Quick iterative scheme for
the calculation of transfer matrices: application to Mo(100). J. Phys. F, 14:1205,
1984.
[92] B. Zhou, H.-Z. Lu, R.-L. Chu, S.-Q. Shen, and Q. Niu. Finite size effects on helical
edge states in a quantum spin Hall system.Phys. Rev. Lett., 101:246807, 2008.
[93] P. A. Lee and T. V. Ramakrishnan. Disordered electronicsystems.Rev. Mod. Phys.,
57:287, 1985.
[94] G. Bergmann. Weak localization in thin films.Phys. Rep., 107:1, 1984.
[95] M. R. Zirnbauer. Super Fourier analysis and localization in disordered wires.Phys.
Rev. Lett., 69:1584, 1992.
[96] Y. Takane. Quantum electron transport in disordered wires with symplectic symme-
try. J. Phys. Soc. Jpn., 73:1430, 2004.
[97] Y. Takane. Conductance of disordered wires with symplectic symmetry: Compari-
son between odd- and even-channel cases.J. Phys. Soc. Jpn., 73:2366, 2004.
[98] I. Zutic, J. Fabian, and S. Das Sarma. Spintronics: Fundamentals and applications.
Rev. Mod. Phys., 76:323, 2004.
[99] Y. Aharonov and D. Bohm. Significance of electromagnetic potentials in the quan-
tum theory.Phys. Rev., 115:485, 1959.
[100] D. Frustaglia, M. Hentschel, and K. Richter. Quantum transport in nonuniform mag-
netic fields: Aharonov-Bohm as a spin switch.Phys. Rev. Lett., 87:256602, 2001.
BIBLIOGRAPHY 216
[101] M. Hentschel, H. Schomerus, D. Frustaglia, and K. Richter. Aharonov-Bohm
physics with spin. I. Geometric phases in one-dimensional ballistic rings. Phys.
Rev. B, 69:155326, 2004.
[102] D. Frustaglia, M. Hentschel, and K. Richter. Aharonov-Bohm physics with spin.
II. Spin-flip effects in two-dimensional ballistic systems. Phys. Rev. B, 69:155327,
2004.
[103] D. Frustaglia and K. Richter. Spin interference effects in ring conductors subject to
Rashba coupling.Phys. Rev. B, 69:235310, 2004.
[104] S. Murakami. Quantum spin Hall effect and enhanced magnetic response by spin-
orbit coupling.Phys. Rev. Lett., 97:236805, 2006.
[105] A. Shitade, H. Katsura, J. Kunes, X. L. Qi, S. C. Zhang,and N. Nagaosa. Quantum
spin Hall effect in a transition metal oxide Na2IrO3. Phys. Rev. Lett., 102:256403,
2009.
[106] A. D. Stone and A. Szafer. What is measured when you measure a resistance? –The
Landauer formula revisited.IBM J. Res. Dev., 32:384, 1988.
[107] J. C. Slonczewski. Conductance and exchange couplingof two ferromagnets sepa-
rated by a tunneling barrier.Phys. Rev. B, 39:6995, 1989.
[108] S. Datta and B. Das. Electronic analog of the electro-optic modulator.Appl. Phys.
Lett., 56:665, 1990.
[109] Y. Aharonov and A. Casher. Topological quantum effects for neutral particles.Phys.
Rev. Lett., 53:319, 1984.
[110] G. Usaj. Edge states interferometry and spin rotations in zigzag graphene nanorib-
bons.Phys. Rev. B, 80:081414(R), 2009.
[111] P. J. Mohr, B. N. Taylor, and D. B. Newell. CODATA recommended values of the
fundamental physical constants: 2006.Rev. Mod. Phys., 80:633, 2008.
BIBLIOGRAPHY 217
[112] T. H. Hansson, V. Oganesyan, and S. L. Sondhi. Superconductors are topologically
ordered.Ann. Phys., 313:497, 2004.
[113] X. L. Qi and S. C. Zhang. The quantum spin Hall effect andtopological insulators.
Phys. Today, 63:33, 2010.
[114] J. E. Moore. Topological insulators: The next generation. Nature Phys., 5:378, 2009.
[115] M. Z. Hasan and C. L. Kane. Colloquium: Topological insulators.Rev. Mod. Phys.,
82:3045, 2010.
[116] X. L. Qi and S. C. Zhang. Topological insulators and superconductors.e-print
arXiv:1008.2026, 2010.
[117] M.-C. Chang and M.-F. Yang. Optical signature of topological insulators.Phys. Rev.
B, 80:113304, 2009.
[118] A. D. LaForge, A. Frenzel, B. C. Pursley, T. Lin, X. Liu,J. Shi, and D. N. Basov.
Optical characterization of Bi2Se3 in a magnetic field: Infrared evidence for mag-
netoelectric coupling in a topological insulator material. Phys. Rev. B, 81:125120,
2010.
[119] N. P. Butch, K. Kirshenbaum, P. Syers, A. B. Sushkov, G.S. Jenkins, H. D. Drew,
and J. Paglione. Strong surface scattering in ultrahigh mobility Bi 2Se3 topological
insulator crystals.Phys. Rev. B, 81:241301(R), 2010.
[120] J. N. Hancock, private communication.
[121] J. Zhe, private communication.
[122] L. D. Landau and E. M. Lifshitz.Electrodynamics of Continuous Media, 2nd Edition.
Pergamon Press, Oxford, 1984.
[123] R. B. Laughlin. Anomalous quantum Hall effect: an incompressible quantum fluid
with fractionally charged excitations.Phys. Rev. Lett., 50:1395, 1983.
BIBLIOGRAPHY 218
[124] S. C. Zhang and J. P. Hu. A four-dimensional generalization of the quantum Hall
effect. Science, 294:823, 2001.
[125] B. A. Bernevig, C. H. Chern, J. P. Hu, N. Toumbas, and S. C. Zhang. Effective
field theory description of the higher dimensional quantum Hall liquid. Ann. Phys.,
300:185, 2002.
[126] A. M. Essin, J. E. Moore, and D. Vanderbilt. Magnetoelectric polarizability and
axion electrodynamics in crystalline insulators.Phys. Rev. Lett., 102:146805, 2009.
[127] M. Levin and A. Stern. Fractional topological insulators. Phys. Rev. Lett.,
103:196803, 2009.
[128] A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Ludwig.Classification of topo-
logical insulators and superconductors in three spatial dimensions. Phys. Rev. B,
78:195125, 2008.
[129] A. Kitaev. Periodic table for topological insulatorsand superconductors.AIP Conf.
Proc., 1134:22, 2009.
[130] J. K. Jain. Incompressible quantum Hall states.Phys. Rev. B, 40:8079, 1989.
[131] X. G. Wen. Edge excitations in the fractional quantum Hall states at general filling
fractions.Mod. Phys. Lett. B, 5:39, 1991.
[132] X. G. Wen. Non-Abelian statistics in the fractional quantum Hall states.Phys. Rev.
Lett., 66:802, 1991.
[133] B. Blok and X. G. Wen. Many-body systems with non-Abelian statistics.Nucl.
Phys. B, 374:615, 1992.
[134] X. G. Wen. Theory of the edge states in fractional quantum Hall effects.Int. J. Mod.
Phys. B, 6:1711, 1992.
[135] X. G. Wen. Projective construction of non-Abelian quantum Hall liquids.Phys. Rev.
B, 60:8827, 1999.
BIBLIOGRAPHY 219
[136] M. Barkeshli and X. G. Wen. Effective field theory and projective construction for
Zk parafermion fractional quantum Hall states.Phys. Rev. B, 81:155302, 2010.
[137] M. Levin and M. P. A. Fisher. Gapless layered three-dimensional fractional quantum
Hall states.Phys. Rev. B, 79:235315, 2009.
[138] S. L. Adler. Axial vector vertex in spinor electrodynamics. Phys. Rev., 177:2426,
1969.
[139] J. S. Bell and R. Jackiw. A PCAC puzzle:π0 → γγ in the sigma model.Nuovo
Cimento A, 60:47, 1969.
[140] E. Witten. OnS-duality in Abelian gauge theory.Sel. Math. New Ser., 1:383, 1995.
[141] A. Kapustin and E. Witten. Electric-magnetic dualityand the geometric Langlands
program.e-print arXiv:hep-th/0604151, 2006.
[142] S. Gukov and E. Witten. Gauge theory, ramification, andthe geometric Langlands
program.e-print arXiv:hep-th/0612073, 2006.
[143] E. Corrigan and D. Olive. Colour and magnetic monopoles. Nucl. Phys. B, 110:237,
1976.
[144] F. D. M. Haldane. Fractional quantization of the Hall effect: A hierarchy of incom-
pressible quantum fluid states.Phys. Rev. Lett., 51:605, 1983.
[145] B. I. Halperin. Statistics of quasiparticles and the hierarchy of fractional quantized
Hall states.Phys. Rev. Lett., 52:1583, 1984.
[146] Y. Ran, H. Yao, and A. Vishwanath. Composite fermions and quantum Hall stripes
on the topological insulator surface.e-print arXiv:1003.0901, 2010.
[147] E. Witten. Dyons of chargeeθ/2π. Phys. Lett. B, 86:283, 1979.
[148] G. Rosenberg and M. Franz. Witten effect in a crystalline topological insulator.
Phys. Rev. B, 82:035105, 2010.
BIBLIOGRAPHY 220
[149] X. G. Wen. Vacuum degeneracy of chiral spin states in compactified space.Phys.
Rev. B, 40:7387, 1989.
[150] X. G. Wen and Q. Niu. Ground-state degeneracy of the fractional quantum Hall
states in the presence of a random potential and on high-genus Riemann surfaces.
Phys. Rev. B, 41:9377, 1990.
[151] X. G. Wen and A. Zee. Topological degeneracy of quantumHall fluids. Phys. Rev.
B, 58:15717, 1998.
[152] M. Sato. Topological discrete algebra, ground-statedegeneracy, and quark confine-
ment in QCD.Phys. Rev. D, 77:045013, 2008.
[153] A. P. Schnyder, S., A. Furusaki, and A. W. W. Ludwig. Classification of topological
insulators and superconductors.AIP Conf. Proc., 1134:10, 2009.
[154] S. Ryu, A. P. Schnyder, A. Furusaki, and A. W. W. Ludwig.Topological insula-
tors and superconductors: tenfold way and dimensional hierarchy. New J. Phys.,
12:065010, 2010.
[155] A. B. Sushkov, G. S. Jenkins, D. C. Schmadel, N. P. Butch, J. Paglione, and H. D.
Drew. Far-infrared cyclotron resonance and Faraday effectin Bi2Se3. Phys. Rev. B,
82:125110, 2010.
[156] G. S. Jenkins, A. B. Sushkov, D. C. Schmadel, N. P. Butch, P. Syers, J. Paglione,
and H. D. Drew. Terahertz Kerr and reflectivity measurementson the topological
insulator Bi2Se3. Phys. Rev. B, 82:125120, 2010.
[157] R. Jackiw and C. Rebbi. Vacuum periodicity in a Yang-Mills quantum theory.Phys.
Rev. Lett., 37:172, 1976.
[158] C. G. Callan, R. F. Dashen, and D. J. Gross. The structure of the gauge theory
vacuum.Phys. Lett. B, 63:334, 1976.
[159] M. M. Vazifeh and M. Franz. Quantization and2π periodicity of the axion action in
topological insulators.Phys. Rev. B, 82:233103, 2010.
BIBLIOGRAPHY 221
[160] S. Raghu, S. B. Chung, X. L. Qi, and S. C. Zhang. Collective modes of a helical
liquid. Phys. Rev. Lett., 104:116401, 2010.
[161] L. Santos, T. Neupert, C. Chamon, and C. Mudry. Superconductivity on the sur-
face of topological insulators and in two-dimensional noncentrosymmetric materi-
als. Phys. Rev. B, 81:184502, 2010.
[162] A. Cortijo. A superconducting instability on the surface of a topological insulator.
e-print arXiv:1012.2008, 2010.
[163] C. Xu. Time-reversal symmetry breaking at the edge states of a three-dimensional
topological band insulator.Phys. Rev. B, 81:020411, 2010.
[164] C. Xu. Quantum critical points of helical Fermi liquids. Phys. Rev. B, 81:054403,
2010.
[165] K.-S. Kim and T. Takimoto. Nambu-Eliashberg theory for multi-scale quantum crit-
icality: Application to ferromagnetic quantum criticality on the surface of three-
dimensional topological insulators.e-print arXiv:1011.4851, 2010.
[166] P. Ghaemi and S. Ryu. Competing orders in the Dirac-like electronic structure and
the non-linear sigma model with topological terms.e-print arXiv:1012.5840, 2010.
[167] J.-H. Jiang and S. Wu. Spin susceptibility and helicalmagnetic order at the edge of
topological insulators due to Fermi surface nesting.e-print arXiv:1012.1299, 2010.
[168] Y. Zhang, Y. Ran, and A. Vishwanath. Topological insulators in three dimensions
from spontaneous symmetry breaking.Phys. Rev. B, 79:245331, 2009.
[169] M. Kargarian, J. Wen, and G. A. Fiete. Competing exotictopological insulator
phases in transition-metal oxides on the pyrochlore lattice with distortion.e-print
arXiv:1101.0007, 2011.
[170] M. Kurita, Y. Yamaji, and M. Imada. Topological insulators from spontaneous
symmetry breaking induced by electron correlation on pyrochlore lattices.e-print
arXiv:1012.0655, 2010.
BIBLIOGRAPHY 222
[171] Z. Wang, X. L. Qi, and S. C. Zhang. Topological order parameters for interacting
topological insulators.Phys. Rev. Lett., 105:256803, 2010.
[172] V. Gurarie. Single-particle Green’s functions and interacting topological insulators.
Phys. Rev. B, 83:085426, 2011.
[173] D. Pesin and L. Balents. Mott physics and band topologyin materials with strong
spin-orbit interaction.Nature Phys., 6:376, 2010.
[174] W. Witczak-Krempa, T. P. Choy, and Y. B. Kim. Gauge fieldfluctuations in three-
dimensional topological Mott insulators.Phys. Rev. B, 82:165122, 2010.
[175] X.-J. Liu, X. Liu, L. C. Kwek, and C. H. Oh. Fractional spin Hall effect in atomic
Bose gases.Phys. Rev. B, 79:165301, 2009.
[176] A. Karch, J. Maciejko, and T. Takayanagi. Holographicfractional topological insu-
lators in2 + 1 and1 + 1 dimensions.Phys. Rev. D, 82:126003, 2010.
[177] B. I. Halperin. Possible states for a three-dimensional electron gas in a strong mag-
netic field.Jpn. J. Appl. Phys. Suppl., 26:1913, 1987.
[178] M. Kohmoto, B. I. Halperin, and Y.-S. Wu. Diophantine equation for the three-
dimensional quantum Hall effect.Phys. Rev. B, 45:13488, 1992.
[179] L. Balents and M. P. A. Fisher. Chiral surface states inthe bulk quantum Hall effect.
Phys. Rev. Lett., 76:2782, 1996.
[180] J. Maciejko, X. L. Qi, A. Karch, and S. C. Zhang. Fractional topological insulators
in three dimensions.Phys. Rev. Lett., 105:246809, 2010.
[181] B. Swingle, M. Barkeshli, J. McGreevy, and T. Senthil.Correlated topological insu-
lators and the fractional magnetoelectric effect.e-print arXiv:1005.1076, 2010.
[182] K. G. Wilson. Confinement of quarks.Phys. Rev. D, 10:2445, 1974.
[183] A. M. Polyakov. Compact gauge fields and the infrared catastrophe.Phys. Lett. B,
59:82, 1975.
BIBLIOGRAPHY 223
[184] T. Banks, R. Myerson, and J. Kogut. Phase transitions in Abelian lattice gauge
theories.Nucl. Phys. B, 129:493, 1977.
[185] J. L. Cardy and E. Rabinovici. Phase structure ofZp models in the presence of aθ
parameter.Nucl. Phys. B, 205:1, 1982.
[186] J. L. Cardy. Duality and theθ parameter in Abelian lattice models.Nucl. Phys. B,
205:17, 1982.
[187] A. Ukawa, P. Windey, and A. H. Guth. Dual variables for lattice gauge theories and
the phase structure ofZ(N) systems.Phys. Rev. D, 21:1013, 1980.
[188] N. Seiberg. Supersymmetry and non-perturbative betafunctions. Phys. Lett. B,
206:75, 1988.
[189] D. D. Dietrich and F. Sannino. Walking in theSU(N). Phys. Rev. D, 75:085018,
2007.
[190] T. Banks and A. Zaks. On the phase structure of vector-like gauge theories with
massless fermions.Nucl. Phys. B, 196:189, 1982.
[191] T. DeGrand. Lattice studies of QCD-like theories withmany fermionic degrees of
freedom.e-print arXiv:1010.4741, 2010.
[192] N. Seiberg. Electric-magnetic duality in supersymmetric non-Abelian gauge theo-
ries. Nucl. Phys. B, 435:129, 1995.
[193] C. Hoyos-Badajoz, K. Jensen, and A. Karch. A holographic fractional topological
insulator.Phys. Rev. D, 82:086001, 2010.
[194] E. Vicari and H. Panagopoulos. Theta dependence ofSU(N) gauge theories in the
presence of a topological term.Phys. Rep., 470:93, 2009.
[195] E. Witten. Theta dependence in the largeN limit of four-dimensional gauge theories.
Phys. Rev. Lett., 81:2862, 1998.
BIBLIOGRAPHY 224
[196] E. Fradkin and S. H. Shenker. Phase diagrams of latticegauge theories with Higgs
fields. Phys. Rev. D, 19:3682, 1979.
[197] P. Hosur, S. Ryu, and A. Vishwanath. Chiral topological insulators, superconductors,
and other competing orders in three dimensions.Phys. Rev. B, 81:045120, 2010.
[198] S. Ryu, J. E. Moore, and A. W. W. Ludwig. Electromagnetic and gravitational
responses and anomalies in topological insulators and superconductors.e-print
arXiv:1010.0936, 2010.
[199] M. Mulligan. Interactions and the theta term in one-dimensional gapped systems.
e-print arXiv:1010.6094, 2010.
[200] G. ’t Hooft, inRecent Developments in Gauge Theories, edited by G. ’t Hooftet al.
(Plenum Press, New York, 1980).
[201] G. ’t Hooft. Magnetic charge quantization and fractionally charged quarks.Nucl.
Phys. B, 105:538, 1976.
[202] J. Preskill, inMagnetic Monopoles, edited by R. Carrigan and P. Trower (Plenum
Press, New York, 1983).
[203] G. ’t Hooft. On the phase transition towards permanentquark confinement.Nucl.
Phys. B, 138:1, 1978.
[204] F. Englert and P. Windey. Quantization condition for ’t Hooft monopoles in compact
simple Lie groups.Phys. Rev. D, 14:2728, 1976.
[205] P. Goddard, J. Nuyts, and D. I. Olive. Gauge theories and magnetic charge.Nucl.
Phys. B, 125:1, 1977.
[206] H. Georgi.Lie Algebras in Particle Physics, 2nd Edition. Perseus Books, Reading,
1999.
[207] M. Nakahara. Geometry, Topology and Physics. Institute of Physics Publishing,
London, 1990.
BIBLIOGRAPHY 225
[208] A. M. Polyakov. Thermal properties of gauge fields and quark liberation.Phys. Lett.
B, 72:477, 1978.
[209] E. Witten. Anti-de Sitter space, thermal phase transition, and confinement in gauge
theories.Adv. Theor. Math. Phys., 2:505–532, 1998.
[210] O. Aharony and E. Witten. Anti-de Sitter space and the center of the gauge group.
J. High Energy Phys., 11:018, 1998.
[211] S. Elitzur. Impossibility of spontaneously breakinglocal symmetries.Phys. Rev. D,
12:3978, 1975.
[212] M. Barkeshli and X. G. Wen. Phase transitions inZN gauge theory and twistedZN
topological phases.e-print arXiv:1012.2417, 2010.
[213] D. Gaiotto and E. Witten. Janus configurations, Chern-Simons couplings, and the
theta-angle inN = 4 super Yang-Mills theory.J. High Energy Phys., 06:097, 2010.
[214] K.-T. Chen and P. A. Lee. Topological insulator and theθ-vacuum in a system with-
out boundaries.e-print arXiv:1012.2084, 2010.
[215] E. Witten. Quantum field theory and the Jones polynomial. Commun. Math. Phys.,
121:351, 1989.
[216] E. P. Verlinde. Fusion rules and modular transformations in 2D conformal field
theory.Nucl. Phys. B, 300:360, 1988.
[217] R. Dijkgraaf and E. P. Verlinde. Modular invariance and the fusion algebra.Nucl.
Phys. Proc. Suppl. B, 5:87, 1988.
[218] E. Tang, J. W. Mei, and X. G. Wen. High temperature fractional quantum Hall states.
Phys. Rev. Lett., 106:236802, 2011.
[219] K. Sun, Z. C. Gu, H. Katsura, and S. Das Sarma. Nearly-flat bands with nontrivial
topology.Phys. Rev. Lett., 106:236803, 2011.
BIBLIOGRAPHY 226
[220] T. Neupert, L. Santos, C. Chamon, and C. Mudry. Fractional quantum Hall states at
zero magnetic field.Phys. Rev. Lett., 106:236804, 2011.
[221] D. N. Sheng, Z. C. Gu, K. Sun, and L. Sheng. Fractional quantum Hall effect in the
absence of Landau levels.Nature Commun., 2:389, 2011.
[222] N. Regnault and B. Andrei Bernevig. Fractional Chern insulator. e-print
arXiv:1105.4867, 2011.
[223] T. Neupert, L. Santos, S. Ryu, C. Chamon, and C. Mudry. Fractional topo-
logical liquids with time-reversal symmetry and their lattice realization.e-print
arXiv:1106.3989, 2011.
[224] M. O. Goerbig. From fractional Chern insulator to a fractional quantum spin Hall
effect.e-print arXiv:1107.1986, 2011.
[225] Y. Li and C. Wu. Three-dimensional topological insulators with Landau levels.e-
print arXiv:1103.5422, 2011.
[226] G. D. Mahan. Many-Particle Physics, 3rd Edition. World Scientific, New York,
2000.
[227] R. Shankar. Bosonization: how to make it work for you incondensed matter.Acta
Phys. Pol. B, 26:1835, 1995.
[228] N. Nagaosa. Quantum Field Theory in Strongly Correlated Electronic Systems.
Springer, Berlin, 1999.
[229] J. L. Cardy.Scaling and Renormalization in Statistical Physics. Cambridge Univer-
sity Press, Cambridge, 1996.
[230] P. Di Francesco, P. Mathieu, and D. Senechal.Conformal Field Theory. Springer,
New York, 1997.
[231] R. Shankar. Solvable model of a metal-insulator transition. Int. J. Mod. Phys. B,
4:2371, 1990.
BIBLIOGRAPHY 227
[232] D. L. Maslov and M. Stone. Landauer conductance of Luttinger liquids with leads.
Phys. Rev. B, 52:R5539, 1995.
[233] E.-A. Kim, private communication.
[234] A. Furusaki and N. Nagaosa. Single-barrier problem and Anderson localization in a
one-dimensional interacting electron system.Phys. Rev. B, 47:4631, 1993.
[235] A. O. Caldeira and A. J. Leggett. Quantum tunnelling ina dissipative system.Ann.
Phys. (N.Y.), 149:374, 1983.
[236] A. Schmid. Diffusion and localization in a dissipative quantum system.Phys. Rev.
Lett., 51:1506, 1983.
[237] M. P. A. Fisher and W. Zwerger. Quantum Brownian motionin a periodic potential.
Phys. Rev. B, 32:6190, 1985.
[238] U. Eckern, G. Schon, and V. Ambegaokar. Quantum dynamics of a superconducting
tunnel junction.Phys. Rev. B, 30:6419, 1984.
[239] M. P. A. Fisher. Quantum phase slips and superconductivity in granular films.Phys.
Rev. Lett., 57:885, 1986.
[240] G. Baym and N. D. Mermin. Determination of thermodynamic Green’s functions.
J. Math. Phys., 2:232, 1961.
[241] A. M. Tsvelik. Quantum Field Theory in Condensed Matter Physics. Cambridge
University Press, Cambridge, 1995.
[242] G. Yuval and P. W. Anderson. Exact results for the Kondoproblem: one-body theory
and extension to finite temperature.Phys. Rev. B, 1:1522, 1970.
[243] R. Landauer. Johnson-Nyquist noise derived from quantum mechanical transmis-
sion. Physica D, 38:226, 1989.
[244] J. Schwinger. Brownian motion of a quantum oscillator. J. Math. Phys., 2:407, 1961.
BIBLIOGRAPHY 228
[245] L. V. Keldysh. Diagram technique for nonequilibrium processes.Zh. Eksp. Teor. Fiz.,
47:1515, 1964 [Sov. Phys. JETP, 20:1018, 1965].
[246] R. de Picciotto, M. Reznikov, M. Heiblum, V. Umansky, G. Bunin, and D. Mahalu.
Direct observation of a fractional charge.Nature, 389:162, 1997.
[247] L. Saminadayar, D. C. Glattli, Y. Jin, and B. Etienne. Observation of thee/3 frac-
tionally charged Laughlin quasiparticle.Phys. Rev. Lett., 79:2526, 1997.
[248] N. Nagaosa.Quantum Field Theory in Condensed Matter Physics. Springer, Berlin,
1999.