Electronic Phenomena in 2D Dirac-like Systems:

Silicene and Topological Insulator Surface States

by

Calvin Jerome Tabert

A Thesis
presented to
The University of Guelph

In partial fulfilment of requirements

for the degree of
Doctor of Philosophy
in
Physics

Guelph, Ontario, Canada

⃝Calvin
c Jerome Tabert, November, 2015
 ABSTRACT

Electronic Phenomena in 2D Dirac-like Systems:

Silicene and Topological Insulator Surface States

Calvin Jerome Tabert Advisor:

University of Guelph, 2015 Professor Elisabeth J. Nicol

With the discovery of graphene, considerable attention has been given to two-
dimensional (2D) physics. Over the past few years, several new systems have been
synthesized. Of particular prominence are topologically nontrivial insulators. Here,
metallic boundary states exist between regions of different topology. While the quan-
tum Hall system has been known to exist since 1980, these systems display topological
behaviour even in the absence of a magnetic field; additionally, their bound states
carry a spin current as opposed to the electric current of the quantum Hall effect. In
this thesis, we explore several novel phenomena in 2D topological Dirac-like systems.
In particular, we consider a low-energy model for a buckled honeycomb lattice which
maps onto the Kane-Mele Hamiltonian; we also examine the Bychkov-Rashba Hamil-
tonian for the surface states of a three-dimensional (3D) topological insulator. In the
former system, we explore spin- and valley-polarized responses in the optical and Hall
conductivities with particular attention to the effects of sublattice asymmetry which
can be tuned by an external electric field. We also consider quantum oscillations and
electron-electron interactions. For the 2D Dirac-like surface states of a 3D topolog-
ical insulator, we restrict our attention to finite magnetic fields and emphasize the
particle-hole asymmetry which is manifest in the optical and Hall conductivities as
well as magnetic oscillations.
 ACKNOWLEDGEMENTS

First and foremost, I would like to thank my advisor, Elisabeth Nicol. Without
her insight and support, this project would not have been possible. Her commitment
to my professional development and academic growth has not only given me the
necessary skills to conduct physics research, but has also made this a very enjoyable
journey. I am particularly appreciative of the travelling opportunities which allowed
me to present my research at several international conferences, network with other
scientists and spend a month of intensive study at the Les Houches Summer School
in France.
I would also like to thank the members of my advisory committee: Jules Car-
botte, Bernie Nickel, Alex Gezerlis and Martin Williams. Their oversight, insight
and feedback were valued. In particular, I would like to thank Jules for his infectious
enthusiasm and for the opportunity to work closely together. In addition, I would like
to thank those who comprised my examination committee, namely Jules Carbotte,
Xiaorong Qin and Tamar Pereg-Barnea.
I must also acknowledge all my friends and family for their support and entertain-
ment. In particular, I owe a debt of gratitude to my wonderful wife for her patience,
understanding and encouragement throughout the duration of my studies.

iii
Contents

1 Introduction 1
1.1 Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Berry Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Landau Levels and the Quantum Hall Effect . . . . . . . . . . 5
1.1.3 TKNN Invariant and the Quantum Hall Effect . . . . . . . . . 8
1.2 Topological Insulators . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.1 Time-Reversal Symmetry and Kramers’ Theorem . . . . . . . 14
1.2.2 Edge States and Z2 Classification . . . . . . . . . . . . . . . . 15
1.3 3D Topological Insulators . . . . . . . . . . . . . . . . . . . . . . . . 17

2 Overview 20

3 Silicene: A Low-Buckled Honeycomb Lattice 22

3.1 Silicene . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.1 Edge States of a Zigzag Nanoribbon . . . . . . . . . . . . . . . 24
3.2 Low-Energy Band Structure . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 Density of States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Berry Phase and Winding Number . . . . . . . . . . . . . . . . . . . 36

4 Optical Conductivity of Silicene 39

4.1 Results for the Longitudinal Conductivity . . . . . . . . . . . . . . . 44
4.2 Response to Circularly-Polarized Light . . . . . . . . . . . . . . . . . 47
4.3 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

iv
5 Spin and Valley Hall Effects in Silicene 52
5.1 Spin and Valley Hall Conductivity . . . . . . . . . . . . . . . . . . . . 53
5.1.1 Charge Neutrality . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.1.2 Finite Chemical Potential . . . . . . . . . . . . . . . . . . . . 59

6 Silicene in a Magnetic Field 62

6.1 Landau Level Dispersion of Silicene . . . . . . . . . . . . . . . . . . . 64
6.2 Density of States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.3 Magneto-Optical Conductivity of Silicene . . . . . . . . . . . . . . . . 71
6.3.1 Longitudinal Conductivity . . . . . . . . . . . . . . . . . . . . 74
6.3.2 Transverse Hall Conductivity . . . . . . . . . . . . . . . . . . 81
6.3.3 Results for Circularly-Polarized Light . . . . . . . . . . . . . . 83
6.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

7 Magnetic Oscillations in Silicene 85

7.1 The de Haas-van Alphen Effect . . . . . . . . . . . . . . . . . . . . . 86
7.2 Magnetization of Silicene and the Hall Effect . . . . . . . . . . . . . . 92
7.3 Magnetization and Magnetic Susceptibility . . . . . . . . . . . . . . . 102
7.3.1 The Zero-Temperature, Clean limit . . . . . . . . . . . . . . . 102
7.3.2 Impurity and Finite Temperature Effects . . . . . . . . . . . . 108
7.4 Magnetic Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.4.1 Impurity and Finite Temperature Effects . . . . . . . . . . . . 117

8 Screening in the Random Phase Approximation 119

8.1 Electron-Electron Interactions . . . . . . . . . . . . . . . . . . . . . . 119
8.1.1 Polarization Function . . . . . . . . . . . . . . . . . . . . . . . 121
8.1.2 Plasmons and the Dielectric Function . . . . . . . . . . . . . . 124
8.2 Electron-Electron Interactions in Silicene . . . . . . . . . . . . . . . . 127
8.3 Loss Function and Plasmons . . . . . . . . . . . . . . . . . . . . . . . 135
8.4 Screening of a Charged Impurity . . . . . . . . . . . . . . . . . . . . . 141

v
9 Magnetic Properties of Topological Insulator Surface States 143
9.1 Surface State Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . 143
9.2 Topological Insulators in a Magnetic Field . . . . . . . . . . . . . . . 146
9.2.1 Density of States . . . . . . . . . . . . . . . . . . . . . . . . . 150
9.3 Magneto-Optical Conductivity . . . . . . . . . . . . . . . . . . . . . . 153
9.4 Magnetization of the Metallic Surface States . . . . . . . . . . . . . . 159
9.4.1 Magnetization and Hall Conductivity . . . . . . . . . . . . . . 159
9.4.2 Magnetic Oscillations . . . . . . . . . . . . . . . . . . . . . . . 162

10 Concluding Remarks 167

A Relativistic Fermions in 2D 168

A.1 Graphene: A Monolayer of Carbon Atoms . . . . . . . . . . . . . . . 168
A.2 Optical Conductivity of Graphene . . . . . . . . . . . . . . . . . . . . 172

B Massless Dirac Fermions in a Magnetic Field 178

B.1 Relativistic LLs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
B.2 Quantum Hall Effect in Graphene . . . . . . . . . . . . . . . . . . . . 184

C Analytic Solution for the Optical Conductivity of Silicene 188

C.1 Longitudinal Conductivity . . . . . . . . . . . . . . . . . . . . . . . . 188
C.2 Transverse Hall Conductivity . . . . . . . . . . . . . . . . . . . . . . 192

D Landau Level Dispersion of Silicene 194

D.1 Energy Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
D.2 Wave-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

E Topological Insulators in a Magnetic Field 201

E.1 Surface States in a Magnetic Field . . . . . . . . . . . . . . . . . . . . 201
E.1.1 Landau levels . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
E.1.2 Wave-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 205
E.2 Analytic Solution for Magnetic Oscillations . . . . . . . . . . . . . . . 207

vi
Bibliography 214

vii
List of Figures

1.1 (a) Classical representation of electrons in a magnetic field moving in

circular orbits. (b) Semiclassical representation of the Landau levels
proposed by Halperin. . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 (a) Idealized view of the band structure of a 2D topological insulator.
(b) Schematic of the helical edge channels which exist at the boundary
of a topological insulator. . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 1D Kramers’ pairs of edge states for (a) topologically trivial and (b)
topologically nontrivial systems. . . . . . . . . . . . . . . . . . . . . 16
1.4 (a) Idealized view of the helical 2D Dirac cone at the surface of a 3D
topological insulator. (b) Schematic of the spin polarization in real
space. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5 Schematic kx = 0 cuts of the bulk and surface bands of (a) Bi2 Se3 and
(b) Bi2 Te3 . A cross sectional cut of the energy states for (c) Bi2 Se3
and (d) Bi2 Te3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3.1 The crystal structure of silicene. . . . . . . . . . . . . . . . . . . . . 23

3.2 (left) Structure of a nanoribbon of the honeycomb lattice with zigzag
edges along the x direction. (right) Next-nearest-neighbour hopping
on the honeycomb lattice. . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Edge states of a zigzag (a) graphene nanoribbon, (b) QSHI with ∆z =
0, buckled honeycomb lattice in the (c) topological (∆so > ∆z ) and (d)
trivial (∆z > ∆so ) phases. . . . . . . . . . . . . . . . . . . . . . . . . 30

viii
3.4 The low-energy band structure of silicene in the presence of an on-site
potential difference between sublattices. . . . . . . . . . . . . . . . . 32
3.5 Schematic representation of the bulk band structure evolution at the
K point as ∆z is increased. . . . . . . . . . . . . . . . . . . . . . . . 33
3.6 The low-energy electronic density of states of silicene. . . . . . . . . 35
3.7 The location of the energy gaps of silicene as a function of electric field
strength. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.1 The low-energy band structure of graphene about a single K point and
the allowed optical transitions. . . . . . . . . . . . . . . . . . . . . . 40
4.2 Longitudinal conductivity of charge-neutral silicene for various electric
fields. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3 Longitudinal conductivity of silicene with finite chemical potential. . 46
4.4 Drude weight as a function of electric field. . . . . . . . . . . . . . . 47
4.5 Circularly-polarized response of silicene in the TI regime. . . . . . . 48
4.6 Circularly-polarized response of silicene in the BI regime. . . . . . . 49

5.1 Schematic of the spin Hall effect. . . . . . . . . . . . . . . . . . . . . 52

5.2 Charge-neutral spin and valley Hall conductivities of a Kane-Mele
QSHI. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.3 Charge-neutral spin Hall conductivity of silicene for finite ∆z . . . . . 55
5.4 Charge-neutral valley Hall conductivity of silicene for finite ∆z . . . . 56
5.5 Charge-neutral (a) spin and (b) valley Hall conductivities of silicene
for varying ∆z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.6 (a) Spin and (b) valley Hall conductivity of silicene for finite chemical
potential. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

6.1 (a) The relativistic Landau level dispersion. (b) The Landau level
evolution as a function of B. . . . . . . . . . . . . . . . . . . . . . . 63
6.2 Schematic representation of the Landau levels in the TI (upper frame)
and BI (lower frame) regimes about the K and K ′ points. . . . . . . 67

ix
6.3 Landau level energies as a function of magnetic field for ∆z = 0.5∆so . 67
6.4 Total electronic density of states (left frames) and the spin dependent
density of states (right frames). . . . . . . . . . . . . . . . . . . . . . 69
6.5 Integrated density of states for the TI and VSPM regimes. . . . . . . 70
6.6 Real part of the longitudinal magneto-optical conductivity of silicene
for varying ∆z and µ = 0. The results are vertically offset by 15 units. 75
6.7 Real part of the longitudinal conductivity of the Kane-Mele model for
a 2D QSHI (∆z = 0) for varying chemical potential. . . . . . . . . . 75
6.8 Real part of the longitudinal conductivity of silicene for varying chem-
ical potential in the (a) TI, (b) VSPM and (c) BI regimes. The results
are vertically offset by (a) 10 units, (b) 16 and (c) 9 units. . . . . . . 77
6.9 Schematic of the allowed transitions between LLs at both valleys in
the TI regime for various values of chemical potential. . . . . . . . . 78
6.10 Schematic of the optical transitions which give rise to a quartet of
valley-spin-polarized absorption lines. . . . . . . . . . . . . . . . . . 80
6.11 Imaginary part of the transverse Hall conductivity of silicene for vary-
ing chemical potential. These results are vertically offset by 10 units. 82
6.12 The absorptive part of the conductivity response to (a) left-handed
and (b) right-handed circularly-polarized light. . . . . . . . . . . . . 83

7.1 (a) k-space orbit of an electron in a magnetic field. (b) Area between
two orbits of different energies bounded by k1 and k2 . . . . . . . . . 88
7.2 Magnetization as a function of µ in the TI phase and VSPM regimes
for gs = 0. Inset: the slope of the magnetization as a function of µ.
Plateaus have been offset from their integer values for clarity. . . . . 96
7.3 The slope of the magnetization which is related to the Hall conductiv-
ity. The charge, spin and valley contributions are given for TI, VSPM
and BI regimes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
7.4 The slope of the magnetization for the TI with Zeeman splitting. The
charge, spin and valley contributions are emphasized. . . . . . . . . . 100

x
7.5 Vacuum contribution to the magnetization as a function of (a) B and
(b) ∆z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.6 (a) Vacuum contribution to the magnetic susceptibility as a function
of ∆z . (b) Vacuum plus regular parts of the susceptibility as a function
of µ for various ∆z . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.7 (a) The effect of finite temperature and impurity scattering on the
susceptibility as a function of ∆z . (b) The effect of finite chemical
potential on the susceptibility. . . . . . . . . . . . . . . . . . . . . . 111
7.8 Schematic illustration of the possible cyclotron orbits in silicene. . . 113
7.9 (a) The cyclotron orbit area as a function of µ for µ between the two
gaps. (b) The two cyclotron orbit areas for µ above both gaps. . . . 114
7.10 Magnetic oscillations as a function of 1/B for varying µ and ∆z . . . 116

8.1 The applicable Feynman diagrams. . . . . . . . . . . . . . . . . . . . 120

8.2 The renormalized Coulomb interaction. . . . . . . . . . . . . . . . . 120
8.3 Bubble diagram: irreducible self-energy in the random phase approxi-
mation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
8.4 Particle-hole excitations from the Fermi sea. . . . . . . . . . . . . . . 124
8.5 Friedel oscillations in the electron density in the presence of a charged
impurity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
8.6 Regions for the (left) K ↑ /K ′ ↓ and (right) K ↓ /K ′ ↑ bands in which
the polarization function has different expressions. . . . . . . . . . . 131
8.7 Imaginary part of the silicene polarization function. . . . . . . . . . 132
8.8 Real part of the silicene polarization function. . . . . . . . . . . . . . 133
8.9 Constant frequency cuts of the real part of the polarization function
as a function of q. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
8.10 Static polarization of silicene. . . . . . . . . . . . . . . . . . . . . . . 134
8.11 Energy loss function of silicene for α = 0.8, ∆so /µ = 1.4 and varying
∆z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

xi
8.12 Energy loss function of silicene for α = 0.8, ∆so /µ = 0.75 and varying
∆z . Inset: the location of the chemical potential relative to the band
structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.13 (a) Frequency cuts of 2παReΠ(ω, q)/µ for α = 0.8, ∆so /µ = 0.375
and ∆z /∆so = 1. The intersection of 2παReΠ(ω, q)/µ with the line of
unit slope (orange) gives the solutions to Eqn. (8.43). (b) Plasmon fre-
quency as a function of q corresponding to the solutions of Eqn. (8.43).
(c) Frequency cuts of the loss function. (d) Zoomed in version of (c). 138
8.14 Low-energy plasmon dispersion of silicene for varying ∆z . . . . . . . 139
8.15 Screening potential of a charged impurity in silicene. . . . . . . . . . 142

9.1 (a) Comparison of ideal-Dirac cones and those of a TI. (b) Pure Schrödinger
band compared to that of the Bychkov-Rashba Hamiltonian. . . . . 144
9.2 Surface states of (a) a TI thin film and (b) TI with magnetic dopants
on one surface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
9.3 Comparison between the (a) non-gapped and (b) gapped band struc-
tures of TI surface states. . . . . . . . . . . . . . . . . . . . . . . . . 147
9.4 LL dispersion for a topological insulator with (a) ∆ = 0, (b) (1 +
g)E0 /2 > ∆ and (c) ∆ > (1 + g)E0 /2. . . . . . . . . . . . . . . . . . 149
9.5 Slope of the magnetization for gapped graphene at the (top) K point
and (middle) K ′ point. (bottom) The full result. . . . . . . . . . . . 150
9.6 Integrated density of states for graphene at the K point and a TI (a)
without and (b) with a gap of ∆ = 0.5 meV. . . . . . . . . . . . . . 151
9.7 ∆ = 0 LLs in (a) graphene and (b) a TI with and without Zeeman
splitting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
9.8 µ = 0 longitudinal conductivity in a TI for varying ∆. . . . . . . . . 155
9.9 Longitudinal conductivity in (a) gapless graphene, and (b) a gapless
and (c) gapped TI for positive and negative µ. . . . . . . . . . . . . 156
9.10 The absorptive part of the transverse Hall conductivity for a TI with
parameters set to match Fig. 9.9(c). . . . . . . . . . . . . . . . . . . 158

xii
9.11 Conductivity response to (a) right- and (b) left-handed circularly-
polarized light. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
9.12 (a) µ dependence of the magnetization for a TI with varying ∆. (b)
The corresponding slope of the magnetization. . . . . . . . . . . . . 161

A.1 Graphene lattice with emphasized triangular sublattices on the A and

B sites. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
A.2 First Brillouin zone of the honeycomb lattice. . . . . . . . . . . . . . 170
A.3 Low-energy dispersion for graphene over and beyond the first Brillouin
zone. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
A.4 (a) Low-energy band structure of graphene and the allowed optical
transitions. (b) Optical conductivity of graphene. . . . . . . . . . . . 177

B.1 (a) The LL dispersion in graphene. (b) The LL evolution as a function

of B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
B.2 The IQHE in graphene. . . . . . . . . . . . . . . . . . . . . . . . . . 187

xiii
 CHAPTER I

INTRODUCTION

1.1 Topology
The electronic band structure of a solid gives the energy states which can be occupied
by an electron of a given momentum. A gap in the energy dispersion (referred to as
the band gap) represents a region of energies where no electron states can exist.
The band gap typically refers to the energy difference between the lowest part of
the conduction band and the top of the valence band. The energy range of the
gap corresponds to the energy required to free an electron in the outer shell of an
atom and allow it to propagate through the solid. Systems which display such a gap
are typically classified as insulators or semiconductors depending on the size of the
gap. Semiconductors form an important class which play an integral role in modern
electronics. While the presence of a band gap is a characterizing feature of such
systems, it becomes important to arrange the systems into topologically equivalent
groups. For example, in the presence of a magnetic field, the electronic band structure
of a two dimensional electron gas (2DEG) is known to form discrete Landau levels (see
Sec. 1.1.2) with a Landau level (LL) spacing of ℏωc . Thus, the system exhibits a band
gap and can be thought of as a form of semiconductor; however, a nonzero transverse
conductivity is measured when the chemical potential is in between levels [1]. Not
only is there a finite conductivity but it is also quantized in integer values of e2 /h.
This phenomenon is known as the integer quantum Hall effect (IQHE) and was the
first indication that insulators and semiconductors require another sub-classification.
This is based on the topology of the bands.
For our purposes, topology can be thought of as the study of equivalence. A

1
standard example is that of a malleable sphere and a torus. One can image deforming
the sphere into the shape of a bowl. One could also image transforming the torus
into a coffee cup where the hole in the torus becomes the handle of the cup. One can
not, however, turn the sphere into a torus without puncturing a hole in it. Thus, the
sphere and bowl are said to be in the same topological class while the torus and cup
are in a separate class. Mathematically, the two classes are distinguished by their
genus (the number of holes in the manifold).
The idea of topological equivalence can be extended to band structures. In regards
to band theory, topological equivalence means that the band structure of one system
may be tuned to that of another system without closing the gap [2, 3]. While an
atomic insulator (electrons bound to atoms in closed shells) and a semiconducting
system appear to be quite different, the Hamiltonian can be tuned to turn the band
structure of one into the other without closing the band gap [2]. Dirac’s relativistic
quantum theory states that the vacuum may also be described by a gapped band
structure [2] (i.e. a conduction band for electrons, a valence band for positrons and
an energy gap for pair production [2]). Thus, the atomic insulator, semiconductor
and vacuum all belong to the same topological class. The quantum Hall system,
however, is topologically different. Indeed, there exist several systems which differ
topologically from the vacuum.
How can these different topologies be classified? It turns out that each band has
an associated integer topological invariant n known as the Chern invariant [2,3]. This
allows for a Z classification of the different topological phases. This Chern invariant
must be a property of the Hamiltonian for it to effectively classify the systems. Indeed,
it can be related to the Berry phase of the Bloch wave-functions.

1.1.1 Berry Phase

Before, deriving an expression for the Chern invariant and showing how it explains
the IQHE, we must establish the theory of Berry phase [4,5]. Consider a Hamiltonian
described by a set of time-dependent parameters R(t) = (R1 , R2 , ...). We introduce

2
the instantaneous eigenstate |n[R(t)]⟩ such that

Ĥ[R(t)] |n[R(t)]⟩ = ϵn [R(t)] |n[R(t)]⟩ . (1.1)

Let us suppose that R(t) changes adiabatically from its value at t = 0. According
to the quantum adiabatic theorem, a system in some state |n[R(0)]⟩ will remain an
instantaneous eigenstate of Ĥ[R(t)] if the evolution is sufficiently slow in time. The
system can, however, pick up a phase θ(t). The total wave-function at t is

|Ψn (t)⟩ = e−iθ(t) |n[R(t)]⟩ . (1.2)

The time evolution of the system is governed by the Schrödinger equation:

d
Ĥ[R(t)] |Ψ(t)⟩ = iℏ |Ψ(t)⟩ . (1.3)
dt

In terms of the instantaneous eigenstates, we have

dθ d
En [R(t)] |n[R(t)]⟩ = ℏ |n[R(t)]⟩ + iℏ |n[R(t)]⟩ . (1.4)
dt dt

Assuming orthonormality, we obtain

d dθ
En [R(t)] − iℏ ⟨n[R(t)]| |n[R(t)]⟩ = ℏ , (1.5)
dt dt

which implies
∫ t ∫ t
1 ′ ′ d
θ(t) = En [R(t )]dt − i ⟨n[R(t′ )]| |n[R(t′ )]⟩ dt′ . (1.6)
ℏ 0 0 dt ′

The first term of Eqn. (1.6) is the familiar dynamical phase and will not be of interest.
The second term is the Berry phase [4]. It is convenient to remove the time dependence
(using d/dt′ = ∇R · dR/dt′ ) and express it as
∫
γn = −i ⟨n(R)| ∇R |n(R)⟩ · dR, (1.7)
C

3
where C specifies a path in parameter space from R(0) to R(t). The total wave-
functions may be written as

∫t
En [R(t′ )]dt′ −iγn
|Ψ(t)⟩ = e− ℏ
i
0 e |n[R(t)]⟩ . (1.8)

From this, we conclude that, during the adiabatic evolution, the state picks up an
addition phase to the usual dynamical phase.
It is convenient to define a quantity known as the Berry connection,

An (R) = −i ⟨n(R)| ∇R |n(R)⟩ , (1.9)

which allows the Berry phase to be written as

∫
γn = dR · An (R). (1.10)
C

The Berry connection is gauge-dependent. Under a gauge transformation

|n(R)⟩ → eiξ(R) |n(R)⟩ , (1.11)

for some smooth, single-valued ξ(R), An (R) transforms in the usual way:

∂
An (R) → An (R) − ξ(R). (1.12)
∂R

As such, γn is changed by
∫
∂
− ξ(R) · dR = ξ[R(0)] − ξ[R(T )], (1.13)
C ∂R

where T is a long time after C is completed. Many people assumed that ξ(R) could
be chosen so that it would cancel γn , thus, making the Berry phase irrelevant [5]. A
problem in this argument is apparent if we consider a closed path C, such that R(0) =
R(T ). Due to the single-valued nature of the eigenstates, |n[R(T )]⟩ = |n[R(0)]⟩. This

4
property must be maintained by a gauge transformation. We require

eiξ[R(0)] |n[R(0)]⟩ = eiξ[R(T )] |n[R(T )]⟩ , (1.14)

which is only true if ξ[R(T )] − ξ[R(0)] = 2πm for m ∈ Z. So, under a closed
path, γn can only be cancelled if it is also an integer multiple of 2π. In general, γn
cannot be gauged away. As γn is gauge-invariant up to an integer multiple of 2π,
exp(iγn ) is absolutely gauge-invariant and may be related to physical observables. In
Appendix A, we work out the famous π Berry phase of graphene.

1.1.2 Landau Levels and the Quantum Hall Effect

To understand the IQHE, the next (and more fundamental) concept which must be
established is the LL dispersion. It is well known, that a particle in a magnetic field
will experience a Lorentz force in a direction perpendicular to the magnetic field and
the particles direction [6]. Thus, the electrons in a 2DEG will undergo circular motion
when a perpendicular B-field is applied. To understand the effect this will have on
the electrons’ energy, consider the kinetic part of the free electron Hamiltonian, that
is,

p̂2
Ĥ = . (1.15)
2m

For simplicity, assume that the electrons are confined to the xy plane and the magnetic
field is applied in the z direction. The B-field will have an associated electromagnetic
vector potential A given by B = ∇ × A. Choosing to work in the Landau gauge, we
may write the vector potential as A = (−By, 0, 0). We then express the mechanical
momentum in terms of the canonical momentum and electromagnetic vector potential
via the Peierls substitution [7]:

p̂ → p̂ − qA/c. (1.16)

5
The Hamiltonian becomes
( )2
1 eBy p̂2y
Ĥ = p̂x − + . (1.17)
2m c 2m

We can see that the spacial variable x does not appear in the Hamiltonian. Since
[y, p̂x ] = 0 and [p̂x , p̂y ] = 0, we have [Ĥ, p̂x ] = 0 which means that px is a good quantum
number. Let us define px ≡ ℏk so p̂x Ψ = ℏkΨ. Making the usual substitution
p̂x → −iℏ∂x , we can express our wave-functions as Ψ(x, y) = exp[ikx]ϕ(y). Therefore,
we have

p̂2
ikx y
+ (ℏk − eBy/c)2
ĤΨ = e ϕ(y)
[ 2m
( )2 ]
p̂2
y e 2 2
B ℏkc
= eikx + y− ϕ(y)
2m 2mc2 eB
[ 2 ]
p̂y mωc2
=e ikx
+ (y − y0 ) ϕ(y),
2
(1.18)
2m 2

where ωc = eB/(mc) and y0 = ℏkc/(eB). We recognize that Eqn. (1.18) has the
form of the simple harmonic oscillator Hamiltonian with a shifted origin y0 . This has
the well known eigenvalues En = ℏωc (n + 1/2). Therefore, the energy dispersion of
a 2DEG in a perpendicular magnetic field is given by discrete levels spaced by ℏωc .
These levels are the LLs. From a classical perspective, we know that the Lorentz
force law will cause electrons in a magnetic field to experience circular motion around
the magnetic field lines. Thus, a simple harmonic oscillator interpretation seems
reasonable as electrons would precess around in the 2D plane. This is schematically
illustrated in Fig. 1.1(a). If we consider a finite sized 2DEG, one can imagine the
electrons at the edge trying to move in circular orbits and bouncing around the edge.
This gives rise to a finite edge current which characterizes the IQHE. We note that,
in this classical interpretation, the electron cannot move in the opposite direction
along the edge due to the Lorentz force law. This alludes to the idea of a topological
difference between the 2DEG in a B-field and the vacuum.
As argued by Halperin [8], if we look at the energy dispersion as a function of y0

6
(or equivalently k as y0 ∝ k), we obtain the results shown in Fig. 1.1(b). Here we

Classically Semi-Classically
ε
B

n=3
n=2
µ
n=1
n=0

y
(a) (b)

Figure 1.1: (a) Classical representation of electrons in a magnetic field moving in

circular orbits. (b) Semiclassical representation of the Landau levels proposed by
Halperin.

see flat LLs in the bulk which are equally spaced by ℏωc . Since the electrons cannot
move into the vacuum, there must be a confining potential at the edge which causes
the LLs to move toward infinite energy. If the chemical potential is between two LLs
in the bulk, at the edge, it will cut through the number of occupied levels [8]. It turns
out that the number of occupied levels is exactly equal to the integer quantization of
the Hall effect.
It is also useful to know the degeneracy of each LL. Recall that the momentum k
in the x direction is a good quantum number. Therefore, along the x axis, the system
can be described by a particle in a box with allowed momentum values k = 2πn/Lx
where n ∈ Z and Lx is the length of the system in the x direction. The maximum
value of k is 2πN/Lx , where N is the number of states at a given energy. We have
y0 = ℏkc/(eB) for the center of oscillation of an electron. The maximum value of y0
which will keep the electron in the sample is Ly (the length of the system in the y

7
direction). Combining these two results, we have

ℏc 2πN
Ly = , (1.19)
eB Lx

which determines the degeneracy for a system of size Lx × Ly . This gives

N eB eB
n≡ = = (1.20)
Lx Ly 2πℏc hc

for the LL degeneracy per unit area.

1.1.3 TKNN Invariant and the Quantum Hall Effect

Aided by the discussion of Secs. 1.1.1 and 1.1.2, let us now develop an understanding
of the IQHE and its relation to topology. In 1980 at the high magnetic field laboratory
in Grenoble, Klaus von Klitzing made the shocking experimental discovery that the
Hall conductivity was exactly quantized in integer multiples of e2 /h [1] (recently, the
quantization has been measured to 1 part in 109 [9]). Five years later, von Klitzing
was awarded the Noble Prize for this beautiful result. The unexpected finding led
theorists to question why such robust quantization was observed independent of the
sample geometry. In 1982, Thouless, Kohmoto, Nightingale and den Nijs showed
that the quantization of the Hall conductivity is related to a topological invariant
(the TKNN invariant) and thus depends on the systems intrinsic band structure [10].
To derive the TKNN invariant, we follow the method of Ref. [3].
Let us calculate the Hall conductivity of a 2D electron system of dimension L × L
confined to the xy plane. We place the system in a magnetic field of strength B
oriented in the z direction and apply an electric field in the y direction. We treat
the E-field as a perturbation potential V = −eEy and employ perturbation theory
to obtain the eigenstates of the perturbed system. As we are working with a weak
E-field, we will be interested in the linear response. The new eigenstates are found

8
in the standard way and are given by

∑ ⟨m| (−eEy) |n⟩

|n⟩E = |n⟩ + |m⟩ + · · · , (1.21)
E n − Em
n̸=m

where En is the energy of the nth LL. We now use these eigenstates to calculate the
expectation value of the longitudinal current density jx = evx /L2 , where vx is the
velocity of the electron in the x direction. That is,

∑ ( ev )
x
⟨jx ⟩E = f (En ) ⟨n|E 2
|n⟩E , (1.22)
n
L

where f (x) is the Fermi distribution function. To first order in E, this becomes
( )
1 ∑ ⟨n| evx |m⟩ ⟨m| (−eEy) |n⟩ + ⟨n| (−eEy) |m⟩ ⟨m| evx |n⟩
⟨jx ⟩E = 2 f (En ) ,
L n En − Em
m̸=n

(1.23)

where we take ⟨n| jx |n⟩ = 0 so that there is no current without the external field. To
solve this, we employ the Heisenberg equation of motion which states

dy 1
= [y, Ĥ], (1.24)
dt iℏ

where dy/dt ≡ vy . Using this, we note

1
⟨m| vy |n⟩ = (En − Em ) ⟨m| y |n⟩ . (1.25)
iℏ

Thus, the current density can be written as

( )
−iℏe2 E ∑ ⟨n| vx |m⟩ ⟨m| vy |n⟩ − ⟨n| vy |m⟩ ⟨m| vx |n⟩
⟨jx ⟩E = f (En ) . (1.26)
L2 n (En − Em )2
m̸=n

9
The Hall conductivity is, therefore, given by
( )
⟨jx ⟩E −iℏe2 ∑ ⟨n| vx |m⟩ ⟨m| vy |n⟩ − ⟨n| vy |m⟩ ⟨m| vx |n⟩
σxy = = f (En ) .
E L2 n (En − Em )2
m̸=n

(1.27)

We now assume that, in the absence of E, the system is in a periodic potential so its
eigenstates may be expressed as the Bloch states |unk ⟩. To proceed, note

1 ∂ Ĥ
⟨unk | vµ |umk′ ⟩ = ⟨unk | |umk′ ⟩ , (1.28)
ℏ ∂kµ

which, by employing the product rule, can be written as

1 ∂ ( ) 1 ∂
⟨unk | Ĥ |umk′ ⟩ − ⟨unk | Ĥ |umk′ ⟩ . (1.29)
ℏ ∂kµ ℏ ∂kµ

We can also rewrite the first term using the same procedure to obtain

1 ∂ ( ) 1 ∂Emk′ 1 ∂
⟨unk | Ĥ |umk ⟩ = ⟨unk |
′ |umk′ ⟩ + ⟨unk | Emk′ |umk′ ⟩ . (1.30)
ℏ ∂kµ ℏ ∂kµ ℏ ∂kµ

It is clear that the first term of the right hand side is zero as the sum is restricted to
m ̸= n. Therefore,

1 ∂
⟨unk | vµ |umk′ ⟩ = (Emk′ − Enk ) ⟨unk | |umk′ ⟩ . (1.31)
ℏ ∂kµ

Using this identity, we find

{
ie2 ∑ ∂ ∂
σxy = f (Enk ) ⟨unk | |umk′ ⟩ ⟨umk′ | |unk ⟩
ℏL 2
′
∂kx ∂ky
n,k,k
m̸=n
}
∂ ∂
− ⟨unk | |umk′ ⟩ ⟨umk′ | |unk ⟩ . (1.32)
∂ky ∂kx

Next, let us examine the first term in the above expression. It can be expanded

10
to give

∑ ∂ ∂
⟨unk | |umk′ ⟩ ⟨umk′ | |unk ⟩ =
m,k′
∂kx ∂ky
∑[ ∂ (
∂
) ]
∂
⟨unk | umk′ ⟩ − ⟨unk | |umk′ ⟩ ⟨umk′ | |unk ⟩ . (1.33)
m,k′
∂kx ∂kx ∂ky

Again, the first term on the right hand side is zero by the restriction m ̸= n. Con-
ducting the sums over m and k′ , we obtain
( )
∂ ∂
− ⟨unk | |unk ⟩ =
∂kx ∂ky
∂ ∂ ∂2
− ⟨unk | |unk ⟩ + ⟨unk | |unk ⟩ . (1.34)
∂kx ∂ky ∂kx ∂ky

The second term of Eqn. (1.32), is found by interchanging kx and ky in the above
derivation. Therefore,
[ ]
ie2 ∑ ∂ ∂ ∂ ∂
σxy = f (Enk ) − ⟨unk | |unk ⟩ + ⟨unk | |unk ⟩ . (1.35)
ℏL2 n,k ∂kx ∂ky ∂ky ∂kx

While this does not appear to be a tractable expression for the Hall conductivity, we
recall that the Berry connection is given by [see Eqn. (1.9)]

An (k) = −i ⟨unk | ∇k |unk ⟩ . (1.36)

Therefore, in the continuum limit,

∫
e2 ∑ d2 k
σxy = f (Enk )∇k × An (k). (1.37)
ℏ n BZ (2π)2

11
We define the Chern number
∫
d2 k
νn = ∇k × An (k)
BZ 2π
I
dk
= · An (k)
∂BZ 2π
1
= γn , (1.38)
2π

where ∂BZ represents the Brillouin zone boundary. Due to the single valued nature
of the wave-functions, the change in phase (γn ) after encircling the Brillouin zone can
only be an integer multiple of 2π. Therefore, γn = 2πn for n ∈ Z. Thus, νn must be
an integer (the Chern number of the nth level) and

e2 ∑
σxy = νn . (1.39)
h n∈occ.

Equation (1.39) shows that the Hall conductivity is quantized in integer multiples of
e2 /h where the Z quantization is given by summing up the Berry phases (scaled by
2π) of all the occupied LLs. This integer is the TKNN invariant.

1.2 Topological Insulators

The topological edge states observed in the IQHE are a beautiful example of the
importance topology plays in condensed matter physics. This interest in topological
classifications of matter caused physicists to search for other theoretical models which
could give rise to robust edge states. In 1988, Haldane proposed a model based on
the 2D honeycomb lattice which included a complex next-nearest-neighbour hopping
amplitude [11]. The complex nature of the hopping manifests itself in a phase which
is governed by the Aharonov-Bohm effect [12, 13]. The Aharonov-Bohm effect states
that if one moves a particle around a closed contour, the phase difference between the
initial and final states is proportional to the magnetic flux enclosed by the loop. Thus,
in the absence of an external magnetic field, a complex hopping integral is possible in
systems which have some sort of intrinsic magnetic background. Therefore, such an

12
effect will only be present in a system which breaks time-reversal (TR) symmetry. In
the Haldane model, nonzero Chern numbers are possible. This means that, while the
system is insulating in the bulk, a gapless spectrum of chiral edge states (i.e. charge
currents) will appear at the boundary. The number of conducting channels is equal
to the sum of the Chern numbers of the occupied bands. This effect is referred to
as the quantum anomalous Hall effect. It is anomalous because, unlike the IQHE, a
quantized Hall conductance is found in the absence of an external magnetic field.
In 2005, Kane and Mele proposed a model for graphene (carbon atoms bonded
together on a 2D honeycomb lattice) with intrinsic spin-orbit coupling which opens
a gap in the band structure [14]. This model is essentially two copies of the Haldane
Hamiltonian with a different sign of the mass term for the two spin species. The
Kane-Mele model preserves TR symmetry and has an overall Chern number of zero;
however, the spin-up (-down) bands have Chern number +1(−1) which results in
two counter-propagating edge states of opposite spin [14]. Therefore, while there is
no charge current circling the edge, there is a pair of helical edge modes (i.e. spin-
up propagating one way while spin-down moves in the other direction). This gives
rise to a spin current at the boundary between different topological systems. This
system is referred to as a quantum spin Hall insulator (QSHI) and represents a 2D
topological insulator. Figure. 1.2(a) gives a schematic representation of the band
structure of a general 2D topological insulator at the edge. The bulk valence and
conduction bands are gapped. If the chemical potential µ sits in the gap, there
is no charge conduction. However, the 1D edge states can each carry a quantum
of conductance. Here the blue band represents spin up while the red band is for
spin down. The helical edge channels are illustrated in Fig. 1.2(b). The spin-up
(blue) and spin-down (red) channels are confined to move either clockwise or counter-
clockwise around the sample. A spin-up fermion cannot backscatter into the counter-
propagating channel without a spin flip. Therefore, these edge currents are robust
against TR symmetry preserving impurities. Recently, signatures of helical edge
conduction have been found in Hgx Cd1−x Te quantum wells of thickness dc > 6.3 nm
where a longitudinal conductance of 2e2 /h was measured when the chemical potential

13
Figure 1.2: (a) Idealized view of the band structure of a 2D topological insulator. (b)
Schematic of the helical edge channels which exist at the boundary of a topological
insulator.

sat in the bulk band gap [15, 16]. This quantization was independent of the sample
width which provides evidence for gapless edge states. We now wish to explore the
robustness of such edge states.

1.2.1 Time-Reversal Symmetry and Kramers’ Theorem

The term “topological insulator” is typically reserved for systems with topologically
protected boundary states which have a gap in the bulk band structure and preserve
TR symmetry. The operator that represents TR symmetry is antiunitary and written
as Θ̂ =exp(iπ Ŝy /ℏ)K̂, where Ŝy is the spin operator and K̂ is the complex conjugation
operator [2, 3]. Generally, Θ̂ has the property that Θ̂2 = (−)2S , where S is the spin
of the particles. Therefore, for fermions, Θ̂2 = −1. This constraint gives rise to
Kramers’ theorem which states that all eigenstates of a TR invariant Hamiltonian
are at least two-fold degenerate [2, 3]. The proof of this is given by contradiction.
For a TR invariant Hamiltonian, [Ĥ, Θ̂] = 0 meaning we can find states which are
simultaneously eigenstates of both the Hamiltonian and the TR operator. This implies
that |χ⟩ and Θ̂ |χ⟩ must have the same energy. Let us assume that |χ⟩ and Θ̂ |χ⟩ give
the same state (i.e. there is no degeneracy). We have |χ⟩ = Θ̂ |χ⟩. This implies that
Θ̂ |χ⟩ = Θ̂2 |χ⟩. To satisfy our assumption, we require Θ̂2 = +1. For fermions, Θ̂2 =
−1. Therefore, a non-degenerate eigenstate is not permitted. For trivial insulators,

14
this simply gives the two-fold spin degeneracy of energy bands.

1.2.2 Edge States and Z2 Classification

For systems with spin-orbit interactions and nontrivial topology, we saw in the pre-
vious section that two gapless counter-propagating helical edge states occur at the
boundary. Kramers’ theorem dictates that the spin-up band, described by the energy
dispersion ε↑ (k), must have a TR partner of opposite spin described by ε↓ (−k) =
ε↑ (k). Here, we have used the fact that TR flips the spin and sends k to −k. There-
fore, at the points on the edge Brillouin zone where k = −k, the bands must cross.
These points of crossing are referred to as time-reversal invariant momenta (TRIMs)
and the TR partners are known as Kramers’ pairs. For a 2D system, the TRIMs of
the 1D edge Brillouin are k = 0, ±π/a, where we note that ±π/a describe the same
momentum.
The existence of these edge states does not make the system a topological insula-
tor. In Fig. 1.3, plots of two different TR symmetric Kramers’ pairs are given over
the 1D edge Brillouin zone (á la Fig. 6 of Ref. [3]). Note that the two halves of the
Brillouin zone ([−π/a, 0] and [0, π/a]) are symmetric under a change in spin. The
left frame shows a topologically trivial insulator as one could imagine pushing the
edge states upward into the bulk conduction band and thus removing all the edge
states. The right frame shows a topologically nontrivial insulator as there is no way
to remove the edge states without cutting the bands. This state is the topological
insulator. Indeed, for any Fermi energy in the bulk gap, if the edge states cross that
energy an even number of times in the half Brillouin zone, the system is necessar-
ily trivial. If there is an odd number of crossings, the edge states are topologically
protected. The 2D bulk continuum states are given by the orange and green bands.
Let us now discuss the topological edge states in greater detail. It is important to
note that these helical modes are protected by TR symmetry (i.e. Kramers’ theorem).
Thus, as long as the system preserves this symmetry, they will be present. They
are also robust against backscattering. For a single pair of edge states, one can not
scatter an electron from the spin-up band to the counter-propagating spin-down band

15
 Topologically Topologically
Trivial Nontrivial
ε ε

k k
-π/a π/a -π/a π/a

(a) (b)

Figure 1.3: 1D Kramers’ pairs of edge states for (a) topologically trivial and (b)
topologically nontrivial systems.

without the use of a magnetic impurity. If there were three pairs of helical edge states,
backscattering could occur between two of the pairs. This would destroy those edge
channels. Thus, there can only be one pair of helical edge states which is robust and
topologically protected.
While the total Chern number is 0 for a topological insulator, it has been shown
that another index can be used to distinguish the topological insulator from the trivial
band insulator [17]. In essence, this index counts whether there is an even (trivial)
or odd (topological) number of band crossings at the edge. As there are only two
possibilities, this index is a Z2 classification which can take the values ν = 0, 1 for
the trivial and nontrivial phases, respectively.
In Chapter 3, we will introduce a specific model for a 2D topological insulator
which maps onto the Kane-Mele Hamiltonian for graphene with intrinsic spin-orbit
coupling [14].

16
1.3 3D Topological Insulators
In the preceding discussion, we limited our attention to 2D systems. Before the Z2
topology was confirmed in 2D, theorists realized that a topological classification of
the insulating phase could be extended to 3D [2, 3]. In fact, the term “topological
insulator” was first introduced in reference to the 3D system [18]. Here, four Z2
invariants are required to characterize the topological class [18–20]. These invariants
are typically written in the form (ν0 ; ν1 ν2 ν3 ). A 3D topological insulator is “strong”
if ν0 = 1 and “weak” if ν0 = 0 and νi = 1 for some i = 1, 2, 3. A “weak” topological
insulator is one in which the surface states are not protected by TR symmetry (ex.
a stack of 2D topological insulators) and can, thus, be localized by disorder. The
“strong” class have TR symmetric surface states. Such insulators are marked by
an odd number of 2D Dirac-like helical states per surface. The first predicted 3D
topological insulator was Bi1−x Sbx [19] which was later confirmed by experiment
[21–24]. The bulk band gap necessary for this to be a topological insulator occurs for
x between 0.09 and 0.23 [25]. Due to the prominent spin-polarized surface states of
Bi (which arise from strong Rashba effects on the surface [26]), this material houses a
complicated surface-state structure [24, 27] and is, therefore, not useful for a detailed
study of the topological surface states. It also has a relatively small band gap of
∼ 30 meV [25] requiring low temperatures to see 2D transport. Further calculations
showed that Bi2 Se3 , Bi2 Te3 and Sb2 Te3 should also be 3D topological insulators [28].
In all three of these systems, the four Z2 invariants are (1;000). This means that the
2D helical Dirac surface states are centred at the Γ point of the hexagonal surface
Brillouin zone [2, 3]. An idealized view of this band structure is shown in Fig. 1.4(a)
(modelled after Fig. 1(d) of Ref. [3]). Aside from the simple surface band structure,
these so-called “second generation” topological insulators are promising candidates
for experimental probing as they are stoichiometric (i.e. pure compounds, not alloys)
making them easier to prepare at higher purity. They also have larger band gaps
(∼ 0.3 eV for Bi2 Se3 and ∼ 0.17 eV Bi2 Te3 ); this allows the topological properties to
be seen at room temperature [3]. It is this generation of topological insulators which

17
 ε

Figure 1.4: (a) Idealized view of the helical 2D Dirac cone at the surface of a 3D
topological insulator. (b) Schematic of the spin polarization in real space.

will be of interest for this thesis.

The surface states of Bi2 Se3 are nearly perfect Dirac cones with a slight curvature
due to a Schrödinger mass contribution. Bi2 Te3 , however, has a more complicated
structure with significantly more curvature (see the kx = 0 cuts in Figs. 1.5(a) and
(b) which are modelled after Fig. 8 of Ref. [3]). Most importantly, the Dirac point sits
below the top of the valance band. This means that the transport properties of the
surface states are difficult to probe for energies near the Dirac point as bulk carriers
will also be involved. There is also significantly more hexagonal warping in Bi2 Te3
than Bi2 Se3 . That is, the energy contours of Bi2 Se3 surface states are nearly circular
while those of Bi2 Te3 are strongly warped [see Figs. 1.5(c) and (d)]. An important
feature of the Dirac states is a Berry phase of π. This means that the real spin rotates
once around the Fermi surface and remains locked perpendicular to the momentum;
thus, backscattering is strongly suppressed. Indeed, the two characterizing features
of topological surface states are spin-momentum locking and a Berry phase of π [2, 3]
(seen in Fig. 1.4). While the single Dirac cone seems to violate the fermion doubling
theorem [29] (which states that Dirac cones come in pairs for TR invariant systems),
it is important to note that the partner cone exists on the opposite surface [2]. In

18
 ε
Bi2Se3
ky
vF
Spin

Γ k
kx

(a) (c)
ky=0
ky
ε
Bi2Te3
ky

vF
Spin

Γ k kx

(b) (d)
ky=0
ky

Figure 1.5: Schematic kx = 0 cuts of the bulk and surface bands of (a) Bi2 Se3 and
(b) Bi2 Te3 . A cross sectional cut of the energy states for (c) Bi2 Se3 and (d) Bi2 Te3 .

Chapter 9, we will introduce the model Hamiltonian which we use to probe the novel
physics. A detailed discussion of topological insulators can be found in Refs. [2, 3]
Throughout this thesis, our primary interest will be on low-energy Dirac physics in
condensed matter systems based on the graphene Hamiltonian. For an introduction
to graphene, the reader is referred to Appendices A and B.

19
 CHAPTER II

OVERVIEW

With the brief introduction to topological insulators presented in the preceding chap-
ter, we are ready to explore the interesting physics of several systems. Before pro-
ceeding, a brief outline of the material is presented:
In Chapter 3, we introduce the 2D model of a buckled-honeycomb lattice (ex.
silicene).
Chapter 4 contains work on the optical conductivity of silicene which was done
in collaboration with Lukas Stille (an undergraduate research assistant) and can be
found in Ref. [30].
[30] L. Stille, C.J. Tabert and E.J. Nicol, Phys. Rev. B 86, 195405 (2012).

In Chapter 5, the work of Chapter 4 is extended to the spin and valley Hall
conductivities and reflects that which is published in Ref. [31].
[31] C.J. Tabert and E.J. Nicol, Phys. Rev. B 87, 235426 (2013).

The results of Chapter 6 are contained in Refs. [32, 33] and relate to the optical
conductivity of silicene in a magnetic field.
[32] C.J. Tabert and E.J. Nicol, Phys. Rev. Lett. 110, 197402 (2013).
[33] C.J. Tabert and E.J. Nicol, Phys. Rev. B 88, 085434 (2013).

In Chapter 7, we discuss the magnetic response of silicene with particular attention

to the quantization of the Hall conductivity, the magnetic susceptibility and the
quantum oscillations present in the magnetization. This summarizes the results of
Ref. [34].

20
 [34] C.J. Tabert, J.P. Carbotte and E.J. Nicol, Phys. Rev. B 91, 035423 (2015).

In Chapter 8 we present the results of Ref. [35] where we include electron-electron

interactions and examine the dynamical polarization function, plasmon dispersion
and the screening of a charged impurity.
[35] C.J. Tabert and E.J. Nicol, Phys. Rev. B 89, 195410 (2014).

We then look at the magnetization, Hall conductivity and magneto-optical re-

sponse of 3D topological insulator surface states in Chapter 9. The combined results
of this chapter can be found in Refs. [36, 37].
[36] C.J. Tabert and J.P.Carbotte, J. Phys.: Condens. Matter 27, 015008 (2015).
[37] C.J. Tabert and J.P.Carbotte, Phys. Rev. B 91, 235405 (2015).

Conclusions follow in Chapter 10.

The remainder of this thesis consists of five useful appendices which contain an in-
troduction to graphene (Appendices A and B) as well as several pertinent derivations
(Appendices C-E) which were used to obtain the results in this thesis.
In all cases, we are interested in fixing the chemical potential which can be ac-
complished by back gating [38]. In particular, for a varying magnetic field, we allow
the electron number to change as we hold the chemical potential at a given value.

21
 CHAPTER III

SILICENE: A LOW-BUCKLED
HONEYCOMB LATTICE

In this thesis, our primary focus will be on low-energy phenomena of Dirac fermions
in buckled honeycomb lattices with intrinsic spin-orbit coupling. That is, we shall
examine systems which map onto the graphene-like Kane-Mele Hamiltonian for the
QSHI (see Appendix A for a review of the low-energy Hamiltonian of graphene). We
will also give particular attention to a sublattice symmetry breaking term. Recently, a
monolayer of silicon atoms which bond on a 2D honeycomb lattice (dubbed silicene),
has been predicted to map onto the Kane-Mele Hamiltonian [39]. As carbon and
silicon reside in the same column on the periodic table of elements, there are many
similarities between graphene and silicene. For definitiveness, we shall refer primarily
to silicene throughout this thesis; although, the results presented herein should apply
to other low-buckled honeycomb lattices (ex. germanene).

3.1 Silicene
Since its first experimental isolation in 2010 [40,41], silicene has garnered considerable
theoretical attention. Like graphene, the silicon atoms bond on a 2D honeycomb
lattice with a reduced nearest-neighbour hopping integral t ∼ 1.6 eV [42] compared
to the carbon equivalent [43,44] (t ∼ 2.8 eV); the larger ionic size of silicon causes the
lattice to buckle due to an sp2 -sp3 hybridization as opposed to the sp2 hybridization
of graphene [39, 45–47]. This causes the A and B sublattices to sit in vertical planes
separated by d ≈ 0.46 Å [39,48]. This low-buckled lattice is illustrated [49] in Fig. 3.1.
Currently, silicene has only been synthesized on metallic substrates [40, 41, 50–53];

22
 B
A

Figure 3.1: The crystal structure of silicene.

although, efforts are under way to obtain the monolayer on an insulating surface.
In early 2015, the first silicene field-effect transistor was reported [54]. The device
was fabricated by synthesizing silicene on a Ag(111) thin film, covering the exposed
monolayer with an Al2 O3 layer, depositing the structure on a SiO2 substrate (silver
film up) and then carving away the silver layer to leave behind two electrodes [54]. The
room temperature mobility of the sample was a mere ∼ 100 cm2 V−1 s−1 and the sample
remained stable for only a couple of minutes; however, this exciting experimental
realization demonstrates that silicene is a feasible 2D nanomaterial [54].
As a result of the buckled lattice, an external perpendicular electric field (Ez )
creates an on-site potential difference (∆z = Ez d) between the A and B sublattices.
Silicene is also expected to have a stronger intrinsic spin-orbit gap than is seen in
graphene [55] with values predicted to be ∆so ≈ 1.55 meV by density functional
theory calculations [39, 45, 46] and ∆so ≈ 7.9 meV in tight-binding calculations [46].
These values can be increased under strain [45, 46]. This larger spin-orbit interaction
makes silicene susceptible to spin manipulation. The resulting band gap near the
two valleys K and K ′ (≡ −K) of the first Brillouin zone provides a “mass” to the
Dirac electrons that can be controlled by the strength of ∆z [39, 48, 56]. Silicene
has also been predicted [39, 56] to undergo a transition from a topological insulator
(TI) to a trivial band insulator (BI) as the strength of ∆z becomes greater than ∆so .

23
At the critical value of ∆z = ∆so (Ez ≈ 17.4 meV/Å), there exists a spin-polarized
Dirac point at each valley; this phase is referred to as a valley-spin-polarized metal
(VSPM). These qualities are also predicted for a monolayer of germanium which is
isostructural to silicene but is expected to exhibit a much stronger spin-orbit band
gap of ∆so ≈ 24−93 meV from first principles [45] and tight-binding calculations [46].

3.1.1 Edge States of a Zigzag Nanoribbon

To explicitly see the topological transition which occurs as ∆z becomes greater than
∆so , we need to look at the edge states of a silicene nanoribbon. For definitiveness,
we will use the geometry shown in the left frame of Fig. 3.2 (i.e. we shall restrict our
attention to zigzag edges). Here, we take the ribbon to be infinite in the x direction
Unit Cell

NA m=N
NB

m=N-1

m=N-2

4A
4B m=4
a2 a1
3A m=3
δ3 δ2 3B
δ1 +
νij= -
2A
2B m=2
y
1A
1B m=1
x

Figure 3.2: (left) Structure of a nanoribbon of the honeycomb lattice with zigzag edges
along the x direction. (right) Next-nearest-neighbour hopping on the honeycomb
lattice.

but to have a finite width of N sites in the y direction. The unit cell is given by the

24
dashed box. The lattice vectors which make up this configuration are
( √ ) ( √ )
a a 3 a a 3
a1 = , , a2 = − , . (3.1)
2 2 2 2

The nearest-neighbour vectors are

a1 + a2
δ1 = − ,
3
2 1
δ2 = δ1 + a1 = a1 − a2 , (3.2)
3 3
1 2
δ3 = δ1 + a2 = − a1 + a2 .
3 3

The Kane-Mele Hamiltonian on which our model is based is given by the tight-binding
expression [14]

∑ ∆so ∑ ∆z ∑ †
Ĥ = −t c†i cj + i √ νij c†i σz cj + ξi ci ci , (3.3)
6 3 ⟨⟨i,j⟩⟩ 2 i
⟨i,j⟩

where c†i (ci ) is the creation (annihilation) operator written in second quantization
notation. The first term is the usual graphene piece which describes nearest-neighbour
hopping with hopping integral t and ⟨i, j⟩ implies summation over nearest neighbours.
The second term accounts for a complex hopping amplitude between next-nearest-
neighbours (⟨⟨i, j⟩⟩) where ∆so is the spin-orbit gap, νij = −νji = ± depending
on whether the hopping is counter-clockwise or clockwise (right frame of Fig. 3.2),
respectively, and σz is the Pauli spin matrix. The last term accounts for the on-site
potential difference between sublattices A and B where ξi = ± for states at A and B,
respectively. To solve Eqn. (3.3) on a finite geometry, we define the operators a†α (m)
[aα (m)] and b†α (m) [bα (m)] which create [annihilate] an electron at site m in unit cell
α for sublattice A and B, respectively [57].

25
 Written in terms of these operators, the graphene term becomes

∑∑[ ] ∑∑[ ]
ĤG = − t a†α (m)bα−1 (m) − t a†α (m)bα+1 (m)
α modd α meven
∑∑[ ]
−t b†α (m)aα (m) + a†α (m + 1)bα (m) + H.c., (3.4)
α m

where H.c. stands for the hermitian conjugate. We then define the Fourier transforms
[58]

1 ∑ ikxmA
aα (m) = √ e α ak (m), (3.5)
Lx k

and

1 ∑ ikxmB
bα (m) = √ e α bk (m), (3.6)
Lx k

where xmA
α (xmB
α ) is the x component of site mA (mB) in unit cell α. Recasting our

Hamiltonian in terms of these operators and using the relation

1 ∑ i(k−k′ )xα
e = δk,k′ , (3.7)
Lx α

we arrive at the expression

∑[
ĤG = − a†k (m)bk (m)K1 + b†k (m)ak (m)K1 + ta†k (m + 1)bk (m) (3.8)
m,k
]
+ tb†k (m)ak (m + 1) ,

where K1 = 2tcos(ka/2). We then define the single-particle state [57, 58]

∑[ ]
|Ψ(k)⟩ = ψmA (k)a†k (m) + ψmB (k)b†k (m) |0⟩ , (3.9)
m

and utilize the Schrödinger equation ĤGk |Ψ(k)⟩ = ε |Ψ(k)⟩. Employing the fermion

26
anticommutation relations

{ }
a†k (m), ak (n) = δm,n , (3.10)

and

{ }
b†k (m), bk (n) = δm,n , (3.11)

with all other anticommutations giving zero, we obtain the coupled equations

εψmA = −K1 ψmB − tψm−1B ,

εψmB = −K1 ψmA − tψm+1A . (3.12)

Solving these equations for ε gives the edge dispersion for a zigzag graphene nanorib-
bon.
Next, let us consider the spin-orbit term

∑ {
even
Ĥso = iλso a†α (m)σ [aα (m + 1) + aα (m − 1) + aα+1 (m) − aα+1 (m + 1)
α,meven

− aα−1 (m) − aα+1 (m − 1)]

+ b†α (m)σ [bα (m + 1) + bα−1 (m) + bα (m − 1) − bα−1 (m + 1)

− bα−1 (m) − bα (m − 1)]} , (3.13)

and

∑ {
odd
Ĥso = iλso a†α (m)σ [aα−1 (m + 1) + aα−1 (m − 1) + aα+1 (m) − aα (m + 1)
α,modd

− aα−1 (m) − aα (m − 1)]

+ b†α (m)σ [bα+1 (m + 1) + bα−1 (m) + bα+1 (m − 1) − bα−1 (m + 1)

− bα+1 (m) − bα (m − 1)]} , (3.14)

√
for even and odd m, respectively, where λso ≡ ∆so /(6 3). Proceeding as before, we

27
arrive at the equations

εψmA = −K3 [ψm+1A + ψm−1A ] + K2 ψmA ,

εψmB = K3 [ψm+1B + ψm−1B ] − K2 ψmB , (3.15)

where K2 = 2λso σsin(ka) and K3 = 2λso σsin(ka/2).

Lastly, consider the sublattice asymmetry term

∆z ∑ [ † ]
Ĥz = aα (m)aα (m) − b†α (m)bα (m) . (3.16)
2 α,m

This yields

∆z
εψmA = ψmA ,
2
∆z
εψmB = − ψmB . (3.17)
2

Combing Eqns. (3.12), (3.15) and (3.17) we obtain the coupled equations
] [
∆z
εψmAσ = −K1 ψmBσ − tψm−1Bσ − K3 [ψm+1Aσ + ψm−1Aσ ] + K2 + ψmAσ , (3.18)
2

and
][
∆z
εψmBσ = −K1 ψmAσ − tΨm+1Aσ + K3 [ψm+1Bσ + ψm−1Bσ ] − K2 + ψmBσ ,
2
(3.19)

where we have restored the spin dependence of the wave-functions, σ = ± for spin
up and down, respectively,
( )
ka
K1 = 2tcos , (3.20)
2

K2 = 2λso σsin (ka) , (3.21)

28
 ( )
ka
K3 = 2λso σsin , (3.22)
2
√
and λso ≡ ∆so /(6 3). To solve these equations, we construct a matrix Ĥσ in the
basis

ψσT = (ψ1Aσ , ψ2Aσ , ..., ψN Aσ , ψ1Bσ , ψ2Bσ , ..., ψN Bσ ), (3.23)

and solve for the eigenvalues Ĥσ ψσ = εψσ . As an example, we have solved the
equations and plotted the eigenvalues in Fig. 3.3 for N = 30 and various values
of ∆so and ∆z . In frame (a), we have the well known result of a zigzag graphene
nanoribbon (∆so = ∆z = 0). We note that there is a zero energy mode which is
spin degenerate. If we include spin-orbit coupling [frame (b)] we get the results of
the Kane-Mele QSHI. Here we have used ∆so /t = 0.6 and only show the results at a
single edge (m = N ). There are helical channels in the gap with blue corresponding
to spin up and red to spin down. At the other edge, the spin labels are reversed.
In this scenario, the system is a topological insulator. For finite ∆z [frames (c) and
(d)] two possible insulating phases are present (note: both edges are plotted). For
∆so > ∆z [frame (c)], there exist a pair of helical edge modes at each edge (one edge
has the channels shifted up in energy while at the opposite edge, they are shifted
down). For ∆z > ∆so [frame (d)] a band inversion occurs and the edge channels are
lost signifying the system is a trivial insulator. We can identify the edge associated
with the helical states by examining the associated eigenvectors and seeing whether
the n = 1 or n = N values are finite.

3.2 Low-Energy Band Structure

If we restrict our attention to the low-energy physics of the bulk, an effective Hamil-
tonian can be written [39] to describe the physics in the vicinity of the two K points

29
 1∆so/t=0 ∆1so/t=0.6
∆z/t=0. ∆z/t=0.
(a) (b)
0.5 0.5
ε/t

ε/t
0 0

-0.5 -0.5

-1 -1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
ka (2π) ka (2π)
Lower Edge
1∆so/t=0.6 ∆1so/t=0.6 Upper Edge
∆z/t=0.24. ∆z/t=0.8.
(c) (d)
0.5 0.5
ε/t

ε/t

0 0

-0.5 -0.5

-1 -1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
ka (2π) ka (2π)

Figure 3.3: Edge states of a zigzag (a) graphene nanoribbon, (b) QSHI with ∆z = 0,
buckled honeycomb lattice in the (c) topological (∆so > ∆z ) and (d) trivial (∆z >
∆so ) phases.

of the hexagonal first Brillouin zone. Written for a single spin and valley,

1 1
Ĥξσ = ℏv(ξkx τ̂x + ky τ̂y ) − ξσ ∆so τ̂z + ∆z τ̂z , (3.24)
2 2

where τ̂i are Pauli matrices associated with the sublattice pseudospin (i.e. written
in the A-B sublattice basis). The two valleys K and K ′ are indexed by ξ = ±,
respectively, with ℏkx and ℏky being the momentum components measured relative

30
to the Dirac points. The real spin of the electrons is given by σ = ± for spin up and
down, respectively. The first term of Eqn. (3.24) is the well-known graphene model for
relativistic Dirac electrons with Fermi velocity v (≈ 5.27 − 6.47 × 105 m/s for silicene
[39, 56]). Including the first two terms, one obtains the Kane-Mele Hamiltonian for
intrinsic spin-orbit coupling [14] leading to a spin-orbit band gap of ∆so . The last
term is associated with the on-site potential difference ∆z . In Ref. [42], a Rashba
spin-orbit coupling is also included; however, it is typically neglected as it is an order
of 10 times smaller in magnitude than ∆so [42]. If the Rashba term is ignored, the
full 8x8 matrix spanning the two K points is block diagonal in 2x2 matrices labelled
by valley ξ = ± and spin σ = ±. These 2x2 matrices are
 
∆ξσ ℏv(ξkx − iky )
Ĥξσ =  , (3.25)
ℏv(ξkx + iky ) −∆ξσ

where ∆ξσ = (1/2)(−ξσ∆so + ∆z ). For convenience, we define |∆++ | ≡ ∆min and

|∆+− | ≡ ∆max as the minimum and maximum band gaps, respectively. The eigenval-
ues of Eqn. (3.25) are

√
Eξσ (k) = ± ℏ2 v 2 k 2 + ∆2ξσ . (3.26)

Several plots of the band structure are shown in Fig. 3.4 where dashed blue curves
correspond to spin-up and solid red curves represent spin-down bands. The upper
frame of Fig. 3.4, shows the result for ∆z /∆so = 0.5 at the K and K ′ points. The
lower frame, demonstrates the closing of the low-energy band gap when ∆z = ∆so .
For ∆z = 0, the band structure is that of the QSHI. Including a finite ∆z spin-splits
the bands which are now characterized by two gaps ∆min and ∆max . While the bands
are spin-split at a given valley, the other K point has the opposite spin labels. As ∆z
is increased but remains less than ∆so , the system is a TI; as a function of ∆z , the
lowest gap decreases while the higher gap increases. At the critical value ∆z = ∆so
the lowest gap closes at each valley and the system is in the VSPM phase. As ∆z is
increased further, the gap reopens with both gaps increasing with ∆z ; the system is

31
 3 3
K K′
2 2

1 1

ε / ∆so
0 0

-1 -1

-2 -2
∆z / ∆so = 0.5 ∆z / ∆so = 0.5
-3 -3
-2 -1 0 1 2 -2 -1 0 1 2
ℏvk / ∆so ℏvk / ∆so
3 3
K K′
2 2

1 1
ε / ∆so

0 0

-1 -1

-2 -2
∆z / ∆so = 1.0 ∆z / ∆so = 1.0
-3 -3
-2 -1 0 1 2 -2 -1 0 1 2
ℏvk / ∆so ℏvk / ∆so

Figure 3.4: The low-energy band structure of silicene in the presence of an on-site
potential difference between sublattices.

now a trivial BI. This behaviour is illustrated in Fig. 3.5. As ∆z passes through the
critical value, a band inversion occurs between the lowest gapped bands [56]. To see
this, return to Eqn. (3.24) and note that at k = 0, [Ĥ, τ̂z ] = 0. Therefore, at the two
K points, the wave-functions are also eigenstates of τ̂z and can be characterized by a
quantum number τ = ± associated with the A and B sublattices. Our Hamiltonian
for a given ξ and σ is now simply
 
∆ξσ 0
Ĥξσ =  , (3.27)
0 −∆ξσ

32
Figure 3.5: Schematic representation of the bulk band structure evolution at the K
point as ∆z is increased.

with eigenvalues ±∆ξσ . It is clear that the eigenvectors uξσ = (uξσ ξσ T

A , uB ) (written in

the A − B basis) are

       
1 0 1 0
u++
1 = , u++
2 = , u+−
3 = , u+−
4 = , (3.28)
0 1 0 1
       
1 0 1 0
u−+
5 = , u−+
6 = , u−−
7 = , u−−
8 = . (3.29)
0 1 0 1

We must now assign these eigenvectors to the appropriate energy bands. Let us begin
with the K point. We know that u++
1 and u++
2 are both associated with spin up. The
vector u++
1 is connected to the +∆ξσ eigenvalue. In the TI regime, this eigenvalue is
at negative energy (i.e. ∆++ < 0). Therefore, we assign u++
1 to the negative spin-up
band. We apply similar arguments to assign the remaining eigenvectors. It is clear
that all the odd numbered uξσ
α will have positive eigenvalues when τ̂z is applied (i.e.

they correspond to the A sublattice); likewise, the even numbered vectors will have
a negative eigenvalue (i.e. they come from the B sublattice). Thus, for a TI, the

33
energy ordering E(σ, τ ) at the K point is

E(−1, 1) > E(1, −1) > 0 > E(1, 1) > E(−1, −1). (3.30)

At K ′ , we find

E(1, 1) > E(−1, −1) > 0 > E(−1, 1) > E(1, −1). (3.31)

In the BI regime, the sign of ∆++ and ∆−− changes. Now assigning the eigenvectors
to the appropriate bands gives

E(−1, 1) > E(1, 1) > 0 > E(1, −1) > E(−1, −1) (3.32)

at K and

E(1, 1) > E(−1, 1) > 0 > E(−1, −1) > E(1, −1) (3.33)

at K ′ . From this, it is clear that the pseudospin label of the two lowest gapped bands
(middle entries in the above equations) has switched. This signifies a band inversion.
Examining the pseudospin label, we note that the τ = −1 bands decrease in energy
as ∆z increases while the τ = 1 bands increase with ∆z . This is understood by
remembering that ∆z creates a potential difference between the A and B sublattices.
It causes the potential on A to be higher than that on B. Therefore, the B sublattice
becomes more favourable for electrons.

3.3 Density of States

Given Eqn. (3.26), we can easily obtain the electronic density of states. The density
of states describes the number of states available at a given energy (ω) for an electron

34
to occupy. The density of states per unit area is

∑∑
N (ω) = δ(ω − Eξσ (k)). (3.34)
ξσ k

Working in the continuum limit,

∑ ∫
1
→ d2 k, (3.35)
k
(2π)2

we obtain the analytic expression [30]

∑
N (ω) = Nξσ (ω), (3.36)
ξσ

where

|ω|
Nξσ (ω) = Θ (|ω| − |∆ξσ |) . (3.37)
2πℏ2 v 2

Plots of the electronic density of states evaluated from Eqn. (3.36) are shown in
Fig. 3.6 for ∆z /∆so = 0.5 and 1.0. The two energy gaps are clearly reflected in the

3 3
∆z / ∆so = 0.5 ∆z / ∆so = 1.0

2 2
N(ω)

1 1

0 0
-2 -1 0 1 2 -2 -1 0 1 2
ω/∆so ω/∆so

Figure 3.6: The low-energy electronic density of states of silicene.

density of states. At the critical value ∆z = ∆so , the density of states increases
linearly out of zero energy until ω = ∆so after which it jumps to a higher value
due to the accessibility of the upper gapped band. It continues to increase linearly

35
afterwards but with twice the slope. For ∆z ̸= ∆so , the density of states is zero
for energies below the lowest gap. At ω = ∆min , a jump occurs. The discontinuity
associated with ω = ∆max remains. For ∆z = 0, only the spin-orbit gap remains and
there is a single jump associated with ∆so .
The ∆z dependence of the gaps is plotted in Fig. 3.7. The regions describing
the trivial insulating regime are coloured pink while the TI is identified by the yel-
low panel. It is clear that the lowest gap (dashed line) decreases in the TI regime
1.5

1
VSPM
∆ / ∆so

0.5

BI TI BI
0
-3 -2 -1 0 1 2 3
∆z / ∆so

Figure 3.7: The location of the energy gaps of silicene as a function of electric field
strength.

(∆z /∆so < 1) until it vanishes in the VSPM phase. In the BI regime (∆z /∆so > 1),
both gaps increase with ∆z with a constant separation of 2∆so . In the following chap-
ter, we shall use the optical conductivity to trace the behaviour of the gaps. With
this, we hope to find signatures of the two insulating phases.

3.4 Berry Phase and Winding Number

It is also illuminating to discuss the Berry phase of silicene. To do so, we first need
to evaluate the wave-functions associated with Eqn. (3.25) which we recast (at the

36
K-point) as
 
∆+σ fk
ĤKσ =  , (3.38)
fk∗ −∆+σ

where fk = ℏv(kx −iky ) ≡ |fk |exp[iArgfk ]. We introduce the following definitions [59]:

∆+σ
cosθ ≡ +
E+σ (k)
|fk |
sinθ ≡ +
E+σ (k)
ϕ ≡ −Argfk , (3.39)

√
α
where E+σ (k) = α ℏ2 v 2 k 2 + ∆2+σ , to obtain
 
−iϕ
cosθ sinθe
+
Ĥ+σ = E+σ (k)  . (3.40)
sinθeiϕ −cosθ

The associated wave-functions are

 
cos(θ/2)
|uk+ ⟩ =  , (3.41)
iϕ
sin(θ/2)e

for α = + and
 
−iϕ
−sin(θ/2)e
|uk− ⟩ =  , (3.42)
cos(θ/2)

for α = −.
To compute the Berry phase, we employ Eqn. (1.7) with R → k. For α = +, we
have
∫
γn = sin2 (θ/2)dk · ∇k ϕ. (3.43)
C

Immediately, we note that the Berry phase is dependent on the contour C. As this

37
quantity becomes important when discussing cyclotron orbits (see Chapter 7), we
will consider a closed equienergy contour C such that θ contributes a constant to the
integrand. Therefore,

γn = 2πsin2 (θ/2)Wc , (3.44)

where
I
dϕ
Wc ≡ =1 (3.45)
C 2π

is the topological winding number about the Dirac point. Equivalently,

γn = πWc (1 − cosθ)
( )
∆+σ
= πWc 1− √ . (3.46)
ℏ v k 2 + ∆2+σ
2 2

We see that for gapped Dirac systems, the Berry phase of a given spin-cone is not
quantised in multiples of π but has an energy dependent contribution in addition to
the topological winding number piece. An identical analysis yields analogous results
(up to a sign) for α = −. This argument is easily extended to the K ′ point. In
general,
 
∆ξσ
γn = πWc 1 − √ , (3.47)
ℏ2 v 2 k 2 + ∆2ξσ

where
I
dϕ
Wc ≡ αξ = αξ. (3.48)
C 2π

The first term of (3.47) is referred to as the topological part of the Berry phase
and is the contribution which manifests itself in the quantum oscillations [60] (see
Chapter 7).

38
 CHAPTER IV

OPTICAL CONDUCTIVITY OF
SILICENE

The dynamical conductivity is a powerful experimental tool for probing the important
energy scales of a system. When a sample is exposed to incident light, some of the
light is reflected while some is transmitted. The system may also absorb light. It
is this process which gives rise to the interesting physics. A photon of frequency
Ω can be absorbed by exciting an electron in the sample. This is best understood
by examining the systems band structure. Figure 4.1 displays the low-energy band
structure of graphene about one of its inequivalent K points. The meeting of the
two cones occurs at charge neutrality. For a finite chemical potential µ, all the states
below µ are occupied while those above are unoccupied. An electron in the lower cone
(valence band) can absorb an energy ℏΩ and be excited to the unoccupied part of
the upper cone (conduction band) leaving behind a hole. Due to the Pauli exclusion
principle, this can only happen for energies greater than 2µ. Such a process is known
as an interband transition. One would, therefore, expect no absorption for energies
below 2µ and a finite response above. In fact, there is another process which can occur.
For almost no energy cost, an electron could be excited from just below µ to just above.
This is an intraband transition and gives rise to the well-known Drude conductivity [6].
Note: we have assumed that the photon cannot transfer momentum to the electron
and, thus, all the transition are vertical. This is a good approximation [61] as the
photon typically has a small wavenumber [∼ 107 m−1 for visible light of wavelength
O(100nm)] in comparison to the dimensions of the Brillouin zone (∼ 1010 m−1 for an
atom spacing of order 1Å). Simply by varying the photon frequency, one can probe

39
Figure 4.1: The low-energy band structure of graphene about a single K point and
the allowed optical transitions.

the important energy scales of the system and obtain information about the band
structure.
For zero temperature, the optical conductivity at photon frequency Ω, can be
evaluated from the Kubo formula [62]. Written in the many-body spectral represen-
tation,
∫ |µ| ∫ [ ]
1 dω d2 k
Reσαβ (Ω) = Tr ĵα Â(k, ω + Ω)ĵβ Â(k, ω) , (4.1)
2Ω |µ|−Ω 2π (2π)2

where µ is the chemical potential, ĵα = ev̂α is the current operator where ℏv̂α =
∂ Ĥ/∂kα , with α, β corresponding to the x and y spatial coordinates. Â(k, ω) is the
spectral function which is related to the Green’s function (Ĝ−1 (z) = z Iˆ − Ĥ) through

∫ ∞
dω Âij (k, ω)
Ĝij (z) = . (4.2)
−∞ 2π z − ω

As an introductory example, in Appendix A.2, the Kubo formula is used to evaluate

40
the optical conductivity of graphene. For silicene, we use Eqn. (3.25), which gives the
Green’s function for a single spin and valley
 
Gξσ
11 (z) Gξσ
12 (z)
Ĝξσ (z) =  , (4.3)
Gξσ ξσ
21 (z) G22 (z)

where

z + 2∆ξσ
Gξσ
11 (z) = , (4.4)
z2 − Eξσ (k)2

ϕ∗ (k)
Gξσ
12 (z) = , (4.5)
z 2 − Eξσ (k)2

ϕ(k)
Gξσ
21 (z) = , (4.6)
z 2 − Eξσ (k)2

and

z − 2∆ξσ
Gξσ
22 (z) = , (4.7)
z2 − Eξσ (k)2

with ϕ(k) = ℏv(ξkx + iky ). Using Eqn. (4.2), we obtain the four spectral function
elements

∑( α∆ξσ
)
Aξσ
11 (k, ω) =π 1− δ(ω + α|Eξσ (k)|), (4.8)
α=±
|Eξσ (k)|

ϕ∗ (k) ∑
12 (k, ω) = −π
Aξσ αδ(ω + α|Eξσ (k)|), (4.9)
|Eξσ (k)| α=±

ϕ(k) ∑ ∗
21 (k, ω) = −π
Aξσ αδ(ω + α|Eξσ (k)|) = Aξσ
12 (k, ω) , (4.10)
|Eξσ (k)| α=±

41
and

∑( α∆ξσ
)
Aξσ
22 (k, ω) =π 1+ δ(ω + α|Eξσ (k)|). (4.11)
α=±
|E ξσ (k)|

We can now calculate the optical conductivity.

First, consider the longitudinal conductivity (α = β = x). We must now evaluate
Tr[v̂x Â(k, ω + Ω)v̂x Â(k, ω)] where
 
0 1
v̂x = ξv  . (4.12)
1 0

Therefore, suppressing the spin and valley labels, we have

[ ]
Tr v̂x Â(k, ω + Ω)v̂x Â(k, ω)

= v 2 [A∗12 (k, ω + Ω)A∗12 (k, ω) + A22 (k, ω + Ω)A11 (k, ω)

+A11 (k, ω + Ω)A22 (k, ω) + A12 (k, ω + Ω)A12 (k, ω)] . (4.13)

To proceed, note that ϕ(k) ∝ ±kx + iky can be expressed in polar coordinates such
that ϕ(k) ∝ e−iθk where θk =arctan(±ky /kx ). Recall that A12 ∝ ϕ(k); therefore,
A∗12 (k, ω + Ω)A∗12 (k, ω) ∝ e2iθk and A12 (k, ω + Ω)A12 (k, ω) ∝ e−2iθk . As there is no
other angular dependence, these terms will go to zero when the angular average is
conducted. We can ignore them in the trace. The only contribution from the trace
is v 2 [A22 (k, ω + Ω)A11 (k, ω) + A11 (k, ω + Ω)A22 (k, ω)]. The absorptive part of the
longitudinal conductivity is then [30]

∫ |µ| ∫ [ ]
ξσ e2 v 2 dω d2 k ξσ ξσ ξσ ξσ
Reσxx = A (k, ω + Ω)A11 (k, ω) + A11 (k, ω + Ω)A22 (k, ω) .
2Ω |µ|−Ω 2π (2π)2 22
(4.14)

42
This can be solved analytically (see Appendix C.1), to give [30]

{ } [ ( )2 ]
ξσ
σxx (Ω) µ2 − ∆2ξσ 1 2∆ξσ
Re = δ(Ω)Θ (|µ| − |∆ξσ |) + 1+ Θ (Ω − Ωc ) ,
σ0 |µ| 4 Ω
(4.15)

where Ωc = max(2|∆ξσ |, 2|µ|), σ0 = e2 /(4ℏ) and Ω > 0. The imaginary part can be
found by the usual Kramers-Kronig transformation
∫ ∞
′′ 2Ω σ ′ (ω)
σ (Ω) = − dω, (4.16)
π 0 ω 2 − Ω2

where σ ′ (Ω) and σ ′′ (Ω) are the real and imaginary parts, respectively, and we have
used the fact that the real part of the conductivity is even in Ω. Therefore, the
imaginary part of the longitudinal conductivity is [30]

{ } [ ( )2 ]
ξσ
σxx (Ω) µ2 − ∆2ξσ 1 2∆ξσ 2∆2ξσ
Im = + 1+ f (Ω) + , (4.17)
σ0 πΩ|µ| 4π Ω πΩΩc

where f (Ω) = ln[|Ω + Ωc |/|Ω − Ωc |]. To account for scattering, the δ-function in
Eqn. (4.15) is replaced by a Lorentzian such that δ(x) → (Γ/π)/(Γ2 + x2 ) where
Γ = 2η is a phenomenological transport scattering rate rather than the quasiparticle
scattering rate η which is present in the broadened spectral functions. If the effects
of scattering are considered, the Kramers-Kronig transformation of δ(Ω) to 1/(πΩ)
will become (Γ/π)/(Γ2 + x2 ) transforming to (Ω/π)/(Ω2 + Γ2 ).
We can also derive an expression for the transverse Hall conductivity by setting
α = x and β = y in Eqn. (4.1). Using
 
0 −i
v̂y = v  , (4.18)
i 0

43
 ξσ
the imaginary part of σxy (Ω) is found by solving

∫ ∫
ξe2 v 2 |µ|
dω d2 k [ ξσ
ξσ
Imσxy (Ω) = A22 (k, ω + Ω)Aξσ
11 (k, ω)
2Ω |µ|−Ω 2π (2π)2
]
−Aξσ
11 (k, ω + Ω)A ξσ
22 (k, ω) . (4.19)

Again, this can be solved analytically (see Appendix C.2) to give [30]
{ }
ξσ
σxy (Ω) ∆ξσ
Im = −ξ Θ(Ω − Ωc ), (4.20)
σ0 Ω

with the Kramers-Kronig related real part [30]

{ }
ξσ
σxy (Ω) ∆ξσ
Re =ξ f (Ω). (4.21)
σ0 πΩ

4.1 Results for the Longitudinal Conductivity

We now present the results for the longitudinal conductivity which we shall use to
identify signatures of the TI, VSPM and BI phases. In the following, we use the
Lorentzian representation of the δ-function with a broadening parameter of η =
0.01∆so . This signifies impurity scattering with an associated transport scattering
rate of 1/τimp = 2η. While all of the interesting physics could be elucidate without
this definition, we use it to present more realistic looking curves.
To begin, let us consider the charge-neutral system (i.e. µ = 0). The associated
conductivity curves are shown in Fig. 4.2 for varying ∆z in the TI, VSPM [frame (a)]
and BI [frame (b)] regimes. The inset shows the evolution of the gaps as a function
of ∆z which governs the onset of interband transitions. For ∆z = 0, there is a single
jump in the conductivity at Ω = ∆so associated with transitions between the gapped
bands. With a finite ∆z , the single feature splits in two. For the TI regime, the
lowest peak moves down in Ω as ∆z is increased. This is due to transitions between
the lowest gapped bands. Note that transitions can only occur between bands of the
same spin index. In the BI regime, both jumps move higher in energy as the electric

44
 2 1.5
µ=0
1 ∆max

∆ / ∆so
Re σxx(Ω) / σo
0.5
∆min

1 0
∆z / ∆so 0 1 2 3
0.00
∆z / ∆so ∆z / ∆so
0.25 1.25
0.50 1.50
0.75 1.75
1.00
(a) (b)
0
0 1 2 3 0 1 2 3 4
Ω / ∆so Ω / ∆so

Figure 4.2: Longitudinal conductivity of charge-neutral silicene for various electric

fields.

field is increased. They maintain a constant separation of 2∆so . In the VSPM phase,
a flat conductivity persists to zero energy. This is due to interband transitions in the
gapless band. This behaviour is identical to graphene; however, the background value
has been halved as only a single spin species contributes at a given valley. The single
jump at 2∆so is due to transitions between the gapped bands. At high energy, all the
curves merge into the universal background value of σ0 . The unique behaviour of the
three regimes should allow for an experimental identification of the TI, VSPM and
BI phases as well as provide a measure of ∆so .
For finite charge doping, we obtain the results of Fig. 4.3 for µ = 0.25∆so and
µ = 0.75∆so . In this case, the lowest energy for absorption is controlled by the greater
of 2µ or 2|∆ξσ |. This is illustrated in frames (a) and (b) for 2|µ| < ∆so and frames
(c) and (d) for 2|µ| > ∆so . Consider the first two frames. As discussed in reference to
Fig. 4.1, for Ω < 2µ, optical transitions are Pauli blocked. Indeed, similar behaviour
is seen between the finite µ case and Fig. 4.2; however, now there is a low-energy
cutoff. That is, for ∆z values such that ∆min < µ [dash-dotted green and dashed red
curves of Figs. 4.3(a) and (b)], the lowest interband transition occurs at Ω = 2µ. Due
to the finite chemical potential, a low-energy Drude response is present for µ > ∆min .
By varying the electric field strength, the Drude response can be switched on and off.
The Drude weight as a function of ∆z is plotted in Fig. 4.4. The Drude weight is given

45
 2
1.5
µ / ∆so = 0.25
∆max
1

∆ / ∆so
Re σxx(Ω) / σo
0.5 ∆min
2µ
1 0
∆z / ∆so 0 1 ∆ /∆ 2 3
z so ∆z / ∆so
0.00
0.25 1.25
0.50 1.50
0.75 1.75
1.00
(a) (b)
0
0 1 2 3 0 1 2 3 4
Ω / ∆so Ω / ∆so
2 1.5
µ / ∆so = 0.75
1 ∆max ∆min

∆ / ∆so
2µ
Re σxx(Ω) / σo

0.5

1 0
0 1 2 3
∆z / ∆so
∆z / ∆so
0.00
0.25 ∆z / ∆so
0.50
0.75 1.25
1.00 1.75
(c) 2.75 (d)
0
0 1 2 3 0 1 2 3 4
Ω / ∆so Ω / ∆so

Figure 4.3: Longitudinal conductivity of silicene with finite chemical potential.

by the spectral weight under the δ-function in Eqn. (4.15) for Ω = 0+ . Let us begin
with the black curve (µ = 0.25∆so ). Here µ is less than ∆min for ∆z < 0.5∆so and
only interband transitions can occur. As ∆z is increased, the minimum gap decreases
and eventually µ becomes greater than ∆min . Now intraband absorption leads to a
Drude response. The maximum absorption occurs in the VSPM phase. After this,
the Drude weight decreases as the lowest gap increases with ∆z . Eventually, the
electric field will be strong enough to cause the Fermi energy to lie in the gap and
intraband transitions no longer occur. For this curve, that occurs at ∆z = 1.5∆so .
Thus, for 2µ < ∆so , a Drude response is only present when ∆z is tuned to place
the minimum gap below the chemical potential. In the opposite regime (2µ > ∆so )
intraband transitions occur even when ∆z = 0. Now a kink occurs at µ = ∆max which
is associated with the maximum gap becoming larger than the chemical potential.

46
 2
TI BI
µ/∆so
1.5 1.00
0.75
µ=∆max 0.50

Drude Weight
0.25

1 µ=∆max

0.5

0
0 1 2 3
∆z/∆so

Figure 4.4: Drude weight as a function of electric field.

4.2 Response to Circularly-Polarized Light

We now wish to examine the response to circularly-polarized that. This type of light
has been suggested as a means to elucidate the valley-spin coupling of silicene [63]
and group-IV dichalcogenides [64, 65]. In these references, the discussion was limited
to specific frequencies. Here, we wish to explore the response over a large range of Ω.
In the circularly-polarized basis, the optical conductivity for right- and left-handed
polarized light is given by σxx (Ω) ± iσxy (Ω), for right- and left-handed polarization,
respectively. The absorptive part is thus

σ± (Ω) = Reσxx (Ω) ∓ Imσxy (Ω), (4.22)

where Reσxx (Ω) and Imσxy (Ω) are given by Eqns. (4.15) and (4.20), respectively. As
we are interested in distinguishing between the different spin and valley species, we
write

ξσ
σ± ξσ
(Ω) = Reσxx (Ω) ∓ Imσxy
ξσ
(Ω). (4.23)

To begin, we plot the valley-separated conductivities in the TI and BI regimes. This

is shown in Figs. 4.5 and 4.6 for µ = 0 and ∆z = 0.75∆so and 1.25∆so , respectively.

47
 In both figures, the top frames shows the response to right-handed polarized light
1.5
∆z/∆so=0.75 K K'

0
σ /σ
1

+
0.5

0
K K'
0
σ /σ

1
-

0.5

0
0 1 2 0 1 2 3
Ω/∆ Ω/∆
so so

Figure 4.5: Circularly-polarized response of silicene in the TI regime.

while the lower frames are for left-handed polarization. The left frames show the
response at the K point; the right frames show the K ′ contribution. For clarity,
we have also coloured the curves to show the spin contributions. Blue is spin up
and red is spin down. In the regions where two spin-species are involved, the total
response at a given valley is shown by the black curve. To construct these figures,
we used the analytic formulae and did not account for scattering. This was done
to make clear the interesting features. Let us begin our analysis with the right-
handed response in the TI regime (upper frames of Fig. 4.5). Here, we have inset
the low-energy band structure and used black arrows to emphasize the dominant
transitions. At the K point (left frame), we see that for energies between Ω1 = 2∆min
and Ω2 = 2∆max the response is entirely due to spin up, albeit, a small contribution.
At Ω2 , a sharp jump is seen associated with spin-down electrons. This arises from
transitions between the two largest gapped bands (emphasized by the arrows in the
inset). While there is a strong spin-down response, there is also a small spin-up

48
 1.5
∆z/∆so=1.25 K K'

0
σ /σ
1

+
0.5

0
K K'
0
σ /σ

1
-

0.5

0
0 1 2 0 1 2 3
Ω/∆ Ω/∆
so so

Figure 4.6: Circularly-polarized response of silicene in the BI regime.

background. By contrast, at the K ′ point, there is a strong discontinuous response

from spin-down electrons at Ω = Ω1 and a very small spin-up response which onsets
at Ω2 . Combining these results, we see that for Ω = 2∆min , a right-handed-polarized
response can be generated from spin-down electrons at K ′ . A large spin-down at
K imbalance can also be generated at Ω = 2∆max ; however, there remains a spin-
up background. Switching the polarization of the light, we note that now strong
spin-up responses are present. Here, at Ω = 2∆min , the conductivity is made of spin-
up and momentum K electrons. Now, the higher energy spin-imbalance is due to
spin-up electrons at K ′ . To understand these “selection rules” note that circularly-
polarized light has an inherent chirality (one value for right-handed and another for
left-handed light) so we expect a particular-handed light to prefer transitions with a
given chirality. For right-handed polarization, we note that the dominant transitions
occur from the negative chiral valence band to the positive chiral conduction band (i.e.
from the B to A (A to B) sublattice bands at K (K ′ ) as discussed in Chapter 3.2). For
left-handed light, a transition from the positive chiral valence band to the negative

49
chiral conduction band is observed [from A to B (B to A) bands at K (K ′ )]. Recalling
the discussion in Chapter 3.2, sublattice pseudospin is only a good quantum number
at the Dirac point. This explains the small background which results from absorption
between the bands of the opposite chirality. Since the chirality of the lowest two bands
switches as we enter the BI regime, we expect a change in the circular dichroism as
we switch insulating phases. Turning our attention to the BI (see Fig. 4.6), we
note that right-handed light has a strong response at the K point while left-handed
light has a dominant contribution from the K ′ point. Again, spin-up at K and
spin-down at K ′ polarized fermions can be excited at Ω = 2∆min . To summarize,
by varying the handedness of the polarized light, spin-up at K and spin-down at
K ′ polarized electrons can be generated in both the TI and BI regimes [30]. Also,
in the TI regime, right- and left-handed light excite primarily spin-down and -up
electrons, respectively. In the BI regime, the right- and left-handed conductivity is
dominantly due to momentum K and K ′ , respectively. Like the longitudinal response,
the location of the jumps tracks the energy gaps and will behave differently in the TI
and BI regimes as ∆z is varied.

4.3 Concluding Remarks

In this chapter, we explored the longitudinal and circularly-polarized optical conduc-
tivities of silicene. We found that the location of the low-energy jumps traced the two
energy gaps which exist for finite ∆z . The lowest peak moved lower in energy in the
TI regime and higher in Ω for the BI. This behaviour should allow for an experimen-
tal determination of the insulating regime as well as ∆so (by identifying the point of
crossover between the phases). We also saw that circularly-polarized light allows for
a valley-spin-polarized response of electrons with (ξ, σ) values of (K, ↑) and (K ′ , ↓).
Throughout this thesis, we will look for similar identifiable trademarks of silicene.
Particular attention will be given to the phenomenon of valley-spin polarized charge
carriers.
Note: for most of this thesis, many-body effects (ex. electron-electron and electron-

50
phonon interactions) are neglected. For the dynamical conductivity of 2D crystals,
some effects of the electron-phonon interaction are illustrated in Refs. [66] and [67]. It
has been suggested [68, 69] that electron-electron interactions could give a frequency
dependent scattering rate in graphene. These interactions are adjusted by the choice
of substrate. In our calculations, we have assumed that such effects are small or made
small by an appropriate choice of substrate or by making the material freestanding.
It has been confirmed experimentally that the single-particle picture works well for
capturing the finite frequency response of similar systems.

51
 CHAPTER V

SPIN AND VALLEY HALL EFFECTS

IN SILICENE

The formulae developed in the previous chapter can be extended to investigate the
spin and valley Hall effects of such systems. The spin Hall effect is a novel phenomenon
in which distinct spin-species flow to opposite transverse edges of a sample when a
longitudinal electric field is applied [70–75]. This spin current will eventually cause a
spin imbalance to occur on the two transverse edges of the sample. This behaviour
is analogous to the classical Hall effect in which charges of opposite sign move to
opposing lateral edges when an electric field is applied in the presence of a magnetic
field. For the charge Hall effect, the charge build up is driven by the Lorentz force.
For the spin Hall effect, a magnetic field is not required. A schematic illustration
of this phenomenon is shown in Fig. 5.1. When an electric field Ex is applied, the
two spin-species flow in opposite directions creating a spin current Jys . Recently,

Figure 5.1: Schematic of the spin Hall effect.

the spin Hall effect has become the subject of great interest [14, 15, 17, 31, 73, 76–84]
due to its potential application in spintronic technology. The ability to manipulate

52
the electron’s spin degree of freedom is pivotal in the field of spintronics. Thus, a
strong spin-orbit interaction is important [14, 84] and silicene appears to be suitable
for spintronic applications [85]. An analogous valley Hall effect exists [86,87] in which
electrons of different valley label flow to opposite transverse edges. This allows for
the possibility of valleytronic devices. Due to the response of silicene to a sublattice
potential difference (∆z ), both the spin and valley Hall effects are predicted in silicene.
While the zero-frequency spin and valley Hall effects of silicene have been theoretically
examined [81, 88], there exists limited work for the finite chemical potential spin Hall
effect [81]. Here, we examine the finite frequency response as a function of chemical
potential and sublattice potential difference.

5.1 Spin and Valley Hall Conductivity

We now wish to explore the intrinsic spin Hall effect in silicene. That is, the effect
that arises in the absence of impurities [81, 83, 89]. To begin, return to the Kubo
formula written in the Green’s function representation [Eqn. (4.1)]

∫ |µ| ∫ [ ]
1 ℏdω d2 k
Reσαβ (Ω) = Tr ĵα Â(k, ω + Ω) ĵβ Â(k, ω) . (5.1)
2Ω |µ|−Ω 2π (2π)2

Being a transverse phenomenon, the spin and valley Hall conductivities are obtained
by solving for σxy (Ω). As in the previous section, the ĵx operator is given by the
usual ĵx = ev̂x for electrical conductivity, where ℏv̂x = ∂ Ĥ/∂kx . Here, the ĵy current
operator is replaced with [66] ĵys = ℏσv̂y /2 and ĵyv = ξv̂y /2 for the spin and valley con-
ductivities, respectively, where v̂y is given by Eqn. (4.18). We can evaluate Eqn. (5.1)
and apply the Kramers-Kronig relation to obtain Imσxy (Ω). Therefore, the real and
imaginary parts of the spin Hall conductivity are [31]

∑ e ∆ξσ
s
Reσxy (Ω) = ξσ f (Ω), (5.2)
ξ,σ
8 πΩ

53
and

∑ e ∆ξσ
s
Imσxy (Ω) = − ξσ Θ(Ω − Ωc ), (5.3)
ξ,σ
8 Ω

respectively, where again, ∆ξσ = (1/2)(−ξσ∆so + ∆z ) (with |∆++ | ≡ ∆min and

|∆+− | ≡ ∆max ), Ωc = max(2|µ|, 2|∆ξσ |) and f (Ω) = ln[|Ω + Ωc |/|Ω − Ωc |]. Simi-
larly, the real and imaginary parts of the valley Hall conductivity are [31]

∑ e ∆ξσ
v
Reσxy (Ω) = f (Ω), (5.4)
ξ,σ
4ℏ πΩ

and

∑ e ∆ξσ
v
Imσxy (Ω) = − Θ(Ω − Ωc ), (5.5)
ξ,σ
4ℏ Ω

respectively.

5.1.1 Charge Neutrality

We begin our analysis by considering the result for the Kane-Mele QSHI (∆z = 0).
The spin and valley Hall conductivities for the charge-neutral (µ = 0) system are
shown in Fig. 5.2. To emphasize the character of the spin Hall response, a schematic
of the transverse spin buildup is included as an inset. In the DC limit (Ω = 0),
s
the previously predicted [81, 88] values for the spin Hall [Reσxy = e/(2π)] and valley
v
Hall [Reσxy = 0] conductivities are obtained. If the system is subjected to photons
of frequency Ω, an increase in the magnitude of the spin Hall response is observed.
For photons of frequency Ω = ∆so , a strong response is observed in the spin Hall
conductivity. Thus, a stronger spin Hall response may be accessible in silicene at
finite frequency tuned to the spin orbit gap. For Ω > ∆so , the spin Hall conductivity
quickly diminishes and subsequently disappears at sufficiently high frequency. For
all frequencies, the valley Hall effect is zero signifying an equal distribution of valley
species on each edge.

54
 Figure 5.2: Charge-neutral spin and valley Hall conductivities of a Kane-Mele QSHI.

For finite ∆z and µ = 0, plots of the real (solid black) and imaginary (dotted
red) parts of the spin Hall conductivity can be seen in Figs. 5.3(a) and (b) for the
TI and BI regimes, respectively. Considering the real part, a positive conductivity
0.6
0.2 BI
0.4
0
0.2
σsxy(Ω) (e/2)

σsxy(Ω) (e/2)

-0.2 0
µ=0, ∆z=0.75∆so

-0.4 -0.2
µ=0, ∆z=1.25∆so
-0.4 Real part
-0.6
Imaginary part
-0.6
-0.8 TI (a) (b)
-0.8
0 1 2 3 4 0 1 2 3 4
Ω/∆so Ω/∆so

Figure 5.3: Charge-neutral spin Hall conductivity of silicene for finite ∆z .

signifies a net spin-up (-down) accumulation in one transverse direction while a neg-
ative conductivity yields a net spin-up (-down) accumulation on the opposite edge.
This is schematically shown by the insets. In the TI regime [Fig. 5.3(a)], the real
part of the conductivity (black curve) is always negative (i.e. spins of the same ori-
entation always accumulate on the same transverse edge). The two sharp features

55
in the conductivity are given by the two values of Ωc (which are denoted Ωmin
c and
Ωmax
c for the first and second feature, respectively). For µ = 0, these represent the
onset of the interband transitions at ∆min and ∆max , respectively. In the BI regime
[Fig. 5.3(b)], there is a sign change in the conductivity which results in a change in
spin accumulation (illustrated by the insets). For low frequency (Ω < Ωmin
c ), the spin

current flows in the opposite direction to the current in the TI regime while for higher
frequencies, the current returns to the direction of the TI phase. For the TI, there is
s
a finite DC response (Ω = 0) for Reσxy (Ω) which, for the case of µ = 0, is the same
value of e/(2π) discussed in Fig. 5.2. By contrast, in the BI regime there is no spin
Hall conductivity at Ω = 0.
The corresponding real (solid black) and imaginary (dashed blue) parts of the
valley Hall conductivity are shown in Figs. 5.4(a) and (b) for the TI and BI regimes,
respectively. As previously mentioned, the valley Hall effect corresponds to electrons
0.8
µ=0, ∆z=0.75∆so K K K (a) 0.8 µ=0, ∆z=1.25∆so BI (b)
0.6 Real part
Imaginary part K K K K K K
0.6
0.4 K' K' K'
σvxy(Ω) (e/ℏ)

σvxy(Ω) (e/ℏ)

0.2 0.4 K' K' K' K' K' K'

0 0.2
K' K' K'
-0.2
0
-0.4
K K K
TI -0.2
0 1 2 3 4 0 1 2 3 4
Ω/∆so Ω/∆so

Figure 5.4: Charge-neutral valley Hall conductivity of silicene for finite ∆z .

from one valley (K or K ′ ) flowing in one direction perpendicular to the longitudinal

current while electrons from the other valley flow in the opposite direction creating a
net transverse valley imbalance. The change in sign of the conductivity seen in the
TI regime [Fig. 5.4(a)] corresponds to a change in the side to which electrons from
one valley flow. Conversely, in the BI regime, electrons from a given valley flow in
only one direction for all frequencies. Again, the sharp features occur at Ω = Ωmin
c

56
 and Ωmax
c .
If we examine both the spin and valley Hall effect for the TI regime [Figs. 5.3(a)
and 5.4(a)] it is apparent that at Ω = Ωmin
c , electrons of a particular spin and valley

label flow to one transverse edge. At Ω = Ωmax

c , electrons of the same spin but
opposite valley index flow to that edge. Likewise, in the BI regime [Figs. 5.3(b) and
5.4(b)] electrons of a particular valley label always flow to one transverse edge while
the spin of the electrons changes between Ω = Ωmin
c and Ωmax
c . This creates the
potential to optically generate a spin and valley polarized accumulation of carriers on
a given transverse edge (i.e. a build up of a particular spin with either valley label or
of a particular valley label and either spin index).
Next, let us consider the effect of varying the sublattice potential difference at
charge neutrality. Figures 5.5(a) and (b) show these results for the spin Hall and
valley Hall conductivities, respectively. As ∆z is increased in the TI regime [dash

∆min ∆max
0.4 (a) 0.8 ∆min (b)

0.6 µ=0
0
Reσvxy(Ω) (e/ℏ)
Reσsxy(Ω) (e/2)

0.4

-0.4 0.2
µ=0
∆z/∆so=0 0
-0.8 ∆z/∆so=0.75 ∆z/∆so=0
∆max ∆z/∆so=1 -0.2 ∆z/∆so=0.75
∆min ∆z/∆so=1.25 ∆min ∆z/∆so=1
-1.2 -0.4
∆z/∆so=1.25
∆min=∆max
0 1 2 3 4 0 1 2 3 4
Ω/∆so Ω/∆so

Figure 5.5: Charge-neutral (a) spin and (b) valley Hall conductivities of silicene for
varying ∆z .

dotted red curve of Fig. 5.5(a)], the single peak seen in the ∆z = 0 case splits into
two, with one peak moving to higher energy and the other to lower energy with
increased ∆z . In the VSPM phase (dashed blue curve) a single feature reappears but
of half the strength of the ∆z = 0 case. This is associated with the lowest gap in the
band structure closing leaving only one gapped band at each valley as opposed to the
two spin-degenerate gapped bands per valley when ∆z = 0. As the system transitions

57
into the BI regime (dash double-dotted green curve), two peaks reappear as a result
of the second gap reopening; however, the low-energy interband feature has switched
sign as a result of the band inversion that occurs upon moving from the TI to BI
regime [56]. Both peaks now move to higher energy with a constant separation of
2∆so as ∆z is increased. Thus, in the TI regime, the spin Hall effect will lead to a spin-
up (-down) accumulation in one transverse direction, while in the BI regime, spin-up
(-down) accumulation can be produced in either direction by tuning Ω. Figure 5.5(b)
shows the corresponding results for the valley Hall conductivity. The application of
∆z introduces a finite response with two strong interband features appearing at Ωmin
c

and Ωmax
c . Here, the switch in direction of the Hall induced imbalance now occurs in
the TI regime while it remains the same in the BI regime. The strong responses in
the spin and valley Hall conductivities occur at the same frequencies, Ωmin
c and Ωmax
c ,
and thus a strong spin Hall and valley Hall effect should occur simultaneously. The
switch in sign of both effects should allow for a transverse accumulation of charge
carriers of particular spin and valley label which can be brought about by tuning the
incident frequency.
For µ = 0, the DC response for both the spin Hall and valley Hall conductivities
are [31, 81, 88]

 e

 − , ∆z < ∆so

 2π






s e
Reσxy (Ω = 0) = − , ∆z = ∆so , (5.6)

 4π








0, ∆z > ∆so

58
 and


 0, ∆z < ∆so








v e
Reσxy (Ω = 0) = , ∆z = ∆so . (5.7)

 4πℏ







 e , ∆ >∆
z so
2πℏ

Thus, a simultaneous DC response for both the spin and valley Hall conductivities is
only present in the VSPM state. Unlike the ∆z = 0 case (where a finite DC spin Hall
v
effect exists but σxy (Ω = 0) = 0), a finite DC valley Hall conductivity is obtained in
the BI regime; however, the DC spin Hall conductivity is now zero. Our results at
finite frequency properly capture this limiting behaviour as Ω → 0.

5.1.2 Finite Chemical Potential

So far, we have only considered the scenario where µ lies in both gaps. Now, let us
discuss the spin Hall effect for finite chemical potential. The spin Hall conductivity for
varying µ is shown in Fig. 5.6(a) and (b) for the TI and BI regime, respectively. The
relative location of µ in the band structure is given by the inset of Fig. 5.6(a). The
0 0.6
∆min
0.4
-0.2
0.2 2|µ|
Reσsxy(Ω) (e/2)

Reσxys (Ω) (e/2)

-0.4

2|µ|
µ
{ 0

-0.2
∆z=1.25∆so
-0.6 µ/∆so=0
2|µ| -0.4
∆z=0.75∆so µ/∆so=0.2
2|µ|
-0.6 µ/∆so=0.6
-0.8 (a) µ/∆so=1.9 (b)
∆min ∆max ∆max
-0.8
0 1 2 3 4 0 1 2 3 4
Ω/∆so Ω/∆so

Figure 5.6: (a) Spin and (b) valley Hall conductivity of silicene for finite chemical
potential.

59
finite-µ results for the valley Hall effect are not shown as they are analogous. For µ in
the energy gap (i.e. |µ| < ∆min ), the system behaves as in the undoped case. When
µ is increased such that it sits in the lowest conduction band (∆min < |µ| < ∆max )
(dashed dotted red and dashed blue curves) the location of the first peak occurs
at 2|µ| as interband transitions are Pauli blocked before this point. Finally, if the
chemical potential is placed through both conduction bands (|µ| > ∆max ) (dashed
double-dotted green curve), only one feature at Ω = 2|µ| is present. Note that as µ is
increased, the strength of the overall response diminishes with the exception of that
associated with ∆min and ∆max should either peak not be Pauli blocked.
In general, accounting for a finite chemical potential, the DC response for the real
part of the spin Hall conductivity can be worked out analytically and is [31]
 [ ]

 e

 sgn(∆++ ) − sgn(∆+− ) , ∆min > |µ|

 4π [ ]

e ∆++
s
Reσxy (Ω = 0) = −1 , ∆min < |µ| < ∆max , (5.8)

 4π |µ|



 e ∆so
 − , ∆max < |µ|
4π |µ|

where sgn(x)=1 for x > 0, 0 for x = 0 and -1 for x < 0. These results are in agreement
with those of Refs. [81,88] for finite µ. The DC response for the real part of the valley
Hall conductivity can also be worked out analytically and is given by [31]
 [ ]

 e

 sgn(∆++ ) + sgn(∆+− ) , ∆min > |µ|

 4πℏ [ ]

e ∆++
v
Reσxy (Ω = 0) = +1 , ∆min < |µ| < ∆max . (5.9)

 4πℏ |µ|



 e ∆z
 , ∆max < |µ|
4πℏ |µ|

Unlike the undoped case, a finite chemical potential |µ| > ∆min allows a non-zero DC
spin and valley Hall conductivity to occur in all insulating regimes. Note that for
|µ| > ∆max , the DC spin Hall effect is controlled by ∆so whereas the DC valley Hall
response is controlled by ∆z . Thus, for large chemical potential, the spin Hall effect
is an intrinsic property of the system while the valley Hall effect is generated and

60
tuned by the sublattice potential difference.

61
 CHAPTER VI

SILICENE IN A MAGNETIC FIELD

In Chapter 1, we saw that the energy dispersion of free electrons in a magnetic field
(B) is a discrete set of highly-degenerate LLs. These levels are equally spaced by
ℏωc and are indexed by a positive integer N . Profound differences are found in the
relativistic case (see Appendix B for a brief review of graphene in a magnetic field).
Most notably, the LLs are given by

√
EN,s = s 2ℏevF2 N B, (6.1)

where N = 0, 1, 2, 3, ... and s = ± (note: by definition, s = + for N = 0). A

striking difference between this and the linear behaviour of a 2DEG is the square root
dependence on N and B. There is also a zero energy LL which remains pinned at
the Dirac point [43,90–95]. This zeroth level has both a hole and particle nature [96].
There are also a set of negative energy levels which are symmetric in energy compared
to the positive levels. This is clearly seen in Fig. 6.1 which we model after Fig. 5 in
Ref. [97]. Figure 6.1(a) depicts the LL formation of a single Dirac point in the presence
of a finite magnetic field. Note that the N = 0 level is formed from both the particle
and hole bands. Such a schematic is useful in visualizing the two different energy
dispersions [97]; however, one cannot simply assign states from the B = 0 band
structure to a particular level. In frame (b), the B dependence of the positive energy
√ √
levels is shown for ℏevF2 = 25.64 meV/ T, which corresponds to vF ≈ 106 m/s.
The LL index is given next to the corresponding curve. Only the s = + band is
shown as the s = − dispersion is mirror symmetric. As one may expect, including a
mass term shifts the N = 0 level away from zero energy [90, 91].

62
 8 7 6 5 4
120
3
N=3, s=+
100
N=2, s=+ 2
ε N=1, s=+
80

Energy (meV)
1
N=0 60
k

N=1, s=- 40
N=2, s=-
N=3, s=- 20

0 0
0 1 2 3 4 5
B (T)

(a) (b)

Figure 6.1: (a) The relativistic Landau level dispersion. (b) The Landau level evolu-
tion as a function of B.

In this chapter, we will explore the LL dispersion which arises from the silicene
Hamiltonian [see Eqn. (3.24)]. We will then look for characterizing features and
novel phenomena in the finite frequency conductivity. It is well known that, in the
presence of incident light, transitions may occur between LLs which are governed
by a set of selection rules. This generates absorption lines in the magneto-optical
conductivity. For graphene, this has been predicted theoretically [90, 91, 93] and
confirmed experimentally [98–105]. Recently, this has also been explored in other
Dirac-like systems [32, 33, 106–108]. Much like gapped graphene [90, 91], the stronger
spin-orbit coupling in silicene causes the N = 0 level to split between ±∆so /2 at
both valleys. Indeed, the massive graphene limit is obtained by taking ∆so = 0 and
∆z /2 = ∆ where the N = 0 level is split between ±∆ at the two K points.

63
6.1 Landau Level Dispersion of Silicene
Let us now explore the form of the low-energy LLs which arise from the silicene
Hamiltonian [see Eqn. (3.24)]:

1 1
Ĥ = ℏv(ξkx τ̂x + ky τ̂y ) − ξσ ∆so τ̂z + ∆z τ̂z . (6.2)
2 2

In principle, there should also be a Zeeman term associated with spin polarization.
This can be included by adding −gs µB Bσ/2 to the Hamiltonian, where gs is the
Zeeman coupling strength, µB = eℏ/(2me ) ≈ 5.78 × 10−2 meV/T is the Bohr mag-
neton with me , the mass of an electron. For silicene, the Zeeman interaction shifts
the levels by O(10−1 meV) for 1T and is typically ignored [42, 56, 63, 109]. As it will
have a negligible effect on the phenomena discussed in this chapter, we too will ignore
it [32, 33]. For a single valley and spin, we are left with evaluating
 
− 12 ξσ∆so + 1
∆ ℏv(ξkx − iky )
Ĥξσ =  2 z . (6.3)
ℏv(ξkx + iky ) 1
2
ξσ∆so − 1
∆
2 z

Again, we shall define

1 1
∆ξσ = − ξσ∆so + ∆z . (6.4)
2 2
To account for an external magnetic field which we orient perpendicular to the lattice
(i.e in the ẑ direction), we introduce the vector potential A defined through B =
∇ × A. We then make the usual Peierls substitution [7]

ℏki → ℏki + eAi , (6.5)

which remains valid provided the lattice constant a is much smaller than the magnetic
√ √
length lB = h/(eB) [110]. In silicene, a ∼ 0.38 nm [39] and lB ∼ (26 nm)/ B(T)
[110]; therefore, this approximation should hold for large fields. By employing “min-
imal coupling” [7], we assume that only the charge distribution couples to the mag-
netic field and not higher multipole moments [7] (such as the electron’s magnetic

64
moment). We have assumed that the pseudo-momentum eigenvalue k can be inter-
preted as a mechanical operator p which can then be converted to a canonical mo-
mentum by minimal coupling. This assumption has accurately captured the magneto-
physics of graphene [98–105, 110]. For convenience, we work in the Landau gauge
A = (−By, 0, 0). Therefore, our low-energy Hamiltonian becomes
 ( [ ] ) 
∆ξσ ℏv ξ kx − eB
y − iky
Ĥξσ =  [ ]
ℏ . (6.6)
ℏv(ξ kx − eB
ℏ
y + iky ) −∆ξσ

Using the fact that Ĥψ = Eψ, where ψ T = (ϕA , ϕB ) is the wave-function with com-
ponents associated with the A and B sublattices, we can solve Eqn. (6.6) (see Ap-
pendix D) to obtain the LL spectrum:
 √
 s ∆2 + 2N E 2 , N = 1, 2, 3, ...
ξσ 1
EN,s
ξσ
= , (6.7)
 −ξ∆ξσ , N =0

√
where s = ± is a band index, and we have used the definition E1 ≡ ℏev 2 B ≈ 12.82
meV for B = 1T. It is important to note that in a sum over s, there is only a single
contribution from the N = 0 levels.
We can also solve for the associated wave-functions at the K and K ′ point [see
Appendix D]. They are [32, 33]:
 
 ⟩ −iAN,s |N − 1⟩
N̄ =  , (6.8)
K
BN,s |N ⟩

and  
 ⟩ −iAN,s |N ⟩
N̄ ′ =  , (6.9)
K
BN,s |N − 1⟩

65
respectively, where [32, 33]
 √

 s |EN,s
ξσ
| + s∆ξσ


 √ , N ̸= 0
AN,s = ξσ
2|EN,s | , (6.10)



 1−ξ
 , N =0
2

and  √

 |EN,s
ξσ
| − s∆ξσ


 √ , N ̸= 0
BN,s = ξσ
2|EN,s | . (6.11)



 1+ξ
 , N =0
2
Examining Eqn. (6.7), we see that the zeroth LL undergoes an important transi-
tion between the TI (∆z < ∆so ) and BI (∆z > ∆so ) regimes. This is shown in the
upper and lower frames of Fig. 6.2, respectively (Note: µ = 0 corresponds to zero
energy). Given the dependence of E0 on ∆ξσ , it is evident that the spin-up N = 0
LLs are at positive energy in the TI regime while the N = 0 spin-down levels are at
negative energy (refer to the upper frame of Fig. 6.2). For the BI, ∆z is greater than
∆so . This results in a change in sign of ∆K↑ and ∆K ′ ↓ [and, therefore, E0 (∆K↑ ) and
E0 (∆K ′ ↓ )]. Thus, the N = 0 spin-up level at the K point and the N = 0 spin-down
level at K ′ switch sign. This is a signature of the aforementioned band inversion
associated with the transition between the TI and BI regimes. In the VSPM state
(not shown) the N = 0 spin-up (-down) level at the K (K ′ ) point sits at zero energy.
This reflects the graphene-like nature of the VSPM.
The LL evolution with varying magnetic field for ∆z = 0.5∆so is shown in Fig. 6.3.
√
Similar to graphene, the N ̸= 0 levels scale as B. Unlike graphene, the N = 0 level
is not pinned at zero energy; however, the four spin- and valley-split N = 0 levels do
not scale with the magnetic field. There location is controlled through the spin-orbit
interaction and perpendicular electric field. The separation between N = 0 levels
at different valleys is adjusted by tuning the electric field; however, the separation
between N = 0 levels at the same valley is fixed at ∆so . For low magnetic field (left
frame), two of the N = 0 levels are higher in energy than several higher numbered

66
 K K'
2 2
2 2
1 1
1 1

ε 0

0
ℏvk
TI 0

-1 -1
-1 -1
-2 -2
-2 -2

K K'
2 2
2 2
1 1
1 1

ε 0

0 ℏvk BI 0

-1 -1
-1 -1
-2 -2
-2 -2

Figure 6.2: Schematic representation of the Landau levels in the TI (upper frame)
and BI (lower frame) regimes about the K and K ′ points.
1.2
20
0.8

0.4 10 ∆z=0.5∆so
ε/∆so
ε/∆so

0 0

-0.4 -10
-0.8
-20
-1.2
0 5 10 15 20 0 2 4 6 8 10
B/∆2s o (G/meV2) B/∆2s o (x103 G/meV2)

Figure 6.3: Landau level energies as a function of magnetic field for ∆z = 0.5∆so .

levels. This behaviour becomes important in the magneto-optical conductivity when

trying to tune the onset of interband transitions.

67
6.2 Density of States
Let us briefly consider the electronic density of states. In general, the number of
electronic states at a particular energy is given by

∑∑
N (ω) = δ[ω − ε(k)]. (6.12)
ξ,σ k

Since B breaks the time-reversal symmetry of the system, k is no longer a good

quantum number. Therefore, we wish to express this as a sum over LL index. We
can write

∑∑
∞
N (ω) = G δ(ω − EN,s ), (6.13)
ξ,σ N =0
s=±

where G is the degeneracy (excluding spin and valley) of each LL, EN,s is given by
Eqn. (6.7) and we must be careful when handling N = 0 (i.e. only s = + contributes).
A calculation identical to that done in Chapter 1.1.2 gives G = eB/(2πℏc). Taking
c = 1, we then write [32, 33]

eB ∑ ∑
∞
N (ω) = δ(ω − EN,s ). (6.14)
2πℏ ξ,σ=± N =0
s=±

In terms of relevant parameters, the degeneracy factor can also be expressed as G =

E12 /(2πℏ2 v 2 ).
To evaluate this numerically, we use the familiar Lorentzian representation of the
√ √
δ-function: δ(x) → (η/π)/(η 2 + x2 ). The results for E1 / B = 12.82 meV/ T, ∆so =
8 meV, B = 1T and a scattering rate of η = 0.2 meV are shown in Fig. 6.4. From top
to bottom, the results are shown for the TI, VSPM and BI regimes, respectively. The
left frames give the total electronic density of states while the right frames show the
individual spin contributions. Again, dashed blue curves correspond to spin up and
the solid red curves are for spin down. As expected, four spin- and valley-polarized
levels are located at the energies of the N =0 LLs (ω = −∆K↓ , ∆K ′ ↓ , −∆K↑ and

68
 80
∆z=4 meV TI
E1/√B=12.82 meV/√T
∆so=8 meV
60 B=1T
η=0.2 meV
N(ω) ([ħv]-2 meV)

K↓ K'↓ K↑ K'↑
40

20

(a) (d)
0
∆z=8 meV
VSPM
60
N(ω) ([ħv]-2 meV)

K↓ K'↓/K↑ K'↑
40

20

(b) (e)
0
∆z=10 meV
BI
60
N(ω) ([ħv]-2 meV)

K↓ K↑ K'↓ K'↑
40

20

(c) (f)
0
-20 -10 0 10 20 -10 0 10 20
ω (meV) ω (meV)

Figure 6.4: Total electronic density of states (left frames) and the spin dependent
density of states (right frames).

69
∆K ′ ↑ ). The band inversion associated with the TI to BI transition is seen in the
shift in onset of two N = 0 levels. In the VSPM phase, the two middle N = 0
peaks coalesce and result in a valley-spin degenerate peak. While the N = 0 levels
are spin- and valley-polarized, all higher features are degenerate [thus, the double
weighting in the total density of states (left frames)]. As the LLs in graphene have
been detected experimentally [111–113], the four low-energy spin- and valley-polarized
levels in silicene should also be observed by scanning tunnelling spectroscopy.
The difference between the insulating and VSPM phase is clearly seen in the
integrated density of states. We define the total integrated density of states up to
energy ωmax as

∑ ∫ ωmax
I(ωmax ) = Nξσ (ω)dω, (6.15)
ξ,σ=± 0

where Nξσ (ω) is given by the single ξσ component of Eqn. (6.14). In Fig. 6.5, I(ωmax )
is plotted as a function of the cutoff energy ωmax . Note: all the parameters are the
same as those used in Fig. 6.4 except now a scattering rate of η = 0.05 meV has been
taken for clarity. Increasing η would cause the edges of the steps to smear. Here,
800
(a) E1/√B=12.82 meV/√T ∆z=4 meV (b) ∆z=8 meV
∆so=8 meV
η=0.05 meV
600
I(ωmax ) ([ħv]-2 meV2)

30
30

400
0
0 10 0
TI 0 10
200
B=0T VSPM
B=1T B=0T
B=1.5T B=1T
0
0 10 20 30 40 0 10 20 30 40 50
ωmax (meV) ωmax (meV)

Figure 6.5: Integrated density of states for the TI and VSPM regimes.

only the TI and VSPM regimes are shown as both insulating phases are analogous.

70
For the TI regime [Fig. 6.5(a)], the two gaps in the B = 0 bands are finite, and equal
to 2 and 6 meV. Three values of magnetic field are used: B = 1T (solid red curve),
B = 1.5T (dashed blue curve), and B = 0 (dash-dotted black curve). For B = 0, all
the electronic states are gapped and I(ωmax ) = 0 for ωmax less than the minimum gap
∆min = 2 meV. A change in slope occurs at ωmax = ∆max = 6 meV (this is highlighted
in the inset). For higher energy, I(ωmax ) ∝ ωmax
2
. This can be traced back to the
linear energy dispersion of the Dirac fermions. As seen in the previous figure, a finite
magnetic field causes the density of states to form a series of LLs. I(ωmax ) remains
zero for ωmax < ∆min = 2 meV after which a vertical step is seen; this is associated
with the occupation of the lowest LL. The height of the step reflects the spectral
weight of the δ-function. The first two steps have the same height while the other
steps are twice as large due to the spin and valley degeneracy of the N ̸= 0 levels.
Note that the energy range between the steps alternates between small and large. As
ωmax increases, I(ωmax ) tends toward its B = 0 value. For B = 1.5T, the height of the
steps has increased due to the B dependence of the LL degeneracy [see Eqn. (6.14)].
The energy range between steps is also changed as it reflects the energy between
neighbouring levels. Similar results are found for the BI regime (most importantly,
I(ωmax ) is also zero for small energies). In Fig. 6.5(b), we show the results for the
VSPM case (∆z = ∆so ). Here, there is only a single gap for B = 0 and the lowest
energy band has closed into a Dirac point. When B = 0 (dash-dotted black curve),
a change in slope is seen for I(ωmax ) at ωmax = ∆max . However, a finite integrated
density of states persists to ωmax → 0. In a finite magnetic field, I(ωmax → 0) ̸= 0,
and is constant in the entire range of ωmax < ∆max . This is a signature of the N = 0
level which remains pinned at zero energy. This different behaviour, can be employed
to monitor the topological phase transition.

6.3 Magneto-Optical Conductivity of Silicene

Let us now move on to the primary focus of this chapter. With the energy dispersion
and wave-functions of silicene in the presence of an external magnetic field, we are now

71
able to calculate the magneto-optical conductivity. We use the formula [32, 33, 106]
(in SI units)
⟨ ⟩⟨ ⟩
′ ′
iℏ ∑ ∑
∞ ∑ fM,s′ − fN,s N̄ s|ĵα |M̄ s M̄ s |ĵβ |N̄ s
σαβ (Ω) = , (6.16)
2πlB ξ,σ=± N,M =0 s,s′ =± EN,s − EM,s′ ℏΩ + EM,s′ − EN,s + iℏ/(2τ )
2

√
where lB = ℏ/(eB), fn,s is the Fermi function for state n in band s, τ is the
relaxation time and ĵα is the current function, equal to (e/ℏ)(∂ Ĥ/∂kα ). As we are
interested in the zero temperature case, fn can be modelled by the Heaviside step
function Θ(µ − En,s ). Our primary task will be to evaluate the appropriate matrix
elements. We begin by considering the longitudinal conductivity (i.e. α = β = x).
First, consider ⟨N̄ s|ĵx |M̄ s′ ⟩ = evξ⟨N̄ s|τ̂x |M̄ s′ ⟩ with τ̂x the usual pseudospin Pauli
matrix. We have,

  
( ) 0 1 −iAN,s |N − 1⟩
⟨M̄ s′ |τ̂x |N̄ s⟩ = iAM,s′ ⟨M − 1| BM,s′ ⟨M |   
1 0 BN,s |N ⟩

= iAM,s′ BN,s δM −1,N − iBM,s′ AN,s δM +1,N , (6.17)

and

⟨N̄ s|τ̂x |M̄ s′ ⟩ = iAN,s BM,s′ δM,N −1 − iBN,s AM,s′ δM −1,N . (6.18)

Therefore,

⟨N̄ s|τ̂x |M̄ s′ ⟩⟨M̄ s′ |τ̂x |N̄ s⟩ = (AM,s′ BN,s )2 δM −1,N + (BM,s′ AN,s )2 δM +1,N

− 2AM,s′ BN,s BM,s′ AN,s δM −1,N δM +1,N .

(6.19)

Note that the last term is never nonzero; thus,

⟨N̄ s|τ̂x |M̄ s′ ⟩⟨M̄ s′ |τ̂x |N̄ s⟩ = (AM,s′ BN,s )2 δM −1,N + (BM,s′ AN,s )2 δM +1,N . (6.20)

72
The Kronecker δ-functions show that optical transitions can only occur between states
of index M and N ±1. These are the well known harmonic oscillator selection rules for
LL transitions [90,91,106]. Using this result in Eqn. (6.16), we obtain the longitudinal
magneto-optical conductivity [32, 33]

iℏe2 v 2 ∑ ∑ ∑ Θ(µ − EM,s′ ) − Θ(µ − EN,s )

∞
σxx (Ω) =
2πlB2
ξ,σ=± N,M =0 s,s′ =±
EN,s − EM,s′
(AM,s′ BN,s )2 δM −1,N + (BM,s′ AN,s )2 δM +1,N
× .
ℏΩ + EM,s′ − EN,s + iℏ/(2τ )
(6.21)

Favouring the explicit real and imaginary representation [32, 33]

{ }
2v 2 ℏeB ∑ ∑ ∑ Θ(µ − EM,s′ ) − Θ(µ − EN,s )
∞
σxx (Ω)
Re =
σ0 π ξ,σ=± N,M =0 s,s′ =±
EN,s − EM,s′
[ ] Γ
× (AM,s′ BN,s )2 δN,M −ξ + (BM,s′ AN,s )2 δN,M +ξ , (6.22)
Γ2 + (ℏΩ + EM,s′ − EN,s )2

and [33]

{ }
2v 2 ℏeB ∑ ∑ ∑ Θ(µ − EM,s′ ) − Θ(µ − EN,s )
∞
σxx (Ω)
Im =
σ0 π ξ,σ=± N,M =0 s,s′ =±
EN,s − EM,s′
[ ] ℏΩ + EM,s′ − EN,s
× (AM,s′ BN,s )2 δN,M −ξ + (BM,s′ AN,s )2 δN,M +ξ , (6.23)
Γ2 + (ℏΩ + EM,s′ − EN,s )2

where σ0 = e2 /(4ℏ) and Γ = ℏ/(2τ ) is a phenomenological transport scattering rate

which is taken to be constant.
Similarly, the real and imaginary parts of the transverse Hall conductivity (α = x
and β = y) are given by [33]

{ }
2v 2 ℏeB ∑ ∑ ∑ Θ(µ − EM,s′ ) − Θ(µ − EN,s )
∞
σxy (Ω)
Re = ξ
σ0 π ′
ξ,σ=± N,M =0 s,s =±
EN,s − EM,s′
[ ] ℏΩ + EM,s′ − EN,s
× (AM,s′ BN,s )2 δN,M −ξ − (BM,s′ AN,s )2 δN,M +ξ , (6.24)
Γ2 + (ℏΩ + EM,s′ − EN,s )2

73
and [32, 33]

{ }
2v 2 ℏeB ∑ ∑ ∑ Θ(µ − EM,s′ ) − Θ(µ − EN,s )
∞
σxy (Ω)
Im =− ξ
σ0 π ξ,σ=± N,M =0 s,s′ =±
EN,s − EM,s′
[ ] Γ
× (AM,s′ BN,s )2 δN,M −ξ − (BM,s′ AN,s )2 δN,M +ξ , (6.25)
Γ2 + (ℏΩ + EM,s′ − EN,s )2

respectively, where we have used ĵy = evτ̂y . Note that Reσxx (Ω) and Imσxy (Ω) corre-
spond to the absorptive parts of the longitudinal and transverse Hall conductivities,
respectively, which means that the absorption peaks shown here will appear as dips
in the experimentally measured transmission [103].

6.3.1 Longitudinal Conductivity

The absorptive part of the longitudinal magneto-optical conductivity is given by

Eqn. (6.22). A numerical evaluation of Eqn. (6.22) is shown in Fig. 6.6 for varying ∆z
at charge neutrality (µ = 0). For generality, we have chosen to scale everything by ∆so .
The relevant parameters are B/∆2so = 65.7G/meV2 and Γ = 0.05∆so , where we again
√ √
use E1 / B = 12.82 meV/ T. In the forthcoming discussion, an interband process
refers to a transition between LLs which arise from different B = 0 bands. Likewise,
an intraband transition is one between levels from the same B = 0 band. For ∆z = 0
(lowest purple curve) there are strong absorptive responses associated with interband
transitions subject to the selection rules. The energy of the first feature is set by
the difference in energy of the N = 0 and 1 LLs, that is, Ω = ∆so + E1,+ (∆z = 0).
As ∆z is increased, each interband feature splits in two. This is a result of all LLs
becoming spin split. The intensities of the peaks are reduced due to a redistribution
of spectral weight between the features. The lowest of the split peaks moves to lower
energy as ∆z is increased which is a signature of the closing of the smallest band
gap of the B = 0 bands. The second split peak moves higher in energy due to the
second band gap increasing. When ∆z = ∆so , the first feature is set by Ω = E1,+
K↑
, or
K ↓ ′
equivalently, Ω = E1,+ . As the system transitions through the VSPM state into the
BI regime, all interband features move to higher energy. This signals the reopening

74
 80 µ/∆so=0
BI 2
B/∆2so=65.7G/meV

60 BI

Reσxx(Ω)/σ0
VSPM
40
TI

20
TI

TI
0
0 1 2 3 4
Ω/∆so

Figure 6.6: Real part of the longitudinal magneto-optical conductivity of silicene for
varying ∆z and µ = 0. The results are vertically offset by 15 units.

of the lowest gap. The variation in interband behaviour is a distinct signature of the
different insulating regimes. By monitoring this behaviour, it should be possible to
experimentally identify the VSPM transition and the size of ∆so .
The effect of adjusting the chemical potential when ∆z = 0 is shown in Fig. 6.7.
Again, the relevant parameters are B/∆2so = 65.7G/meV2 and Γ = 0.05∆so . This
30
2
B/∆2so=65.7G/meV
25
∆z/∆so=0
µ/∆so=0
20 Intraband µ/∆so=1.2
Reσxx(Ω)/σ0

µ/∆so=1.7
µ/∆so=2.3
15

10 Interband

0
0 1 2 3 4
Ω/∆so

Figure 6.7: Real part of the longitudinal conductivity of the Kane-Mele model for a
2D QSHI (∆z = 0) for varying chemical potential.

75
system corresponds to the Kane-Mele Hamiltonian for a QSHI. Even though only a
single gap appears in the band structure, these results differ from those shown in
Ref. [90, 91] for graphene with an asymmetry gap as that system is a band insulator
(one spin-degenerate N = 0 LL per valley) as opposed to a topological insulator
(two spin-split N = 0 levels per valley). This difference affects the spin and valley
contributions to the low-energy features seen in Refs. [90, 91]. Including a finite
chemical potential that ensures the Fermi energy lies between all the N = 0 and 1
LLs (dashed red curve), causes the lowest interband feature of the µ = 0 system (solid
black curve) to decrease in intensity by a factor of two as half the spectral weight
is shifted to a low-energy intraband peak [90, 91]. However, as opposed to Ref. [90],
where the two lowest peaks are valley-polarized, here the two lowest features of the red
curve are spin-polarized [33] as marked by the red arrows. This means that the lowest
peak is associated entirely with spin-up electrons while the next peak is solely due to
spin-down electrons. These two peaks are made of an equal mixture of valley label.
When µ is situated between the N = 1 and 2 levels (dash-dotted blue curve), the
interband absorption peak associated with transitions to and from the N = 0 levels
disappears due to Pauli blocking. The spectral weight of the next highest response is
diminished as interband transitions to the N = 1 level are Pauli blocked. With higher
µ, a similar redistribution of interband to intraband spectral weight is observed. The
√
low-energy intraband signature moves to lower energy as a result of the N spacing
between LLs which causes adjacent levels to be closer together at higher energy.
The effect of varying the chemical potential for finite ∆z is shown in Fig. 6.8(a),
(b) and (c) for the TI (∆z /∆so = 0.75), VSPM (∆z /∆so = 1) and BI (∆z /∆so = 1.25)
regimes, respectively. In all cases, B/∆2so = 65.7G/meV2 and Γ = 0.05∆so . A finite
∆z spin splits the LLs which allows for a much richer structure in the magneto-
optical conductivity. Similar to the case of ∆z = 0, moving the Fermi energy causes
a redistribution of spectral weight from interband to intraband responses. The low-
energy transitions which yield the results of Fig. 6.8(a) are marked by arrows in
Fig. 6.9. The LLs at each valley are represented by circles coloured blue and red for
spin up and down, respectively. Horizontal lines are drawn between the two valleys

76
 50 2
50 2
B/∆2so=65.7G/meV (a) B/∆2so=65.7G/meV (b)
∆z=0.75∆so ∆z= ∆so
µ/∆so=0 µ/∆so=0
40 µ/∆so=0.4 40 µ/∆so=1.2
µ/∆so=1.1 µ/∆so=1.9
µ/∆so=1.9
Reσxx(Ω)/σ0

Reσxx(Ω)/σ0
30 30
K↑/K'↓
K'↑ K↑ K'↓ K↓
K'↑ K↓
20 20

K↑ K'↓

10 10

TI VSPM
0 0
0 1 2 3 4 0 1 2 3 4
Ω/∆so Ω/∆so
50 2
B/∆2so=65.7G/meV (c)
∆z=1.25∆so
µ/∆so=0
40 µ/∆so=0.4
µ/∆so=1.25
µ/∆so=2.25
Reσxx(Ω)/σ0

30

K'↑ K'↓ K↑ K↓

20
K'↓ K↑

10

BI
0
0 1 2 3 4
Ω/∆so

Figure 6.8: Real part of the longitudinal conductivity of silicene for varying chemical
potential in the (a) TI, (b) VSPM and (c) BI regimes. The results are vertically offset
by (a) 10 units, (b) 16 and (c) 9 units.

to signify the total LL contribution in the system. These lines are dashed blue, solid
red and dash-dotted purple lines for spin-up, spin-down and spin-degenerate levels,
respectively. The LL index is marked on the far right and the various values of
chemical potential used in Fig. 6.8(a) are given by the solid cyan line. The lowest
energy transitions for each value of chemical potential are represented by coloured
arrows where transitions are only permitted between like spin states. The arrows

77
 K K'
N=2++
N=2+
N=1+
N=1
N=0
N=0
µ N=0
N=0
N=1--
N=1-
N=2-
N=2
TI (∆z=0.75∆so)

Figure 6.9: Schematic of the allowed transitions between LLs at both valleys in the
TI regime for various values of chemical potential.

are ordered by increasing length such that a change in length corresponds to a new
absorption feature in the conductivity. Thus, in reference to Fig. 6.8(a), the first peak
of the solid black curve is associated with the first two black arrows in Fig. 6.9, and
the second, third and fourth features are represented by the second, third and fourth
pairs of black arrows, respectively. For the dashed red curve, the first red arrow
yields the first feature while the second arrow gives rise to the second peak. The next
absorption line results from the next two arrows while the remaining pairs of arrows
yield the final two features, respectively. With respect to the dash-dotted blue curve,
the first four blue arrows result in the first four peaks, respectively, while the next two
features come from the reaming two pairs of arrows. The strong low-energy signature
of the solid green curve comes from the first two arrows with the two higher-energy
peaks being represented by the remaining pairs of arrows, respectively.
Returning to Fig. 6.8(a), when µ is placed between the N = 0 spin-up levels at
the K and K ′ points (dashed red curve) such that it is above three N = 0 levels
(i.e. |∆K↑ | < µ < |∆K ′ ↑ |), the lowest interband feature redistributes its spectral
weight between itself and a low-energy intraband peak. These intraband transitions
result entirely from spin-up electrons at the K point while the remaining interband
response is associated entirely with spin-down electrons at K ′ [33]. If µ is now sit-

78
uated between all the N = 0 and 1 LLs [i.e. |∆K ′ ↑ | < µ < ε1 (∆K↑ )], the second
interband feature redistributes its spectral weight into a lower-energy intraband sig-
nature. These intraband transitions result entirely from spin-up electrons at K ′ and
the remaining interband transitions are associated with spin-down at K. The other
two spin- and valley-polarized responses remain; for this value of µ, there are four
robust spin- and valley-polarized peaks making it possible to generate charge carriers
of definite spin and valley label [32, 33]. The two lowest features are associated with
intraband transitions while the upper two result from interband transitions. The spin-
K ↑ ′ K ↓ ′
and valley-polarized responses onset at Ω = E1,+ − ∆K ′ ↑ , E1,+ − ∆K ′ ↓ , E1,+
K↑
− ∆K↑
and E1,+
K↓
− ∆K↓ for spin up and down at K ′ , and spin up and down at K, respectively.
While valley-spin polarization is predicted in the response to circularly-polarized light
when B = 0 [30, 63], it is limited to only two valley-spin-polarized species which can
only be selected by changing the insulating regime or handedness of the polarized
light [30, 114] (see Chapter 4). Here, robust valley-spin polarization is present in
both insulating regimes even in the longitudinal response [32,33] allowing any valley-
spin-polarized response to be isolated by tuning the incident photon frequency. The
transitions which give rise to this polarized quartet are schematically illustrated in
Fig. 6.10 for the TI (upper) and BI (lower) regimes. The location of these polarized
peaks in frequency space can be tuned by ∆z and/or B. These polarized features
should be visible in experiment as the magnetic and electric field values required to
observe them are well within experimental limits. Aside from B and Ez , the deter-
mining factor in the onset frequency of the polarized response is the size of the spin
orbit gap, ∆so . For ∆so = 1.55 meV, the curves shown here correspond to a B of
∼ 0.15T and for ∆so = 7.9 meV, B ≈ 4.1T. In the former case, the splitting of the
peaks in the polarized quartet is ∼ 1.5 meV, a resolution easily achieved in broad-
band optics. In this case, the quartet sits in the range of 1 − 2 THz where broadband
experiments can be done [115]. If B = 1T, such peaks will shift to the far infrared
in the range of 17 − 20 meV. If instead ∆so is as large as 8 meV, the splitting of the
quartet will be about 3-5 meV for B = 1T and the quartet will fall in the range of
13 − 25 meV. Conductivity experiments on graphene have spanned the range from

79
 K K'
2 2
2 2
1 1
1 1
µ
ε 0

0
ℏvk
TI 0

-1 -1
-1 -1
-2 -2
-2 -2

K K'
2 2
2 2
1 1
1 1
µ
ε 0

0 ℏvk BI 0

-1 -1
-1 -1
-2 -2
-2 -2

Figure 6.10: Schematic of the optical transitions which give rise to a quartet of valley-
spin-polarized absorption lines.

THz to eV [98–101, 104, 116] with B∼ 0 to 18T [98–101, 104, 116] and resolutions of
order less than a meV; therefore, experiments on silicene should observe this polarized
behaviour. As µ is increased further, only one intraband signature remains which is
associated with the semiclassical cyclotron resonance frequency. The spectral weight
of this feature increases with increasing µ.
In the VSPM phase [Fig. 6.8(b)] the potential for four spin- and valley-polarized
responses is no longer present as two of the N = 0 LLs are now at zero energy due
to the closing of the lowest gap of the B = 0 bands. The conductivity curves look
similar to the results for the TI regime; however, when the chemical potential is
placed between all the N = 0 and 1 LLs (dash-dotted blue curve), there are only
two valley-spin-polarized responses. For this value of chemical potential, the lowest
feature of the µ = 0 system remains and is made of an equal mixture of spin and

80
valley species. This absorption signature is associated equally with interband and
intraband transitions. By examining the LLs which contribute to the degenerate
N = 0 level (see Fig. 6.2) it is apparent that the spin-up contribution is associated
with an intraband transition while the spin-down piece results from an interband
transition. This is similar to the behaviour of the N = 0 LL in graphene [91]. Again,
there is a spectral weight redistribution to a strong intraband response for increased
µ. A double peak feature is present in the intraband response as seen by the lowest
feature in the solid green curves of Fig. 6.8. This results from the LLs being spin
split; thus, intraband transitions between different spin levels at a given valley are not
of equal energy. The separation between the double peaks decreases as µ increases.
The conductivity in the BI regime [see Fig. 6.8(c)] is analogous to that of the TI
regime; however, in this case, the spin and valley labels of the two middle spin- and
valley-polarized responses switch due to the band inversion. While the relative onset
inverts, the separation between the two middle peaks always remains at 2|∆K↑ | for
∆z ̸= 0 and, thus, decreases with increasing ∆z in the TI regime until the polarization
is lost in the VSPM state. The separation then increases with increasing ∆z in the
BI regime, however, the spin and valley labels switch. Aside from the four spin-
and valley-polarized peaks, all other transitions are made of an equal spin and valley
mixture.
The onset frequency of transitions is determined by the energy difference between
LLs and is therefore controlled by the magnetic and electric fields. The other deter-
mining factor is the strength of the spin-orbit gap. Thus, a careful tuning of B and
∆z should allow for a determination of ∆so .

6.3.2 Transverse Hall Conductivity

The absorptive part of the transverse Hall conductivity is given by Eqn. (6.25). The
numerical results for ∆z = 0.75∆so and varying chemical potential are shown in
Fig. 6.11. Again, B/∆2so = 65.7G/meV2 and Γ = 0.05∆so . The values of µ were
chosen to correspond with those in Fig. 6.8(a) for Reσxx (Ω). Clearly, Imσxy (Ω) is
related to the longitudinal response. However, not all the features in the longitudinal

81
 50 2
B/∆2so=65.7G/meV
∆z=0.75∆so
µ/∆so=0
40 µ/∆so=0.4
µ/∆so=1.1
µ/∆so=1.9

Im σxy(Ω)/σ0
30
K'↑ K↑ K'↓ K↓

20
K↑ K'↓

10

0
0 1 2 3 4
Ω/∆so

Figure 6.11: Imaginary part of the transverse Hall conductivity of silicene for varying
chemical potential. These results are vertically offset by 10 units.

conductivity have counterparts in the transverse Hall conductivity. The only features
that are present in Imσxy (Ω) come from transitions between EN,s′ and EN ±1,+ when
the transition from EN ±1,s′ to EN,+ is Pauli blocked. Therefore, the only possible
transitions which give nontrivial features in the absorptive part of the transverse Hall
conductivity are those associated with transitions involving N = 0. For example,
when µ = 0, Imσxy (Ω) is zero for all Ω as no transitions meet the criteria for a finite
Hall response. When µ = 0.4∆so , the only finite Hall response is due to transitions
between the spin-up N = 0 and 1 LLs at the K point, and N = 0 and 1 spin-down
levels at K ′ as the equal energy transitions from N = −1 to 0 spin-down levels at K ′ ,
and N = −1 to 0 spin-up levels at K are Pauli blocked (see Fig.6.9). For µ = 1.1∆so ,
four transitions meet the criteria for a finite response and, hence, four features are
present in the conductivity. As was seen in Reσxx (Ω), four valley-spin-polarized
absorption lines can be generated. Mathematically, this behaviour is clearly seen by
examining Eqn. (6.25). Noting the minus sign between the two Kronecker δ-function
terms, we see that there will be no contribution when the two prefactors are the same.
The transverse Hall conductivity is important when considering circularly-polarized

82
light which we will now discuss.

6.3.3 Results for Circularly-Polarized Light

The optical response to circularly-polarized light is determined by σxx (Ω) ± iσxy (Ω)
for right- (+) and left-handed (−) polarization [93]. The absorptive part of the
conductivity is
Reσ± (Ω) = Reσxx (Ω) ∓ Imσxy (Ω). (6.26)

This is easily evaluated with Eqns. (6.22) and (6.25). The conductivities for left-
and right-handed polarization are shown in Fig. 6.12(a) and (b), respectively. Here,

20 (a) ∆z=0.5∆so
µ=3.0∆so
Reσ_ (Ω)/σ0

10

0
(b)
20
Reσ (Ω)/σ0

+
10

0
0 2 4 6 8 10 12 14 16 18 20
Ω/∆so

Figure 6.12: The absorptive part of the conductivity response to (a) left-handed and
(b) right-handed circularly-polarized light.

∆z /∆so = 0.5, B/∆2so = 657G/meV2 and µ/∆so = 3.0. The value of chemical
potential was chosen so that the Fermi energy lies between the N = 0 and 1 LLs.
This means that a quartet of spin- and valley-polarized absorption lines are present in
both Reσxx (Ω) and Imσxy (Ω). To clearly separate the features, a larger magnetic field
is used (ten times greater than those previously employed). As it is only the quartet
of polarized peaks that meets the criteria for a nonzero transverse Hall response, they
double in spectral weight for the left-handed response and are not present for right-
handed polarization. In all cases, the higher-energy features are identical to those of

83
the longitudinal response.

6.4 Concluding Remarks

The valley-spin polarization in the magneto-optical conductivity is of particular in-
terest when compared to that of the B = 0 response to circularly-polarized light (see
Chapter 4.2). For no magnetic field, it is possible to generate charge carries with
either valley-spin label K ↑ or K ′ ↓. In the TI regime, this was done with either
left- or right-handed polarized light, respectively. In the BI regime it was done with
either right- or left-handed polarized light, respectively. Thus, to generate two types
of polarized charge carriers requires a tuning of the handedness of the light or a
changing of the insulating regime. With the application of a magnetic field, however,
valley-spin-polarized charge carriers can be produced with any spin and valley label
and can be obtained in the longitudinal response as well as with circularly-polarized
light. The polarization of the charge carriers can be changed simply by changing
the frequency of the incident light and the onset frequency can be controlled by B
and/or ∆z . These polarized peaks should be visible in experiment as the magnetic
and electric field values required to observe them are well within experimental limits.
Experiments may also provide a measure of ∆so , for instance, by examining when the
two middle polarized peaks of the polarized quartet become superimposed.
In principle, one should include the effects of interacting electrons. While such
interactions have been discussed in detail for graphene [110, 117], magneto-optical
experiments in graphene have been well-described by the single-particle picture. This
description worked well up to very large magnetic fields of ∼ 16T [98–103,110]. There-
fore, for modest magnetic fields of O(1T), electron-electron interactions are not ex-
pected to modify our main results. Our calculations have also assumed freestanding
silicene or, equivalently, silicene on an insulating substrate. Currently silicene is
fabricated on metallic substrates (i.e. silver [40, 118]), but efforts are under way to
find alternative substrates (ex. ZrB2 [53]). Like graphene, the choice of insulating
substrate is not expected to qualitatively affect our results.

84
 CHAPTER VII

MAGNETIC OSCILLATIONS IN
SILICENE

In a magnetic field, metals are known to magnetize. This will be done in one of two
ways: either by an attempt to align the spins with the direction of the magnetic field
(a paramagnetic response), or by moving in such a way as to induce a magnetic field
in the opposite direction to try and minimize its effect (a diamagnetic response). The
diamagnetism of metals is due to the formation of LLs which result from the precession
of electrons around the field lines. The quantity that describes this response is the
magnetization (M ). In essence, the magnetization provides a measure of the torque
(τ ) a system will experience when placed in a magnetic field (B). This is governed
by the simple relation

1
τ = M × B, (7.1)
N

where N is the density of magnetic moments in the system. It is also useful to examine
the magnetic susceptibility (χ = ∂M/∂B) which relates the degree of magnetization
to the magnetic field through

M = χB. (7.2)

In this chapter, we examine some of the interesting properties of this fundamental

quantity.

85
7.1 The de Haas-van Alphen Effect
In 1930, de Haas and van Alphen measured the magnetic field dependence of the
magnetization of a bismuth sample [119, 120]. Using fields of order kG, they found
oscillations in M/B. The usefulness of this curious observation went largely unnoticed
until Onsager’s work in 1952 [121]. Since the 1960s, this oscillating behaviour has
also been observed in the magnetic susceptibility of many metals. When plotted as a
function of 1/B, a remarkable periodic dependence is seen in χ. Often times, two or
more periods are superimposed. This phenomenon is directly tied to the LL dispersion
of the system and the corresponding orbit areas. As will be discussed below, this so-
called de Haas-van Alphen (dHvA) effect has become a dominant and powerful tool
for probing the shape of the Fermi surface. To establish the simple theory behind
this phenomenon, we will follow the derivation of Ashcroft and Mermin [6].
To begin, let us consider the semiclassical motion of an electron in a uniform
magnetic field B. The semiclassical equations which govern the electron’s motion are

1 ∂E(k)
ṙ = v(k) = , (7.3)
ℏ ∂k

and

−e
ℏk̇ = v(k) × B. (7.4)
c

As a consequence of the above equations, the component of k along B and the elec-
tron’s energy E(k) are constants of the motion. Thus, in k-space, electrons move
along curves which are determined by the intersection of planes perpendicular to B
and constant energy surfaces. Let us now define the projection of the real space orbit
in a plane perpendicular to B:

r⊥ = r − (B̂ · r)B̂. (7.5)

86
It follows from Eqn. (7.4) that

eB [ ] eB
B̂ × ℏk̇ = − ṙ − (B̂ · ṙ)B̂ = − ṙ⊥ . (7.6)
c c

This can be integrated to give

ℏc
r⊥ (t) − r⊥ (0) = − B̂ × [k(t) − k(0)] . (7.7)
eB

By definition, r⊥ is perpendicular to B̂; therefore, k is simply (up to a scale factor)

r⊥ rotated by 90 ◦ about B̂.
We are also interested in the rate at which these orbits are traversed. To begin,
consider an orbit of energy E which moves in a plane perpendicular to the magnetic
field. The time required to travel between two points k1 and k2 which lie on the orbit
is
∫ ∫  
t2 k2  dt 
t2 − t1 = dt =   dk. (7.8)
 dk 
t1 k1

We can eliminate the inverse k̇ by employing Eqns. (7.3) and (7.4). This yields
∫
ℏ2 c k2
dk
t2 − t1 = , (7.9)
eB k1 |(∂E/∂k)⊥ |

where (∂E/∂k)⊥ is the projection of ∂E/∂k onto the plane with normal vector B̂. For
clarity, let us discus the geometric interpretation of (∂E/∂k)⊥ . We begin by defining
the vector ∆(k) which exists in the plane of the orbit and is perpendicular to the
orbit at a given point k. This vector also joins k to another orbit of energy E + ∆E
in the same plane as the original. This is illustrated in Fig. 7.1(b). For small ∆E,
( )
∂E ∂E
∆E = · ∆(k) = · ∆(k). (7.10)
∂k ∂k ⊥

Recalling that ∂E/∂k is perpendicular to a constant energy surface, we note that

87
 kz
B

ky
dk k2
k2 k
k1 ∆(k)
kz
ε k1
ky
ε ε+∆ε
ε+∆ε
kx
kx (a) (b)

Figure 7.1: (a) k-space orbit of an electron in a magnetic field. (b) Area between two
orbits of different energies bounded by k1 and k2 .

(∂E/∂k)⊥ is perpendicular to the orbit and thus parallel to ∆(k). Therefore,

( ) 
 ∂E 
∆E =   ∆(k), (7.11)
∂k ⊥ 

which allows Eqn. (7.9) to be rewritten as

∫
ℏ2 c 1 k2
t2 − t1 = ∆(k)dk. (7.12)
eB ∆E k1

Clearly, the integral of ∆(k) is nothing more than the area of the plane between the
two orbits E and E + ∆E from the points k1 to k2 , which we denote A12 . Taking the
limit ∆E → 0, we obtain

ℏ2 c ∂A12
t2 − t1 = . (7.13)
eB ∂E

For a closed loop (k1 = k2 ), t2 − t1 ≡ T is the period of the orbit and (∂/∂E)A12 ≡
(∂/∂E)A where A is the area in k-space which is enclosed by the orbit. Thus, we

88
arrive at the simple expression

ℏ2 c ∂A
T = . (7.14)
eB ∂E

This can be written in the form of the free electron result,

2π 2πmc
T = = , (7.15)
ωc eB

by defining an effective cyclotron mass

ℏ2 ∂A
m∗ = . (7.16)
2π ∂E

Since an electron precesses about the Fermi surface in planes perpendicular to the
magnetic field, by varying the orientation of B, a simple map of the Fermi surface
can be obtained.
Let us now consider a collection of free electrons in a cubical box of side length L.
For a magnetic field in the z direction, the orbital energy levels are found by solving
the 3D Schrödinger equation in the presence of B = B ẑ which is a trivial extension
of the work done in Chapter 1. The energies are

ℏ2 kz2
En (kz ) = + (n + 1/2)ℏωc , (7.17)
2m

where

eB
ωc = (7.18)
mc

is the cyclotron frequency, n ∈ Z+ gives the level number and kz = 2πnz /L (for nz ∈
Z) takes the same values as the B = 0 result (since the Lorentz force has no component
in the direction parallel to B). As previously discussed, each level is highly degenerate
with degeneracy eBL2 /(hc) for a single spin species. The degeneracy reflects the
classical behaviour of an electron with a given energy and kz which spirals around a
line parallel to ẑ with arbitrary x and y coordinates. Applying Bohr’s correspondence

89
principle (which states that the energy difference between two adjacent levels is h
times the frequency of the semiclassical motion) to the levels n and n + 1 with a given
kz , we find

h
En+1 (kz ) − En (kz ) = , (7.19)
T (En , kz )

where

ℏ2 c ∂A(E, kz )
T (En , kz ) = (7.20)
eB ∂E

is the period of the semiclassical orbit of energy En and momentum kz , with A(E, kz )
being the enclosed k-space area of the orbit. Therefore,

∂A(E, kz ) 2πeB
(En+1 − En ) = . (7.21)
∂E ℏc

Let us now restrict our attention to En of order EF . In this regime, the energy differ-
ence between neighbouring levels (ℏωc ) is typically much smaller than the energies of
the levels. We can, thus, make the approximation

∂A(E, kz ) A(En+1 ) − A(En )

= , (7.22)
∂E En+1 − En

to obtain the simple expression

2πeB
A(En+1 ) − A(En ) = . (7.23)
ℏc

From this, we note that classical orbits for the same kz at neighbouring allowed
energies differ in their enclosed area by the fixed amount

2πeB
∆A = . (7.24)
ℏc

Equivalently, at large n, the enclosed area of a cyclotron orbit at an allowed E and

90
kz is given by

A(En , kz ) = (n + γ)∆A, (7.25)

where γ is the Maslov index which is independent of n and is a fully quantum me-
chanical effect which arises due to the single-valuedness of the wave-function [122]
(γ = 1/2 for the present system). Equation (7.25) is the famous Onsager result [121].
Having established that electrons in a magnetic field move in helical orbits with a
frequency that is related to the orbit area, we can now explore why a beating pattern
is seen in the magnetization at low field values.
It turns out that the dHvA effect is rooted in the oscillating behaviour of the
density of states [N (E)]. As previously discussed, the density of states has a sharp
peak when the energy E is equal to that of a LL (En ). The density of states at the
Fermi energy EF determines many electronic properties of metals. In the pure limit,
N (EF ) is singular when the magnetic field causes the energy of a LL on the Fermi
surface to satisfy the Onsager relation:

2πeB
A(EF ) = (n + γ) . (7.26)
ℏc

As n runs through all the positive integers, this condition is satisfied at regular inter-
vals of 1/B such that
( )
1 2πe 1
∆ = . (7.27)
B ℏc A(EF )

This implies that oscillations as a function of 1/B should be present in all quantities
that depend on the density of states at the Fermi energy (nearly all metallic properties
at zero temperature). With this basic understanding of the magnetization, we can
now examine the magnetic response of silicene.

91
7.2 Magnetization of Silicene and the Hall Effect
It is well known that the magnetization of Dirac-like materials is anomalous [123–128].
For charge-neutral graphene, the magnetic susceptibility displays a divergence [97,
126–129] as a function of the inverse magnetic field. This singularity is removed
by finite temperature [128], the formation of a gap [126, 129] leading to massive
Dirac fermions, disorder effects [130], or by charge doping away from the neutrality
point [126, 130]. As previously established, the low-buckled honeycomb lattice of
silicene is characterized by massive Dirac fermions except for the VSPM where one
of the gaps is closed. Here, we study the magnetic properties of this system [131];
particular emphasis is given to the changes in magnetization as the band structure is
tuned between the nontrivial and trivial topological phases.
In this section, we will examine the IQHE in silicene which we derive by the Streda
relation. For the transverse conductivity, the off-diagonal current response is

Ji = σH ϵij Ej , (7.28)

where ϵij = ± for ij = xy and yx, respectively. Therefore Jx = σH Ey and Jy =

−σH Ex . Employing the continuity equation

∂ρ
= −∇ · J , (7.29)
∂t

where ρ and J are the charge and current densities, respectively, and the Maxwell
equation

∂B
∇×E =− , (7.30)
∂t

we arrive at

∂ρ ∂B
= σH , (7.31)
∂t ∂t

92
which implies that

∂ρ
σH = . (7.32)
∂B

This is the Streda relation which relates the Hall conductivity to the B dependence of
the charge density. If we restrict our attention to the change in internal energy (dΩ)
as B and electron number (N ) are allowed to vary, we obtain the thermodynamic
relation

dΩ = −MdB + µdN, (7.33)

where M is the magnetization and µ is the chemical potential. This yields the two
relations

∂Ω ∂Ω
M=− , and µ= . (7.34)
∂B ∂N

Therefore,

∂M ∂ ∂Ω ∂ ∂Ω
=− , and 1= . (7.35)
∂µ ∂B ∂µ ∂N ∂µ

Using ∂Ω/∂µ = N , we obtain

∂M ∂N
=− . (7.36)
∂µ ∂B

Defining the magnetization per unit area M ≡ M/A and using the fact that ρ = −en
where n ≡ N/A is the electron density per unit area, we obtain the simple relation

∂M
σH = e . (7.37)
∂µ

This states that the Hall conductivity (up to a factor of e) is given by the µ derivative
of the magnetization.
We now begin our discussion of the magnetic response of silicene by examining

93
the grand thermodynamic potential [126]
∫ ∞ ( )
Ω(T, µ) = −T N (ω)ln 1 + e(µ−ω)/T dω, (7.38)
−∞

where T is the temperature and N (ω) is the density of states. Note that we have taken
kB = 1. In the absence of impurity scattering, the density of states in a magnetic
field is given by a series of Dirac δ-functions located at the various LL energies [see
Chapter 6]. Allowing for a finite Zeeman interaction, these levels are determined by
the low-energy Hamiltonian [see Eqn. (3.24)]

1 1 1
Ĥξσ = ℏv(ξkx τ̂x + ky τ̂y ) − ξσ ∆so τ̂z + ∆z τ̂z − gs µB Bσ, (7.39)
2 2 2

where gs is the Zeeman coupling strength which we take to be 23 to elucidate the

interesting features; µB = eℏ/(2me ) ≈ 5.78 × 10−2 meV/T is the Bohr magneton
with me , the mass of an electron. As argued in Chapter 6, for silicene, the Zeeman
energy is small and usually ignored [32, 33, 42, 56, 63, 109]. Here, we wish to include
its effects. Applying the standard techniques, Eqn. (7.39) gives the LL dispersion [see
Eqn. (6.7)]
 √
 − 1 gs µB Bσ + s ∆2 + 2N E 2 , N = 1, 2, 3, ...
2 ξσ 1
EN,s
ξσ
= , (7.40)
 −ξ∆ξσ − 12 gs µB Bσ, N =0

√
where, again, s = ± is a band index and E1 ≡ ℏev 2 B ≈ 12.82 meV for B = 1T.
Recall: in a sum over s = ±, there is only a single contribution from the N = 0 levels.
Returning to the density of states, for a single valley and spin,
 
eB  ( ) ∑ ( )
∞
ξσ 
Nξσ (ω) = δ ω − E0
ξσ
+ δ ω − E N,s  . (7.41)
h N =1
s=±

As we shall see, the quantities of interest depend on a derivative of the grand potential
∫∞
with respect to B; therefore, for convenience, we choose to add (µ/2) −∞ N (ω)dω to

94
Eqn. (7.38). Since the integral of the density of states over all energies gives the total
number of states (which is independent of B), this term will not contribute to the
magnetization (−∂Ω/∂B) [34, 36]. At zero temperature, Eqn. (7.38) then becomes
∫ ( ∫ ∫
0
µ) µ
µ ∞
Ω(µ) = ω− N (ω)dω + (ω − µ)N (ω)dω + N (ω)dω. (7.42)
−∞ 2 0 2 0

For silicene,
∫ ∞ ∫ 0
eB
Nξσ (ω)dω = Nξσ (ω)dω − Υ, (7.43)
0 −∞ h

where (for realistic values of Zeeman splitting)



 −σ ∆z < ∆so



 ξ − σ ∆z = ∆so , gs = 0

Υ= 2 . (7.44)



 ξ ∆z = ∆so , gs > 0



ξ ∆z > ∆so

Therefore [34],
∫ µ ∫ 0
eBµ
Ωξσ (µ) = (ω − µ)Nξσ (ω)dω − Υ+ ωNξσ (ω)dω. (7.45)
0 2h −∞

The final term (which does not depend on µ) gives the vacuum contribution and will
simply provide a constant background to the µ dependence of the magnetization. In
the absence of impurity scattering, when summed over ξ and σ, the first two terms
of Eqn. (7.45) give [34]
[ ]
eB ∑ ( ξσ ) ( ) ( ) ∑Nc ( ) ( )
Ω̃(µ) = E0 − µ Θ µ − E0 Θ E0 +
ξσ ξσ
EN,+ − µ Θ µ − EN,+
ξσ ξσ
h ξ,σ=± N =1
eBµ
− Υ, (7.46)
2h

where we have used Eqn. (7.41) for the density of states, Nc is a large value associated
with the band cutoff and we have assumed that all the s = − states are negative. It

95
is important to note that Θ(0) ≡ 1/2 as only half the δ-function situated at ω = 0 is
integrated. A plot of the magnetization derived from Eqn. (7.46) is shown in Fig. 7.2
for ∆z = 4 (solid black) and 8 meV (dashed red). In both cases, a jagged saw-tooth

dM(µ)/dµ (e/h)
2 E1/√B=12.82 meV/√T
∆so=8 meV
1
B=1T
0

-1
40
0 2 4 6 8 10
µ (meV)
M(µ) (e/h meV)

20

0
∆z=4 meV TI
∆z=8 meV VSPM

-20
0 10 20 30
µ (meV)

Figure 7.2: Magnetization as a function of µ in the TI phase and VSPM regimes for
gs = 0. Inset: the slope of the magnetization as a function of µ. Plateaus have been
offset from their integer values for clarity.

oscillation is seen. For ∆z = 4 meV, the system is in the TI phase and, thus, two
N = 0 LLs are at positive energy. For low chemical potential, two kinks are seen in
M (µ) at µ = 2 and 6 meV associated with the spin-up N = 0 levels at K and K ′ ,
respectively. For the VSPM, only one LL exists for ω > 0 and, hence, only one kink is
observed at µ = 8 meV. When µ is greater than the N = 0 levels, the usual saw-tooth
behaviour is present. The jumps occur at the LL energies of all the N > 0 levels. Two
teeth exist close in energy due to the spin-splitting of the levels. The magnetization
in the BI regime is similar to that of the TI; however, the two low-energy kinks are
associated with the spin-down and -up N = 0 levels at K ′ . One may wonder why
the magnetization changes when the chemical potential lies between LLs. This issue
was resolved by Bremme et al. who showed that this peculiar result is due to the

96
edge-state contribution [132].
As noted earlier, the slope of the magnetization is of particular interest as it
is related to the robust quantization of the Hall conductivity through the Streda
formula [133] ∂M (µ)/∂µ = (1/e)σH . The slope of the magnetization in the TI and
VSPM regimes is shown in the inset of Fig. 7.2. For the TI, no LLs exist at zero
energy and thus the Hall conductivity is zero until µ reaches the energy of the lowest
level. At this energy, it steps up by one unit of e2 /h until the next level at which
point it increments by another e2 /h. For the VSPM, there does exist a level at charge
neutrality and thus the slope of the magnetization is finite for zero chemical potential.
The BI behaves similar to the TI.
The quantization of ∂M/∂µ can be seen by taking the µ derivative of Eqn. (7.46).
For µ > 0, we find [34]
[ ]
∂ Ω̃ eB ∑ ( ) ( ) ∑ Nc ( ) Υ
= −Θ µ − E0ξσ Θ E0ξσ − Θ µ − EN,+
ξσ
− . (7.47)
∂µ h ξ,σ=± N =1
2

The slope of the magnetization is then [34]

[ ]
∂ ∂ Ω̃ ∂M e ∑ ( ) ( ) ∑ Nc ( ) Υ
− ≡ = Θ µ − E0ξσ Θ E0ξσ + Θ µ − EN,+
ξσ
+ .
∂B ∂µ ∂µ h ξ,σ=± N =1
2

(7.48)

For the TI and BI regimes in the absence of Zeeman splitting, the Hall conductivity
σH = (e2 /h)ν has filling factors [34] ν = 0, ±1, ±2, ±4, ±6, ...; while, the VSPM
has filling factors ν = ±1, ±2, ±4, ±6, .... If ∆z = 0, the familiar values of gapped
graphene (ν = 0, ±2, ±6, ±10, ...) are retained. For ∆so = 0 and ∆z = 0, the famous
graphene values ν = ±2, ±6, ±10, ... are observed (see Appendix B for a review of the
Hall effect in graphene).
For simplicity, let us continue our analysis of the Hall conductivity for gs = 0.
Figures 7.3(a), (d) and (g) show the slope of the magnetization in the TI (∆z = 4
meV), VSPM (∆z = 8 meV) and BI (∆z = 14 meV) regimes, respectively, for B = 1T.
The valley-split LL structure is schematically shown in the inset (again dashed blue

97
 TI
20 K K' 4
ε 2 2
2 2
(b) 0↓K' 0↑K'

dM(µ)/dµ (e/h)
1 1
1 1
2 0↓K
15 0
0
0 0
dM(µ)/dµ (e/h) 0 1↑K1↓K'
10 -1 -1 -2 Spin up
-1 -1
-2
-2
-2
-2
0↑K Spin down
-4
5 0↑K' (c) 0↑K 0↑K'
0↑K

dM(µ)/dµ (e/h)
0↓K' 2 0↓K 0↓K'
0↓K
0 0
1↑K1↓K'
∆z=4 meV -2 K
-5 (a) K'
-4
-20 -10 0 10 20 30 40 -20 -10 0 10 20
µ (meV) µ (meV)
VSPM
20 K K' 4
ε 2 2
2 2
(e) 0↓K' 0↑K 0↑K'

dM(µ)/dµ (e/h)
1 1
1 1
2 0↓K
15 0
0 0 0
0 1↑K1↓K'
dM(µ)/dµ (e/h)

10 -1 -1 -2 Spin up
-1 -1
-2 -2 Spin down
-2 -2

0↑K' -4
5 0↓K' 0↑K (f) 0↓K' 0↑K 0↑K'
dM(µ)/dµ (e/h)
↓
0K 2 0↓K

0 0
1↑K1↓K'
∆z=8 meV -2 K
-5 (d) K'
-4
-20 -10 0 10 20 30 40 -20 -10 0 10 20
µ (meV) µ (meV)
BI E1/√B=12.82 meV/√T
20 ε 2
K K'
2 ∆so=8 meV 4
2 2 (h) 0↓K' 0↑K'
dM(µ)/dµ (e/h)

B=1T
1
1
1
1
2 0↓K 0↑K
15 0

0
0 0
0 1↑K1↓K'
dM(µ)/dµ (e/h)

10 -1 -1 -2 Spin up
-1 -1
-2 -2 Spin down
-2 -2
-4
5 0↑K' 0↑K 0↑K'
0↓K' (i)
dM(µ)/dµ (e/h)

↑
↓ 0K 2 0↓K
0K
0 0
1↑K1↓K'
∆z=14 meV -2 K
-5 (g) 0↓K' K'
-4
-20 -10 0 10 20 30 40 -20 -10 0 10 20
µ (meV) µ (meV)

Figure 7.3: The slope of the magnetization which is related to the Hall conductivity.
The charge, spin and valley contributions are given for TI, VSPM and BI regimes.

lines represent spin up and solid red lines are spin down). µ = 0 is given by the dotted
black line. The µ locations of several key levels are emphasized in the figures by the

98
index Nξσ . The aforementioned step structure is clearly visible in addition to the
µ = 0 step that characterizes the VSPM. We now want to look at the spin and valley
contributions to σH (the charge Hall effect is made by adding the individual spin or
valley contributions). In the TI regime, the single spin and valley pieces of the Hall
effect are given in frames (b) and (c), respectively. Frames (e) and (f), and (h) and
(i) give the spin and valley contributions in the VSPM and BI regimes, respectively.
In analysing these results, it is important to note that a negative Hall conductivity
corresponds to a net edge current in the opposite direction to that of the positive Hall
effect. Thus, while at µ = 0 the charge Hall conductivity in the TI regime is zero,
there is a net spin Hall effect with a spin-up conductance of 2e2 /h in the negative
′
direction [34]. For µ between the two spin-up N = 0 LLs (E0K↑ < µ < E0K ↑ ), only the
spin-down electrons from K contribute to the Hall effect. For negative µ between the
′
two spin-down N = 0 levels (E0K↓ < µ < E0K ↓ ), only spin-up electrons from K ′ are
involved in the finite Hall conductivity. Therefore, a finite charge and spin Hall effect
is present [34]. All higher steps are valley-spin degenerate. For the VSPM, the first
µ > 0 plateau is polarized in spin down and momentum K; while, the lowest negative
µ step has the opposite polarization. The finite charge conductivity of the BI has
the same polarization as the TI. However, in contrast to the topologically nontrivial
phase, the zero Hall effect near µ = 0 is accompanied by a finite valley Hall effect
with a K-point edge conduction of 2e2 /h in the positive direction.
Let us now explore the effect of Zeeman splitting on the Hall response. For
simplicity, we will focus on the TI regime. A Zeeman interaction breaks the valley
degeneracy of the N > 0 levels and the Hall conductivity can take any integer value
of e2 /h. This is shown in Figure 7.4(a) for B = 1T and gs = 23. The steps associated
with the N = 0 levels are shifted [higher(lower) in µ for spin up(down)]. The steps
at higher µ are split into four integer plateaus which are close in energy. For gs = 0,
the four steps reduce to two steps of height 2e2 /h [see Fig. 7.3(a)]. The inset shows
a valley-split schematic of the LL energies for a TI. The effect of Zeeman splitting
in gapped graphene has previously been examined [134]. For graphene with gap
∆, there are two spin-degenerate N = 0 LLs at −ξ∆. Zeeman interactions split

99
 TI
20 K K' 6
ε 2
2
22
(b) 1↑K'

dM(µ)/dµ (e/h)
1 ↓ ↑
11 4 0K' 0K' ↑
1
0↓K 1K
15 0 2
0
0 1↓K
0 0 ↓
1K'
dM(µ)/dµ (e/h)

10 -1 Spin up
-1 -1 -1 -2
-2 -2 -2
-2 0↑K Spin down
-4
5 0↑K' (c) 1↓K'
0↑K

dM(µ)/dµ (e/h)
↓
↓0K' 4 0↑K 0↑K' ↑
1K
0K 0↓K 0↓K'
2
0 1↓K
E1/√B=12.82 meV/√T 0 ↑
∆so=8 meV gs=23 1K'
-5 (a) ∆z=4 meV B=1T -2 K
K'
-4
-20 -10 0 10 20 30 40 -20 -10 0 10 20 30
µ (meV) µ (meV)

Figure 7.4: The slope of the magnetization for the TI with Zeeman splitting. The
charge, spin and valley contributions are emphasized.

the degeneracy and four N = 0 plateaus are seen [134]. The four-fold valley-spin-
degenerate N > 0 levels become spin-split and a double step structure is observed
[134]. This is equivalent to the limit ∆so = 0 and ∆z /2 = ∆. Note that these results
are for T = 0. A finite temperature will smear the edges of the steps on an energy scale
associated with kB T , but will leave the quantization unaffected. Therefore, a very
low temperature is required to see the fine structure due to the Zeeman interaction
which is seen above the central quartet of plateaus.
The spin and valley contributions to the Hall effect are given in frames (b) and (c),
respectively. Identical valley-spin polarization is seen for the N = 0 levels as when
gs = 0. However, for finite Zeeman splitting, the quartet of steps associated with
each of the N = 1, 2, 3, ... levels also provides a spin and valley imbalance. While the
ν = ±2 Hall effect is valley-spin degenerate, the first two steps of the N = 1 levels
provide a valley-spin-polarized response [34]. This is lost when µ becomes greater
than the remaining N = 1 levels. The pattern continues for higher values of N .
To summarize, the total Hall conductivity at small µ is zero for the TI and BI;
however, a finite spin Hall effect is present in the TI regime while an analogous
valley Hall effect is found in the BI regime. The finite charge Hall conductivity

100
is also accompanied by valley-spin Hall effects for various values of µ. Including
Zeeman interactions results in a four step feature associated with each of the N > 0
levels. This also permits a valley-spin imbalance. By contrast, when |µ| is small
in the VSPM phase, the total Hall conductivity is finite and carries the sign of the
chemical potential. Spin and valley Hall effects are also seen. Realistically, the
Zeeman interaction will shift the LLs by much less than the inter-level spacing [33].
Therefore, in what follows, we ignore the Zeeman term as its effects are negligible.
As the filling factors and valley contributions to the Hall effect calculated from
the magnetization are in contrast to what is predicted by Tahir and Schwingenschlögl
for the DC Hall conductivity [135], these results were verified by direct calculation of
σxy (Ω → 0) given by Eqn. (6.24). To derive Eqn. (6.24), we used the Kubo formula
and, thus, Eqn. (6.24) represents an independent check. In the DC limit (Ω → 0) for
µ > 0, we find [33]

( ) ( )
e 2 ∑ ∑
∞ Θ µ − E ξσ
m,s − Θ µ − E ξσ
n,+
σxy (µ) = − E12 ξ ( )2
h ξ,σ=± n,m=0 En,+
ξσ
− Em,s
ξσ
s=±
[ ]
× (Am,s Bn,+ )2 δn,m−ξ − (Bm,s An,+ )2 δn,m+ξ , (7.49)

where [32, 33] (see Chapter 6)

 √

 s |En,s
ξσ
| + s∆ξσ


 √ , n ̸= 0
An,s =
 2|E ξσ
n,s | , (7.50)

 −

 1 ξ
, n=0
2

and  √

 |En,s
ξσ
| − s∆ξσ


 √ , n ̸= 0
Bn,s =
 2|E ξσ
n,s | , (7.51)



 1 + ξ
, n=0
2
which gives the same quantized plateaus as Eqn. (7.48) after multiplication by a factor

101
of e.

7.3 Magnetization and Magnetic Susceptibility

7.3.1 The Zero-Temperature, Clean limit

We now wish to return our focus to the magnetization. To do this, we choose to work
with the relativistic form of the grand potential [126]:
∫ ∞ ( )
ω−µ
Ω(T, µ) = −T N (ω)ln 2cosh dω. (7.52)
−∞ 2T

For our purposes, this form is equivalent to Eqn. (7.38) [36]. To see this, note that
Equation (7.52) can be rewritten as
∫ ∫ ∫
∞ ( ) 1 ∞
1 ∞
Ω(T, µ) = −T (µ−ω)/T
N (ω)ln 1 + e dω + µN (ω)dω − ωN (ω)dω.
−∞ 2 −∞ 2 −∞

(7.53)

The first term on the right hand side has the form of the usual non-relativistic grand
potential [see Eqn. (7.38)] which is typically the appropriate starting point. Recall
that the integral of the density of states over all energies gives the total number
of states which must be independent of B. Therefore, the second term is half the
chemical potential times the total number of states in our bands. Thus, it will not
contribute to the magnetization which is given by the first derivative of Ω(T, µ) with
respect to B at fixed chemical potential, i.e., M (T, µ) = −[∂Ω(T, µ)/∂B] µ . Since we
have particle-hole symmetry (when we sum over spin and valley), the final term is
zero. As a consequence of this, we will only report on the total magnetization and
not examine the single spin and valley contributions.
Equation (7.52) has been worked out analytically for gapped graphene at T = 0
[126] and may be extended to our model. Mapping our density of states to Ref. [126],
we write the grand potential as the sum of a regular, vacuum and oscillating piece.

102
For µ ≥ 0, the three contributions are [34]
[
eB ∑ µ(µ2 − 3∆2ξσ )
Ωreg (µ) = − Θ (µ − |∆ξσ |) −|∆ξσ | (7.54)
2h ξ,σ=± 3E12
( )]
1 ∆2ξσ
−2 3/2
E1 ζ − , 1 + ,
2 2E12

[ ( )] ( )
eB ∑ Λ∆2ξσ 1 ∆2ξσ 1
Ωvac (µ = 0) = − √ 2 + |∆ξσ | + 2 E1 ζ − , 1 +
3/2
2
+O ,
2h ξ,σ=± πE1 2 2E1 Λ

(7.55)

and
( )
eB E12 ∑ ∑ 1
∞
πk(µ2 − ∆2ξσ )
Ωosc (µ) = cos , (7.56)
2h µ ξ,σ=± k=1 (πk)2 E12

respectively (see Eqns. (6.3), (A4) and (8.7) in Ref. [126]), where ζ(−1/2, x) is the
Hurwitz zeta function and Λ is a large ultraviolet cutoff. Applying the B derivative,
the magnetization becomes the sum of [34]

e ∑
Mreg (µ) = − Θ (µ − |∆ξσ |) C(∆ξσ , B), (7.57)
h ξ,σ=±

e ∑
Mvac (µ = 0) = C(∆ξσ , B), (7.58)
h ξ,σ=±

and
( )
e E12 ∑ ∑ 1
∞
πk(µ2 − ∆2ξσ )
Mosc (µ) = − cos
h µ ξ,σ=± k=1 (πk)2 E12
( )
e ∑ µ2 − ∆2ξσ ∑ 1
∞
πk(µ2 − ∆2ξσ )
− sin 2
, (7.59)
h ξ,σ=± 2µ k=1
πk E 1

103
where
( ) ( )
|∆ξσ | 3E1 1 ∆2ξσ ∆2ξσ 1 ∆2ξσ
C(∆ξσ , B) = + √ ζ − ,1 + − √ ζ ,1 + , (7.60)
2 2 2 2E12 2 2E1 2 2E12

and we have used the relation

d
ζ(s, x) = −sζ(s + 1, x). (7.61)
dx

Comparing Eqns. (7.57) and (7.58), it is clear that for µ > |∆ξσ |, the regular and
vacuum contributions to the magnetization cancel with each other and only the os-
cillating piece remains. For small B fields, we use the expansion [126]

a1−s 1 s
ζ(s, 1 + a) ≈ − s+ (7.62)
s − 1 2a 12a1+s

for large a to obtain [34]

∑ e
Mvac (µ = 0) ≈ − M (∆ξσ , B) , (7.63)
ξ,σ=±
h

where


 E12

 , ∆ξσ ̸= 0

 6|∆ξσ |
M (∆ξσ , B) = , (7.64)

 ( )

 3E 3

 √ ζ
1
, ∆ξσ = 0
4 2π 2

and ζ(x) is the Riemann zeta function. Again, the regular part is the same as the
vacuum contribution up to a factor of −Θ(µ − |∆ξσ |). Plots of the vacuum magne-
tization as a function of B and ∆z are given in Figs. 7.5(a) and (b), respectively.
The solid curves correspond to the result of Eqn. (7.58). These are compared with
Eqn. (7.63). We begin by discussing Fig. 7.5(a). For ∆so = ∆z = 0, we recover
the graphene limit. In this case, the full solution for Mvac (solid black curve), given
by Eqn. (7.58), agrees perfectly with Eqn. (7.63) (dotted yellow). This is expected

104
 0 E1/√B=12.82 meV/√T 0
∆so=8 meV
∆z=0 meV TI
Eqn. (7.63)
-4 ∆z=8 meV VSPM

Mvac(∆z) (e/h meV)

Mvac(Β) (e/h meV)
Eqn. (7.63) -10

-8
B=0.01T
Eqn. (7.63)
-20
B=0.1T
-12 ∆so=0 meV Eqn. (7.63)
Eqn. (7.63) B=1T
Eqn. (7.63)
(a) (b)
-16 -30
0 0.1 0.2 0.3 0.4 0.5 0 10 20 30 40
B (T) ∆z (meV)

Figure 7.5: Vacuum contribution to the magnetization as a function of (a) B and (b)
∆z .

as no approximations were made when deriving Eqn. (7.63) from Eqn. (7.58) for
√
∆ξσ = 0. As a function of B, Mvac ∝ − B via the definition of E1 . This depen-
√
dence is evident in the figure. The − B dependence also dominates in the VSPM
case. This is due to the single graphene-like band at each valley. For low B, the
agreement between Eqns. (7.58) and (7.63) is good over the range of B shown. This
results from the fact that the energy scale associated with these small values of B
is largely sampling the linear band (where no approximations were needed) and not
the massive one. Thus, as B → 0, the dashed red curve [Eqn. (7.63)] is in excellent
agreement with the full result of Eqn. (7.58). The curve is reduced by approximately
a factor of two when compared to the graphene case. This reflects the fact that there
is only a single spin-polarized linear band at each K-point. With increasing field,
the massive bands begin to contribute and Eqn. (7.63) increasingly deviates from
the full result. The energy scale for such deviations is controlled by ∆2ξσ ≈ 2E12 .
Lastly, consider the Kane-Mele QSHI given by ∆z = 0 but finite ∆so [solid green
curve in Fig. 7.5(a)]. In this case, the result of Eqn. (7.63) depends on the approx-
imation a ≡ ∆2ξσ /(2E12 ) = ∆2so /(2E12 ) ≫ 1 and predicts Mvac ∝ −B. This approx-
imation is only appropriate for small B ≪ ∆2so /(2ℏev 2 ). Hence, the relatively poor
agreement between the solid green (exact) and dash-double-dotted purple [approxi-

105
mate Eqn. (7.63)] curves. Therefore, we expect the simple formulae to be robust as
B → 0. This makes it highly relevant when calculating the magnetic susceptibility
(∂M/∂B|B→0 ) which exists in the zero-B limit. For large B, the full expression should
be applied.
Let us now turn our attention to Fig. 7.5(b). Here the vacuum magnetization
is shown as a function of ∆z for three values of magnetic field. Recall that the TI
regime is for ∆z < ∆so = 8 meV and the BI case corresponds to ∆z > ∆so . Hence, the
divergent features seen at ∆z = ∆so are occurring as the VSPM phase is approached.
When obtaining the vacuum contribution from the full result (solid curves), no diver-
gence occurs but rather a cusp appears in the magnetization that indicates the critical
point between the TI and BI; this marks the VSPM. As understood from the preced-
ing discussion surrounding Fig. 7.5(a), for very low B and small ∆z , the approximate
Eqn. (7.63) works reasonably well even as ∆z is varied away from zero. As ∆z → ∆so ,
however, the 1/|∆ξσ | factor in Eqn. (7.63) diverges. This is due to a breakdown in
the approximation ∆2ξσ ≫ 2E12 that was used in Eqn. (7.62). While Eqn. (7.63) works
well only for very small magnetic fields and ∆z away from the VSPM critical point, it
also is seen to work well for very large ∆z where ∆2ξσ ≫ 2E12 is again satisfied. From
this frame, it appears that the agreement between Eqns. (7.58) and (7.63) is terrible
as one approaches the VSPM. This is in contrast to what is seen in Fig. 7.5(a). To
reconcile this discrepancy, note that the data point ∆z = ∆so in Fig. 7.5(b) evaluated
by Eqn. (7.63) is not shown, but would appear as a discontinuous point near the cusp
at ∆z = ∆so in good agreement with the full solution [34]. In summary, one must be
careful when using the simple formula of Eqn. (7.63) beyond its intended region of
validity determined by (∆so − ∆z )2 ≫ 2ℏev 2 B. We also note that the VSPM state is
identified by a cusp (not a divergence) in the vacuum magnetization as the electric
field is varied.
Let us now calculate the magnetic susceptibility (−∂ 2 Ω/∂B 2 |B→0 = χ). Differ-
entiating Eqn. (7.63) with respect to B and taking the limit B → 0, we obtain the

106
vacuum susceptibility [34]

e ∑
χvac = − X (∆ξσ ), (7.65)
h ξ,σ=±

where

 ℏev 2

 , ∆ξσ ̸= 0

 6|∆ξσ |

X (∆ξσ ) = . (7.66)

 √ ( )

 3 ℏev 2 3

 lim √ ζ , ∆ξσ = 0
B→0 8 2π B 2

As is well understood in the context of graphene [97, 123, 125, 128], when ∆ξσ = 0,
√
Eqn. (7.65) diverges as 1/ B for B → 0. For finite ∆ξσ → 0, χvac diverges [129]
as 1/|∆ξσ |. A plot of the vacuum susceptibility as a function of ∆z is given in
Fig. 7.6(a). The result of differentiating Eqn. (7.58) [solid black curve] is compared
with Eqn. (7.65) [dashed green curve]. Here, agreement is found between the full
0

0
-200
χvac(∆z) (e/h meV/T)

χ (µ) (e/h meV/T)

∆so=8meV -20
-400
Eqn. (7.65)

∆so=8 meV
-600
-40 ∆z=0 meV TI
∆z=4 meV TI
∆z=8 meV VSPM
-800
∆z=10 meV BI
(a)
TI BI -60
(b)
-1000
0 2 4 6 8 10 0 2 4 6 8 10
∆z (meV) µ (meV)

Figure 7.6: (a) Vacuum contribution to the magnetic susceptibility as a function of

∆z . (b) Vacuum plus regular parts of the susceptibility as a function of µ for various
∆z .

result and that obtained from the approximate equation for Mvac (as expected since
B → 0). Unlike the magnetization, there is a singularity at ∆z = ∆so . This arises

107
from the massless Dirac fermions of the lowest band. Since the susceptibility is
inversely proportional to the effective mass [97, 123–129], a divergent diamagnetic
response occurs when the VSPM phase is approached [34, 131]. As the two gaps
increase (and the effective mass of the electrons becomes larger), the susceptibility
decreases in magnitude. We note that the slope of the susceptibility has a different
sign in the two insulating regimes; this may be used to distinguish the two insulating
phases [34].
Combining the vacuum and regular parts of the susceptibility, we obtain [34]

e ∑
χ=− X (∆ξσ ) [1 − Θ(µ − |∆ξσ |)] . (7.67)
h ξ,σ=±

This is plotted in Fig. 7.6(b) as a function of µ. Like gapped graphene [130, 136], the
susceptibility is nonzero when µ is in the gap. This is due to the constant contribution
provided by the filled valence band [136]. For ∆z = 0 (solid black curve), there is only
one gap and the susceptibility is given by its vacuum value until µ = ∆so /2. As the
chemical potential enters the conduction band, a step occurs and the susceptibility is
zero. This agrees with the result of doped graphene [129, 136]. For finite ∆z ̸= ∆so
(dashed blue and dash-double-dotted green curves), the system is characterized by
two gaps. For µ less than the minimum gap, the magnitude of the susceptibility is
maximized. Two steps are seen at the values of the gaps ∆min and ∆max . Again,
for µ greater than the maximum gap, the susceptibility is zero. In the VSPM phase
(dash-dotted red curve), there is only one gap and, thus, a single step.
Analysing this behaviour of the susceptibility should provide a probe of the im-
portant energy scales of the system [34]. By varying ∆z , the VSPM transition can
be identified [131] and, thus, ∆so can be determined. Additionally, by varying µ, the
two gaps ∆min and ∆max can be extracted [34].

7.3.2 Impurity and Finite Temperature Effects

So far, we have worked entirely in the clean limit (i.e. no temperature or impurities).
To account for the LL broadening that results from impurity scattering (Γ) and finite

108
temperature, we convolve the magnetization with the scattering and temperature
functions

Γ
PΓ (ω − µ) = , (7.68)
π[(ω − µ)2 + Γ2 ]

and

∂nF 1
PT (ω − µ) = − = ( ), (7.69)
∂ω 4T cosh2 ω−µ
2T

respectively, where nF is the Fermi distribution function. For simplicity, we have

assumed (as done in Ref. [126] and other works) that Γ is the same for all LLs. In
principle, the impurity scattering rate will depend on the type of impurities [137].
There exist more complicated descriptions of impurity scattering; in particular, see
Ref. [136] which deals explicitly with the diamagnetism of disordered graphene. Ac-
counting for Γ and T , the magnetization is defined by
∫ ∞ ∫ ∞
M (µ, Γ, T ) = PΓ (ω − µ)PT (ω ′ − ω)M (ω ′ )dω ′ dω, (7.70)
−∞ −∞

where M (ω ′ ) is obtained by setting µ = ω ′ in the Γ = 0 and T = 0 result [Eqns. (7.57)-

(7.59)].
To begin, consider the effect of impurity scattering on the vacuum-plus-regular
piece of the magnetization. Taking T = 0 and combining Eqns. (7.57) and (7.58),
Eqn. (7.70) gives [34]

∫ |∆ξσ |
e ∑ Γ
M (µ, Γ) = C(∆ξσ , B) dω
−|∆ξσ | π[(ω − µ) + Γ ]
h ξ,σ=± 2 2

e ∑
= C(∆ξσ , B)DΓ , (7.71)
h ξ,σ=±

where
[ ( )]
1 2|∆ξσ |Γ
DΓ = arctan . (7.72)
π µ2 − ∆2ξσ + Γ2

109
For our low B expression [see Eqn. (7.63)], C(∆ξσ , B) should be replaced with the
approximate M(∆ξσ , B). For the vacuum-plus-regular part of the susceptibility [see
Eqn. (7.67)],

e ∑
χ(Γ) = − X (∆ξσ )DΓ . (7.73)
h ξ,σ=±

This is in agreement with Eqn. (6) of Ref. [130] which deals with the gapped graphene-
limit |∆ξσ | → ∆. For both the magnetization and susceptibility, the damping factor
DΓ has a dependence on chemical potential and becomes smaller as µ increases. This
means that as µ increases, the magnetic response decreases.
The same procedure can be applied for finite temperature. Equation (7.70), with
the combined result of Eqns. (7.57) and (7.58), gives [34]

∫ |∆ξσ | ( )
e ∑ ∂nF
M (µ, T ) = C(∆ξσ , B) − dω
h ξ,σ=± −|∆ξσ | ∂ω
e ∑
= C(∆ξσ , B)DT , (7.74)
h ξ,σ=±

where

sinh(β|∆ξσ |)
DT = , (7.75)
cosh(β|∆ξσ |) + cosh(βµ)

β = T −1 and Γ has been set to zero. While this is a general result for the system,
several limiting cases are known and discussed below. For ∆ξσ → 0, sinh(β|∆ξσ |) ≈
β|∆ξσ | and

β|∆ξσ |
DT ≈ . (7.76)
1 + cosh(βµ)

The magnetic susceptibility is

e ∑
χ(T ) = − X (∆ξσ )DT . (7.77)
h ξ,σ=±

110
Employing Eqn. (7.76), the susceptibility in the zero-gap limit (∆ξσ → 0) is

e ∑ ℏev 2 β
χ(T ) = − 2 , (7.78)
h ξ,σ=± 12 cosh (βµ/2)

which is in agreement with Refs. [123, 125, 128]. Returning to Eqn. (7.75) and taking
zero chemical potential,

DT = tanh (β|∆ξσ |) , (7.79)

and

e ∑
χ(T ) = − X (∆ξσ )tanh (β|∆ξσ |) . (7.80)
h ξ,σ=±

This agrees with Eqn. (31) in Ref. [131].

The charge-neutral (µ = 0) susceptibility at finite temperature and impurity scat-
tering is shown in Fig. 7.7(a) as a function of ∆z . At T = Γ = 0 (solid black curve),
0 0

-100
-50
χ(∆z) (e/h meV/T)
χ(∆z) (e/h meV/T)

∆so=8 meV
-200 µ=0 meV
T,Γ=0 meV ∆so=8 meV
T=0.1 meV
-100
T,Γ=0 meV
T=0.5 meV
-300 Γ=0.2 meV
µ=0 meV
Γ=0.5 meV µ=0.5 meV
µ=1 meV
-150
-400

(a) TI BI (b) TI BI
-500 -200
0 4 8 12 16 20 0 4 8 12 16 20
∆z (meV) ∆z (meV)

Figure 7.7: (a) The effect of finite temperature and impurity scattering on the sus-
ceptibility as a function of ∆z . (b) The effect of finite chemical potential on the
susceptibility.

a singularity exists at the critical value ∆so = ∆z . This divergence is removed by

the inclusion of temperature or impurity effects [128]. The curves associated with T

111
(dash-dotted blue) or Γ (dotted magenta) equal to 0.5 meV are very similar. While
the singularity has been smeared, a clear minimum is present at the location of the
phase transition. Thus, the VSPM boundary should still be experimentally observ-
able provided that the LL broadening is kept low [34]. Returning to Eqn. (7.67),
we note that finite µ will also remove the singularity. This is shown in Fig. 7.7(b).
There is an abrupt vertical drop in the absolute value of χ as the magnitude of ∆ξσ is
reduced below the value of µ. For ∆max > µ > ∆min , the susceptibility remains finite
and is given by χ = −[e/h][ℏev 2 /(3∆max )]. This value originates entirely from the
spin-valley branches of the Dirac dispersion which have the maximum gap ∆max [34].
This abrupt change in χ can be used to determine the size of µ (i.e. the doping away
from charge neutrality). The determining structures involved in such a measurement
are most prominent for small values of chemical potential.

7.4 Magnetic Oscillations

So far, we have ignored the oscillating part of the magnetization which will super-
impose itself on the aforementioned results. In fact, for µ larger than both gaps,
only the oscillating piece is nonzero. Recall our earlier discussion of the semiclassi-
cal behaviour of an electron in a magnetic field where we found that the density of
states was maximized at regular intervals of 1/B (see Eqn. (7.27) where here we take
c = 1). This implies that the magnetization (which depends on the density of states)
should oscillate accordingly. Indeed, semiclassically, it is known that, Mosc oscillates
according to [138–140]
( [ ])
ℏA(µ)
k
Mosc ∝ sin 2πk −γ , (7.81)
2πeB

where A(µ) is the area of a cyclotron orbit at the Fermi energy (A(µ) = πkF2 ).
γ = 1/2 − |Wc |/2 is a phase offset related to the Maslov index of the harmonic
oscillator (1/2) and the topological winding number (Wc ) associated with the Berry
phase [60,141]; therefore, γ = 1/2 for a 2DEG and 0 for relativistic Dirac-like systems

112
 ε

µ3, two orbits

µ2, one orbit

µ1, no orbit
k

Figure 7.8: Schematic illustration of the possible cyclotron orbits in silicene.

with topological Berry phase π. k indexes the oscillating harmonic. Typically, we are
only interested in the fundamental case of k = 1 which can be extracted from the
data. Silicene, being comprised of low-energy relativistic fermions, has a topological
Berry phase of π [see Eqn. (3.47)] yielding γ = 0. A remarkable difference between
this system and the familiar case of gapped graphene is that the spin splitting of the
bands allows for two, one, or no cyclotron orbit(s) to exist at a given energy (see
Fig. 7.8 where the positive energy bands at the K point are shown). In silicene, the
Fermi momentum for the band labelled by ξ and σ is
√
µ2 − ∆2ξσ
kFξσ = , (7.82)
ℏ2 v 2

where only real solutions are considered. Therefore,

[ ]
π (∆ so − ξσ∆z )2
Aξσ (µ) = 2 2 µ2 − . (7.83)
ℏv 4

Indeed, taking only the leading term of Eqn. (7.59), we note [34]

( [ ])
e ∑ µ2 − ∆2ξσ ∑ 1
∞
ℏAξσ (µ)
Mosc (µ) = − sin 2πk −γ , (7.84)
h ξ,σ=± 2µ k=1
πk 2πeB

113
 TI BI (a) TI BI (b)

1/4 Amax(µ)
K↑, K'↓ µ>1/2
A(µ) (π/[ħv∆ so] 2) µ=1/2 µ2
2
µ<1/2 µ -1/4 Amax(µ)
K↑, K'↓

A(µ) (π/[ħv∆ so] 2)

Amin(µ)
K↓, K'↑

µ2

0 ∆z 0 ∆z
0 1-2µ 1 1+2µ 2 0 1 2µ-1 1+2µ

Figure 7.9: (a) The cyclotron orbit area as a function of µ for µ between the two
gaps. (b) The two cyclotron orbit areas for µ above both gaps.

where γ = 0 and Aξσ (µ) is given by Eqn. (7.83). Equation (7.83) is plotted in
Fig. 7.9(a) for µ̄ ≡ µ/∆so ≤ 1/2 as a function of ∆
¯ z ≡ ∆z /∆so . The results are

shown for the lowest gapped bands (spin up at K and spin down at K ′ ) since, for
µ̄ ≤ 1/2, only one band is occupied. For µ̄ < 1/2 (dash-dotted curve), the lowest
band is occupied (and there is an associated cyclotron orbit) if 1 − 2µ̄ < ∆
¯ z < 1 + 2µ̄.

As µ is increased, the range of ∆z grows for which A(µ) is nonzero. The maximum
orbit area is fixed at µ̄2 and occurs at ∆z = ∆so (the VSPM phase). µ̄ = 1/2 (solid
curve) corresponds to the maximum range of ∆z for which A(µ) > 0 when only one
¯ z = 0 and 2. In all
band is occupied. The lowest band remains occupied between ∆
cases, there is perfect symmetry between the TI and BI regimes. Schematically, the
regions of ∆z which give a finite area are those which ensure µ sits in the µ2 position
of Fig. 7.8. µ = µ1 corresponds to no cyclotron orbits. Next, consider µ̄ > 1/2.
This corresponds to µ = µ3 in Fig. 7.8. Now both bands are occupied. A(µ) as a
¯ z is shown in Fig. 7.9(b). The maximum area (solid curve) comes from
function of ∆
the spin-up electrons at K and the spin-down electrons at K ′ . The other spin-valley
combinations make up the minimum area (dashed curve). For Amax (µ), the maximum
¯ z remains at 1 + 2µ̄. 1 − 2µ̄ now occurs for negative ∆
value of ∆ ¯ z ; thus, for ∆z = 0,

Amax (µ) is finite at a value of µ̄2 − 1/4. The maximum still occurs in the VSPM phase
¯ z = 2µ̄ − 1 and is equal to Amax (µ)
and retains its value of µ̄2 . Amin (µ) persists until ∆

114
at ∆z = 0. There is perfect symmetry between the maximum and minimum areas
¯ z are considered. In that region, the spin and valley labels of
if negative values of ∆
¯ z < 2µ̄ − 1, there are two orbits
the two areas are interchanged. It is clear that for ∆
which will contribute to the magnetic oscillations. For ∆z = 0, the areas are equal
and will contribute an overall degeneracy to Mosc . For finite ∆z , the difference in the
two orbits increases with the electric field strength.
To see the interaction between the two areas, we shall consider the total oscillating
magnetization [Eqn. (7.84)]. We restrict our attention to the lowest harmonic (k = 1)
as it contains all the necessary features and clearly emphasizes the interesting physics.
Therefore [34],
{ ( [ ])
e µ2 − ∆2min ℏAmin (µ)
Mosc (µ) = − sin 2π
h πµ 2πeB
( [ ])}
µ − ∆max
2 2
ℏAmax (µ)
+ sin 2π , (7.85)
πµ 2πeB

where Amin ≡ A+− and Amax ≡ A++ . Equation (7.85) illustrates that Mosc is con-
structed from two sine waves with 1/B frequencies of
[ ]
ℏAmin (µ) π (∆ so + ∆ z )2
ωmin = = µ2 − , (7.86)
e ℏev 2 4

and
[ ]
ℏAmax (µ) π (∆ so − ∆ z )2
ωmax = = µ2 − . (7.87)
e ℏev 2 4

Therefore, for µ between the two gaps (µ2 in Fig. 7.8), there is only one characteristic
dHvA frequency. For µ above both gaps (µ3 in Fig. 7.8), there are two frequencies
and interference will be observed [142–144]. When the difference between Amin (µ)
and Amax (µ) is small, a strong beating is present. Looking at Fig. 7.9(b), this will
occur when ∆z is very small (but nonzero) or µ is very large (µ ≫ ∆max ).
The beating of the quantum oscillations is shown in Fig. 7.10 for varying µ and
∆z . Figure 7.10(a) shows Mosc as a function of 1/B for µ = 15 meV and ∆z = 0, 2

115
 E1/√B=12.82 meV/√T ∆so=8 meV µ=15 meV E1/√B=12.82 meV/√T ∆so=8 meV
30 ∆z=0 meV 24 µ=4 meV
∆z=2 meV (c) TI µ=7 meV
∆z=4 meV µ=15 meV
∆z=4 meV
20
16

Mosc(B) (e/h meV)

Mosc(B) (e/h meV)

10

8
0

-10 0

-20
(a) -8
-30 24
µ=7 meV (d) VSPM µ=4 meV
∆z=0 meV ∆z=8 meV µ=9 meV
10 ∆z=2 meV µ=15 meV
∆z=4 meV 16

Mosc(B) (e/h meV)

Mosc(B) (e/h meV)

∆z=6 meV
5

8
0

-5
0

-10
(b) -8
-15
40 50 60 70 80 90 40 50 60 70 80 90
1/B(T) 1/B(T)

Figure 7.10: Magnetic oscillations as a function of 1/B for varying µ and ∆z .

and 4 meV (green, purple and black curves, respectively). Here µ is above both gaps.
For ∆z = 0 meV, only one gap is present and strong dHvA oscillations are observed.
For ∆z = 2 meV, the bands have become spin split and a beating phenomenon
emerges [142, 144] in addition to the usual oscillations. For ∆z = 4 meV, the beating
persists; however, the beat frequency has increased. This phenomenon is a signature
of a small difference between the two orbit areas which we control here with the
magnitude of the external electric field. For smaller values of µ, increasing ∆z can
lead from a single fundamental frequency to a beating regime and back to a single
frequency as ∆z becomes large enough that µ is only above the smaller gap [see
Fig. 7.10(b)].
The magnetic oscillations for fixed ∆z and varying µ are shown in frames (c) and
(d) for the TI and VSPM, respectively. For µ = 4 meV (dash-dotted black curve) only

116
a single frequency is present. For the dashed curves, both bands contribute and two
characteristic frequencies become visible. They are ωmax = 0.860T and ωmin = 0.249T
for ∆z = 4 meV and µ = 7 meV. As µ is increased further (solid curves), the two
cyclotron orbits become closer (ωmax = 4.22T and ωmin = 3.61T for ∆z = 4 meV
and µ = 7 meV) and beating is clearly observed. The beating is more pronounced
for the TI because the bands at a given energy are closer in momentum. Analogous
results are found for the BI. Due to the linear-Dirac band, the VSPM displays dHvA
oscillations for all µ ̸= 0.

7.4.1 Impurity and Finite Temperature Effects

In the presence of impurities, quantum oscillations are damped by a Dingle factor

which depends on the scattering probability. Again, this effect is included by con-
volving Mosc (µ) with PΓ (µ − Γ) as seen in Sec. 7.3.2. For gapped graphene, this has
been worked out analytically [126] and can be extended to the doubly gapped case of
silicene. In the presence of impurities and at finite temperature (see Eqn. (8.17) of
Ref. [126] which we repeat here for completeness),

( )
eBE12 ∑ 1
∞
πk(µ2 − ∆2ξσ − Γ2 )
Ωξσ
osc (µ, Γ, T ) = cos RD RT , (7.88)
2hµ k=1 (πk)2 E12

with the Dingle factor

( )
µΓ
RD = exp −2πk 2 , (7.89)
E1

and temperature factor

kλ
RT = , (7.90)
sinh(kλ)

where

2π 2 T µ
λ= . (7.91)
E12

117
For both impurity and temperature effects, the damping of the oscillations is depen-
dent on µ making the dHvA effect difficult to observe for large values of chemical
potential [126]. It is clear that by comparing the amplitudes of the oscillations at
fixed B and various temperatures, one can extract vF and thus the effective cyclotron
frequency (ℏωc = E12 ) and mass (mc = eB/ωc ). In addition, if one then compares the
oscillation amplitude for varying B, the scattering rate Γ is easily extracted.

118
 CHAPTER VIII

SCREENING IN THE RANDOM

PHASE APPROXIMATION

So far, all the phenomena discussed in this thesis have been in the context of the
independent electron approximation. That is, we assume that the fermions in the
system do not interact with each other. While this has been shown to be a good
approximation for the material in the previous chapters, we now wish to examine the
effect electron-electron interactions have on silicene. Our primary focus will be on
screening and plasmon excitations.

8.1 Electron-Electron Interactions

When two electrons are present in a vacuum, they repel each other via the Coulomb
potential V (r) ∝ e2 /r. If another electron is added, each particle now experiences two
of these potentials. If 1023 electrons are present, calculating the interaction between
electrons becomes a highly non-trivial task. If we include the positive background
due to the lattice of ions, each electron also experiences an attractive force. To
account for this complicated set of interactions, Bohm and Pines [145–147] proposed
that a re-normalized Coulomb interaction be considered. Thus, an electron is seen
to interact via a single screened potential; this greatly simplifies the problem. While
this approximation is not always accurate, it has been shown to be quite effective
in describing condensed matter systems [148]. In terms of Feynman diagrams, the
bare Coulomb interaction is assumed to be renormalized by the creation of an infinite
series of particle-hole pairs. This is referred to as the random phase approximation
(RPA) [145–148].

119
 To include electron-electron interactions via a renormalized Coulomb interaction,
we will employ the well-established Feynman diagrams. We use the following repre-
sentations:
G0(ω,q) Bare Green's Function

v(q) Bare Coulomb Interaction

u(q) Renormalized Coulomb Interaction

Π(ω,q) Irreducible Self-Energy

Figure 8.1: The applicable Feynman diagrams.

Diagrammatically, the renormalized Coulomb potential is given by Fig. 8.2.

=
Figure 8.2: The renormalized Coulomb interaction.

We must now make an assumption on the form of the irreducible self-energy.

We will use the RPA proposed by Bohm and Pines where we take the particle-hole
“bubble” as the self-energy (see Fig. 8.3). Mathematically, the Dyson series for the

= Π0(ω,q)

Figure 8.3: Bubble diagram: irreducible self-energy in the random phase approxima-
tion.

120
renormalized Coulomb potential becomes

u(q) = v(q) − 2v(q)Π0 (ω, q)u(q), (8.1)

where the factor of two is for spin, and

∫
i ∑ dε
Π0 (ω, q) = G0 (k, ε)G0 (k − q, ε − ω). (8.2)
V k 2π

Equation (8.1) can be rearranged to give

v(q)
u(q) = . (8.3)
1 + 2v(q)Π0 (ω, q)

We note that the screened Coulomb potential is written in terms of the bare potential
divided by a renormalizing factor. We denote this factor as the generalized dielectric
function

ε(ω, q) = 1 + 2v(q)Π0 (ω, q). (8.4)

Thus, electron-electron interactions are fully determined once the polarization func-
tion is known.

8.1.1 Polarization Function

We now evaluate Π0 (ω, q) for the electron gas as we will use it for comparison later.
If we substitute for the bare Green’s functions, we obtain
∫
i ∑ dε
Π0 (ω, q) = G0 (k, ε)G0 (k − q, ε − ω)
V k 2π
∫ [ ]
i ∑ dε Θ(kF − k) Θ(k − kF )
= +
V k 2π ε − εk − i0+ ε − εk + i0+
[ ]
Θ(kF − |k − q|) Θ(|k − q| − kF )
× + . (8.5)
ε − ω − εk−q − i0+ ε − ω − εk−q + i0+

121
It is clear that each term in the above expression goes as 1/ε2 . Therefore, we can
solve the ε integral with contour techniques and will have convergence regardless of
whether we define the contour in the upper or lower plane. Thus, all terms with the
the pole on the same side of the real axis must be zero. We are left with
∫ {
i ∑ dε Θ(kF − k)Θ(|k − q| − kF )
Π0 (ω, q) =
V k 2π (ε − εk − i0+ )(ε − ω − εk−q + i0+ )
}
Θ(k − kF )Θ(kF − |k − q|)
+ . (8.6)
(ε − εk + i0+ )(ε − ω − εk−q − i0+ )

We choose to close the contour for the first term above the real axis and that of the
second below and get
{ }
1 ∑ Θ(kF − k)Θ(|k − q| − kF ) Θ(k − kF )Θ(kF − |k − q|)
Π0 (ω, q) = − .
V k εk−q − εk + ω − i0+ εk−q − εk + ω + i0+
(8.7)

Letting k → k − q,
{ }
1 ∑ Θ(kF − k)Θ(|k + q| − kF ) Θ(k − kF )Θ(kF − |k + q|)
Π0 (ω, q) = − .
V k εk+q − εk − ω − i0+ εk+q − εk − ω + i0+
(8.8)

From now on, we will restrict ourselves to two dimensions. Consider the electron
gas εk = ℏ2 k 2 /(2m) for ω = 0. Working in the continuum limit

∑ ∫
V
→ 2 d2 k, (8.9)
k
4π

the polarization function is

∫ { }
1 Θ(kF − k)Θ(|k + q| − kF ) Θ(k − kF )Θ(kF − |k + q|)
Π0 (0, q) = 2 2
dk − .
4π εk+q − εk − i0+ εk+q − εk + i0+
(8.10)

122
Letting k + q → −k and using ε−k = εk , the second term of the integrand becomes

Θ(kF − k)Θ(|k + q| − kF )
. (8.11)
εk − εk+q + i0+

We can see that the imaginary parts of both terms are equal and the real parts are
negatives of each other. Thus,
∫
1 Θ(kF − k)Θ(|k + q| − kF )
Π0 (0, q) = 2 d2 k . (8.12)
2π εk+q − εk

Adding
∫
1 Θ(kF − k)Θ(kF − |k + q|)
F (q) = 2 d2 k = −F (q) = 0, (8.13)
2π εk+q − εk

we obtain
∫
1 Θ(kF − k)
Π0 (0, q) = 2 d2 k . (8.14)
2π εk+q − εk

This can be solved to give


 m

 , q ≤ 2kF
  2πℏ
√
2 
Π0 (0, q) = m  4kF2  . (8.15)

 1− 1− , q > 2k

 2πℏ2 q2
F

This is the Lindhard function.

To see the physical significance of Π0 (0, q), return to Eqn. (8.14). In the discrete
limit,

2 ∑ 1
Π0 (0, q) = . (8.16)
Ω εk+q − εk
|k|<kF
|k+q|>kF

This describes all the possible virtual transitions from below to above the Fermi
surface. Clearly, this will be maximized when the denominator is minimized. If we
take the usual Fermi circle in 2D, Eqn. (8.16) corresponds to bubbling off particle-hole

123
pairs (see Fig. 8.4). If |q| < 2kF , particle-hole pairs of nearly zero energy difference

q<2kF q>2kF
Fermi Surface Fermi Surface Fermi Surface

k+q q q
k+q q
k+q
k k k

2kF 2kF 2kF

Figure 8.4: Particle-hole excitations from the Fermi sea.

can be created (i.e. transitions from just below to just above the Fermi energy). These
have strong implications on the Lindhard function as they cause the denominator to
approach zero. If |q| > 2kF , these type of pairings are no longer possible. Thus,
Π0 (0, q) is greatly diminished.
This function is the determining factor in the screening of a charged impurity. In
real space, electrons will attempt to distribute themselves evenly. If an impurity of
charge +Ze is added at the origin of our coordinate system, electrons near r = 0 will
surround the impurity (being attracted to its opposite charge). The electrons which
originally sat further away will then see a greater concentration of negative charge at
the origin and will move away. This depletes an area of negative charge which then
attracts electrons from further away. This same pattern continues until the electron
density is no longer a homogeneous distribution but oscillates throughout space. This
is schematically shown in Fig. 8.5. These are known as Friedel oscillations.

8.1.2 Plasmons and the Dielectric Function

Let us now discuss collective charge oscillations in the 2DEG. Plasmons are longitu-
dinal plasma oscillations where the electrons in a system oscillate together at some

124
 Electron Density

Screening Density Normal Density

r
+Ze

Figure 8.5: Friedel oscillations in the electron density in the presence of a charged
impurity.

frequency ωp known as the plasma frequency. In general, the plasmon frequency is

determined by the zeros of the complex dielectric function ε(ω, q).
To solve for the plasmon excitation in the 2DEG, return to Eqn. (8.4) and seek
the solution

1 + 2v(q)Π0 (ω, q) = 0. (8.17)

In 2D, the Fourier transform of the Coulomb potential is v(q) = 2πe2 /q. It is often
sufficient to solve for the low q behaviour of the plasmon. Therefore, let us consider
the limit q → 0, and ω > ℏ2 kF q/m. Return to Eqn. (8.8) in the continuum limit,
∫ { }
1 Θ(kF − k)Θ(|k + q| − kF ) Θ(k − kF )Θ(kF − |k + q|)
Π0 (ω, q) = 2 2
dk − ,
4π εk+q − εk − ω − i0+ εk+q − εk − ω + i0+

and Taylor expand the first term for small q. This gives

lim Θ(kF − k)Θ(|k + q| − kF ) ≈ lim Θ(kF − k)Θ(k + qcosθ − kF ), (8.18)

q→0 q→0

where we require cosθ ≥ 0 for this product to be finite. Expanding further, we obtain

lim Θ(kF − k)Θ(|k + q| − kF ) ≈ δ(k − kF )qcosθΘ(cosθ). (8.19)

q→0

125
Likewise, the second term gives

lim Θ(k − kF )Θ(kF − |k + q|) ≈ lim Θ(k − kF )Θ(kF − k − qcosθ), (8.20)

q→0 q→0

where we require cosθ ≤ 0 for this product to be finite. Expanding further, we obtain

lim Θ(k − kF )Θ(kF − |k + q|) ≈ −δ(k − kF )qcosθΘ(−cosθ). (8.21)

q→0

Combining these results and keeping up to first order in q, we have

∫ { }
m kF qcosθΘ(cosθ) kF qcosθΘ(−cosθ)
Π0 (ω, q) = 2 2 dθ + , (8.22)
2ℏ π 2kF qcosθ − ε̃ − i0
2 + 2kF qcosθ − ε̃2 + i0+

where ε̃2 ≡ 2mω/ℏ2 . Expanding in terms of real and imaginary parts, we notice that
for ω > ℏ2 kF q/m, ImΠ0 (ω, q → 0) = 0 and we obtain
[ ]
m 1
ReΠ0 (ω, q → 0) = 1− √ , (8.23)
2πℏ2 1 − 4kF2 q 2 /ω̄ 2

where ω̄ ≡ ε̃2 > 2kF q. In this limit, the plasmon frequency is found by solving
[ ( )]
4πe2 m 1
1+ 1− √ = 0. (8.24)
q 2πℏ2 1 − 4kF2 q 2 /ω̄ 2

In terms of the electron density 2πn = kF2 , the plasma frequency in 2D is given
by [149]

2πne2
ωp2 = q (8.25)
m

for q → 0.
Before moving on to an examination of electron-electron interactions in silicene,
we wish to present a brief description of the principle behind detecting these oscilla-
tions in experiment. A basic plasmon experiment involves subjecting a sample to an
incident electron of energy ℏω. If ℏω is sufficient, the Coulomb field of the electron

126
causes a plasma oscillation to begin. The energy of that oscillation (ℏωp ) must be
removed from the electron. Therefore, after exciting a plasmon, the electron leaves
with energy ℏ(ω − ωp ). The energy loss spectrum is then analysed. It will show peaks
at multiples of the plasma energy. Peaks in the loss function [Imε−1 (ω, q)] correspond
to zeros in the complex dielectric function.

8.2 Electron-Electron Interactions in Silicene

We now wish to examine the polarization function in silicene. We will compare
our results to the 2DEG when appropriate. Π(ω, q) has been studied extensively in a
variety of 2D systems beyond the usual electron gas [150]. For example, in addition to
the works on graphene-related systems both with [151–153] and without [117,154–161]
an energy gap, there has been work on semiconducting systems, such as quantum
wells, including spin-orbit coupling [162–168]. In the RPA, the dielectric function is
given by

ε(ω, q) = 1 − v(q)Π(ω, q), (8.26)

where v(q) = 2πα/q is the 2D Coulomb potential and α = e2 /(ℏε0 v) is the effective
fine structure constant in which e is the elementary charge and ε0 is the bare dielectric
constant. Note that we are using slightly different notation than that in the previous
section (the spin information is absorbed into Π(ω, q) as well as a minus sign). To
access information about collective excitations and many-body effects in silicene, we
need to calculate the dynamical polarization function Π(ω, q). In what follows, factors
of ℏ and v are ignored.
In the one-loop approximation, the polarization function is found by solving [62,

127
151–153, 155, 157, 159, 169, 170]

1 ∑
∫ ∑ ( k · (q + k) + ∆2σξ
)
2 ′
Π(ω, q) = 2 dk 1 + λλ
8π σ,ξ=±1 λ,λ′ =±1
Ek Ek+q
nF (λEk ) − nF (λ′ Ek+q )
× , (8.27)
λEk − λ′ Ek+q − ω − i0+
√
where Ek = ℏ2 v 2 |k|2 + ∆2ξσ [see Eqn. (3.26)]. In what follows, ∆ξσ ≡ (1/2)| −
ξσ∆so + ∆z | (i.e. it is implicitly positive) and we will frequently make use of the
definitions ∆max ≡ ∆+−/−+ and ∆min ≡ ∆++/−− (which correspond to the maximum
and minimum gaps at each valley, respectively). Here, we work at zero temperature
so that the Fermi functions nF (z) can be replaced by step functions. Since the
polarization function depends on the absolute value of the chemical potential (µ) and
given the general relation [152] Π(−ω, q) = Π(ω, q)∗ , we only present the results for
µ > 0 and ω > 0. Eqn. (8.27) can be solved analytically [35, 151, 152] to give

∑ [ σξ ]
Π(ω, q) = Π0 (ω, q)Θ(∆σξ − µ) + Πσξ
1 (ω, q)Θ(µ − ∆ σξ ) , (8.28)
σ,ξ=±1

where Πσξ σξ σξ
α (ω, q) = ReΠα (ω, q) + iImΠα (ω, q). If µ < ∆σξ the imaginary part of the

polarization is [35]
( )
q2 4∆2σξ
ImΠσξ
0 (ω, q) =− √ 1+ 2 Θ(ω 2 − q 2 − 4∆2σξ ), (8.29)
16 ω − q
2 2 ω − q 2

and the Kramers-Kronig related real part is [35]

{ [
q2 ∆σξ q 2 − ω 2 − 4∆2σξ q 2 − ω 2 − 4∆2σξ
ReΠσξ
0 (ω, q) =− + Θ(q − ω)arccos 2
4π q 2 − ω 2 4|q 2 − ω 2 |3/2 ω − q 2 − 4∆2σξ
√ ]}
(2∆σξ + ω 2 − q 2 )2
− Θ(ω − q)ln . (8.30)
|ω 2 − q 2 − 4∆2σξ |

128
For µ > ∆σξ , the real and imaginary parts are [35]



 0, 1A

 ( )



 G 2µ−ω
, 2A

 (
<
) q
( )



 G 2µ+ω
+ G 2µ−ω
, 3A

 < <

 ( q
) ( q
)

 G< 2µ−ω − G< 2µ+ω

 , 4A
µ  ( q
) ( q )
1 (ω, q) = −
ReΠσξ + f (ω, q) × G> 2µ+ω − G> 2µ−ω , 1B , (8.31)
2π 

q
( ) q



 G> 2µ+ω , 2B

 ( ) q
( )



 G 2µ+ω
− G ω−2µ
, 3B


>
( q
)
>
( q
)



 G> ω−2µ + G> 2µ+ω , 4B



 ( q
) ( q )
 G 2µ+ω − G 2µ−ω , 5B
0 q 0 q

and
 ( ) ( )

 2µ+ω
− 2µ−ω

 G> G > , 1A


q
( ) q



 G> 2µ+ω , 2A


q



 0, 3A





 0, 4A

ImΠ1 (ω, q) = −f (ω, q) ×
σξ
0, 1B , (8.32)

 ( )



 −G< 2µ−ω
, 2B

 q



 π(2 − x20 ), 3B





 π(2 − x20 ), 4B




 0, 5B

where

q2
f (ω, q) = √ , (8.33)
16π |q 2 − ω 2 |

√
4∆2σξ
x0 = 1+ , (8.34)
q2 − ω2

129
 √
G< (x) = x x20 − x2 − (2 − x20 )arccos(x/x0 ), (8.35)

√
G> (x) = x x2 − x20 − (2 − x20 )arccosh(x/x0 ), (8.36)

and

√ ( √ )
G0 (x) = x x − x0 − (2 − x0 )arcsinh x/ −x0 .
2 2 2 2
(8.37)

The regions specified in Eqns. (8.31) and (8.32) are given by

√
1A : ω <µ− (q − kFσξ )2 + ∆2σξ
√ √
2A : ±µ ∓ (q − kF ) + ∆σξ < ω < −µ + (q + kFσξ )2 + ∆2σξ
σξ 2 2
√
3A : ω < −µ + (q − kFσξ )2 + ∆2σξ
√
4A : −µ + (q + kFσξ )2 + ∆2σξ < ω < q
√ √
1B : q < 2kFσξ , q 2 + 4∆σξ2
< ω < µ + (q − kFσξ )2 + ∆2σξ , (8.38)
√ √
2B : µ + (q − kFσξ )2 + ∆2σξ < ω < (q + kFσξ )2 + ∆2σξ
√
3B : µ + (q + kFσξ )2 + ∆2σξ < ω
√ √
4B : q > 2kFσξ , q 2 + 4∆σξ2
< ω < µ + (q − kFσξ )2 + ∆2σξ
√
5B : q < ω < q 2 + 4∆2σξ

where

√
kFσξ = µ2 − ∆2σξ . (8.39)

These results reduce to those of gapped graphene [151, 152] in the limit ∆σξ → ∆.
The regions given by Eqn. (8.38) are shown in Fig. 8.6 for ∆min (left frame) and ∆max
(right frame) when ∆so /µ = 0.6 and ∆z /∆so = 1.5.
In Fig. 8.7, ImΠ(ω, q) is shown for ∆so /µ = 1.4 and varying ∆z (as shown in the
inset, the chemical potential is situated above both gaps). The results are scaled
by N (µ) = 2µ/π (i.e. density of states of graphene at µ). The white regions signify

130
Figure 8.6: Regions for the (left) K ↑ /K ′ ↓ and (right) K ↓ /K ′ ↑ bands in which
the polarization function has different expressions.

ImΠ(ω, q) = 0. These results are essentially the sum of two gapped systems with gaps
∆min and ∆max . As one might expect, when ∆z is applied such that ∆z < ∆so , the ω >
q regions splits into two with one moving toward the ω = q line and the other moving
away as ∆z increases [35]. When ∆z = ∆so , the system is in the VSPM phase and
ImΠ(ω, q) is not gapped. As ∆z is increased further, the two gaps in the ω > q region
grow with ∆z . Again, the differing behaviour of the three regimes, may allow for an
experimental determination of ∆so . As noted in reference to the 2DEG, the imaginary
part of the polarization function plays an important role in determining the plasmon
dispersion. The regions in (q, ω) space of non-zero polarization (i.e. the particle-
hole continuum) correspond to regions in which collective oscillations are damped.
This quantity is also related to the optical conductivity through [148, 157, 171, 172]
σ(ω) = limq→0 ie2 ωΠ(ω, q)/q 2 . Therefore, the behaviour of ImΠ(ω, q) along the q = 0
line is directly related to the absorptive part of the optical conductivity (refer to
Chapter 4).
ReΠ(ω, q) is plotted in Fig. 8.8 with the same ∆so /µ and ∆z as Fig. 8.7. Several
frequency cuts of Fig. 8.8(b) are shown in Fig. 8.9. While the imaginary part of the
polarization determines the damping of the plasmon, the real part of the dynamical
polarization is used to determine the location of the plasmon branch in (q, ω) space.
We shall return to this point later.

131
 Figure 8.7: Imaginary part of the silicene polarization function.

The static limit (ω = 0) of ReΠ(ω, q) is of particular importance as it determines

the screened potential of a charge impurity. For ω → 0, ImΠ(ω, q) → 0 and the static
polarization is given by [35]

µ ∑
Π(0, q) = − [F (q)Θ(∆σξ − µ) + G(q)Θ(µ − ∆σξ )] , (8.40)
2π σ,ξ=±1

where [152]
√
∆σξ q 2 − 4∆2σξ q2
F (q) = + arcsin , (8.41)
2µ 4qµ q 2 + 4∆2σξ

132
 Figure 8.8: Real part of the silicene polarization function.

and
√
q 2 − 4(kFσξ )2
q 2 − 4∆2σξ
G(q) = 1 − Θ(q − 2kFσξ )  −
2q 4qµ
√ 
q − 4(kF )
2 σξ 2

×arctan   . (8.42)
2µ

Π(0, q) is plotted in Fig. 8.10 for ∆so /µ = 1.4 and varying ∆z . We begin by
discussing the graphene limit (∆so /µ = 0) which is shown for comparison (solid
black curve). The static polarization of graphene is well known [154, 157, 159, 161,
173, 174]. A monolayer of graphene has a linear static polarization function of

133
 2
∆so/µ=1.4
∆z/∆so=0.25 6
ω/µ=2
1 ω/µ=2.5
ω/µ=3

-ReΠ(q,ω)/N(µ)
-ReΠ(q,ω)/N(µ)

4
µ
0

2
ω/µ=0
-1 ω/µ=0.5
ω/µ=1
ω/µ=1.5
(a) 0 (b)
-2
0 1 2 3 4 0 1 2 3 4
q/µ q/µ

Figure 8.9: Constant frequency cuts of the real part of the polarization function as a
function of q.
2
∆so/µ=1.4
∆z/∆so=0
∆z/∆so=0.25
{ µ
1.5 ∆z/∆so=1
- ReΠ(ω=0,q) / N(µ)

∆z/∆so=1.25
∆z/∆so=2.5
∆z/∆so=3.5
1
2√1-(∆max/µ)2
2√1-(∆min/µ)2

0.5

∆so/µ=0

0
0 1 2 3 4
q/µ

Figure 8.10: Static polarization of silicene.

Π(0, q) = −N (µ)πq/(8kF ). This is associated with interband transitions. There is

also an intraband contribution of the form Π(0, q) = −N (µ)[1−πq/(8kF )] for q ≤ 2kF
√
and Π(0, q) = −N (µ)[1 − (1/2) 1 − 4kF2 /q 2 − (q/4kF )sin−1 (2kF /q)] for q > 2kF .
Combining the two results gives a constant polarization of −N (µ) for q ≤ 2kF which
is typical of metallic screening [174, 175]. For large q, the interband term results

134
in a linear contribution which signifies insulating-like screening [174]. Returning to
Fig. 8.10, the dashed red and dashed-dotted blue curve correspond to µ > ∆max such
that the chemical potential is above the gap(s). Like graphene and the 2DEG, for
√
q/µ < 2 1 − (∆max /µ)2 , the polarization is constant. After this value, there is a de-
crease in the magnitude of the polarization function before the interband transitions
cause it to increase linearly. The solid green and dash-double-dotted orange curves
correspond to ∆min < µ < ∆max (i.e. µ lies between the two gaps). In this case, the
magnitude of the polarization begins at half the graphene value and shows a general
√
increasing trend with a small dip at q/µ = 2 1 − (∆min /µ)2 . Finally, the dash-
double-dotted purple and dotted magenta curves result from the chemical potential
∑
sitting in both gaps. In this case, Π(0, 0) = 0 and Π(0, q → 0) ∼ − σξ q 2 /(12π∆σξ ).
In all cases, as ∆z is increased, the magnitude of the polarization at a given q is less
than or equal to that of lower ∆z . When using the inset of Fig. 8.10, it is important to
note that the chemical potential is fixed and it is the band structure that is varying.
The inset merely shows where the chemical potential sits relative to the two bands.
We now wish to discuss several physical consequences of Π(ω, q).

8.3 Loss Function and Plasmons

As we saw in Section 8.1.2, the plasmon dispersion ωp (q) calculated in the RPA is given
by the zeros of the complex dielectric function. Equivalently, we can look for poles of
the loss function Imε−1 (ω, q). A plot of Imε−1 (ω, q) is given in Fig. 8.11 for α = 0.8,
∆so /µ = 1.4 and ∆z set to match Figs. 8.7 and 8.8. For graphene, the fine structure
constant has been experimentally determined for a variety of substrates [176]. On
SiC, it is reported as α ≈ 4.4, for h-BN α ≈ 7.7 and for Quartz α ≈ 17.7. The value
of α used herein was chosen to elucidate interesting features on a tractable energy
scale and not to mimic an experimental value. Another set of plots of Imε−1 (ω, q) is
shown in Fig. 8.12 for α = 0.8, ∆so /µ = 0.75 and varying ∆z . This set of parameters
will be used for our detailed examination of the plasmon branch. In each case, the
red branch corresponds to the pole of Imε−1 (ω, q) and thus describes the plasmon

135
Figure 8.11: Energy loss function of silicene for α = 0.8, ∆so /µ = 1.4 and varying
∆z .

excitation. In the (q, ω) domain for which ImΠ(ω, q) = 0 (white regions of Fig. 8.7),
the plasmon is undamped; however, when the imaginary part of the polarization is
non-zero, collective oscillations are quickly damped. As the polarization enters in
the denominator of the loss function (i.e. ε−1 ∝ 1/[1 − v(q)Π(ω, q)]), the plasmon
dispersion is not simply the sum of two gapped systems [35].
A similar situation of spin-split bands is found in MoS2 (see Ref. [170]) although
the details are quite different. Due to the large band gap in the MoS2 band structure,
the plasmons damp out in the intraband portion of the particle-hole continuum [170]
instead of the interband damping seen in graphene. In silicene, intraband damping
is observed for most values of ∆z ; however, for ∆z ≈ ∆so , the plasmon branch damps

136
Figure 8.12: Energy loss function of silicene for α = 0.8, ∆so /µ = 0.75 and varying
∆z . Inset: the location of the chemical potential relative to the band structure.

out in the interband region of the particle-hole continuum due to the closing of the
∆min gap [35]. In addition, as our model Hamiltonian is particle-hole symmetric,
there is no difference for the plasmon dispersion curve for the particles and holes as
found for MoS2 .
It is also useful to derive an approximate expression for the plasmon branch. To
do this, we must find the zeros of the dielectric function ε(ωp − iγ, q), where γ is the
plasmon decay rate. For weak damping, it is sufficient to solve [151–153, 157, 175],

q − 2παReΠ(ω, q) = 0, (8.43)

137
where the decay rate of the plasmon is determined by

ImΠ(ωp , q)
γ= . (8.44)
(∂/∂ω)ReΠ(ω, q)|ω=ωp

As we are primarily interested in the low q behaviour (which exists when ImΠ(ω, q) =
0 and thus γ = 0), we only report the results of Eqn. (8.43). Several constant ω slices
of 2παReΠ(ω, q)/µ are shown in Fig. 8.13(a) for α = 0.8, ∆so /µ = 0.75 and ∆z /∆so =
1. The solutions of Eqn. (8.43) are given by the intersection of 2παReΠ(ω, q)/µ versus

∆so/µ=0.75 ∆z/∆so=1 α=0.8

ω/µ=0.25 ω/µ=0.75
ω/µ=0.5 ω/µ=1
8 0.5
6 (a) (b)
(2πα)Re Π(q,ω)/µ

0.4
4 ωp/µ
2 ωo/µ 0.3
0 0.2
-2
0.1
-4
-6 0
0 0.4 0.8 1.2 1.6 0 0.2 0.4
q/µ q/µ
20 1
(c) (d)
15 0.8
-Im ε (q,ω)

0.6
-1

10
0.4
5
0.2
0 0
0 0.4 0.8 1.2 1.6 0 0.4 0.8 1.2 1.6
q/µ q/µ

Figure 8.13: (a) Frequency cuts of 2παReΠ(ω, q)/µ for α = 0.8, ∆so /µ = 0.375
and ∆z /∆so = 1. The intersection of 2παReΠ(ω, q)/µ with the line of unit slope
(orange) gives the solutions to Eqn. (8.43). (b) Plasmon frequency as a function of q
corresponding to the solutions of Eqn. (8.43). (c) Frequency cuts of the loss function.
(d) Zoomed in version of (c).

q/µ with the line of unit slope [177] (solid orange line). We can see that for low ω

138
(first two curves), the function intersects the line three times and, therefore, there
are three solutions at those frequencies (ωo ). This is shown in Fig. 8.13(b). The two
larger q contributions are not physical [152, 153] since Eqn. (8.43) is only exact when
ImΠ(ω, q) = 0. These values fall in the region when ImΠ(ω, q) is finite and thus
represent spurious solutions. We can see this by examining cuts of the loss function
shown in Fig. 8.13(c) and (d) where we note that for a given ω, there is only one pole
and thus only one plasmon branch. The physical solution is denoted ωp .
Using Eqn. (8.28), the numerical solutions to Eqn. (8.43) are obtained and are
shown in the left and right frames of Fig. 8.14 for ∆so /µ = 0.6 and 1.8, respectively
[35]. For ∆so /µ = 0.6, the chemical potential is above both gaps for all chosen values

∆z/∆so=0
∆z/∆so=0.25
∆z/∆so=0.75
∆z/∆so=1.0 α=0.8
1 ∆z/∆so=1.25
∆z/∆so=1.75

∆so/µ=0.6
0.75 µ{
Δz
ωp/µ

0.5

0.25
µ { Δz

∆so/µ=1.8
0
0 0.2 0.4 0 0.2 0.4 0.6
q/µ q/µ

Figure 8.14: Low-energy plasmon dispersion of silicene for varying ∆z .

of ∆z (see the inset). Here, the frequency of the plasmon branch decreases for a given
q as the external electric field is increased. For ∆so /µ = 1.8, the non-zero ∆z values
result in ∆min < µ < ∆max . In this regime, the plasmon frequency increases with ∆z
in the TI regime (while the lower band gap is decreasing) until it reaches a maximum
in the VSPM phase and then decreases with increased ∆z in the BI regime (while both
band gaps are increasing with increased ∆z ). Once µ < ∆min , collective oscillations
are not present as no states exist at the chemical potential. If the sublattice potential

139
difference could be tuned without substantially changing properties of the system
(such as electron-electron interactions), the location of the plasmon branch could
be manipulated. This may be of interest in the field of plasmonics. In addition, by
varying ∆z at fixed ∆min < µ < ∆max , the change in behaviour of the plasmon branch
could allow for an experimental determination of ∆so [35].
The plasmon behaviour can be captured analytically in the limit q ≪ ω ≪ µ. If
µ > ∆max , Taylor expanding Eqn. (8.43) for small q and ω gives [35]
v [
u
∑ ( )2 ]
ωp (q) u αq ∆
≈t
σξ
1− , (8.45)
µ 2µ σ,ξ=±1 µ

or, equivalently,
v [
u ( )2 ( )2 ]
ωp (q) u αq ∆ ∆
≈t
max min
2− − . (8.46)
µ µ µ µ

Written in terms of the individual Fermi momenta of the two spin split bands,
√
ωp (q) αq [ min 2 ]
≈ (k F ) + (k max 2
F ) , (8.47)
µ µ3
√ √
where kFmax = µ2 − ∆2max and kFmin = µ2 − ∆2min . In this form, it is clear that the
plasmon dispersion is dependent on the Fermi momentum of the individual spin-split
bands. If ∆min < µ < ∆max , the first order approximation ωp (q) is given by [35]
√ ( )
ωp (q) αq ∆2min
≈ 1− 2 . (8.48)
µ µ µ

As plasma excitations can only be sustained if there are states available to oscillate,
there is no plasmon branch if the chemical potential lies in both gaps.

140
8.4 Screening of a Charged Impurity
Finally, we wish to examine the screening of a charged impurity. In the RPA, the
static potential of a screened impurity of charge Q is given by [150, 153, 170, 178]
∫ ∞
Q J0 (qr)
Φ(r) = dq , (8.49)
ε0 0 ε(0, q)

where J0 (x) is the zeroth Bessel function of the first kind. The Lighthill theorem
[170, 179] implies that the non-analytic points of the dielectric function yield the
asymptotic behaviour of the screening potential. Examining Fig. 8.10, we see that
singular points arise in the derivative of the static polarization when q = 2kFmax and
2kFmin . Thus, sinusoidal Friedel oscillations are observed in Φ(r) [170].
Equation (8.49) can be evaluated numerically where we note that the value of
Φ(r) is primarily determined by the long-wavelength (small q) behaviour of the static
polarization function [Eqn. (8.40)]. The numerical evaluation of Eqn. (8.49) is shown
in Fig. 8.15 for α = 0.8, ∆so /µ = 0.6 and varying ∆z [35]. The upper left frame
(∆z = 0) contains the results of gapped graphene [152, 153]. That is, we have an
oscillatory function that decays as 1/r3 for small r and 1/r2 for large r. When a finite
sublattice potential difference is included and µ > ∆max , a beating effect is observed
in addition to the usual Friedel oscillations. The characteristic length between beats
changes as a function of ∆z . As a result, for small ∆z , which corresponds to the two
gaps being close together in energy, the characteristic wavelength of the beats is long.
As ∆z is increased, the separation between beats decreases due to the difference in
the two kF values becoming large. For sufficiently large ∆z , the beating becomes hard
to discern. This effect is only observed if µ is above both gaps.
Beating of the Friedel oscillations is also seen in other systems with split bands,
such as MoS2 [170]. This also results from the two kF values associated with the two
split bands which yield kinks in the static polarization function.

141
 0.06 ∆so/µ=0.3 ∆z/∆so=0 ∆z/∆so=0.1
0.04 α=0.8

0.02
(µr) Φ(r)

0
2

-0.02
-0.04
0.06 ∆z/∆so=0.3 ∆z/∆so=0.5
0.04
0.02
(µr) Φ(r)

0
2

-0.02
-0.04
0.06 ∆z/∆so=0.75 ∆z/∆so=1.25
0.04
0.02
(µr) Φ(r)

0
2

-0.02
-0.04
0 20 40 60 0 20 40 60 80
µr µr

Figure 8.15: Screening potential of a charged impurity in silicene.

142
 CHAPTER IX

MAGNETIC PROPERTIES OF
TOPOLOGICAL INSULATOR
SURFACE STATES

9.1 Surface State Hamiltonian

As discussed in Chapter 1, a 3D topological insulator is an insulator which pos-
sesses a bulk band gap and a metallic spectrum of topologically protected helical
surface states. In this chapter, we will restrict our attention to the Dirac spectrum
of the “second generation” TIs (ex. Bi2 Se3 , Bi2 Te3 ). These systems have a single
Dirac cone centred at the Γ point of each hexagonal surface Brillouin zone. Unlike
graphene (which exhibits particle-hole symmetry), the surface states of a TI display
band bending and take an hourglass shape. The valence band displays significantly
more outward bending than the inward bending of the conduction band. As previ-
ously mentioned, we account for this effect by adding a Schrödinger mass term to the
dominant ideal-Dirac Hamiltonian.
In the absence of a magnetic field, the helical Dirac fermions which characterize
a TI are well described by the Bychkov-Rashba Hamiltonian [180, 181]

ℏ2 k 2
Ĥ = + ℏvF (kx σ̂y − ky σ̂x ), (9.1)
2m

where σ̂x and σ̂y are the usual Pauli spin matrices and k is the momentum measured
relative to the Γ point of the surface Brillouin zone. The first term is the parabolic
Schrödinger piece for describing an electron with effective mass m. The second term

143
describes massless Dirac fermions which move with a Fermi velocity vF . Solving
Eqn. (9.1) yields the energy dispersion

ℏ2 k 2
E± (k) = ± ℏvF k. (9.2)
2m

A schematic of the energy bands is shown in Fig. 9.1. Figure 9.1(a) shows the surface-

Figure 9.1: (a) Comparison of ideal-Dirac cones and those of a TI. (b) Pure
Schrödinger band compared to that of the Bychkov-Rashba Hamiltonian.

state band structure in the TI regime (i.e. the Dirac term dominates). Here, a
momentum cutoff must be applied as we are mapping a continuum Hamiltonian onto
a lattice model. If not handled correctly, the model allows the valence band to bend
back over and eventually cross the zero-energy plane. This is not physical in the
context of TIs. The blue cones correspond to the pure Dirac limit (m → ∞) while
the red dispersion results from a non-infinite mass. In Fig. 9.1(b), the spintronic limit
is shown (i.e. the Schrödinger term dominates). The blue parabola corresponds to
the pure Schrödinger case (vF → 0) while the two offset red parabolas result from

144
a finite vF . It should be clear from this figure that, while the same Hamiltonian
describes both curves, in the confines of a Brillouin zone, they are very different
and lead to different physics. One must note that the low-energy Hamiltonian does
not allow these two regimes to be connected continuously; but, it does describe each
separately given appropriate values of the two characterizing parameters vF and m.
For Bi2 Te3 , detailed band structure calculations [28, 182] give: vF = 4.3 × 105 m/s,
and m = 0.09me where me is the bare mass of an electron. For the typical spintronic
materials, vF is reduced to O(103 )m/s [183, 184] while m remains at ∼ 0.09me .
It is also possible for a gap to be introduced at the Dirac point of the surface states.
This can be obtained by using thin films of TI material [3]. For TI slabs of thickness
< O(100 − 1000)nm, the states from the top and bottom surfaces hybridize and the
band structure becomes gapped [3]. The sign of the gap is different between the
two surfaces which has important ramifications of the LL dispersion [185]. A recent
experiment on Bi2 Se3 has shown that a gap can also be introduced to the surface
states by including magnetic dopants which break time-reversal symmetry [186]. In
this instance, a gap of ∆ ∼ 7 meV was obtained. These two effects are shown in
Figs. 9.2(a) and (b) for a thin film and magnetically doped surface, respectively.
Here the LL dispersion (coloured rings) which results from an external magnetic field
is overlaid on the B = 0 band structure. To account for this effect, we include a gap
term in our single-surface Hamiltonian to obtain

ℏ2 k 2
Ĥ = + ℏvF (kx σ̂y − ky σ̂x ) + ∆σ̂z . (9.3)
2m

Diagonalizing Eqn. (9.3), gives the energy dispersion

√
ℏ2 k 2
E± (k) = ± (ℏvF k)2 + ∆2 . (9.4)
2m

A comparison between the non-gapped and gapped topological band structure is

shown in Figs. 9.3(a) and (b), respectively.
To experimentally verify topological surface states, one must prove spin-momentum
locking (typically done with spin-ARPES) and a Berry phase of π. Both Shubnikov-

145
 <0
(a)

>0 Top Surface

N=0
2∆
TI Slab

2∆ N=0

Bottom Surface

>(1+g)E0/2 <(1+g)E0/2
(b)

Magnetic Dopant
N=0 B
2∆ N=0 2∆

TI

Figure 9.2: Surface states of (a) a TI thin film and (b) TI with magnetic dopants on
one surface.

de Hass and de Hass-van Alphen oscillations provide a powerful tool to verify the
latter feature. In this chapter, we will examine the magnetization of these surface
states and explore the dHvA effect.

9.2 Topological Insulators in a Magnetic Field

We must now explore how a magnetic field affects the surface states. Again, we will
work in the Landau gauge [A = (0, Bx, 0)] and make the usual Peierls substitution [7]

ℏk → ℏk + eA. (9.5)

146
 ε

k 2∆

(a) (b)

Figure 9.3: Comparison between the (a) non-gapped and (b) gapped band structures
of TI surface states.

We are also interested in including a Zeeman interaction which can be done by adding
−gs µB Bσz /2 to Eqn. (9.3), where gs is the Zeeman coupling strength and µB =
eℏ/(2me ) ≈ 5.78 × 10−2 meV/T is the Bohr magneton. Following the same procedure
as before (see Chapter 6) we arrive at the LL dispersion (see Appendix E.1.1)
 √ [ ]2


 E0 N + s 2N E 2 + ∆ − E0 (1 + g) N = 1, 2, 3, ...
1
EN,s = 2 , (9.6)

 E
 0
(1 + g) − ∆ N =0
2
√
where E1 ≡ ℏv eB/ℏ, E0 ≡ ℏeB/m and g = gs m/(2me ) is the renormalized Zeeman
√ √
coupling coefficient. For a typical TI, E1 / B ∼ 10.4 meV/ T, E0 /B ∼ 1.1 meV/T
and gs ∼ 8. It is important to note that in a sum over s, the N = 0 level will only
contribute once. The corresponding wave-functions are (Appendix E.1.2)
 
↑
CN,s |N − 1⟩
|ψ⟩ =  , (9.7)
↓
CN,s |N ⟩

147
where [37]
 √

 −s EN,+ − E0 N + s[∆ − (1 + g)E0 /2] N = 1, 2, 3, ...

↑
CN,s = 2(EN,+ − E0 N ) , (9.8)


 0 N =0

and
 √

 EN,+ − E0 N − s[∆ − (1 + g)E0 /2]
 N = 1, 2, 3, ...
↓
CN,s = 2(EN,+ − E0 N ) . (9.9)


 1 N =0

From the wave-functions, we immediately see that the N = 0 LL is entirely populated

↑
by spin-down electrons (i.e. CN,s = 0). To see the spin texture of the remaining levels,
we compute the average

ℏ
⟨ŝz ⟩ = ⟨ψ| σ̂z |ψ⟩ . (9.10)
2

This gives [37]


 s [∆ − (1 + g)E0 /2]

 √ N ̸= 0
ℏ
⟨ŝz ⟩ = 2E1
2
N + [∆ − (1 + g)E 0 /2] 2
. (9.11)
2
 −1 N =0

Clearly, an ŝz spin texture is quickly lost for N ̸= 0.

Now, return to Eqn. (9.6). Examining the N = 0 level reveals that for ∆ = 0,
the LL sits at positive energy. For finite ∆, the level remains above zero as long
as (1 + g)E0 /2 > ∆. For ∆ > (1 + g)E0 /2, the zeroth level is situated at negative
energy. Also, in contrast to graphene and silicene (where spin is a good quantum
number), a finite Zeeman term does not split the levels but simply renormalizes their
energy. Again, as opposed to silicene, the N ̸= 0 levels are not simply shifted by
±gs µB B/2. The zeroth level is translated in the usual way since it has a definite
spin-down polarization. A schematic plot of the low-energy LLs in a TI is shown in

148
Fig. 9.4. Particular attention is given to the relative location of the N = 0 level. The
various LL energies are given by the coloured circles (blue for conduction levels, red
for valence levels and purple for N = 0) and are overlaid on the B = 0 band structure.
An important feature of the LLs (and a fundamental focus of this chapter) is that,
∆=0 (1+g)E0/2>∆ (1+g)E0/2<∆

3+ 3+ 3+
2+ 2+ 2+
1+ 1+ 1+

0 0 ε=0
0

1- 1- 1-
2- 2- 2-
3-- 3- 3-
4 4- 4-

(a) (b) (c)

Figure 9.4: LL dispersion for a topological insulator with (a) ∆ = 0, (b) (1+g)E0 /2 >
∆ and (c) ∆ > (1 + g)E0 /2.

like the B = 0 band structure, there is strong particle-hole asymmetry. In graphene,

particle-hole symmetry is present. If a gap is included, the N ̸= 0 LLs remain
symmetric. The zeroth level is shifted down in energy at K and up in energy at K ′ ;
including both valleys returns the symmetry. When examining the Hall conductivity
in gapped graphene, the system appears insulating at charge neutrality. In fact, both
valleys are metallic but result in edge currents in opposite directions. This is clearly
seen in Fig. 9.5 where the quantization of the Hall conductivity has been obtained
from the magnetization (this is the silicene limit of ∆so = 0 and ∆z /2 → ∆ [see
Chapter 7]). The upper two frames show the individual valley contributions for K
and K ′ , respectively, while the lower frame shows the full result. We have chosen
√ √
E1 / B = 25.64 meV/ B which is characteristic of graphene. In the top frame, we
show the effect of Zeeman splitting; as spin is a good quantum number, a Zeeman

149
 4

0 K
K, gs=16
-4
4

dM(µ)/dµ (e/h)
0
K'
-4

E1/√B=25.64 meV/√T
4 B=1T
∆=2 meV
0
K+K'
-4

-40 -20 0 20 40
µ (meV)

Figure 9.5: Slope of the magnetization for gapped graphene at the (top) K point and
(middle) K ′ point. (bottom) The full result.

interaction simply shifts the spin-up levels down in energy and the spin-down levels
up by the same amount. To compare our results for the TI with an idealized Dirac
system, we will continually refer to the K point result of graphene which is similar
to the limit of E0 → 0.

9.2.1 Density of States

Let us now compare the density of states of graphene with that of a TI. In particular,
we shall examine the integrated density of states. In a magnetic field, the density of
states is given by a series of δ-functions which are peaked at the various LL energies.
For a TI,
 
eB  ∑
∞

N (ω) = δ (ω − E0,+ ) + δ (ω − EN,s ) , (9.12)
h N =1
s=±

150
where EN,s is given by Eqn. (9.6). Similarly, for graphene
 
eB  ( ) ∑ ( )
∞
ξσ 
δ ω − E0,+ + δ ω − EN,s
ξσ
Nξσ (ω) = , (9.13)
h N =1
s=±

where EN,s
ξσ
is given by Eqn. (6.7) in the limit ∆so = ∆z = 0. To compare the two
results, we define the total integrated density of states up to energy ωmax as [see
Eqn. (6.15)]
∫ ωmax
I(ωmax ) = N (ω)dω. (9.14)
0

A plot of I(ωmax ) as function of the cutoff energy ωmax is shown in Fig. 9.6. For
the graphene result, we have included the two-fold spin degeneracy; thus, the steps
are of double weight. Figure 9.6(a) shows the gapless results for graphene at the

Graphene K
E1/√B=10.4 meV/√T 100 E1/√B=10.4 meV/√T
80 ∆=0.5 meV
B=1T, gs=0, ∆=0 gs=0 B=1T
gs=8
I(ωmax ) ([ℏvF]-2 meV2)

I(ωmax ) ([ℏvF]-2 meV2)

80
60 E0/B=1.1 meV/T
Graphene K ∆=0.5 meV, g=0
60 ∆=0.5 meV, g=8
TI, E0/B=1.1 meV/T
40
40

20
20
(a) (b)
0 0
-20 -10 0 10 20 -20 -10 0 10 20
ωmax(meV) ωmax(meV)

Figure 9.6: Integrated density of states for graphene at the K point and a TI (a)
without and (b) with a gap of ∆ = 0.5 meV.

K point (solid black) and a TI (dashed red) in the absence of Zeeman interactions.
For graphene, the N = 0 LL is pinned at zero energy [43, 90–95] and particle-hole
symmetry is present for all N . For the TI, the N = 0 level sits at positive energy
and clear particle-hole asymmetry exists for the N ̸= 0 levels [37]; that is, the valence
levels sit closer to ωmax = 0 than the corresponding conduction levels. In Fig. 9.6(b),

151
a finite gap is included and Zeeman splitting is explored. For graphene with gs = 0
(solid black), the gap moves the N = 0 level to negative energy at the K point
and breaks particle-hole symmetry. With a finite Zeeman term (dash-double-dotted
green), the steps split into two (emphasized by the green circles). This is characteristic
of systems with spin-degeneracy. For a TI (dashed red), the gap shifts the energy
levels and a finite Zeeman interaction (dash-dotted blue) does not split the steps but
further renormalizes their energy. A plot of the LLs for ∆ = 0 and varying gs is
shown in Fig. 9.7. For graphene at K [frame (a)], the gs = 0 levels are symmetric.
(a) Graphene K

gs=8

gs=0

-30 -20 -10 0 10 20 30

Energy
E1/√B=10.4 meV/√T
E0/B=1.1 meV/T
B=1T, ∆=0
(b) TI

gs=8

gs=0

-30 -20 -10 0 10 20 30

Energy

Figure 9.7: ∆ = 0 LLs in (a) graphene and (b) a TI with and without Zeeman
splitting.

A finite Zeeman splitting causes the spin-down levels (red) to increase in energy by
gs µB B/2 while the spin-up levels (blue) decrease by the same amount. This splitting

152
does not depend on N and creates a series of spin-polarized levels. The TI [frame (b)]
is particle-hole asymmetric and a Zeeman term causes the conduction levels (pink)
to increase in energy while the valence levels (purple) decrease. Unlike graphene, the
N ̸= 0 levels are not spin-polarized and the splitting is not uniform; for larger N , the
renormalization becomes less prominent.

9.3 Magneto-Optical Conductivity

We now wish to briefly examine the effect of particle-hole asymmetry on the magneto-
optical conductivity of TI surface states. For a more thorough discussion, the reader
is referred to Ref. [107]. Again, we work in the one-loop approximation. Therefore,
the optical conductivity σαβ (Ω) is given by the Kubo formula
⟨ ⟩⟨ ⟩
′ ′
iℏ ∑
∞
fM,s′ − fN,s N̄ s|ĵα |M̄ s M̄ s |ĵβ |N̄ s
σαβ (Ω) = , (9.15)
2
2πlB N,M =0
EN,s − EM,s′ ℏΩ + EM,s′ − EN,s + iℏ/(2τ )
s,s′ =±

where fn,s is the Fermi function for state n in band s, τ is the relaxation time and
ĵα = ev̂α ≡ (e/ℏ)(∂ Ĥ/∂kα ) is the current operator. We work at zero temperature so
fn can be replaced by the Heaviside step function Θ(µ − En,s ). Here, the necessary
velocity operators are

ℏ ℏ a† − a
v̂x = kx + vF σ̂y = i √ + vF σ̂y , (9.16)
m m 2lB

and
[ ]
ℏ eBx ℏ a† + a
v̂y = ky + − vF σ̂x = √ − vF σ̂x . (9.17)
m ℏ m 2lB

153
Using Eqns. (9.7), (9.16) and (9.17), we can calculate the appropriate matrix elements
and obtain the real and imaginary parts of the longitudinal conductivity [37]

{ }
E12 ∑ fM,s′ − fN,s
∞
σxx (Ω) Γ
Re = (9.18)
e2 /ℏ 2π N,M =0 EN,s − EM,s′ (Ω + EM,s′ − EN,s )2 + Γ2
s,s′ =±

× [F(N s; M s′ )δN,M −1 + F(M s′ ; N s)δM,N −1 ],

and
{ }
E12 ∑ fM,s′ − fN,s
∞
σxx (Ω) Ω + EM,s′ − EN,s
Im = (9.19)
e2 /ℏ 2π N,M =0 EN,s − EM,s′ (Ω + EM,s′ − EN,s )2 + Γ2
s,s′ =±

× [F(N s; M s′ )δN,M −1 + F(M s′ ; N s)δM,N −1 ],

respectively, where Γ ≡ ℏ/(2τ ) and

[ ) ]2
′ ↑ ↓ E 0 (√ ↑ ↑
√ ↓ ↓
F(N s; M s ) ≡ CM,s ′ CN,s −√ N CM,s′ CN,s + N + 1CM,s′ CN,s . (9.20)
2E1

Likewise, the real and imaginary parts of the transverse Hall conductivity are [37]

{ }
E12 ∑ fM,s′ − fN,s
∞
σxy (Ω) Ω + EM,s′ − EN,s
Re = (9.21)
e2 /ℏ 2π N,M =0 EN,s − EM,s′ (Ω + EM,s′ − EN,s )2 + Γ2
s,s′ =±

× [F(N s; M s′ )δN,M −1 − F(M s′ ; N s)δM,N −1 ],

and
{ }
E12 ∑ fM,s′ − fN,s
∞
σxy (Ω) −Γ
Im = (9.22)
e2 /ℏ 2π N,M =0 EN,s − EM,s′ (Ω + EM,s′ − EN,s )2 + Γ2
s,s′ =±

× [F(N s; M s′ )δN,M −1 − F(M s′ ; N s)δM,N −1 ].

For the gapped graphene results, the reader is referred to Eqns. (6.22)-(6.25). The
appropriate limit is given by ∆so = 0 and ∆z /2 = ∆. As previously mentioned, a

154
momentum cutoff must be applied to the band structure to ensure that the valence
band does not bend back across the zero energy axis. In the presence of a magnetic
field, a corresponding cutoff on N must be applied to ensure that EN,− < 0.
We begin our discussion by consider the effect of finite ∆ on the absorptive part
of the longitudinal response. This is shown in Fig. 9.8 for a TI where Reσxx (Ω) is
plotted for µ = 0, g = 0 and three values of ∆: 0 (solid black), 0.5 meV (dashed blue)
and 1 meV (dash-dotted red). The inset shows a schematic plot of the lowest LLs for

1.5 ε N=1+
E1/√B=10.4 meV/√T
E0/B=1.1 meV/T
B=1T, g=0
µ=0 N=0

1
Re σxx(Ω) [e /ℏ]

N=1-
2

µ=0
∆=0
∆=0.5 meV
0.5 ∆=1 meV

0
0 10 20 30 40 50
Ω (meV)

Figure 9.8: µ = 0 longitudinal conductivity in a TI for varying ∆.

the different ∆’s. The lowest-energy optical transition is marked by the arrows. For
∆ = 0, the N = 0 LL sits at E0 /2 (for g = 0). Therefore, the lowest transition occurs
between the N = 1, s = − and N=0 levels. For finite ∆ less than E0 /2, the N = 0
level is still at positive energy but its magnitude has decreased; this causes the lowest
absorption peak to move down in Ω. As ∆ becomes larger than E0 /2, the N = 0
level moves to negative energy and the first transition is now between the zeroth level
and the N = 1 LL of the conduction band. Now, the frequency of the first transition
continues to increase with ∆ as E0,+ is pushed further down in energy. Higher optical
transitions are present and occur in pairs. This is a signature of the particle-hole
asymmetry as the energy for the EN,− to EN ±1,+ transition is not the same as the
EN ±1,− to EN,+ transition [37]. For graphene, the two sets of split higher-energy peaks

155
in Fig. 9.8 would coalesce into a single line. The scale of the splitting is set by the
Schrödinger magnetic energy E0 . This energy increases linearly with magnetic field
and inversely with decreasing m.
Next, we consider a finite chemical potential and g = 0. Figure 9.9 shows the
effect of positive and negative µ. In frame (a), the results of gapless graphene at the

30 N=4+
N=3+
E1/√B=10.4 meV/√T N=2+
20
B=1T,3 gs=0, ∆=0
N=1+
Graphene K 10
Energy

2.5 µ=5 meV 0 N=0

µ=-5 meV
-10
2 N=1-
-20 N=2-
30 N=3+
N=3-
1.5 N=4-
Re σxx(Ω) [e /ℏ]

-30 N=2+
20
2

N=1+

1 10
Energy N=0
0
0.5
-10
N=1-
(a)
N=2--
0 -20 N=3
N=4--
30 N=3+ N=5 (c)
-30
E1/√B=10.4 meV/√T
N=2+
E0/B=1.1 meV/T 20
B=1T, g=0 N=1+
10 E1/√B=10.4 meV/√T
∆=0 1 1 E0/B=1.1 meV/T
Energy

Re σxx(Ω) [e /ℏ]

µ=5 meV N=0 B=1T, g=0

0
2

µ=-5 meV
∆=0.5 meV
-10 - µ=0
N=1
Re σxx(Ω) [e /ℏ]

µ=-14 meV
N=2--
2

-20 N=3 µ=14 meV

0.5 N=4-- 0.5
N=5
-30

(b)
0 0
0 10 20 30 40 50 0 10 20 30 40 50
Ω (meV) Ω (meV)

Figure 9.9: Longitudinal conductivity in (a) gapless graphene, and (b) a gapless and
(c) gapped TI for positive and negative µ.

K point are shown; due to the particle-hole symmetry, the response is identical for
±µ. The optical transitions which lead to this set of absorption lines are shown in
the inset where the arrows are colour-coded to correspond to the appropriate curve.

156
Note that the positions of the EN,− LLs are at the negatives of the EN,+ levels. For
positive µ, the five arrows on the left (black) apply; while, for negative µ, it is the
five green arrows on the right which are relevant. In both cases, each transition has a
one-to-one correspondence with the other set. Figure 9.9(b), displays the results for
a TI when ∆ = 0. Here the finite Schrödinger term breaks particle-hole symmetry.
This is clear from the transitions shown in the inset. In this case, even for ∆ = 0, the
negative energy levels do not mirror the positive energy set. The N = 0 level is no
longer at zero energy but rather has been pushed to positive energy E0 /2. The EN,+
energy is also larger than |EN,− |. The black arrow between N = 0 and N = 1+ (which
applies to the lowest line for positive µ) is longer by E0 /2 than the arrows between
the N = 1− and N = 0 level which is the first transition when µ is negative. Thus,
in Fig. 9.9(b), the first peak of the negative µ response (purple) is lower in Ω than
the corresponding positive µ peak (black). The results for a gapped TI are shown in
frame (c) of Fig. 9.9 with the corresponding optical transitions shown in the inset.
Again, an obvious asymmetry is present between the ±µ regimes. The asymmetry is
now much larger than that shown in frame (b) for two reasons: the finite gap adds
asymmetry; the larger value of chemical potential enhances it is well. An additional
effect which needs to be emphasized is that one of the absorption lines for µ < 0
(blue) is missing in the first set of split peaks shown at higher energy. Of the four
peaks, the first, third and fourth apply to all three curves (µ = 0 and µ = ±14 meV),
but the second is only present in the black and red curves. The missing transition
is indicated by a blue × in the inset. It cannot occur because the N = 1− level is
unoccupied.
The absorptive part of the transverse Hall conductivity is also of interest as it is
involved in the response to circularly polarized light. A plot is shown in Fig. 9.10 for
the same parameters as Fig. 9.9(c). We note that when the optical transition is from
the N th level in the valence band to the (N − 1)th LL, the response is negative; while
the transition from the N th to (N + 1)th level is positive [37]. The shading under the
dashed blue curve for negative µ again helps to emphasize the missing peak around
∼ 36.6 meV which occurs for µ = 0 (black) and µ = 14 meV (red). The other three

157
 1.5
E1/√B=10.4 meV/√T
1 E0/B=1.1 meV/T
B=1T, g=0

0.5

Im σxy(Ω) [e /ℏ]
2
0

-0.5
∆=0.5 meV
µ=0
-1 µ=-14 meV
µ=14 meV

-1.5
0 10 20 30 40 50
Ω (meV)

Figure 9.10: The absorptive part of the transverse Hall conductivity for a TI with
parameters set to match Fig. 9.9(c).

peaks exist for all µ considered here.

We now wish to discuss the response to circularly polarized light which is given
by σxx (Ω) ± iσxy (Ω) for right- and left-handed polarization, respectively. Therefore,
the absorptive part is

Reσ± (Ω) = Reσxx (Ω) ∓ Imσxy (Ω). (9.23)

This is readily evaluated by utilizing Eqns. (9.18) and (9.22). A plot of the circular
dichroism is given in Fig. 9.11 for a gapped TI [again, the parameters are set to
match Fig. 9.9(c)]. Frames (a) and (b) show the response to right- and left-handed
polarized light, respectively. Right-handed light selects out the transitions between
the N th level in the valence band and the (N − 1)th LL in the conduction band. Left-
handed light selects the N to N + 1 transitions. Right circularly polarized light shows
a low-energy absorption peat at µ = 0 (solid black) which moves to lower energy in the
dashed blue curve for µ = −14 meV [frame (a)]. Such a peak is missing for µ = +14
meV (dash-dotted red). Instead, this peak is present for left circular polarization
[frame (b)] while the other two values of µ have no such absorption feature. The
pairs of peaks shown at higher energy in both frames are present for all three values

158
 E1/√B=10.4 meV/√T
3
E0/B=1.1 meV/T
(a) N→N-1 B=1T, g=0 N→N+1 (b)

∆=0.5 meV
µ=0
Re σ+ (Ω) [e /ℏ]

Re σ- (Ω) [e /ℏ]
µ=-14 meV
2 µ=14 meV
2

2
1

0
0 10 20 30 40 0 10 20 30 40 50
Ω (meV) Ω (meV)

Figure 9.11: Conductivity response to (a) right- and (b) left-handed circularly-
polarized light.

of µ for right-handed light. For left-handed polarization, the higher peak in the pair
is present for all µ considered here, the lower one is missing for µ = 14 meV (dashed
blue). This can be traced to the forbidden transition shown in the inset of Fig. 9.9(c)
(marked by the blue ×).

9.4 Magnetization of the Metallic Surface States

9.4.1 Magnetization and Hall Conductivity

Again, our discussion of the magnetization of the surface states begins with the grand
thermodynamic potential
∫ ∞ ( )
Ω(T, µ) = −T N (ω)ln 1 + e(µ−ω)/T dω, (9.24)
−∞

where T is the temperature (in units of kB ) and N (ω) is given by Eqn. (9.12). In
this chapter, we are only interested in the magnetization which depends on a B
derivative of the grand potential; for convenience, we add the B-independent term

159
 ∫∞
(µ/2) −∞
N (ω)dω to Eqn. (9.24). The integral of the density of states over all energies
gives the total number of states (a quantity which must be independent of B). Thus,
this additional term will not contribute to the magnetization (−∂Ω/∂B). At zero
temperature, Eqn. (9.24) becomes
∫ ( ∫ ∫
0
µ) µ
µ ∞
Ω(µ) = ω− N (ω)dω + (ω − µ)N (ω)dω + N (ω)dω. (9.25)
−∞ 2 0 2 0

For a gapped TI [37],

∫ ∞ ∫ 0
eB
N (ω)dω = N (ω)dω + Υ, (9.26)
0 −∞ h

where


 −1, (1 + g)E0 /2 < ∆


Υ= 0, (1 + g)E0 /2 = ∆ . (9.27)



 1, (1 + g)E0 /2 > ∆

Therefore,
∫ µ ∫ 0
eBµ
Ω(µ) = (ω − µ)N (ω)dω + Υ+ ωN (ω)dω. (9.28)
0 2h −∞

The last term in the above expression is independent of µ and gives the vacuum
contribution. It will simply provide a constant background to the µ dependence of
the magnetization and is therefore ignored. Keeping only the µ dependence, the first
two terms of Eqn. (9.28) yield [37]
[
eB Υµ
Ω̃(µ) = sgn(E0,+ ) (E0,+ − µ) Θ (|µ| − |E0,+ |) Θ (sgn(E0,+ )µ) +
h 2
∑
∞ ∑ ∞ ]
+ (EN,+ − µ) Θ (µ − EN,+ ) − (EN,− − µ) Θ (EN,− − µ) , (9.29)
N =1 N =1

where again, we require all the s = − states to be negative. Note that Θ(0) ≡ 1/2 as
only half the delta-function situated at ω = 0 is integrated.

160
 As discussed in Chapter 7, the slope of the magnetization is of particular impor-
tance as it is related to the quantization of the Hall conductivity through the Streda
relation [133] (σH = e∂M/∂µ). The quantization of the slope can be seen by noting
that
[
∂Ω eB Υ
=− sgn(E0,+ )Θ (|µ| − |E0,+ |) Θ (sgn(E0,+ )µ) −
∂µ h 2
∑
∞ ]
+ Θ (µ − EN,+ ) − Θ (EN,− − µ) . (9.30)
N =1

Thus, the slope of the magnetization is [37]

[
∂M ∂ ∂Ω e Υ
=− = sgn(E0,+ )Θ (|µ| − |E0,+ |) Θ (sgn(E0,+ )µ) −
∂µ ∂B ∂µ h 2
∑
∞ ]
+ Θ (µ − EN,+ ) − Θ (EN,− − µ) . (9.31)
N =1

A plot of M (µ) calculated from Eqn. (9.29) through M = −∂Ω/∂B is given in

Fig. 9.12(a). The corresponding slope [see Eqn. (9.31)] is shown in Fig. 9.12(b). Here,
g is taken to be 1 and the effect of varying ∆ is emphasized. A saw-tooth oscillation

E1/√B=10.4 meV/√T ∆=0

12 E0/B=1.1 meV/T ∆=0.5 meV 3
∆=1 meV
B=1T, g=1 ∆=2 meV
2
8 1
M(µ) (e/h meV)

dM(µ)/dµ (e/h)

0
4 -1

-2 ∆
0
-3

-4
(a) (b)
-4
-5
-20 -10 0 10 20 -20 -10 0 10 20
µ (meV) µ (meV)

Figure 9.12: (a) µ dependence of the magnetization for a TI with varying ∆. (b) The
corresponding slope of the magnetization.

161
pattern is present in M (µ) with the location of the teeth sitting at the various LL
energies. A half-integer quantization is present in the slope which is characteristic
of Dirac systems [96]. Therefore, the Hall conductivity is σH = (e2 /h)ν with filling
factors [36, 37] ν = ±1/2, ±3/2, ±5/2, .... For zero gap (solid black curve), the Hall
conductivity at µ = 0 has ν = −1/2. This results from the finite value of the N = 0
level [36]. As ∆ is increased, the location of the N = 0 step moves lower in µ. At
µ = (1 + g)E0 /2 − ∆, the step occurs at zero chemical potential; for larger ∆, the
small |µ| value of σH is given by ν = 1/2. Again, asymmetry is seen between the
negative and positive µ regimes. These results have been verified by taking the DC
limit (Ω → 0) of Reσxy (Ω) [see Eqn. (9.21)].
To compare this with gapped graphene, the reader is referred to Eqn. (7.48).
Again, the appropriate limit is ∆so = 0 and ∆z /2 → ∆. In Fig. 9.5, we present the
√ √
slope of the magnetization for graphene with ∆ = 2 meV, E1 / B = 25.64 meV/ T
(a characteristic value for graphene) and B = 1T. Note that the spin-degeneracy is
included. For a single spin species, the Hall response is characterized by a half-integer
filling factor. The location of the N = 0 LL is symmetric between the two valleys
(sitting at −∆ for K and ∆ for K ′ ). As a result, the total system near µ = 0 is
insulating (σH = 0) since the individual valleys contribute equal and opposite edge
channels. Here, a Zeeman interaction (dashed purple curve in the upper frame) splits
the steps in the conductivity into two.

9.4.2 Magnetic Oscillations

We now wish to examine the quantum oscillations which give rise to the dHvA effect
at low B fields. To begin, return to Eqn. (9.12) and express it as
 
eB d  ∑
∞

N (ω) = Θ (ω − E0,+ ) + Θ (ω − EN,s ) . (9.32)
h dω N =1
s=±

162
The oscillating part of the density of states can be extracted by applying the Poisson
formula

∑
∞ ∫ ∞ ∞ ∫
∑ ∞
1
F (N ) = − F (0) + F (x)dx + 2 F (x)cos(2πkx)dx. (9.33)
N =1
2 0 k=1 0

Following arguments given by Suprunenko et al. [139], this gives,

{
eB d 1 1
N (ω) = Θ(ω − |E0,+ |) − Θ(ω + |E0,+ |)
h dω 2 2
[ ]
∑∞
1
+ [Θ(ω + |E0,+ |) + Θ(ω − |E0,+ |) − Θ(ω − ωmin )] x1 + sin(2πkx1 )
πk
[ ]} k=1
∑∞
1
+ Θ(ω − ωmin ) x2 + sin(2πkx2 ) , (9.34)
k=1
πk

where
( )2
E2 E0 E0
ωmin =− 1 − (1 + g) − ∆ , (9.35)
2E0 2E12 2

and
√ ( )2
E12 ω E14 2E12 ω 1+g ∆
xi = + + (−)i + + − , (9.36)
E02 E0 E04 E03 2 E0

with i = 1, 2. This generalizes Eqn. (8) of Ref. [139] to include a gap and Zeeman
splitting. For E12 ≫ E02 and m → ∞ [36, 37],
[ ]
ω2 ω bE0 ωbE0
x1 ≈ 2
1− 2
− 2
+ , (9.37)
2E1 mvF 2mvF (mvF2 )2

where

[∆ − (1 + g)E0 /2]2
b≡ . (9.38)
E02

163
To lowest order in 1/(mvF2 ), the k th harmonic of the oscillating part of the magneti-
zation is (see Appendix E.2) [37]
[ ]
e (µ2 − ∆2 ) µ2 − ∆2
k
Mosc (µ) ≈− 1+ sin (2πkx1 ) , (9.39)
2πkh µ 2µmvF2

where
( )
µ [ 2 ]
1− µ − ∆2
mvF2 ∆(1 + g)
x1 ≈ 2
+ . (9.40)
2E1 2mvF2

Comparing Eqn. (9.40) to the customary [138]

ℏA(µ)
x1 = − γ, (9.41)
2πeB

the coefficient of the E1−2 ∝ 1/B term in Eqn. (9.40) is in fact related to the area of
the cyclotron orbit:
( )
π [µ2 − ∆2 ] µ
A(µ) = πkF2 ≈ 1− . (9.42)
ℏ2 vF2 mvF2

The remainder is the phase shift which is independent of B. It is [37],

∆(1 + g)
γ=− . (9.43)
2mvF2

In the pure Dirac limit (m → ∞, ∆ → 0), A(µ) reduces to the correct value [138,139]
of πµ2 /(ℏ2 vF2 ). For finite m and ∆ = 0, we obtain a correction [36] to A(µ) of
−[πµ2 /(ℏ2 vF2 )][µ/(mvF2 )]. Clearly, the gapless limit has a phase shift of 0 associ-
ated with a topological Berry phase of π [36, 187]. The amplitude of the quantum
oscillations [in units of −e/(2πkh)] is [37]
[ ]
(µ2 − ∆2 ) µ 2 − ∆2
AM = 1+ , (9.44)
µ 2µmvF2

164
which properly reduces to the results of Sharapov et al. [126] when m → ∞, i.e.
AM = (µ2 − ∆2 )/µ (see their Eqn. (8.10) in the pure limit), and to the results
of Tabert and Carbotte [36] when ∆ = 0, i.e. AM = µ[1 + µ/(2mvF2 )] [see their
Eqn. (49)]. Wright and McKenzie [140] applied the semiclassical quantization method
of Onsager [121]. This was augmented by including a first correction in the band
structure for the energy shift due to the magnetic response of the system (as discussed
by Fuchs et al. [60]). They obtained a phase offset in the magnetic oscillations which,
to within a sign, reduces to our result [Eqn. (9.43)] for large m and when the Zeeman
interaction is neglected (i.e. g = 0). The sign discrepancy is due to the chirality
difference between our Hamiltonians.
In the semiclassical approximation, the phase offset is 1/2 for non-relativistic
Schrödinger particles and 0 for relativistic Dirac fermions. Including a finite mass
and magnetic gap, the offset is changed to γ = −∆(1 + g)/(2mvF2 ). This gives
the relativistic result when either ∆ = 0 or m → ∞. Except for Zeeman factor
g, our quantum mechanically derived results agree with those of Ref. [140] where
semiclassical arguments where employed [3, 60, 97, 138, 188–190].
In Ref. [60], Fuchs et al. begin with the Bohr-Sommerfeld quantization condition
originally suggested by Onsager [121]. They include a correction of −M(k) · B to
the band energies due to the orbital magnetic moment M(k) induced by the external
magnetic field B. They apply this to a two band model of gapped graphene and
show that the phase offset is zero in the magnetic oscillations. This is the case even
though the Berry phase for a complete cyclotron orbit is not π. Instead, it is equal
to πWc (1 − ∆/µ) [191] where Wc is the topologically-invariant winding number. In
Ref. [60], it is referred to as the topological part of the Berry phase. They find that
γ = 1/2 − |Wc |/2 = 0 [see their Eqn. (35)] where only the topological part of the
Berry phase is relevant. In this model, −|Wc |/2 cancels the Maslov index contribution
of 1/2 [60] giving γ = 0. Wright and McKenzie [140] extend these semiclassical
arguments to the particle-hole asymmetric case when a Schrödinger mass is also
included. In that case, the result for the phase offset is given by their Eqn. (26).
To lowest order in 1/(mvF2 ), this reduces to ∆/(2mvF2 ). Here, we have derived this

165
result (up to a sign) without requiring any semiclassical-quantization arguments, and
further find its generalization to include the Zeeman term.

166
 CHAPTER X

CONCLUDING REMARKS

In this thesis, we have examined several low-energy phenomena in 2D Dirac-like sys-

tems. We began by exploring the optical and magnetic properties of a low-buckled
honeycomb lattice with intrinsic spin-orbit coupling (ex. silicene, germanene,...)
which map onto an effective Kane-Mele Hamiltonian for a quantum spin Hall in-
sulator. We illustrate how a sublattice potential difference (which can be generated
by an electric field) can lead to valley-spin-polarized charge carriers. Throughout
this work, we emphasize the difference between the two insulating regimes (topolog-
ical and trivial) which are accessed by tuning the electric field. While most of our
work focusses on the single-particle picture, in Chapter 8, we explore the effect of
electron-electron interactions.
For the final chapter, we consider the magnetic properties of the particle-hole
asymmetric surface states of a 3D topological insulator. Particular attention is given
to the quantized Hall conductivity and the magnetic oscillations which exist for low
magnetic fields. We provide a derivation of the de Haas-van Alphen effect with-
out requiring the Onsager relation and compare this to the results obtained by the
approximate semiclassical-quantization condition.
Currently, 2- and 3D topological insulators are the subject of great interest as the
low-energy Dirac-like electrons are predicted to be promising candidates for new tech-
nological applications. As new topological materials are identified and synthesized,
many more novel phenomena are sure to be observed.

167
 APPENDIX A

RELATIVISTIC FERMIONS IN 2D

In this thesis, we explore several select phenomena in Dirac-like systems. In condensed

matter physics, the most famous (and fundamental) example is graphene (carbon
atoms which bond on a 2D honeycomb lattice). Indeed, it is this system upon which
the models discussed herein are based.

A.1 Graphene: A Monolayer of Carbon Atoms

The novel low-energy band structure of graphene has been known since 1947 [192].
Due to its 2D structure, it was originally believed to be unstable in the free state [193].
However, in 2004 Andre Geim and Konstantin Novoselov made the first experimental
discovery of this wonder material. The linear energy dispersion which results in
graphene’s relativistic nature was shown to arise from a simple nearest-neighbour-
hopping-tight-binding Hamiltonian. At low energy, this Hamiltonian maps onto a
Dirac Hamiltonian for massless fermions with Fermi velocity vF ∼ 106 m/s [194].
The crystal lattice of graphene (see Fig. A.1), is a textbook example of a Bravais
lattice with a two point basis. Clearly, graphene is comprised of two triangular
sublattices labelled A and B. Any primitive unit cell will contain two atoms and thus
result in two energy bands with different sublattice index. This system is described
by the primitive lattice vectors
( √ ) ( √ )
a a 3 a a 3 √
a1 = , , a2 = − , , where a = |a1 | = |a2 | = 3acc , (A.1)
2 2 2 2

with acc ≈1.42Å the carbon-carbon distance [195] (therefore, a ≈ 2.46Å). The

168
 B

a2 a1
B
A A
δ3 δ2
δ1

Figure A.1: Graphene lattice with emphasized triangular sublattices on the A and B
sites.

nearest-neighbour vectors are

a1 + a2
δ1 = − ,
3
2 1
δ2 = δ1 + a1 = a1 − a2 , (A.2)
3 3
1 2
δ3 = δ1 + a2 = − a1 + a2 .
3 3

In reciprocal space, this results in a hexagonal first Brillouin zone which contains two
inequivalent K points labelled K and K ′ ≡ −K (see Fig. A.2). Taking a3 = (0, 0, 1),
the corresponding reciprocal lattice vectors are
( )
2π 2π
b1 = ,√ , (A.3)
a 3a
( )
2π 2π
b2 = − , √ ,
a 3a

which gives the K point coordinates K = (4π/(3a), 0).

Assuming that the valence electrons are tightly bound to the atom from which
they come, a nearest-neighbour-tight-binding Hamiltonian can be written [43,44]. At

169
 b1+b2

b2 K' b1

-b1 -b2

-b1-b2
Figure A.2: First Brillouin zone of the honeycomb lattice.

low energy, this Hamiltonian maps onto an effective model-Hamiltonian Ĥ. In the
basis ψ = (ψKA , ψKB , ψK ′ A , ψK ′ B )T ,
 
ĤK 0
Ĥ =  , (A.4)
0 ĤK ′

with the associated 2D Weyl Hamiltonians for a single valley

∗
ĤK = ℏvF k · τ = −ĤK ′, (A.5)

where k is the momentum measured relative to the Dirac point and τ are the usual
Pauli matrices for the sublattice pseudospin of the system. The energy eigenvalues
are

EG (k) = ±ℏvF |k|. (A.6)

A plot of the low-energy band structure around the first Brillouin zone is shown in
Fig. A.3. These cones (which meet at charge neutrality) are known as Dirac cones.

170
 K
K'

Figure A.3: Low-energy dispersion for graphene over and beyond the first Brillouin
zone.

It is also useful to solve for the eigenfunctions. The momentum space wave-
functions are [38]
 
ik·r 1 
1
ΨK
s,k = e √  ≡ eik·r us (k), for ĤK (k) = ℏvF τ · k, (A.7)
2 iθ
se k

and
 
ik·r 1 
1
ΨK
′
s,k = e √  ≡ eik·r vs (k), for ĤK ′ (k) = −ℏvF τ ∗ · k, (A.8)
2 −se−iθk

where θk = arctan(ky /kx ) and s = ± for the eigenenergies EG (k) = ±ℏvF |k|, re-
spectively. An important characteristic of the wave-functions is their chirality. For
massless fermions, this is identical to the helicity which is defined as the projection
of the particles pseudospin on its momentum [137]

1 k
ĥ = τ · . (A.9)
2 |k|

We can see that the helicity operator has the same τ · k dependence as the Hamilto-
nian. The wave-functions are thus eigenstates of helicity [137],

1
ĥψK (r) = ± ψK (r). (A.10)
2

171
At the K ′ point, the sign is merely inverted. An important consequence of this is that,
at a given phase, the electrons near the Dirac points will have an opposite helicity to
the holes. Likewise, an electron at k will have opposite chirality to one at −k. As
this is a conserved quantity, backscattering is strongly suppressed.
The helicity of graphene is directly related to the Berry phase which can be cal-
culated from our wave-functions and Eqn. (1.7):
∫
γn = −i ⟨n(R)| ∇R |n(R)⟩ · dR. (A.11)
C

As an example, we will compute γn using the K point wave-function. Our contour C

will be represented in momentum space, leading us to the expression
I
γn = −i ⟨us (k)| ∇k |us (k)⟩ · dk, (A.12)
C

where C encircles the K point in the counter-clockwise direction. Therefore,

 
I
1( ) 0
γn = −i 1 se−iθk   · dk
C 2 iθk dθk
sie dk
I
1
= dθk
2 C

= π, (A.13)

where C runs from θk = 0 to 2π. We have arrived at the famous result that graphene
has a Berry phase of π. Physically, this means that the pseudospin of the electrons
(which is parallel to the momentum), wraps once around the Dirac point.

A.2 Optical Conductivity of Graphene

For comparison with Chapter 4, we now wish to establish the well-known conductivity
results of graphene. As K and K ′ ≡ −K have a large separation in momentum space,
it is sufficient to examine the physics about a single K point and include a degeneracy
factor of gv = 2. The energy bands of graphene are also spin-degenerate; thus, spin

172
is included by a degeneracy factor gs = 2. The optical conductivity of graphene is
given by
∫ ∞
gs gv e2 dω
σαβ (Ω) = [f (ω − µ) − f (ω + Ω − µ)]
2Ω −∞ 2π
∫ [ ]
d2 k
× Tr v̂ α Â(k, ω + Ω)v̂ β Â(k, ω) , (A.14)
(2π)2

where f (x) is the Fermi-Distribution function:

1 1
f (x) = , β= . (A.15)
eβx +1 kB T

At zero temperature, β → ∞ and

f (x) = Θ(−x), (A.16)

where Θ(x) is the Heaviside step function which is 1 for x > 0 and zero otherwise.
Therefore,
∫ ∞ ∫ µ
dω dω
[f (ω − µ) − f (ω + Ω − µ)] → . (A.17)
−∞ 2π µ−Ω 2π

To evaluate Eqn. (A.14), we need the spectral function A(k, ω) which is defined
through
∫ ∞
dω Â(ω)
Ĝ(z) = , (A.18)
−∞ 2π z − ω

where Ĝ(z) is the retarded single particle Green’s function given by

Ĝ−1 (z) = z Iˆ − Ĥ. (A.19)

Here, Iˆ is the identity matrix of dimension dim(Ĥ). At the K point,

 
z −ℏvF (kx − iky )
Ĝ−1 (z) =  . (A.20)
−ℏvF (kx + iky ) z

173
This gives
 
1 z ℏvF (kx − iky )
Ĝ(z) =  . (A.21)
z2 − |ℏvF k|2
ℏvF (kx + iky ) z

From this, we see that

[ ]
1 1 1
G11 (z) = + . (A.22)
2 z − ℏvF |k| z + ℏvF |k|

Given Eqn.(A.18), we find

A11 (k, ω) = A22 (k, ω) = π[δ(ω − ℏvF |k|) + δ(ω + ℏvF |k|)]. (A.23)

Identically,

ℏvF (kx − iky )

G12 (z) = G∗21 (z) =
z 2 − |ℏvF k|2
[ ]
ℏvF (kx − iky ) 1 1 1
= − , (A.24)
ℏvF |k| 2 z − ℏvF |k| z + ℏvF |k|

which yields the spectral elements

ℏvF (kx − iky )

A12 (k, ω) = A∗21 (k, ω) = [δ(ω − ℏvF |k|) − δ(ω + ℏvF |k|)]. (A.25)
ℏvF |k|

Note that we are treating z as real number so that G∗ (z) and A∗ (z) merely interchange
ℏvF (kx + iky ) and ℏvF (kx − iky ). For convenience, we will denote EG (k) = ℏvF |k|.
To find the velocity operator, we begin by specifying the current density operator

e ∂ Ĥ
ĵ = − . (A.26)
ℏ ∂k

Employing Eqn. (A.5), we find

 
∂ Ĥ 0 1
= ℏ vF  , (A.27)
∂kx 1 0

174
and
 
∂ Ĥ 0 −i
= ℏ vF  . (A.28)
∂ky i 0

We know [6], j = −ev; therefore,

 
0 1
v̂x = vF  , (A.29)
1 0

and  
0 −i
v̂y = vF  . (A.30)
i 0

For the longitudinal conductivity, α = β = x and we are left with

Tr[v̂x Â(k, ω + Ω)v̂x Â(k, ω)] =

vF2 [A∗12 (k, ω + Ω)A∗12 (k, ω) + 2A11 (k, ω + Ω)A11 (k, ω) + A12 (k, ω + Ω)A12 (k, ω)]
(A.31)

as the integrand of Eqn. (A.14).

Due to the symmetry in k, we can write
∫ ∫ 2π ∫ kC
d2 k dθ kdk
= , (A.32)
(2π)2 0 2π 0 2π

where kC is a momentum cutoff characteristic of the band width. Here, it is convenient

to note that ℏvF (kx + iky ) can be expressed as ℏvF keiθ , where θ is the polar angle in
momentum space about the K point. Using this notation, we see that

A∗12 (k, ω + Ω)A∗12 (k, ω) ∝ [ℏvF (kx + iky )]2 ∝ e2iθ ,

and
A12 (k, ω + Ω)A12 (k, ω) ∝ [ℏvF (kx − iky )]2 ∝ e−2iθ .

175
Since there is no other angular dependence, these terms will average to zero in the θ
integral. We now have
∫ µ ∫ 2π ∫ kC
gv gs e2 vF2 dω dθ kdk
σxx (Ω) = 2A11 (k, ω + Ω)A11 (k, ω). (A.33)
2Ω µ−Ω 2π 0 2π 0 2π

For the k integral, we recall EG (k) = ℏvF k ≡ ε, thus

∫ kC ∫ WC
εdε
kdk → , (A.34)
0 0 ℏ2 vF2

where WC is the band cutoff which we take to be large. It is important to note that
we have taken ℏ = 1 in the conversion from energy to momentum. Therefore, an ℏ
should be include with all the ω terms. Restoring the ℏ to dω, we obtain
∫ µ ∫ WC
σxx (Ω) 4
= dω εdε[δ(ω + Ω − ε)δ(ω − ε) + δ(ω + Ω + ε)δ(ω + ε)
σ0 Ω µ−Ω 0

+ δ(ω + Ω + ε)δ(ω − ε) + δ(ω + Ω − ε)δ(ω + ε)],

(A.35)

where σ0 ≡ e2 /(4ℏ) and we have set EG (k) = ε. This can be solved numerically by
using the Lorentzian form of the δ-function. That is,

1 η
δ(x) = lim , (A.36)
η→0 π η + x2
2

where the broadening parameter η manifests itself as an effective transport scattering

rate of 1/τimp = 2η due to the convolution of the two δ-functions in the conductivity
formula [196]. It is also possible to derive an analytic expression [197] which has the
simple form
σxx (Ω)
= 4|µ|δ(Ω) + Θ(Ω − 2|µ|). (A.37)
σ0
The conductivity [Eqn. (A.35)] with broadening η = 0.007 meV is shown in Fig. A.4(b).
Figure A.4(a) shows a low-energy Dirac cone which we use to discuss the conductiv-
ity. At charge neutrality [dashed blue curve of Fig. A.4(b)], all the states below the
Dirac point are occupied while those above are empty. When a photon of energy

176
Figure A.4: (a) Low-energy band structure of graphene and the allowed optical tran-
sitions. (b) Optical conductivity of graphene.

Ω = 2ε interacts with the sample, it can excite an electron from an occupied state
at energy −ε to an available state at ε. These excitations between Dirac cones are
known as interband transitions. For a finite chemical potential µ, all states below µ
are occupied; thus, the lowest energy interband transition occurs for Ω = 2µ with all
lower energy transitions forbidden by Pauli blocking. For a finite µ [solid red curve of
Fig. A.4(b)], it is possible for an electron just below the chemical potential to access
a state just above. This is known as an intraband transition and is a “zero” energy
response. This is the contribution from the δ-function in Eqn. (A.37) and is modelled
by the Drude conductivity. Above Ω = 2µ, the conductivity is flat at a universal
background value of σ0 which can be traced back to the linear energy dispersion of
graphene.
The interband transitions correspond to transitions between bands of different
chirality (i.e. different sublattice index). This is expected as we are inducing a
current which results from electrons hopping between nearest neighbours (i.e. A-to-
B and B-to-A). Mathematically, this can be seen from the off-diagonal form of the
velocity matrix which only couples A and B sites. This behaviour has been extensively
examined and confirmed experimentally [198–201].

177
 APPENDIX B

MASSLESS DIRAC FERMIONS IN A

MAGNETIC FIELD

In Chapter 1, we introduced the IQHE and discussed its relation to topology. We

found that in a magnetic field, electrons move in circular orbits and are well described
by a simple harmonic oscillator Hamiltonian. This gives rise to a series of highly
degenerate LLs with an energy spacing of ℏωc which are characterized by an integer
index N ∈ Z+ . In that discussion, we assumed a 2DEG with the usual quadratic
B = 0 energy dispersion. In this thesis, we are primarily interested in the behaviour
of relativistic fermions as are found in graphene. We now wish to develop the famous
result for relativistic LLs, and their implication on the Hall conductivity.

B.1 Relativistic LLs

As discussed in Chapter A.2, the low-energy physics of graphene is well described
by a simple nearest-neighbour tight-binding Hamiltonian. In the vicinity of the two
K points (K and K ′ ) of the hexagonal first Brillouin zone, the energy dispersion is
linear and an effective Dirac Hamiltonian can written [see Eqn. (A.5)]. That is,

∗
ĤK = ℏvF k · τ = −ĤK ′, (B.1)

where k is the momentum measured relative to the Dirac point and τ = (τ̂x , τ̂y ) is a
vector of Pauli matrices associated with the pseudospin of the system. In a magnetic
field B = B ẑ, we can define a magnetic vector potential A = (−By, 0, 0) which
is related to B through the relation B = ∇ × A. The magnetic field changes the

178
momentum operators by the usual Peierls substitution

e
ℏk → ℏk + A, (B.2)
c

where ℏk is the momentum for B = 0, e is the charge of the electron and c ≡ 1 is the
speed of light. Therefore, our Hamiltonian for a single valley becomes
 [ ( ) ] 
0 ℏvF ξ kx − eBy
− iky
Ĥξ =  [ ( ) ]
ℏ , (B.3)
ℏvF ξ kx − eBy
ℏ
+ iky 0

where ξ = ± for the K and K ′ point, respectively, and we have expressed Ĥξ in the
sublattice basis (ϕA , ϕB )T . To solve for the LL spectrum, we employ Ĥψ = Eψ where
ψ T = (ϕA , ϕB ) is the wave-function with components associated with the A and B
sublattices. Using
[ ( ) ]
eBy
Ĥξ = ℏvF ξ kx − τ̂x + ky τ̂y , (B.4)
ℏ

and applying the Hamiltonian relation twice, we find

[( )2 ]
eBy eB
Ĥξ2 = ℏ2 vF2 kx − + ky2 − ξτ̂x τ̂y [y, ky ] , (B.5)
ℏ ℏ

where we have used τ̂x τ̂y = −τ̂y τ̂x and the identity matrix is implicit. Using the
familiar position-momentum commutation relation [y, p̂y ] = iℏ, our wave-function ψ,
and Ĥ 2 ψ = E 2 ψ, we find
[( )2 ]
eBy eB
ℏ2 vF2 kx − + ky2 + ξ ϕA = E 2 ϕA , (B.6)
ℏ ℏ

and
[( )2 ]
eBy eB
ℏ2 vF2 kx − + ky2 − ξ ϕB = E 2 ϕB . (B.7)
ℏ ℏ

179
The first two terms of Eqn. (B.6) can be rearranged into

( )2
ℏkx
2
Ĥ12 = vF2 e2 B 2 y− + ℏ2 vF2 ky2 , (B.8)
eB

which has the form of the simple harmonic oscillator Hamiltonian where 2m = vF−2
and ωc = 2vF2 eB. This may be replaced with its eigenvalue ℏωc (N + 1/2). Therefore,
Eqn. (B.6) gives

√
EN,s
A
= s 2ℏevF2 N B + ℏevF2 B(1 + ξ), (B.9)

where N is the LL index and s = ± is a band index associated with the B = 0

conduction and valence bands, respectively. Likewise, Eqn. (B.7) gives

√
EN,s
B
= s 2ℏevF2 N B + ℏevF2 B(1 − ξ). (B.10)

Comparing these two solutions, we note that at the K point (ξ = 1), there is an N = 0
LL at zero energy which is entirely associated with the B sublattice. Similarly, at K ′
(ξ = −1), the N = 0 level at zero energy is entirely due to the A sublattice. We can
combine the two results to find the total dispersion

√
EN,s = s 2ℏevF2 N B, (B.11)

where N = 0, 1, 2, 3, ... and each level will have a four-fold valley and spin degeneracy.
Examining Eqn. (B.11), we note that there are several fundamental, and striking
differences between the relativistic LLs and those of the usual 2DEG discussed in
Chapter 1. The first notable difference is that the LLs are no longer equally spaced
by eB/ℏ. Here, the N dependence falls under a square root, and, thus, the level
spacing decreases as N increases. Secondly, the LLs are not a linear function of B
√
but evolve as B. While for the 2DEG, all the LLs were at positive energy, here,
negative energy states are present. Finally, in graphene there is a level which is
pinned at zero energy [43, 90–95]. This is particularly interesting. In zero magnetic

180
field, the graphene band structure forms a pair of Dirac cones which meet at zero
energy. Thus, there are no states which exist for E = 0. With the application of a
magnetic field, a highly degenerate level is pinned at zero energy. The single spin and
valley degeneracy of this state is eB/h [see Eqn. (1.20)]. Figure B.1(a) shows the LL

8 7 6 5 4
120
3
N=3, s=+
100
N=2, s=+ 2
ε N=1, s=+
80

Energy (meV)
1
N=0 60
k

N=1, s=- 40
N=2, s=-
N=3, s=- 20

0 0
0 1 2 3 4 5
B (T)

(a) (b)

Figure B.1: (a) The LL dispersion in graphene. (b) The LL evolution as a function
of B.

formation in the presence of a finite B field; it depicts the formation of discrete levels
when a magnetic field is included (reproduced from Fig. 6.1). The results are shown
for a single K point and are identical at the other valley. We note that the N = 0
level is special in that it is made from equal contributions of particles and holes [96].
√ √
Figure B.1(b) shows the B dependence of the levels for ℏevF2 = 25.64 meV/ T,
which corresponds to vF ≈ 106 m/s. The LL index is given next to the corresponding
curve. Only the s = + band is shown as the s = − dispersion is mirror symmetric.
We can also solve for the corresponding eigenfunction solutions to Eqn. (B.3). For
convenience, let us define the two ladder operators

( ) √
eBy 2eB
kx − − iky = −i a, (B.12)
ℏ ℏ

181
and
( ) √
eBy 2eB †
kx − + iky = i a. (B.13)
ℏ ℏ

When acting on a Fock state of the harmonic oscillator (denoted by |N ⟩ for N =

√ √
0, 1, 2, ...), a |N ⟩ = N |N − 1⟩. Likewise, a† |N ⟩ = N + 1 |N + 1⟩. Our Hamilto-
nian for a single valley is then
 √ 
0 −iℏvF 2eB
a
∗
ĤK = −ĤK ′ =
 √ ℏ . (B.14)
2eB †
iℏvF ℏ
a 0

Our wave-function is given in the sublattice basis by

 
ϕA
ψ= . (B.15)
ϕB

We can write the individual components of ψ as Fock states of a simple harmonic

oscillator. We define ϕB = β |N ⟩, and ϕA = α |M ⟩. Therefore, applying our Hamil-
tonian to ψ, we obtain
√
2eB
−iℏvF aβ |N ⟩ = EN,s α |M ⟩ , (B.16)
ℏ

and
√
2eB †
iℏvF a α |M ⟩ = EN,s β |N ⟩ , (B.17)
ℏ

which implies that M = N − 1, and

s
β= α. (B.18)
−i

182
Employing the normalization condition ⟨N | N ⟩ = 1, our wave-function at K becomes

 −is 
√ |N − 1⟩
 2 
 ⟩  
N̄ = , (B.19)
K  
 1 
√ |N ⟩
2
 ⟩
where N̄ denotes a Fock state of the 2 × 2 harmonic oscillator Hamiltonian with
eigenenergy EN,s . Implicitly, we have assumed that N ̸= 0. For that case, we recall
that the N = 0 level at K was due entirely to electrons from the B sublattice. Thus,
 
0
|0̄⟩K =  . (B.20)
|0⟩

Following the same procedure for K ′ , we obtain the wave-functions:

 
 ⟩ AN,s |N − 1⟩
N̄ = , (B.21)
K
BN,s |N ⟩

and
 
 ⟩ AN,s |N ⟩
N̄ ′ =  , (B.22)
K
BN,s |N − 1⟩

where

 −is

 √ , N = 1, 2, 3, ...

 2
AN,s = , (B.23)



 1 − ξ,

N =0
2

183
and

 1

 √ , N = 1, 2, 3, ...

 2
BN,s = . (B.24)




 1 + ξ, N =0
2

It is important to note that in a sum over s, the N = 0 level will only contribute
once.

B.2 Quantum Hall Effect in Graphene

We are now in a position to derive the famous IQHE for relativistic fermions. For
concreteness, we will continue to use the model of graphene in which two valleys and
spin species need to be included. For zero frequency and momentum transfer, the
one-loop Kubo formula for a single spin can be written
⟨ ⟩⟨ ⟩
′ ′
ℏeB ∑∞ ∑ fN,s − fM,s′ N̄ s|ĵα |M̄ s M̄ s |ĵβ |N̄ s
σαβ = −i , (B.25)
h N,M =0 s,s′ ,ξ=± EN,s − EM,s′ EM,s′ − EN,s + iℏ/(2τ )

where ℏ/(2τ ) = Γ is a transport scattering rate, fN,s is the Fermi distribution function,
and ĵα = −evF σ̂α , where α = x, y [see Appendix A.2]. For the Hall conductivity,
α = x and β = y. We will use the relation

1 1 1 − tanh(x/2)
= = , (B.26)
ex +1 1 + tanh(x/2) 2
1+
1 − tanh(x/2)

so
( ) ( )
1 β(EN,s − µ) 1 β(EM,s′ − µ)
fN,s − fM,s′ = − tanh + tanh , (B.27)
2 2 2 2

184
where β = 1/(kB T ). Therefore, without loss of generality, we can define
( )
1 β(EN,s − µ)
fN,s → f¯N,s = − tanh . (B.28)
2 2

Using our wave-functions, we obtain

ℏeB 2 2 ∑ ∑ f¯N,s − f¯M,s′ EM,s′ − EN,s − iΓ

∞
σxy = e vF
h N,M =0 s,s′ =±
EN,s − EM,s′ (EM,s′ − EN,s )2 + Γ2
{
× |AK N,s | |BM,s′ | δM,N −1 − |AM,s′ | |BN,s | δM,N +1
2 K 2 K 2 K 2

}
K′ 2 K′ 2 K′ 2 K′ 2
−|AN,s | |BM,s′ | δM,N +1 + |AM,s′ | |BN,s | δM,N −1 , (B.29)

where we have explicitly shown the valley contributions. We will only concern our-
selves with the clean limit Γ = 0, and the observable (real) part of the conductivity.
Therefore, we are tasked with evaluating

ℏeB 2 2 ∑ ∑ f¯N,s − f¯M,s′

∞
σxy =− e vF (B.30)
h N,M =0 s,s′ =±
(EN,s − EM,s′ )2
{
× |AK N,s | |BM,s′ | δM,N −1 − |AM,s′ | |BN,s | δM,N +1
2 K 2 K 2 K 2

}
K′ 2 K′ 2 K′ 2 K′ 2
−|AN,s | |BM,s′ | δM,N +1 + |AM,s′ | |BN,s | δM,N −1 ,

where AN,s and BN,s are given by Eqns. (B.23) and (B.24), respectively. After some
straightforward and tedious algebra (and duly noting that the N = 0 level only
contributes for a single value of s and s′ ), we arrive at the expression
{ ∞ [ ( ) ( )]
2e2 ∑ β(µ − EN,+ ) β(µ − EN,− )
σxy = tanh + tanh
h 2 2
N =1
( )}
β(µ − E0 )
+tanh , (B.31)
2

185
where we have restored the spin degeneracy of two. At zero temperature, this becomes
 

 

2e2 ∑
∞
σxy = sgn(µ − E0 ) + sgn(µ − EN,s ) , (B.32)
h  N =1


s=±

where sgn(x) returns the sign of the argument (sgn(0) = 0). This has the form

e2
σxy = ν, (B.33)
h

where ν = 4(sN + 1/2) is the filling factor with s and N indexing the highest filled
LL. The filling factor is an index which counts how many LLs are occupied. A
positive (negative) filling factor corresponds to the chemical potential in the particle
(antiparticle) region of the dispersion. Consider positive µ: as each N ̸= 0 LL
is crossed, the filling factor increments by 4 due to the two-fold valley and spin
degeneracies. For N = 0, recall that the level is particle-hole like and thus only a
two-fold degeneracy is included. For the 2DEG, ν = 2N with N = 1, 2, 3, ..., and is
zero for all energies below the first level. Thus, the Hall conductivity is characterized
by positive-even integer multiples of e2 /h. For graphene, the negative energy states
allow for a negative Hall effect and the filling factors ν = ±2, ±6, ±10, .... Unlike the
2DEG, where the lowest LL occurs at ℏωc , graphene has a level pinned at E = 0 and,
thus, a finite Hall conductivity persists to µ → 0. A plot of Eqn. (B.31) as a function
√ √
of µ for B = 0.1T, ℏevF2 = 25.64 meV/ T, and kB T = 0.1 meV (solid black) and
0.5 meV (dashed blue) is shown in Fig. B.2. The steps in the conductivity occur at
the individual LL energies. Finite temperature broadens the LLs; therefore, as the
temperature is increased, LL mixing makes the quantization of the Hall conductivity
difficult to observe.

186
 16 B=0.1T ν=14
kBT=0.1 meV
kBT=0.5 meV ν=10
σxy(µ) (e /h)

8 ν=6
2

ν=2
0 ν=-2

ν=-6
-8

-16 -8 0 8 16 24
µ (meV)

Figure B.2: The IQHE in graphene.

187
 APPENDIX C

ANALYTIC SOLUTION FOR THE

OPTICAL CONDUCTIVITY OF
SILICENE

C.1 Longitudinal Conductivity

To calculate the analytic solution for the real part of the longitudinal conductivity,
recall Eqn. (4.14):

∫
ξσ e2 |µ| dω
Reσxx (Ω) = (C.1)
2Ω |µ|−Ω 2π
∫ [ ]
d2 k 2 ξσ
× ξσ ξσ ξσ
v A22 (k, ω + Ω)A11 (k, ω) + A11 (k, ω + Ω)A22 (k, ω) ,
(2π)2

where Aξσ ξσ
11 (k, ω) and A22 (k, ω) are given by Eqns. (4.8) and (4.11), respectively. We
√
will use the definition ε ≡ ℏ2 v 2 k 2 + ∆2ξσ . In what follows, the ξσ subscripts on ∆ξσ
shall be implicit. Note that

dε2 dε
= 2ε = 2(ℏv)2 k, (C.2)
dk dk

and
∫ ∫ 2π ∫ kc
2
dk= dθ kdk, (C.3)
0 0

188
where kc is a “high” energy band cutoff. Restoring the unit of ℏ associated with dω,
we can recast Eqn. (C.1) as

∫ |µ|
ξσ e2
Reσxx (Ω) = 2 dω (C.4)
8π ℏΩ |µ|−Ω
∫ ε(kc ) [ ]
× ξσ ξσ ξσ ξσ
ε A22 (ε, ω + Ω)A11 (ε, ω) + A11 (ε, ω + Ω)A22 (ε, ω) dε.
|∆|

Multiplying out the spectral functions given by Eqns. (4.8) and (4.11) and defining
σ0 = e2 /(4ℏ), we obtain

{ } ∫
ξσ
σxx (Ω) 1 |µ|
Re = dω (C.5)
σ0 Ω |µ|−Ω
∫ ε(kc ) {[ ]
∆2
× εdε 1 − 2 [δ(ω + Ω − ε)δ(ω − ε) + δ(ω + Ω + ε)δ(ω + ε)]
|∆| ε
[ ] }
∆2
+ 1 + 2 [δ(ω + Ω − ε)δ(ω + ε) + δ(ω + Ω + ε)δ(ω − ε)] .
ε

The first term we recognize as the intraband piece and the second term as the inter-
band contribution.
First, consider the intraband term:
{ ξσ
}
σxx (Ω)
Re ≡ σintra = (C.6)
σ0 intra
∫ ∫ ε(kc ) [ ]
1 |µ| ∆2
dω εdε 1 − 2 [δ(ω + Ω − ε)δ(ω − ε) + δ(ω + Ω + ε)δ(ω + ε)] .
Ω |µ|−Ω |∆| ε

The δ-function products can be rewritten to give

189
 ∫ |µ| ∫ ]
ε(kc ) [
1 ∆2
σintra = lim δ(Ω) dω εdε 1 − 2 [δ(ω − ε) + δ(ω + ε)]
Ω→0 Ω |µ|−Ω |∆| ε
∫ |µ| { [ ]
1 ∆2
= lim δ(Ω) dω ω 1 − 2 Θ (ω − |∆|)
Ω→0 Ω |µ|−Ω ω
[ ] }
∆2
−ω 1 − 2 Θ (−ω − |∆|)
ω
∫ |µ| [ ]
1 ∆2
= lim δ(Ω) dω |ω| − Θ (|ω| − |∆|)
Ω→0 Ω |µ|−Ω |ω|
[ 2 ] |µ|
1 ω 
= lim δ(Ω) − ∆ ln|ω| Θ (|µ| − |∆|)
2
Ω→0 Ω 2 |µ|−Ω
[ 2 ( )]
1 µ − (|µ| − Ω) 2
|µ| − Ω
= lim δ(Ω) 2
+ ∆ ln Θ (|µ| − |∆|) . (C.7)
Ω→0 Ω 2 |µ|

As we are taking the limit of Ω → 0, we Taylor expand the log term about Ω/|µ| = 0
to obtain
[ ]
−Ω + 2µ ∆2
σintra = lim δ(Ω) − Θ (|µ| − |∆|) . (C.8)
Ω→0 2 |µ| − Ω

We can now take the limit of Ω → 0 as all Ω terms are well defined. This gives

µ2 − ∆2
σintra = δ(Ω) Θ (|µ| − |∆|) . (C.9)
|µ|

So,
{ }
ξσ
σxx (Ω) µ2 − ∆2ξσ
Re = δ(Ω)Θ (|µ| − |∆ξσ |) . (C.10)
σ0 intra |µ|

Next, let us return to Eqn. (C.5) and calculate the interband contribution to the
conductivity:
{ } ∫
ξσ
σxx (Ω) 1 |µ|
Re ≡ σinter = dω
σ0 inter Ω |µ|−Ω
∫ ε(kc ) [ ]
∆2
× εdε 1 + 2 [δ(ω + Ω − ε)δ(ω + ε) + δ(ω + Ω + ε)δ(ω − ε)] . (C.11)
|∆| ε

190
Conducting the integral over ε, we obtain
∫ |µ| { [ ]
1 ∆2
σinter = dω ω 1 + 2 [−δ(2ω + Ω)Θ (−ω − |∆|)
Ω |µ|−Ω ω
+δ(2ω + Ω)Θ (ω − |∆|)]}
{[ ][ ( ) ( )
1 Ω 4∆2 Ω Ω
= 1+ 2 Θ (Ω − 2|∆|) Θ − − |µ| + Ω Θ |µ| +
Ω 2 Ω 2 2
( ) ( )]}
Ω Ω
−Θ (−Ω − 2|∆|) Θ − − |µ| + Ω Θ |µ| + . (C.12)
2 2

Using the fact that δ(2ω + Ω) = (1/2)δ(ω + Ω/2), and restricting our attention to
positive frequency Ω, we find
{ [ ] ( ) ( )}
1 Ω 4∆2 Ω Ω
σinter = 1 + 2 Θ (Ω − 2|∆|) Θ − |µ| Θ |µ| +
Ω 4 Ω 2 2
[ 2
] ( )
1 4∆ Ω
= 1 + 2 Θ (Ω − 2|∆|) Θ − |µ| . (C.13)
2 Ω 2

Thus, we can write

{ } [ ]
ξσ
σxx (Ω) 1 4∆2ξσ
Re = 1+ Θ (Ω − Ωc ) , (C.14)
σ0 inter 4 Ω2

where Ωc = max(2|∆ξσ |, 2|µ|). Collecting Eqns. (C.10) and (C.14), we obtain the
analytic result for the real part of the longitudinal optical conductivity for positive
frequency:

{ } [ ( )2 ]
ξσ
σxx (Ω) µ2 − ∆2ξσ 1 2∆ξσ
Re = δ(Ω)Θ (|µ| − |∆ξσ |) + 1+ Θ (Ω − Ωc ) .
σ0 |µ| 4 Ω
(C.15)

191
C.2 Transverse Hall Conductivity
ξσ
The imaginary part of σxy (Ω) is found by solving

∫ |µ|
ξσ ξe2
Imσxy (Ω) = (C.16)
2Ω |µ|−Ω
∫
dω d2 k 2 [ ξσ ]
× v A22 (k, ω + Ω)Aξσ
11 (k, ω) − Aξσ
11 (k, ω + Ω)Aξσ
22 (k, ω) ,
2π (2π)2

where, again, Aξσ ξσ

11 (k, ω) and A22 (k, ω) are give by Eqns. (4.8) and (4.11), respectively,

and we imply the ξσ subscripts on ∆ξσ throughout the derivation. As in Sec. C.1, we
work in the continuum limit and, after multiplying out the spectral functions, obtain
{ } ∫
ξσ
σxy (Ω) ξ |µ|
Im = dω
σ0 Ω |µ|−Ω
∫ ε(kc ) ( )
2∆
× εdε [δ(ω + Ω + ε)δ(ω − ε) − δ(ω + Ω − ε)δ(ω + ε)] . (C.17)
|∆| ε

Conducting the ε integral,

{ } ∫
ξσ
σxy (Ω) |µ|
2ξ∆
Im = dω [δ(2ω + Ω)Θ (ω − |∆|)
σ0 Ω |µ|−Ω

−δ(2ω + Ω)Θ (−ω − |∆|)] . (C.18)

Using δ(2ω + Ω) = (1/2)δ(ω + Ω/2),

{ } ∫ [ ( )
ξσ
σxy (Ω) |µ|
ξ∆ Ω
Im = dω δ ω + Θ (ω − |∆|)
σ0 Ω |µ|−Ω 2
( ) ]
Ω
−δ ω + Θ (−ω − |∆|) . (C.19)
2

192
For the ω integral, we note that ω = −Ω/2 and require |µ| − Ω < ω < |µ|. Therefore,
{ } [ ( ) ( ) ( )
ξσ
σxy (Ω) ξ∆ Ω Ω Ω
Im = Θ − − |∆| Θ − − (µ − Ω) Θ µ +
σ0 Ω 2 2 2
( ) ( ) ( )]
Ω Ω Ω
−Θ − |∆| Θ − − (µ − Ω) Θ µ +
2 2 2
ξ∆
= [Θ (−Ω − 2|∆|) Θ (Ω − 2µ) Θ (Ω + 2µ)
Ω
−Θ (Ω − 2|∆|) Θ (Ω − 2µ) Θ (Ω + 2µ)] . (C.20)

Again, we restrict our attention to Ω > 0, so

{ }
ξσ
σxy (Ω) ξ∆
Im =− Θ (Ω − 2|∆|) Θ (Ω − 2µ) Θ (Ω + 2µ)
σ0 Ω
ξ∆
=− Θ (Ω − Ωc ) . (C.21)
Ω

Thus,
{ }
ξσ
σxy (Ω) ∆ξσ
Im = −ξ Θ (Ω − Ωc ) , (C.22)
σ0 Ω

where Ωc = max(2|∆ξσ |, 2|µ|).

193
 APPENDIX D

LANDAU LEVEL DISPERSION OF

SILICENE

D.1 Energy Dispersion

In a magnetic field, the low-energy physics of silicene for a single spin and valley is
well described by the Hamiltonian of Eqn. (6.6). Zeeman interactions can be included
by adding a term −σZ I,
ˆ where Z = gs µB B/2 corresponds to the strength of the

interaction and Iˆ is the identity matrix. Therefore, our Hamiltonian becomes

 ( [ ]) 
∆ξσ − σZ ℏv ξ kx − eB
− iky
y
Ĥξσ =  [ ]
ℏ . (D.1)
ℏv(ξ kx − eB
ℏ
y + iky ) −∆ξσ − σZ

The corresponding wave-functions are written in the A-B sublattice basis (i.e. Ψ =
(ϕA , ϕB )T ). To proceed, let us define the two ladder operators
√ ([ ] )
2ℏv eB
−i a = ℏv kx − y − iky , (D.2)
lB ℏ

and
√ ([ ] )
2ℏv † eB
i a = ℏv kx − y + iky , (D.3)
lB ℏ
√
where lB = ℏ/(eB). For the N th Fock state of the simple harmonic oscillator,
√ √
a|N ⟩ = N |N − 1⟩ and a† |N ⟩ = N + 1|N + 1⟩. Our Hamiltonian at the K point

194
is now
 √ 
∆Kσ − σZ −i 2ℏv
a
ĤKσ =  √
lB , (D.4)
2ℏv †
i lB
a −∆Kσ − σZ

and at the K ′ point

 √ 
∆K ′ σ − σZ −i l2ℏv a†
ĤK ′ σ =  √
B . (D.5)
i 2ℏv
lB
a −∆ K′σ − σZ

Therefore, the eigenvalue equation (ĤΨ = EΨ) at the K point yields the two coupled
equations
√
2ℏv
(∆Kσ − σZ)ϕA − i aϕB = EϕA , (D.6)
lB

and
√
2ℏv †
(−∆Kσ − σZ)ϕB + i a ϕA = EϕB . (D.7)
lB

Equation (D.6) gives

√
−i 2ℏv/lB
ϕA = aϕB , (D.8)
E − ∆Kσ + σZ

which can be substituted into Eqn. (D.7) to give

2ℏ2 v 2 /lB
2
a† aϕB + (−∆Kσ − σZ)ϕB = EϕB . (D.9)
E − ∆Kσ + σZ

But a† a + 1/2 has the form of the simple harmonic oscillator Hamiltonian. This has
eigenvalues which go like N + 1/2 for an eigenstate ϕB ∝ |N ⟩. Therefore, we have

2eBℏv 2
N ϕB = (E + ∆Kσ + σZ)ϕB . (D.10)
E − ∆Kσ + σZ

195
Solving for E, we obtain

√
EN,s
Kσ
= −σZ + s 2v 2 ℏeBN + ∆2Kσ , (D.11)

where s = ± is a band index.

Implicit in our derivation was the assumption that N ̸= 0. At the K point, if
N = 0, a† ϕA = 0 since ϕA ∝ |N − 1⟩ as it is related to ϕB ∝ |N ⟩ by the operation
ϕA ∝ aϕB [see Eqn. (D.8)]. Therefore, Eqn. (D.7) gives

(−∆Kσ − σZ)ϕB = EϕB . (D.12)

So,

E0Kσ = −∆Kσ − σZ. (D.13)

At K ′ , ĤΨ = EΨ yields
√
2ℏv †
(∆K ′ σ − σZ)ϕA − i a ϕB = EϕA , (D.14)
lB

and
√
2ℏv
(−∆K ′ σ − σZ)ϕB + i aϕA = EϕB . (D.15)
lB

Equation (D.15) gives

√
i 2ℏv/lB
ϕB = aϕA , (D.16)
E + ∆K ′ σ + σZ

which we substitute into Eqn. (D.14) to get

2ℏ2 v 2 /lB
2
a† aϕA = (E − ∆K ′ σ + σZ)ϕB . (D.17)
E + ∆K ′ σ + σZ

196
We now use a† aϕA = N ϕA for ϕA ∝ |N ⟩. Therefore, solving for E, we obtain

√
′
EN,s
K σ
= −σZ + s 2v 2 ℏeBN + ∆2K ′ σ , (D.18)

where s = ± is a band index. Again, we assumed N ̸= 0. In that case, a† ϕB = 0 as

it is related to ϕA ∝ |N ⟩ by ϕB ∝ aϕA [see Eqn. (D.16)]. Thus, Eqn. (D.14) gives

′
E0K σ = ∆K ′ σ − σZ. (D.19)

Consolidating our results for both valleys, we obtain the low-energy LL dispersion
of silicene:
 √
 − 1 gs µB Bσ + s ∆2 + 2N E 2 , N = 1, 2, 3, ...
2 ξσ 1
EN,s
ξσ
= , (D.20)
 −ξ∆ξσ − 12 gs µB Bσ, N =0

√
where E1 ≡ ℏev 2 B and in a sum over s, the N = 0 level is only counted once.

D.2 Wave-Functions
To solve for the corresponding wave-functions at the K point, return to Eqns. (D.6)
and (D.7). For simplicity, we take Z = 0. As we saw earlier, ϕA ∝ aϕB . By definition
we had ϕB ∝ |N ⟩ and so ϕA ∝ |N − 1⟩. Let us first consider N ̸= 0 and take our
wave-function to be
 
 ⟩ α |N − 1⟩
ΨK = N̄ K =  . (D.21)
β |N ⟩

Equations (D.6) and (D.7) give

√
2ℏv √
∆Kσ α |N − 1⟩ − i β N |N − 1⟩ = Eα |N − 1⟩ , (D.22)
lB

197
and
√
2ℏv √
−∆Kσ β |N ⟩ + i N α |N ⟩ = Eβ |N ⟩ , (D.23)
lB
√ √
where we used a|N ⟩ = N |N − 1⟩ and a† |N ⟩ = N + 1|N + 1⟩. From Eqn. (D.23),
we see that

E + ∆Kσ
α= √ √ β. (D.24)
i( 2ℏv/lB ) N

We can substitute this into our wave-function and normalize to get

 
 ⟩ −iAN,s |N − 1⟩
N̄ = , (D.25)
K
BN,s |N ⟩

where √
s |EN,s
Kσ
| + s∆Kσ
AK
N,s = √ , (D.26)
2|EN,s |
Kσ

and √
|EN,s
Kσ
| − s∆Kσ
BN,s
K
= √ . (D.27)
Kσ
2|EN,s |

For N = 0, AK
0 = 0 and B0 = 1 as ϕA (N = 0) = 0.
K

At the other valley, our wave-function has the form

 
 ⟩ α |N ⟩
ΨK ′ = N̄ K ′ =  . (D.28)
β |N − 1⟩

Equations (D.14) and (D.15) then become

√
2ℏv √
∆K ′ σ α |N ⟩ − i β N |N ⟩ = Eα |N ⟩ , (D.29)
lB

198
and
√
2ℏv √
−∆K ′ σ β |N − 1⟩ + i α N |N − 1⟩ = Eβ |N − 1⟩ , (D.30)
lB
√ √
where again we used the relations a|N ⟩ = N |N − 1⟩ and a† |N ⟩ = N + 1|N + 1⟩.
Equation (D.30), gives

E + ∆K ′ σ
α= √ √ β. (D.31)
i( 2ℏv/lB ) N

Substituting this into our wave-function and normalizing, we find

 
 ⟩ −iAN,s |N ⟩
N̄ ′ =  , (D.32)
K
BN,s |N − 1⟩

where
′ ′
AK
N,s = AN,s
K→K
, (D.33)

and
′ ′
BN,s
K
= BN,s
K→K
. (D.34)

′ ′
For N = 0, AK
0 = 1 and B0 = 0 as ϕB (N = 0) = 0.
K

Collecting our results, we arrive at the final expression for our wave-functions:
 
 ⟩ −iAN,s |N − 1⟩
N̄ = , (D.35)
K
BN,s |N ⟩

and  
 ⟩ −iAN,s |N ⟩
N̄ ′ =  , (D.36)
K
BN,s |N − 1⟩

199
respectively, where [32, 33]
 √

 s |EN,s
ξσ
| + s∆ξσ


 √ , N ̸= 0
AN,s = ξσ
2|EN,s | , (D.37)



 1−ξ
 , N =0
2
 √

 |EN,s
ξσ
| − s∆ξσ


 √ , N ̸= 0
BN,s = 2|EN,s |
ξσ
, (D.38)



 1+ξ
 , N =0
2
and
1 1
∆ξσ = − ξσ∆so + ∆z . (D.39)
2 2

200
 APPENDIX E

TOPOLOGICAL INSULATORS IN A
MAGNETIC FIELD

E.1 Surface States in a Magnetic Field

E.1.1 Landau levels

The surface-state Hamiltonian of a topological insulator in the presence of a small

gap ∆ is

ℏ2 k 2
Ĥ = + ℏvF (kx σ̂y − ky σ̂x ) + (∆ − Z) σ̂z , (E.1)
2m

where Z = gµB B/2 is the strength of the Zeeman interaction in a magnetic field B.
For convenience, we define

1
β ≡ ∆ − gµB B
2
E0
=∆−g . (E.2)
2

As usual, the effect of a finite magnetic field is included by defining the magnetic
vector potential A and employing the Peierls substitution ℏk → ℏk +eA/ℏ. Working
in the Landau gauge, A = (0, Bx, 0). Therefore, our Hamiltonian becomes
[ ( )2 ] [ ( ) ]
ℏ2 eBx eBx
Ĥ = 2
k + ky + + ℏvF kx σ̂y − ky + σ̂x + β σ̂z . (E.3)
2m x ℏ ℏ

201
To solve Eqn. (E.3), we will first break the Hamiltonian into parts. Let us consider
the Schrödinger term:
[ ( )2 ]
ℏ2 2eB eBx
ĤS = kx2 + ky2 + ky x + . (E.4)
2m ℏ ℏ

We define the ladder operators

[ ]
† lB x + x0
a ≡ √ −ikx + 2
, (E.5)
2 lB

and
[ ]
lB x + x0
a ≡ √ ikx + 2
, (E.6)
2 lB
√
where lB ≡ ℏ/(eB) and x0 ≡ ky lB
2
. On a Fock state (|N ⟩) of the simple Harmonic
oscillator Hamiltonian, these operators have the property:

√
a |N ⟩ = N |N − 1⟩ , (E.7)

and

√
a† |N ⟩ = N + 1 |N + 1⟩ . (E.8)

We note that the Hamiltonian can be recast as

[ ( )2 ]
ℏ2 x 2
x 0 x x
ĤS = kx2 + 40 + 2 2 2 + 2
2m lB lB lB lB
[ ]
ℏ2 1
= kx2 + 4 (x + x0 )2 . (E.9)
2m lB

202
By the definition of the ladder operators,

2
( )
† lB (x + x0 )2 i
aa= 2
kx + 2
− 2 [x, kx ]
2 lB lB
2
( )
lB (x + x0 )2 1
= 2
kx + 2
− 2 , (E.10)
2 lB lB

where we use the familiar commutators [x, px ] = iℏ and [y, px ] = 0. This gives
[ ]
ℏ2 † 1
ĤS = 2
a a+ . (E.11)
mlB 2

Next, consider the relativistic term:

[ ( ) ]
eBx
ĤD = ℏvF kx σ̂y − ky + σ̂x
ℏ
 x0 + x 
0 ikx +
 2
lB 
= −ℏvF  x0 + x 
−ikx + 0
l2
 B
√
2 0 a 
= −ℏvF . (E.12)
lB a† 0

Using these relations, the total Hamiltonian can be written as

 [ ] √ 
ℏ2 † 1 2
 ml2 a a + 2 + β −ℏvF a 
 B √ [ lB ] .
Ĥ =   (E.13)
2 † ℏ2 1
−ℏvF a 2
a† a + −β
lB mlB 2

For convenience, we recast this in terms of E0 ≡ ℏ2 /(mlB

2
) and E1 ≡ ℏvF /lB .
To solve the corresponding energy dispersion, recall that Eqn. (E.13) is written in
the spin basis. Our wave-functions will be of the form
 
ϕ↑
ψ= . (E.14)
ϕ↓

203
Applying the usual relation Ĥψ = Eψ, we arrive at the two equations
( [ ] )
† 1 √
E0 a a + + β ϕ↑ − 2E1 aϕ↓ = Eϕ↑ , (E.15)
2

and
( [ ] )
† 1 √
E0 a a + − β ϕ↓ − 2E1 a† ϕ↑ = Eϕ↓ . (E.16)
2

It is convenient to note that ℏω(a† a + 1/2) is the Hamiltonian for the simple har-
monic oscillator which has eigenvalues ℏω(N + 1/2) for a Fock state |N ⟩. Therefore,
Eqn. (E.16) can be written as
( [ ] )
1 √
E − E0 N + + β ϕ↓ = − 2E1 a† ϕ↑ , (E.17)
2

for ϕ↓ ∝ |N ⟩ and thus ϕ↑ ∝ |N − 1⟩ (which is related to ϕ↓ by the creation operation

a† ϕ↑ ). Using this result in Eqn. (E.15), we find

( [ ] ) ( √ )
1 √ − 2E1
E0 N − 1 + + β ϕ↑ − 2E1 aa† ϕ↑ = Eϕ↑ . (E.18)
2 E − E0 [N + 1/2] + β

By Eqns. (E.7) and (E.8), aa† ϕ↑ = N ϕ↑ for ϕ↑ ∝ |N − 1⟩. Therefore,

( [ ] ) ( )
1 2N E12
E0 N − 1 + +β + − E = 0. (E.19)
2 E − E0 [N + 1/2] + β

This is easily solved to give the LL dispersion

√ [ ]2
E0
E = E0 N + s 2N E12 + ∆− (1 + g) , (E.20)
2

where N > 0 indexes the Fock state. Implicit in our derivation was the assumption
that N ̸= 0. If N = 0, ϕ↑ ∝ |N − 1⟩ would need to be zero as |−1⟩ is not defined.

204
Therefore, Eqn. (E.16) gives
( [ ] )
1
E0 N + − β ϕ↓ = Eϕ↓ , (E.21)
2

but N = 0, so

E0
E= − β. (E.22)
2

We now have an expression for the LL dispersion. It is

 √ [ ]2

 E0
 E0 N + s 2N E12 + ∆− (1 + g) , N = 1, 2, 3, ...
EN,s = 2 . (E.23)

 E0
 (1 + g) − ∆, N =0
2

Here, it is important to note that in a sum over s, the N = 0 level will only contribute
once.

E.1.2 Wave-Functions

It is also instructive to look at the wave-functions. Now that we know the form of the
spinor components, let us take ϕ↑ = α |N − 1⟩ and ϕ↓ = γ |N ⟩ and consider N ̸= 0.
Equation (E.17) becomes
( [ ] )
1 √ √
E − E0 N + + β γ |N ⟩ = − 2E1 N α |N ⟩ . (E.24)
2

Therefore,
[ ]
E − E0 N + 12 + β
α= √ √ γ. (E.25)
− 2E1 N

Imposing the normalization condition ψ 2 = α2 + γ 2 = 1, we obtain

( [ ] )2 
E − E0 N + 2 + β
1
 √ √ + 1 γ 2 = 1. (E.26)
− 2E1 N

205
Using Eqn. (E.23), the argument in the brackets can be written as

(E − E0 N + β − E0 /2)2 2(E − E0 N )
+1= , (E.27)
2
2N E1 E − E0 N − β + E0 /2

which implies
√
E − E0 N − β + E0 /2
γ=
2(E − E0 N )
√
EN,+ − E0 N − s(β − E0 /2)
= , (E.28)
2(EN,+ − E0 N )

and thus
√
E − E0 (N + 1/2) + β EN,+ − E0 N − s(β − E0 /2)
α= √ √
− 2E1 N 2(EN,+ − E0 N )
√
EN,+ − E0 N + s(β − E0 /2) EN,+ − E0 N − s(β − E0 /2)
= −s √
2N E1 2(EN,+ − E0 N )
√
EN,+ − E0 N + s(β − E0 /2)
= −s . (E.29)
2(EN,+ − E0 N )

For N = 0, recall that ϕ↑ = 0 =⇒ ϕ↓ = |0⟩. Combining these results, we obtain the

wave-functions
 
↑
CN,s |N − 1⟩
|ψ⟩ =  , (E.30)
↓
CN,s |N ⟩

where
 √

 EN,+ − E0 N + s(β − E0 /2)
 −s , N = 1, 2, 3, ...
↑
CN,s = 2(EN,+ − E0 N ) , (E.31)


 0, N =0

206
  √

 EN,+ − E0 N − s(β − E0 /2)
 , N = 1, 2, 3, ...
↓
CN,s = 2(EN,+ − E0 N ) , (E.32)


 1, N =0

and β = ∆ − gE0 /2. As our wave-function spinor is written in real-spin space, it

is interesting to observe that for N = 0, the amplitude is entirely associated with
↑
spin-down (i.e. CN,s = 0).

E.2 Analytic Solution for Magnetic Oscillations

For positive ω, the oscillating part of the density of states is

eB d
k
Nosc (ω) = Θ(ω − |E0,+ |)sin(2πkx1 ), (E.33)
πkh dω

where
√
E2 ω E2 2ωE0 E04 b
x1 = 12 + − 12 1+ + 4, (E.34)
E0 E0 E0 E12 E1

with

¯2
∆
b≡ , (E.35)
E02

and
[ ]2
1
∆ ≡ ∆ − E0 (1 + g) .
˜ 2
(E.36)
2

For E12 /E02 ≫ 1 and m → ∞,

ω2 [ ω ] bE0 ωbE0
x1 ≈ 2
1 − 2
− 2
+ , (E.37)
2E1 mv 2mv (mv 2 )2

207
where

bE0 E0 [∆ − E0 (1 + g)/2]2
− = −
2mv 2 2mv 2 E02
[ ( )2 ]
1 E 0
=− ∆2 − E0 ∆(1 + g) + (1 + g)2 . (E.38)
2mv 2 E0 2

The last term in the above expression goes like E0 ∝ B and will therefore drop out
in the limit B → 0 [36]. Thus

bE0 ∆2 ∆(1 + g)
− 2
= − 2
+ , (E.39)
2mvF 2E1 2mvF2

where the last term is constant in B and will therefore contribute a constant phase
to the quantum oscillations. We now have
[ ]
1 [ ω ][ 2 ] ∆(1 + g) ω
x1 ≈ 1− ω −∆ +
2
1− . (E.40)
2E12 mv 2 2mvF2 mvF2

To lowest order in 1/(mvF2 ),

[ ]
1 ω [ 2 ] ∆(1 + g)
x1 ≈ 2
1− 2
ω − ∆2 + . (E.41)
2E1 mvF 2mvF2

To compute the magnetization, return to the grand thermodynamic potential and

keep only the terms which will contribute to the oscillations. That is,
∫ µ
Ωkosc (µ) = (ω − µ)Nosc
k
(ω)dω. (E.42)
0

Using our density of states result,

∫ µ
eB d
Ωkosc (µ) = (ω − µ) Θ(ω − |E0,+ |)sin(2πkx1 )dω
πkh 0 dω
[ µ ∫ µ ]
eB 
= (ω − µ)Θ(ω − |E0,+ |)sin(2πkx1 ) − Θ(ω − |E0,+ |)sin(2πkx1 )dω
πkh 0
∫ µ 0
eB
=− Θ(ω − |E0,+ |)sin(2πkx1 )dω. (E.43)
πkh 0

208
The magnetization is given by M = −∂Ω/∂B; therefore, we need
[ ]
∂x1 1 ω [ 2 ]
=− 2 1− 2
ω − ∆2 . (E.44)
∂B 2E1 B mvF

The magnetization is then given by

∫ [ [ ] ]
e µ
πk ω [ 2 ]
k
Mosc (µ) = sin(2πkx1 ) − 2 1 − ω − ∆ cos(2πkx1 ) dω.
2
πkh |∆|
¯ E1 mvF2
(E.45)

In the limit of interest (E0 → 0), ∆

¯ → ∆. To solve the integral, define

[ ]
ω [ 2 ]
y ≡ 1− 2
ω − ∆2
mvF
[ ]
3ω 2 ∆2
=⇒ dy = 2ω − + dω
mvF2 mvF2
dy
=⇒ dω = . (E.46)
3ω 2 ∆2
2ω − +
mvF2 mvF2

For m → ∞, y ≈ ω 2 − ∆2 so, to a first order correction in 1/m,

[ √ ]
y + ∆2 [ 2 ]
y ≈ 1− 2
ω − ∆2 . (E.47)
mvF

Upon further expansion, we obtain

v [ ]
u √
u y + ∆ 2
ω ≈ ty 1 + + ∆2 , (E.48)
mvF2

and
[ √ ]
y + ∆ 2
ω2 ≈ y 1 + 2
+ ∆2 . (E.49)
mvF

209
Therefore,
[ √ ]
1 1 y + ∆2
≈ √ 1+ . (E.50)
3ω 2 ∆2 2 y + ∆2 mvF2
2ω − +
mvF2 mvF2

The magnetization is thus,

∫ α { ( [ ]) ( [ ])}
e y πky y
k
Mosc (µ) ≈ sin 2πk +δ − 2 cos 2πk +δ
2πkh 2E12 E1 2E12
0
[ ]
1 1
× √ + dy. (E.51)
y + ∆2 mvF2

where
[ ]
µ [ 2 ]
α≡ 1− 2
µ − ∆2 , (E.52)
mvF

and

∆(1 + g)
δ≡ . (E.53)
2mvF2

Next, define y ≡ αx to obtain

∫ { }
e 1
α α { ( ) ( )}
k
Mosc (µ) ≈ dx √ + sin ax + δ̄ − axcos ax + δ̄ ,
2πkh 0 αx + ∆2 mvF2
(E.54)

where a ≡ πkα/E12 , δ̄ ≡ 2πkδ, and again we work to lowest order in 1/(mvF2 ). First,
consider
∫
{ (1 ) ( )} dx
I1 ≡ sin ax + δ̄ − axcos ax + δ̄ √
0 x + ∆2 /α
∫ 1 ∫ 1
( ) dx x d ( )
= sin ax + δ̄ √ − √ dx sin ax + δ̄ . (E.55)
0 x + ∆2 /α 0 x + ∆2 /α dx

Integrating the second term by parts and defining ω ≡ x + ∆2 /α and b ≡ δ̄ − a∆2 /α,

210
we obtain
∫ 1+∆2 /α [ ] ( )
3 ∆2 /α sin a + δ̄
I1 = sin (aω + b) √ + dω − √ . (E.56)
∆2 /α 2 ω 2ω 3/2 1 + ∆2 /α

To solve the remaining integral, we use the definition of the Fresnel sine and cosine
integrals:
∫ z ( )
1 2
S(z) ≡ sin πt dt, (E.57)
0 2

and
∫ z ( )
1 2
C(z) ≡ cos πt dt, (E.58)
0 2

respectively. We note that

sin (aω + b) = sin(aω)cosb + cos(aω)sinb. (E.59)

We have
∫ 1+∆2 /α
dω
sin (aω + b) √ (E.60)
∆2 /α ω
√ [ (√ ) √
 1+∆2 /α
(√ ) √
 1+∆2 /α
]
2π 2a  2a 
= S ω √ cosb + C ω √ sinb .
a π ∆2 /α π ∆2 /α

Once evaluated at the limits, we will have Fresnel functions with arguments pro-
√
portional to a. We are interested in the limit a → ∞. We use the asymptotic
expansions of the Fresnel integrals for a → ∞:
( )
( √ ) 1 1 1 2
S γ a ≈ − √ cos πγ a , (E.61)
2 πγ a 2

and
( )
( √ ) 1 1 1 2
C γ a ≈ + √ sin πγ a . (E.62)
2 πγ a 2

211
Therefore, Eqn. (E.60) is proportional to 1/a and is negligible in the limit of interest.
We now return to Eqn. (E.56) and consider

∫ 1+∆2 /α 1+∆2 /α
dω 2 
sin (aω + b) 3/2 = − √ sin(aω + b) (E.63)
∆2 /α ω ω ∆2 /α
[ (√ ) (√ ) ] √
√  1+∆2 /α
2a 2a 
− 2 2πa S ω sinb − C ω cosb √ .
π π ∆2 /α

Applying the Fresnel expansions for a → ∞, this term is zero. We arrive at the
simple result

sin(a + δ̄)
I1 ≈ − √ . (E.64)
1 + ∆2 /α

Next, we consider
∫ 1 {
( ) ( )}
I2 ≡ sin ax + δ̄ − axcos ax + δ̄ dx
∫0 1 ∫ 1
( ) d ( )
= sin ax + δ̄ dx − xdx sin ax + δ̄ . (E.65)
0 0 dx

Integrating the second term by parts, we obtain

2 ( ) 1 ( )
I2 = − cos ax + δ̄  − sin a + δ̄ . (E.66)
a 0

In the limit a → ∞,

I2 ≈ −sin(a + δ̄). (E.67)

Combing Eqns. (E.64) and (E.67), the magnetization becomes

[ ]
e α α
k
Mosc (µ) ≈− √ + sin(a + δ̄). (E.68)
2πkh α + ∆2 mvF2

212
Written in terms of the original variables,
[ ]
e (µ2 − ∆2 ) µ2 − ∆2
k
Mosc (µ) ≈− 1+ sin (2πkx1 ) , (E.69)
2πkh µ 2µmvF2

where
( )
µ [ 2 ]
1− 2
µ − ∆2
mvF ∆(1 + g)
x1 ≈ 2
+ . (E.70)
2E1 2mvF2

213
Bibliography

[1] K. von Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980).

[2] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010).

[3] Y. Ando, J. Phys. Soc. Jpn. 82, 102001 (2013).

[4] M. V. Berry, Proc. R. Soc. London, Ser. A 392, 45 (1984).

[5] B. A. Bernevig and T. L. Hughes, Topological Insulators and Topological Su-

perconductors (Princton University Press, New Jersey, 2013).

[6] N. W. Ashcroft and D. N. Mermin, Solid State Physics, 1 ed. (Thomson Learn-
ing, Toronto, 1976).

[7] N. A. Doughty, Lagrangian Interaction, 1 ed. (Westview Press, Boulder, 1990).

[8] B. I. Halperin, Phys. Rev. B 25, 2185 (1982).

[9] K. von Klitzing, Philos. Trans. R. Soc. London, Ser. A 363, 2203 (2005).

[10] D. J. Thouless, M. Kohomoto, M. P. Nightingale, and D. den Nijs, Phys. Rev.

Lett. 49, 405 (1982).

[11] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988).

[12] W. Ehrenberg and R. E. Siday, Proc. Phys. Soc. B 62, 8 (1949).

[13] Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959).

[14] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005).

[15] M. König et al., Science 318, 766 (2007).

[16] A. Roth et al., Science 325, 294 (2009).

214
[17] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005).

[18] J. E. Moore and L. Balents, Phys. Rev. B 75, 121306(R) (2007).

[19] L. Fu, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 98, 106803 (2007).

[20] R. Roy, Phys. Rev. B 76, 045302 (2009).

[21] A. A. Taskin and Y. Ando, Phys. Rev. B 80, 085303 (2009).

[22] P. Roushan et al., Nature 460, 1106 (2009).

[23] D. Hsieh et al., Science 323, 919 (2009).

[24] A. Nishide et al., Phys. Rev. B 81, 041309(R) (2010).

[25] B. Lenoir, M. Cassart, J.-P. Michenaud, H. Scherrer, and S. Scherrer, J. Phys.

Chem. Solids 57, 89 (1996).

[26] T. Hirahara et al., Phys. Rev. B 76, 153305 (2007).

[27] D. Hsieh et al., Nature 452, 970 (2008).

[28] H.-J. Zhang et al., Nature Phys. 5, 438 (2009).

[29] H. Nielssen and N. Ninomiya, Phys. Lett. 130B, 389 (1983).

[30] L. Stille, C. J. Tabert, and E. J. Nicol, Phys. Rev. B 86, 195405 (2012).

[31] C. J. Tabert and E. J. Nicol, Phys. Rev. B 87, 235426 (2013).

[32] C. J. Tabert and E. J. Nicol, Phys. Rev. Lett. 110, 197402 (2013).

[33] C. J. Tabert and E. J. Nicol, Phys. Rev. B 88, 085434 (2013).

[34] C. J. Tabert, J. P. Carbotte, and E. J. Nicol, Phys. Rev. B 91, 035423 (2015).

[35] C. J. Tabert and E. J. Nicol, Phys. Rev. B 89, 195410 (2014).

[36] C. J. Tabert and J. P. Carbotte, J. Phys.: Condens. Matter 27, 015008 (2015).

215
[37] C. J. Tabert and J. P. Carbotte, Phys. Rev. B 91, 235405 (2015).

[38] D. S. L. Abergel, V. Apalkov, J. Berashevich, K. Ziegler, and T. Chakraborty,

Adv. in Phys. 59, 261 (2010).

[39] N. D. Drummond, V. Zólyomi, and V. I. Fal’ko, Phys. Rev. B 85, 075423

(2012).

[40] B. Lalmi et al., Appl. Phys. Lett. 97, 223109 (2010).

[41] P. De Padova et al., Appl. Phys. Lett. 96, 261905 (2010).

[42] M. Ezawa, Phys. Rev. Lett. 109, 055502 (2012).

[43] G. W. Semenoff, Phys. Rev. Lett. 53, 2449 (1984).

[44] T. J. Bernatowicz et al., Astrophys. J. 472, 760 (1996).

[45] C.-C. Liu, W. Feng, and Y. Yao, Phys. Rev. Lett. 107, 076802 (2011).

[46] C.-C. Liu, H. Jiang, and Y. Yao, Phys. Rev. B 84, 195430 (2011).

[47] E. Cinquanta et al., J. Phys. Chem. C 117, 16719 (2013).

[48] Z. Ni et al., Nano Lett. 12, 113 (2012).

[49] [Crystal structure plotted using VESTA], K. Momma, and F. Izumi, J. Appl.
Crystallogr. 44, 1272 (2011).

[50] P. De Padova, C. Quaresima, B. Olivieri, P. Perfetti, and G. Le Lay, Appl.

Phys. Lett. 98, 081909 (2011).

[51] P. Vogt et al., Phys. Rev. Lett. 108, 155501 (2012).

[52] C.-L. Lin et al., Appl. Phys. Express 5, 045802 (2012).

[53] A. Fleurence et al., Phys. Rev. Lett. 108, 245501 (2012).

[54] L. Tao et al., Nature Nano. 10, 227 (2015).

216
[55] S. Konschuh, M. Gmitra, and J. Fabian, Phys. Rev. B 82, 245412 (2010).

[56] M. Ezawa, New J. Phys. 14, 033003 (2012).

[57] K. Wakabayashi, M. Fujita, H. Ajiki, and M. Sigrist, Phys. Rev. B 59, 8271
(1999).

[58] K. Wakabayashi, S. Ken-ichi, T. Nakanishi, and T. Enoki, Sci. Technol. Adv.

Mater. 11, 054504 (2010).

[59] J. N. Fuchs, 2013, arXiv:1306.0380.

[60] J. Fuchs, F. Piechon, M. Goerbig, and G. Montambaux, Eur. Phys. J. B 77,

351 (2010).

[61] M. Jablan, H. Buljan, and M. Soljačić, 19, 11236 (2011).

[62] G. D. Mahan, Many Particle Physics (Plenum, New York, 1990).

[63] M. Ezawa, Phys. Rev. B 86, 161407(R) (2012).

[64] D. Xiao, G. B. Liu, W. Feng, X. Xu, and W. Yao, Phys. Rev. Lett. 108, 196802
(2012).

[65] T. Cao et al., Nature Commun. 3, 887 (2012).

[66] Z. Li and J. P. Carbotte, Phys. Rev. B 86, 205425 (2012).

[67] J. P. Carbotte, E. J. Nicol, and S. G. Sharapov, Phys. Rev. B 81, 045419

(2010).

[68] J. P. Carbotte, J. P. F. LeBlanc, and E. J. Nicol, Phys. Rev. B 85, 201411(R)

(2012).

[69] J. P. F. LeBlanc, J. P. Carbotte, and E. J. Nicol, Phys. Rev. B 84, 165448

(2011).

[70] M. I. Dyakonov and V. I. Perel, Phys. Lett. A 35, 459 (1971).

217
[71] J. E. Hirsch, Phys. Rev. Lett. 83, 1834 (1999).

[72] S. Murakami, N. Nagaosa, and S. C. Zhang, Science 301, 1348 (2003).

[73] J. Sinova et al., Phys. Rev. Lett. 92, 126603 (2004).

[74] Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Awschalom, Science 306,

1910 (2004).

[75] J. Wunderlich, B. Kaestner, J. Sinova, and T. Jungwirth, Phys. Rev. Lett. 94,
047204 (2005).

[76] B. A. Bernevig, T. L. Hughes, and S. C. Zhang, Science 314, 1757 (2006).

[77] S. Murakami, Phys. Rev. Lett. 97, 236805 (2006).

[78] C. X. Liu, T. L. Hughes, X. L. Qi, K. Wang, and S. C. Zhang, Phys. Rev. Lett.
100, 236601 (2008).

[79] M. Duckheim, D. L. Maslov, and D. Loss, Phys. Rev. B 80, 235327 (2009).

[80] D. A. Abanin, R. V. Gorbachev, K. S. Novoselov, A. K. Geim, and L. S. Levitov,

Phys. Rev. Lett. 107, 096601 (2011).

[81] A. Dyrdal and J. Barnaś, Phys. Status Solidi RRL 6, 340 (2012).

[82] M. Tahir, A. Manchon, K. Sabeeh, and U. Schwingenschlögl, Appl. Phys. Lett.

102, 162412 (2013).

[83] P. Wenk, S. Kettemann, and G. Bouzerar, Phys. Rev. B 86, 075441 (2012).

[84] D. Pesin and A. H. MacDonald, Nature Materials 11, 409 (2012).

[85] Y. Wang et al., Nano 07, 1250037 (2012).

[86] D. Xiao, W. Yao, and Q. Niu, Phys. Rev. Lett. 99, 236809 (2007).

[87] A. Rycerz, J. Tworzydlo, and C. W. J. Beenakker, Nature Phys. 3, 172 (2007).

218
 [88] M. Tahir and U. Schwingenschlögl, Appl. Phys. Lett. 101, 132412 (2012).

[89] A. Kavokin, G. Malpuech, and M. Glazov, Phys. Rev. Lett. 95, 136601 (2005).

[90] V. P. Gusynin, S. G. Sharapov, and J. P. Carbotte, Phys. Rev. Lett. 98, 157402
(2007).

[91] V. P. Gusynin, S. G. Sharapov, and J. P. Carbotte, J. Phys.: Condens. Matter

19, 026222 (2007).

[92] G. Semenoff and F. Zhou, J. High Energy Phys. 2011, 1 (2011).

[93] A. Pound, J. P. Carbotte, and E. J. Nicol, Phys. Rev. B 85, 125422 (2012).

[94] A. Pound, J. P. Carbotte, and E. J. Nicol, Phys. Rev. B 84, 085125 (2011).

[95] A. Pound, J. P. Carbotte, and E. J. Nicol, Europhys. Lett. 94, 57006 (2011).

[96] V. P. Gusynin and S. G. Sharapov, Phys. Rev. Lett. 95, 146801 (2005).

[97] A. Raoux, M. Morigi, J.-N. Fuchs, F. Piéchon, and G. Montambaux, Phys.

Rev. Lett. 112, 026402 (2014).

[98] M. L. Sadowski, G. Martinez, M. Potemski, C. Berger, and W. A. de Heer,

Phys. Rev. Lett. 97, 266405 (2006).

[99] M. L. Sadowski, G. Martinez, M. Potemski, C. Berger, and W. A. de Heer,

Solid State Commun. 143, 123 (2007).

[100] Z. Jiang et al., Phys. Rev. Lett. 98, 197403 (2007).

[101] R. S. Deacon, K.-C. Chuang, R. J. Nicholas, K. S. Novoselov, and A. K. Geim,

Phys. Rev. B 76, 081406 (2007).

[102] P. Plochocka et al., Phys. Rev. Lett. 100, 087401 (2008).

[103] M. Orlita and M. Potemski, Semicond. Sci. Technol. 25, 063001 (2010).

[104] E. A. Henriksen et al., Phys. Rev. Lett. 104, 067404 (2010).

219
[105] M. Orlita et al., Phys. Rev. Lett. 107, 216603 (2011).

[106] W.-K. Tse and A. H. MacDonald, Phys. Rev. B 84, 205327 (2011).

[107] Z. Li and J. P. Carbotte, Phys. Rev. B 88, 045414 (2013).

[108] P. E. C. Ashby and J. P. Carbotte, Phys. Rev. B 87, 245131 (2013).

[109] M. Tahir and U. Schwingenschlögl, Appl. Phys. Lett. 101, 132412 (2012).

[110] M. O. Goerbig, Rev. Mod. Phys. 83, 1193 (2011).

[111] G. Li and E. Y. Andrei, Nature Phys. 3, 623 (2007).

[112] D. L. Miller et al., Science 324, 924 (2009).

[113] E. Y. Andrei, G. Li, and X. Du, Rep. Prog. Phys. 75, 056501 (2012).

[114] Z. Li and J. P. Carbotte, Phys. Rev. B 86, 205425 (2012).

[115] T. Mori et al., Phys. Rev. B 77, 174515 (2008).

[116] I. Crassee et al., Phys. Rev. B 84, 035103 (2011).

[117] V. N. Kotov, B. Uchoa, V. M. Pereira, F. Guinea, and A. H. Castro Neto, Rev.

Mod. Phys. 84, 1067 (2012).

[118] B. Aufray et al., Appl. Phys. Lett. 96, 183102 (2010).

[119] W. J. de Haas and P. M. van Alphen, Leiden Comm. , 208d,212a (1930).

[120] W. J. de Haas and P. M. van Alphen, Leiden Comm. , 220d (1932).

[121] L. Onsager, Philosophical Magazine 43, 1006 (1952).

[122] J. B. Keller, Ann. Phys. 4, 180 (1958).

[123] J. W. McClure, Phys. Rev. 104, 666 (1956).

[124] H. Fukuyama, Prog. Theor. Phys. 45, 704 (1971).

220
[125] S. A. Safran and F. J. DiSalvo, Phys. Rev. B 20, 4889 (1979).

[126] S. G. Sharapov, V. P. Gusynin, and H. Beck, Phys. Rev. B 69, 075104 (2004).

[127] G. Gómez-Santos and T. Stauber, Phys. Rev. Lett. 106, 045504 (2011).

[128] A. Principi, M. Polini, G. Vignale, and M. I. Katsnelson, Phys. Rev. Lett. 104,
225503 (2010).

[129] M. Koshino and T. Ando, Solid State Commun. 151, 1054 (2011).

[130] M. Nakamura, Phys. Rev. B 76, 113301 (2007).

[131] M. Ezawa, Eur. Phys. J. B 85, 363 (2012).

[132] L. Bremme, T. Ihn, and K. Ensslin, Phys. Rev. B 59, 7305 (1999).

[133] L. Smrcka and P. Streda, J. Phys. C: Solid State Phys. 10, 2153 (1977).

[134] V. P. Gusynin, V. A. Miransky, S. G. Sharapov, and I. A. Shovkovy, Phys.

Rev. B 74, 195429 (2006).

[135] M. Tahir and U. Schwingenschlögl, Scientific Reports 3, 1075 (2013).

[136] M. Koshino and T. Ando, Phys. Rev. B 75, 235333 (2007).

[137] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K.

Geim, Rev. Mod. Phys. 81, 109 (2009).

[138] I. A. Luk’yanchuk and Y. Kopelevich, Phys. Rev. Lett. 93, 166402 (2004).

[139] Y. Suprunenko, E. V. Gorbar, S. G. Sharapov, and V. M. Loktev, Fiz. Nizk.

Temp. 34, 1033 (2008).

[140] A. R. Wright and R. H. McKenzie, Phys. Rev. B 87, 085411 (2013).

[141] L. M. Roth, Phys. Rev. 145, 434 (1966).

[142] S. K. Islam and T. K. Ghosh, J. Phys.: Condens. Matter 26, 335303 (2014).

221
[143] A. Mawrie, T. Biswas, and T. K. Ghosh, J. Phys.: Condens. Matter 26, 405301
(2014).

[144] V. Y. Tsaran and S. G. Sharapov, Phys. Rev. B 90, 205417 (2014).

[145] D. Bohm and D. Pines, Phys. Rev. 82, 625 (1951).

[146] D. Pines and D. Bohm, Phys. Rev. 85, 338 (1952).

[147] D. Bohm and D. Pines, Phys. Rev. 92, 609 (1953).

[148] H. Bruus and K. Flensberg, Many-body quantum theory in condensed matter

physics (Oxford University Press, Oxford, 2005).

[149] B. M. Santoyo and D. del Castillo-Mussot, Rev. Mex. de Fisica 39, 640 (1993).

[150] F. Stern, Phys. Rev. Lett. 18, 546 (1967).

[151] P. Pyatkovskiy, J. Phys.: Conference Series 129, 012006 (2008).

[152] P. Pyatkovskiy, J. Phys.: Condens. Matter 21, 025506 (2009).

[153] A. Scholz, T. Stauber, and J. Schliemann, Phys. Rev. B 86, 195424 (2012).

[154] K. W.-K. Shung, Phys. Rev. B 34, 979 (1986).

[155] E. V. Gorbar, V. P. Gusynin, V. A. Miransky, and I. A. Shovkovy, Phys. Rev.

B 66, 045108 (2002).

[156] T. Ando, J. Phys. Soc. Jpn. 75, 074716 (2006).

[157] B. Wunsch, T. Stauber, F. Sols, and F. Guinea, New J. Phys. 8, 318 (2006).

[158] Y. Barlas, T. Pereg-Barnea, M. Polini, R. Asgari, and A. H. MacDonald, Phys.

Rev. Lett. 98, 236601 (2007).

[159] E. H. Hwang and S. Das Sarma, Phys. Rev. B 75, 205418 (2007).

[160] T. Stauber, P. San-Jose, and L. Brey, New J. Phys. 15, 113050 (2013).

222
[161] R. Roldán and L. Brey, Phys. Rev. B 88, 115420 (2013).

[162] M. Pletyukhov and V. Gritsev, Phys. Rev. B 74, 045307 (2006).

[163] S. M. Badalyan, A. Matos-Abiague, G. Vignale, and J. Fabian, Phys. Rev. B

79, 205305 (2009).

[164] S. M. Badalyan, A. Matos-Abiague, G. Vignale, and J. Fabian, Phys. Rev. B

81, 205314 (2010).

[165] T. Kernreiter, M. Governale, and U. Zülicke, New J. Phys. 12, 093002 (2010).

[166] J. Schliemann, Europhys. Lett. 91, 67004 (2010).

[167] J. Schliemann, Phys. Rev. B 84, 155201 (2011).

[168] A. Scholz, T. Dollinger, P. Wenk, K. Richter, and J. Schliemann, Phys. Rev. B

87, 085321 (2013).

[169] R. Sensarma, E. H. Hwang, and S. Das Sarma, Phys. Rev. B 82, 195428 (2010).

[170] A. Scholz, T. Stauber, and J. Schliemann, Phys. Rev. B 88, 035135 (2013).

[171] A. Scholz and J. Schliemann, Phys. Rev. B 83, 235409 (2011).

[172] J. P. Carbotte, J. P. F. LeBlanc, and E. J. Nicol, Phys. Rev. B 85, 201411(R)

(2012).

[173] J. Gonzlez, F. Guinea, and M. Vozmediano, Nuclear Physics B 424, 595 (1994).

[174] S. Yuan, F. Jin, R. Roldán, A.-P. Jauho, and M. I. Katsnelson, Phys. Rev. B
88, 195401 (2013).

[175] A. L. Fetter and J. D. Walecka, Quantum Theory of Many-Particle Systems

(Dover, New York, 2003).

[176] C. Hwang et al., Scientific Reports 2, 590 (2012).

[177] J. P. F. LeBlanc and J. P. Carbotte, Phys. Rev. B 89, 035419 (2014).

223
[178] O. V. Gamayun, Phys. Rev. B 84, 085112 (2011).

[179] M. J. Lighthill, Introduction to Fourier Analysis and Generalized Functions

(Cambridge University Press, Cambridge, 1958).

[180] Y. A. Bychkov and E. I. Rashba, JETP Lett. 39, 78 (1984).

[181] Y. A. Bychkov and E. I. Rashba, J. Phys. C: Solid State Phys. 17, 6039 (1984).

[182] C.-X. Liu et al., Phys. Rev. B 82, 045122 (2010).

[183] J. Fabian, A. Matos-Abiague, C. Ertler, P. Stano, and I. Zutic, Acta Physica

Slovaca 57, No.4,5 565 (2007).

[184] I. Žutić, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004).

[185] H.-Z. Lu, W.-Y. Shan, W. Yao, Q. Niu, and S.-Q. Shen, Phys. Rev. B 81,
115407 (2010).

[186] Y. L. Chen et al., Science 329, 659 (2010).

[187] M. Koshino and T. Ando, Phys. Rev. B 77, 115313 (2008).

[188] A. A. Taskin, Z. Ren, S. Sasaki, K. Segawa, and Y. Ando, Phys. Rev. Lett.
107, 016801 (2011).

[189] G. P. Mikitik and Y. V. Sharlai, Phys. Rev. B 85, 033301 (2012).

[190] K. Kishigi and Y. Hasegawa, Phys. Rev. B 90, 085427 (2014).

[191] Z. Li and J. P. Carbotte, Phys. Rev. B 89, 165420 (2014).

[192] P. R. Wallace, Phys. Rev. 71, 622 (1947).

[193] E. Fradkin, Phys. Rev. B 33, 3263 (1986).

[194] P. Avouris, Z. Chen, and V. Perebeinos, Nature Nano. 2, 605 (2007).

[195] L. Pauling, The Nature of the Chemical Bond (Cornell Univ. Press, NY, 1960).

224
[196] E. J. Nicol and J. P. Carbotte, Phys. Rev. B 77, 155409 (2008).

[197] V. P. Gusynin and S. G. Sharapov, Phys. Rev. B 73, 245411 (2006).

[198] Z. Q. Li et al., Nature Phys. 4, 532 (2008).

[199] A. B. Kuzmenko, E. van Heumen, F. Carbone, and D. van der Marel, Phys.
Rev. Lett. 100, 117401 (2008).

[200] R. R. Nair et al., Science 320, 1308 (2008).

[201] K. F. Mak et al., Phys. Rev. Lett. 101, 196405 (2008).

225