Incommensurability

effects on a Second
Order Topological
Insulator

Raul Liquito
Mestrado em Física
Departamento de Física e Astronomia
2023

Supervisor
Eduardo V. Castro, Professor Auxiliar, Faculdade de Ciências da
Universidade do Porto
Acknowledgements
The last five years were not without its challenges. With that being said, I owe the realization of this
work to the influence of several individuals, to whom I would like to convey my heartfelt gratitude.
Firstly, to my supervisor Prof. Eduardo V. Castro, whose help, orientation, availability and knowledge
have proportionated significant educational moments. Beyond the confines of this thesis, I would also like
to express my appreciation to the Professor for his guidance and support during sensitive situations that
arose over the past two years. It is always a pleasure to learn and discuss physics with the Professor.
I also feel obligated to thank Miguel Gonçalves, for his contributions to this work, whether it be in his
amazing availability or knowledgeable insights in condensed matter physics. This work is as much my
own as it is Professor’s and Miguel’s.
Secondly I would like to thank all my course colleagues with special emphasis on Nicolau, Júlio, Diogo,
Patrı́cia, Luı́s, João Pinho, João Silva and Ricardo for the several physics discussions and unforgettable
joyful moments.
To my parents, whom I owe everything, thank you for all the support and encouragement, especially
in these challenging past few years.
Last, but not least, to Sofia. You illuminate my days with brightness and joy, showing me a part of
myself I did not know existed. Thank you!

i
Abstract
Symmetry-protected quantum phases in topological systems have recently been extended to higher-order.
Concurrently, the intricate localization properties of quasi-periodic systems and their impact on topo-
logical insulators have gathered significant attention. Nonetheless, the characterization of quasi-periodic
higher-order topological insulators remains unexplored. In this thesis, we bridge these two areas by
studying the effects of quasi-periodic modulations on the prototypical Benalcazar-Bernevig-Hughes (BBH)
model for higher-order topology at half filling.
In order to build our intuition on the interplay between quasi-periodicity and topology, we start with
a first-order topological insulator. Thus, we characterize a Chern insulator with Quadratic Band Crossing
Points (QBCP) under the effect of a 1D quasi-periodic modulation, where it is shown that clean limit Chern
phases (C = ±2) are robust to QP modulations. However, stronger modulations can induce topological
phase transitions in gap closing and opening fashion into topological regimes (C = ±1) not accessible
in the clean limit. Afterwards we discuss the BBH model and its disordered version to gain insight on
the effects of translational symmetry breaking on higher-order topological phases. The main results are
for the chiral Quasi-Periodic Quadrupole Insulator (QPQI) where the effects of quasi-periodicity in the
BBH model are vastly characterized, thus performing the first study of a QP higher-order topological
insulator. We demonstrate that clean limit quadrupole insulating (QI) phases with zero-energy corner
modes are robust to quasi-periodic modulations. Moreover, we encounter the first gapped QP-induced
QI phase (QPQI) described by intricate dynamics between the zero-energy corner modes and edge states
that disperse with varying phase shifts. A gapless QPQI regime was also encountered with localized states
at the Fermi level, similar to higher-order topological Anderson insulating (HOTAI) phases. This study
is accompanied by a detailed classification of topological, spectral and localization properties throughout
the several observed phases.
In this thesis, we broaden the scope of chiral QI phases by incorporating QP modulations in the BBH
model, providing verification that QI phases are robust to QP modulations. Furthermore, we demonstrate
the emergence of QPQI phases with dynamics absent in the clean limit. This expansion results in a diverse
phase diagram, encompassing QI topological phases, trivial and Anderson insulators, and even critical
metals.

ii
Resumo
As fases quânticas protegidas por simetria em sistemas topológicos foram recentemente estendidas para
uma ordem superior. Ao mesmo tempo, as complexas propriedades de localização de sistemas quase
periódicos e o seu impacto em isoladores topológicos têm atraı́do atenção significativa. No entanto, o
efeito de quase-periodicidade em isoladores topológicos de ordem superior permanece inexplorado. Nesta
tese fazemos a ponte entre estas duas áreas, estudando os efeitos de modulações quase periódicas no
modelo prototı́pico de Benalcazar-Bernevig-Hughes (BBH) para topologia de ordem superior com meio
preenchimento.
Para construir a nossa intuição sobre a interação entre quase periodicidade e topologia, começamos
com um isolalador topológico de primeira ordem. Assim, caracterizamos um isolador Chern com pontos
de cruzamento de banda quadrática (QBCP) sob o efeito de uma modulação quase periódica 1D, no qual
é mostrado que as fases Chern do limite limpo (C = ±2) são robustas a modulações QP. No entanto,
modulações mais fortes podem induzir transições de fase topológicas, em abertura e fecho do hiato de
energia, para regimes topológicos (C = ±1) não acessı́veis no limite limpo. Posteriormente discutimos o
modelo BBH e sua versão desordenada para estudar o efeito da quebra de simetria de translação em fases
topológicas de ordem superior. Os principais resultados são para o isolador de momento quadrupolar
quase periódico quiral (QPQI) no qual os efeitos da quase periodicidade no modelo BBH são amplamente
caracterizados, realizando assim o primeiro estudo de um isolador topológico QP de ordem superior.
Demonstramos que as fases isoladoras de momento quadrupolar de limite limpo (QI) caracterizadas por
modos de canto de energia zero, são robustas a modulações quase periódicas. Além disso, encontramos a
primeira fase QI induzida por QP (QPQI) descrita por uma dinâmica complexa entre os modos de canto
de energia zero e estados de fronteira que dispersam com mudanças de fases do potencial. Um regime
QPQI sem hiato de energia também foi encontrado com estados localizados no nı́vel de Fermi, semelhante
a fases isoladores topológicas de Anderson de ordem superior (HOTAI). Este estudo é acompanhado por
uma classificação detalhada das propriedades topológicas, espectrais e de localização ao longo das diversas
fases observadas.
Nesta tese, apresentamos novas fases QI quirais incorporando modulações QP no modelo BBH,
fornecendo a verificação de que as fases QI são robustas a modulações quase-perioódicas. Além disso,
demonstramos o surgimento de fases QPQI com dinâmica ausente no limite limpo. Esta expansão resulta
em um diagrama de fases diversificado, abrangendo fases topológicas QI, isoladores triviais e de Anderson
e até metais crı́ticos.

iii
Contents

List of Tables vi

List of Figures x

1 Introduction 1
1.1 Topological Insulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The Chern Number and AQHE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Disorder and Topological Insulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Quasi-Periodicity and Topological Insulators . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 higher-order Topological Insulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Discrete Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6.1 Time Reversal Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6.2 Particle-Hole Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6.3 Chiral Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Methods 8
2.1 Wannier Functions, Wannier Centers and Boundary Hamiltonian through Wilson Loops . 8
2.1.1 Polarization in 1D crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 Polarization in 2D crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.3 Wannier Bands and Nested Wilson loop . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Real Space Multipole Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Localization Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.1 Inverse Participation Ratio (IPR) and Scaling . . . . . . . . . . . . . . . . . . . . . 16
2.3.2 Transfer Matrix Method (TMM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Spectral Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Kernel Polynomial Expansion (KPM) and the Density of States (DOS) . . . . . . 20
Chebyshev Polinomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Expansion of Dirac Delta in Chebyshev Series . . . . . . . . . . . . . . . . . . . . . 21
Chebyshev Polynomials and Operators . . . . . . . . . . . . . . . . . . . . . . . . . 21
Stochastic Evaluation of the Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Density of States (DOS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4.2 The Recursive method for computing the Local DOS (LDOS) . . . . . . . . . . . . 23
2.5 Boundary Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

iv
CONTENTS CONTENTS

2.6 Numerical Simulations on Quasi-Periodic Systems . . . . . . . . . . . . . . . . . . . . . . 26

3 Quasi-Periodic modulation and first-order Topology 29

3.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4 Benalcazar-Bernevig-Hughes (BBH) Model in 2 dimensions 34

4.1 Clean Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2 Chiral Disorder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 42

5.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.3 Topological and Spectral Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.4 Localization properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.5 Opening the Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

6 Conclusion and Outlook 57

A Wilson Loops - Supplementary Calculations 59

A.1 ⟨0| γk,n c†q+∆k,α |0⟩ and ⟨0| cq,α γk†′ ,m |0⟩ correlators . . . . . . . . . . . . . . . . . . . . . . . 59
A.2 Polarization of Wannier Sectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

B Chiral Symmetry and Bulk Multipole Moments 64

C Tridiagonal Matrixes 67
C.1 Lanczos Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
C.2 Inverting Tridiagonal Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

D kx -dependent Model 70

E Corner Mode Hybridization 72

F Complementary Results 74

Bibliography 78

v
List of Tables

2.1 First 17 numbers of the Fibonacci sequence, each new number is obtained from the sum of
the two that precede it. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

vi
List of Figures

1.1 Hall resistivity ρxy and longitudinal resistivity ρxx as a function of magnetic filed strength (T ).
Image taken from [1]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

3.1 Band structure for different values of topological mass B. . . . . . . . . . . . . . . . . . . 30

3.2 Berry curvature (Ω(k)) for different values of topological mass B and L = 1000. The scale
of the color maps differs between Fig. (a) and (b). In (a) the scale is symmetric, while in
(b) the scale is definite and positive. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 (left) Chern number as a function of B and W for a system size of L = 21 and 20 different
twist realizations. The grayed-out areas are gapless. (right) Energy gap as a function of B
and W for Ly = 1597. No twist realizations were considered. The gray lines represent the
lines at which a topological phase transition occurs. . . . . . . . . . . . . . . . . . . . . . 32
3.4 TMM results at Fermi level (E = 0) at the kx where the gap closes. Obtained for a 1D
system size of N = 30000 such that ϵ ≈ 1%. The grid discretization is 300 × 300. . . . . . 32
3.5 TMM bulk phase diagram at B = −3 for the entire range of the valence band, E ∈ [−7, 0]
and kx ∈ [−π, π]. The insets follow the same procedure for a smaller region (E ∈ [−1, 0]
and kx ∈ [−π, −π + 0.2]) with a grid discretization of 600 × 600. . . . . . . . . . . . . . . 33

4.1 Tight binding scheme of the BBH model. The dotted “connections” are negative in relation
to γ and λ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2 Band Structure of the 2D BBH model for different values of the hopping ratio γ/λ and
with δ = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3 (left) Energy Spectrum as a function of γ/λ obtained by ED for a system with open
boundaries and L = 30. The gap does not close properly at |γ/λ| = 1 due to finite-size
effects. In red, the energy corner modes. (right) Electron charge density for γ = δ = 10−3 . 36
4.4 (a) qxy as a function of the hopping γ computed via real-space methods for a system size
of L = 10. (b) P = 4 |px py | invariant computed via nested Wilson loops for a system size
of L = 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.5 (a) qxy and P as a function of W . The P invariant was computed for a system size of
L = 200 and averaged over 200 disorder configurations. qxy was obtained via real space
methods, with 1000 disorder configurations for γ = 1.1, L = 10 and 20 while for L = 30
500 configurations were considered. (b) Spectral gap (∆E) as a function of W for γ = 1.1,
L = 200 and averaged over 150 disorder configurations. . . . . . . . . . . . . . . . . . . . 39

vii
LIST OF FIGURES LIST OF FIGURES

4.6 DOS as a function of E calculated through KPM for a system of size L = 1000, M = 5000
moments, R = 10 disorder and random vectors configurations for several values of disorder
strength. (a) Full DOS. (b) Zoomed-in view around Fermi level. . . . . . . . . . . . . . . 39
4.7 LDOS for E = 0 as a function of position r = (x, y) calculated using the recursive method
for a system size L = 40. (left) W = 2.6 (phase II) and 500 disorder configurations. (right)
W = 3.4 (phase III) and 1000 disorder configurations. . . . . . . . . . . . . . . . . . . . . 40
4.8 Normalized localization length (ξ/M ) as a function of W at Fermi level, obtained with
TMM for γ = 1.1 for different transversal sizes M . The size of the longitudinal direction
N was chosen such that ϵ < 1%. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

5.1 (a) Quadrupole moment (qxy ) in the (γ, W ) plane, computed for a system size L = 13,
with θi = π and averaged over 100 random phase shifts realizations. (b) The logarithm
spectral gap (log10 (∆E)) in the (γ, W ) plane for a system size L = 34, with θi = 0 and
averaged over 50 realizations of phase shifts. A minimum value of the spectral gap was
chosen (∆E < 10−3 ) to highlight all the gapped and gapless phases appearing in the phase
diagram. The dotted white lines are constant γ and W cuts for γ = 0.5, 1.1 and W = 3,
respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2 Spectral gap (∆E) as a function of W computed for an even system size L = 144 =⇒
θi = 0 and averaged over 50 phase shifts realizations. (a) γ = 0.5. (b) γ = 1.1. . . . . . . 44
5.3 qxy and P as a function of W . The P invariant was computed for a system size of L = 610
with 115 averages over phase shifts. qxy was obtained via real-space methods with 100
averages over phase shifts. For even L, θx = θy = 0; for odd L θx = θy = 0. (a) γ = 0.5.
(b) γ = 1.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.4 The DOS at zero-energy (ρ(0)) as a function of W . The following KPM parameters were
used: L = 987, M = 5000, R = 10. (a) γ = 0.5. (b) γ = 1.1. . . . . . . . . . . . . . . . . 46
5.5 Clean limit DOS as a function of E. The following KPM parameters were used: L = 987,
M = 4000, R = 10 (a) γ = 0.5. (b) γ = 1.1. . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.6 DOS as a function of E for γ = 1.1 and different values of quasi-periodic potential strength
(W ). The following KPM parameters were used: L = 987, M = 2000, R = 10. . . . . . . . 47
5.7 TMM at Fermi level (E = 0) for several transversal system sizes (M ). For odd M we add
a twits of θM = π. The size of the longitudinal direction was chosen such that the relative
error ϵ < 1%. (a) γ = 0.5. (b) γ = 1.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.8 IP R (a,b) and IP Rk (c,d) calculations for different system sizes obtained with exact diag-
onalization methods. The IP R and IP Rk are averaged over 100 phase twists, shifts and
over the first 8 eigenstates with energies closest to E = 0. (a,c) γ = 0.5. (b,d) γ = 1.1. . . 49
5.9 D2 (a,b) and D2k (c,d) calculations for different system sizes obtained from the IP R and
IP Rk results respectively. (a,c) γ = 0.5. (b,d) γ = 1.1. The red curves were estimated
making a fit in log − log scale for all L = {34, 55, 89, 144, 233, 377}, while the dotted gray
lines were obtained only considering the biggest three system sizes L = {144, 233, 377}. . . 50

viii
LIST OF FIGURES LIST OF FIGURES

5.10 (a-b) Corner occupation probability (Pocc ) as a function of W for the first 2 zero-energy
states. The results were obtained for a system size of L = 377 and averaged over 200 phase
shifts. (c-d) Corner charge (Q̄) as a function of W for several system sizes and averaged
over 100 phase shifts and over the 4 corners. A small δ = 10−5 was used in both cases to
split the degeneracy of the four-fold corner modes. (a,c) γ = 0.5. (b,d) γ = 1.1. . . . . . . 52
5.11 Ratio of open boundary DOS (ρOBC ) and periodic boundaries DOS (ρP BC ) as a function
of energy for W in different phases and γ = 1.1. The grayed-out regions are bulk gaps
opened by the QP modulation and were estimated by setting a threshold for the PBC DOS
(ρP BC < 10−3 ). The following KPM parameters were used: L = 987, M = 5000, R = 10. 53
5.12 (a) Standard deviation (σE=0 ) of the energies of the corner modes as a function of the
phase shifts (ϕx , ϕy ). (b) Logarithm of edge state energy (log10 (Eedge )) as a function of the
phase shifts (ϕx , ϕy ). Results obtained for W = 3, γ = 0.5 for a system size L = 233. . . 54
5.13 (a) Standard deviation (σE=0 ) of the energies of the corner modes as a function of the
phase shifts (ϕx , ϕy ). (b) Logarithm of edge state energy (log10 (Eedge )) as a function of the
phase shifts (ϕx , ϕy ). Results obtained for W = 3.8 and γ = 0.2 in the QPQI phase for a
system size L = 377. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.14 IP R of the edge, outer and inner states plotted from left to right respectively for L = 377. 55

E.1 Probability density (|ψ(R)|2 ) of the 6 eigenstates around Fermi level for γ = 1.1,W = 2.2
(phase II in Fig. 5.3) for one realization of random shifts and with δ = 10−7 to split
the degeneracy of the four-fold corner modes. Each sub-figure is a different system size,
increasing from top to bottom. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

F.1 DOS as a function of energy (E) for γ = 0.5 and different values of quasi-periodic potential
strength (W ). The following parameters were used in the KPM: system size L = 987,
number of Chebyshev momenta M = 2000 and number of averages over traces and phase
shifts R = 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
F.2 Ratio of open boundary DOS and periodic boundaries DOS as a function of energy for W
in different phases and γ = 0.5. The grayed out regions are bulk gaps opened by the QP
modulation. The following KPM parameters were utilized: L = 987, M = 5000, R = 10. . 74
F.3 (a) qxy and P as a function of W . The P invariant was computed for a system size of
L = 610 with 100 averages over phase shifts. qxy was obtained via real space methods with
100 averages over phase shifts. For even L, θx = θy = 0; for odd L θx = θy = 0. (b)
Spectral gap as a function of W computed for an even system size L = 144 =⇒ θi = 0
and averaged over 50 phase shifts realizations. (c-d) Fractal dimension averaged out over
100 phase twists, shifts and over the first 8 eigenstates with energies closest to E = 0. (e)
Corner occupation probability (Pocc ) as a function of W for the first 2 zero-energy states.
The results were obtained for a system size of L = 377 and averaged over 100 phase shifts.
(f) Corner charge (Q̄) as a function of W for several system sizes and averaged over 100
phase shifts and over the 4 corners. (e-f) l δ = 10−5 . . . . . . . . . . . . . . . . . . . . . . 76

ix
LIST OF FIGURES LIST OF FIGURES

F.4 Standard deviation (σE=0 ) of the energies of the corner modes as a function of the phase
shifts (ϕx , ϕy ) and logarithm of edge state energy (log10 (Eedge )) as a function of the phase
shifts (ϕx , ϕy ). Results obtained for W = 3.8 and γ = 0.2 in the QPQI phase for system
sizes L = {89, 144, 233} as indicated. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

x
Chapter 1

Introduction

Condensed matter is an area of physics concerned with studying properties of materials. Its diverse
phenomenon range from descriptions of phase transitions to quantum descriptions of superconductivity,
ferromagnetic and antiferromagnetic phases, condensates and many other interesting and complex proper-
ties. It is safe to say that without our understanding of condensed matter physics, most of the technology
that surrounds us, would not be possible.
This said, we do a brief recap of important
hallmarks of condensed matter. In 1900 Paul
Drude proposed the first theoretical model for a
classical electron gas on a solid metal. This model
allowed for the calculation of several quantities in
terms of the mean free time (τ ), revealing its suc-
cess with the explanation of the Hall effect, discov-
ered fifty years earlier by Edwin Hall. This physi-
cist showed that a transverse voltage to the applied
current appeared in a conductor when it was sub-
jected to a perpendicular magnetic field (B).
A century passed, and in 1980, Klaus von Kl-
itzing and collaborators [2] discovered that the Hall
conductance was quantized to integers of e2 /h,
with this new behavior being robust to system con-
taminations and deformations. The linear depen-
dence of the Hall resistivity (ρxy ) at low fields gives
way to a plateau insulating behavior, accompa-
nied by zero longitudinal resistivity (ρxx ), as shown Figure 1.1: Hall resistivity ρxy and longitudinal resistiv-
in figure 1.1. Another remarkable physicist that ity ρxx as a function of magnetic filed strength (T ). Image
taken from [1].
shaped the course of physics, Lev Landau made
contributions in the field of quantum mechanics,
where he developed the theory of Landau quantization. Landau theorized that the band structure of ma-
terials, when subjected to a uniform perpendicular magnetic field, degenerated into energy levels, known

1
1 Introduction 2

as Landau levels. This theory set the fundamentals needed for the understanding of the quantum Hall
effect (QHE).
It was not until 1982 that Thouless and collaborators [3], shed some light on the underlying mechanism
e2
behind the QHE. The Hall conductivity σxy = hn was not proportional to a “random” integer but to a
quantized topological invariant that emerges from band topology, the first Chern number, to be discussed
later. Another six years were needed after Thouless’s new insights on QHE, for Haldane to propose a
model [4], where quantum Hall phases could be observed in the absence of a magnetic field. The anomalous
quantum Hall effect (AQHE) was born. All this inspiring work performed in the 70s and 80s paved the
way to the field of topology in condensed matter physics.

1.1 Topological Insulators

Topology is an area of mathematics that studies “spaces”1 and the way that they are linked through
slow and continuous deformations. From a topology perspective, a donut and a mug are exactly the
same object, which we call a torus, which is any closed surface that contains one single handle (or, more
colloquially, one hole). We can adiabatically deform a mug into a donut and vice versa. However, we
cannot deform a donut into a sphere since it has no holes. In this manner, the number of holes in a
manifold emerges as a topological invariant, the genus. Topological phases of matter share the same
phenomenology. A system in a given topological phase can be adiabatically deformed into any new
configuration, keeping its topological properties, as long as the system remains in the same topological
phase. For this reason, topological phases of matter are robust to the introduction of deformations and
impurities in the system. A new characterization of phases of matter emerges, having in mind concepts of
topology. In these novel phases, a quantized bulk property is the hallmark of a finite topological invariant.
In crystalline systems with translation invariance, the inherent topological behavior of systems comes
from the underlying band structure. A fundamental consequence of topological phases in gapped band
structures is the existence of gapless conducting states at the interfaces, where the topological invari-
ant changes. For 2 dimensional systems, these edge states were first analyzed by Halperin in 1982,[5].
These electronic states are chiral since they only propagate in one direction along the topological vacuum
boundary. This is the celebrated bulk boundary correspondence [6, 7]. In 1D, there are also interesting
examples of topological systems, with a paradigmatic example being the tight binding model of poly-
acetylene proposed by Su, Schrieffer and Heeger more commonly referred to as the SSH model [8]. This
model presents a quantized bulk electronic polarization alongside zero-energy gapless boundary states.
Due to their exotic insulating properties, they have gained particular attention in research [6, 9, 10, 11].
In contrast with the Landau paradigm, where phase transitions are associated with symmetry break-
ing, topological phases are associated with the presence of a given symmetry of the system or the absence
thereof. Time reversal symmetry (T ), particle-hole2 symmetry (C) and chiral symmetry3 (Π), turn out
to be of major importance, which became clear in 2008 when Schnyder and collaborators completed the
work done by Altland and Zirnbauer [12], linking the presence of these discrete symmetries to topological
1
This can be any structure in a generalized dimension, in condensed matter physics the relevant space is usually the first
Brillouin zone (FBZ), although, some topological invariants are defined in more complex and not so trivial geometries.
2
Also known as charge conjugation
3
Also known as sub-lattice symmetry

2
1 Introduction 3

constrains [13]. Besides discrete symmetries, a material can have a set of crystalline symmetries that
give rise to topological character. These types of materials are known as topological crystalline insulators
[14, 15], which have their own set of topological invariants.

1.2 The Chern Number and AQHE

In order to gain some intuition on the characterization of topological systems, we dive succinctly into the
2D Chern insulators.
A 2D gapped system with a valence and a conduction band is considered. In these types of systems, if
time reversal symmetry is broken, the topological invariant is the Chern number. Let u−

k be the Bloch
states of the valence band, then these states define the Berry connection of the band A(k) = i u− −
k |∇k uk .
We can now ask, what phase does a state accumulate when we take it adiabatically across a given path
over the first Brillouin zone (FBZ). The Abelian Wilson loop over a path C in the FBZ is defined by:
 ˛ 
2
W[C] = exp i A(k) · d k . (1.2.1)
C

The exponent of the Wilson loop is known as the Berry phase (or geometric phase) [16]. Making use of
Stokes’ theorem, the line integral over C = ∂S can be rewritten as a surface integral over the surface S,
˛ ˆ ˆ
2
γ= A(k) · d k = ∇ × A(k) · dS = Ω(k) · dS. (1.2.2)
∂S S S

The surface integral form reveals itself to be more illuminating regarding how this mathematical object
“sees” the band structure of the system. A new quantity is then introduced Ω(k) = ∇ × A(k), the Berry
curvature. For 2 dimensional systems, if the integration is performed over the entire FBZ, the topological
invariant of the system is defined, the first Chern number C ∈ Z,
ˆ
1
C= Ω(k) · dS. (1.2.3)
2π F BZ

The Chern number shares an intimate relation with the transversal conductivity of the system, with
e2
σxy = h C. Chern insulators need to have broken-time reversal symmetry. This concludes the bridge
between topological phases of matter and the anomalous quantum Hall effect. In other dimensions, the
Berry phase is identically defined, though the associated topological invariant may be different. These
topological invariants are quantized by symmetries, and because they are Gauge invariant, manifest as
observables of the system. In the case of 1D systems, such as the SSH model, one possible topological
invariant is the winding number ν, that quantizes the bulk electronic polarization of the system. The
particular choice of topological invariant comes from the underlying symmetries we choose to fix a priori.
In the case of multiple filled bands and if the Berry Connection is non-diagonal, a non-Abelian formulation
is required.

3
1 Introduction 4

1.3 Disorder and Topological Insulators

We now take a look at the effects of a disordered potential. At first glance, disorder might seem an
awkward addition to our system. As it turns out, disorder is omnipresent in experimental realizations
since it can be understood as lattice deformations or impurities. In the clean limit, crystalline symmetry
is present, and so the Bloch Hamiltonian fully describes the system. However, with the introduction of
disorder, spatial symmetries are broken. For this reason, disordered metallic systems manifest a finite
resistivity, even at T = 0, with the resistivity rising as disorder increases. In fact, we can paint an intuitive
picture of what happens when a local disordered potential is added: a part of the wave is scattered from
k → −k.4 If we keep adding scattering centers randomly distributed, destructive interference of the
waves will occur in most of the lattice but for one small region where all the spectral weight of the wave
function is concentrated, or more simply put, the wave function becomes confined to a small region of
space. This localization phenomenon was first discovered by P. W. Anderson in 1958 [17] and is referred
to as Anderson localization. A material is said to be an Anderson insulator if its insulating capabilities
are provided by the existence of Anderson localized states at the Fermi level, even if the system is gapless.
Later contributions to disordered electronic systems showed that for the orthogonal symmetry class
at D ≤ 2, any amount of disorder leads to Anderson localization [18, 19, 20]. This is a surprising result
that shows disorder cannot be treated as perturbations to Bloch waves since the crystalline momenta k is
no longer a good quantum number. The same cannot be said about 3-dimensional systems, whereas, as a
function of energy, the states change from being localized to extended. This energy is called the mobility
edge of the system. For a critical value of disorder (Wc ) all the spectrum becomes Anderson localized. In
this thesis, we will focus on low-dimensional systems with D ≤ 2.
As stated, topological properties are symmetry-protected, so, as long as the effective disordered
Hamiltonian preserves the relevant symmetries, topological phases can occur. In fact, disorder plays
a key role on the QHE, since the plateaus of constant Hall resistivity only appear in the presence of
disorder. Moreover, the original experiment was realized on an contaminated sample [2]. This shows
that in the presence of weak disorder, topological properties are robust. In these regimes, a system in
a topological phase exhibits Anderson localized states. However, the topology is entirely carried by a
few critical states5 . Topological phase transitions can also be induced by disorder, since disorder can
lead to a gap-closing transition [21]. Furthermore, in the presence of disorder, gapless topological phases
are usually observed, with the topological phase transitions (TPT) occurring through the “levitation and
annihilation” mechanism [22, 23, 24, 25]. In this manner, disorder-induced topological phases are referred
to as topological Anderson insulators (TAI) [26].

1.4 Quasi-Periodicity and Topological Insulators

Disordered electronic systems are proficiently classified, either with experimental realizations of TAI,
numerical work or analytical proofs regarding localization properties and their interplay with topology.
Another class of systems that break translational invariance are quasi-periodic systems. Aubry and
4
For 1-dimensional (1D) materials.
5
A critical sate is an intermediary characterization between extended and localized, more concretely is a state with a
fractal dimension smaller than 1 but not localized

4
1 Introduction 5

André proposed the first instance of these types of systems, which displayed a quasi-periodic modulation
in a simple 1-dimensional (1D) tight binding chain [27]. In contrast with Anderson disorder, extended,
critical and localized phases can arise even in 1D [28, 29, 30]. Away from one-dimensional systems, higher-
dimensional models also display diverse localization phenomena, gaining special attention on the interplay
between Moiré physics and localization [31, 32, 33]. An example of exotic behavior in uncorrelated quasi-
periodic systems is the existence of mixed localized and extended spectra in 2-dimensional (2D) systems
[34].
Quasi-periodicity also shows its potential in topological matter, where it unveils new topological
behavior [35, 36] with far richer localization properties than their disordered cousins. Different types of
Chern insulating phases have been shown to be induced by quasi-periodic modulations [37, 38].
These types of systems were first considered in cold atoms experiments, while more recently they have
gained attention in Moiré materials and in many-body localization phenomenon [39, 40]. Furthermore,
systems with quasi-periodic modulations can be realized on widely different platforms, including optical
lattices [41, 42], photonic [43, 44] and phononic [45, 46] metamaterials, and moiré materials [47, 48].

1.5 higher-order Topological Insulators

In the last six years, higher-order topological insulators (HOTIs) have drawn great interest among con-
densed matter physicists. Unlike the more conventional first-order topological insulators, for which bulk-
boundary correspondence guarantees the existence of D − 1 dimensional boundary states, HOTIs are
characterized by boundary states with lower dimensionality. In particular, an mth order topological in-
sulator has D − m dimensional gapless boundary modes for a D dimensional system [49]. For example,
a second-order topological insulator (SOTI) in a 2-dimensional system has zero-energy corner modes.
In these higher-order generalizations, bulk boundary correspondence holds, as the presence of a bulk
quantized property manifests in gapless boundary modes. In contrast with first-order TIs that exhibit
a trivial boundary, HOTIs exhibit a more intricate bulk-boundary correspondence, featuring successive
lower-dimensional TIs in their boundaries. This complexity gives rise to a nested hierarchy of topological
insulators. Consequently, an mth order topological insulator is characterized by an (m − 1)th order TI in
the boundary. Moreover, symmetries are equally important for higher-order topology, with HOTIs phases
protected by either discrete symmetries6 or by crystalline symmetries.
The first theorized model was introduced in 2017 by Bernevig, Benalcazar and Hughes [49](BBH
model) and reveals to be a clever higher-dimensional generalization of the known SSH model. It acts
as the building block for most of the newly proposed HOTIs. The family of BBH models consist of
a 2 dimensional SOTI and a 3 dimensional third-order topological insulator (TOTI). The first one is
characterized by gapless corner states with a quantized bulk quadrupole moment, boundary dipole moment
and corner charges. As for the BBH TOTI, it displays a quantized bulk octupole moment, quadrupole
boundary moment, hinge dipole moment and corner charges. This boundary hierarchy paints a clear
picture of classical electromagnetism, with the emergence of canonical boundary signatures of quadrupole
and octupole moments.
Since the introduction of the BBH model, new higher-order topological models have been proposed,
6
The tenfold way no longer applies in these higher-order generalizations.

5
1 Introduction 6

such as its equivalent model [50], the breathing kagome lattice as presented in Refs. [51, 52], twisted Moiré
superlattice’s [53], generalizations of the Haldane model [54] and many others. Considerable efforts have
been made to unveil the symmetries behind the protection of these topological phases and to understand
their robustness against weak disorder. Furthermore, these new topological regimes have been experimen-
tally realized in phononic metamaterials [55], electric circuits [56, 57] and even real solid-state materials
[58]. Some disordered HOTIs have been studied [59, 60] and have been experimentally observed [57].
Although some work has been performed on quasicrystals with higher-order phases [61, 62], the effects of
quasi-periodicity in higher-order topology remain vastly unstudied.

1.6 Discrete Symmetries

In this section, we make some quick remarks about the three different discrete symmetries that allow for
the definition of each symmetry class in the tenfold way [12, 13]. We follow the book on topology by
Miguel Araújo and Pedro Sacramento [63].

1.6.1 Time Reversal Symmetry

Time reversal is defined by the following condition: for every ψ that satisfies the Schrödinger equation,
then T ψ ∗ also satisfies the Schrödinger equation with the substitution t → −t, where T is unitary. In this
manner, we can write:

iℏ∂t ψ = Hψ ↔ −iℏ∂t ψ ∗ = H∗ ψ ∗ ↔ −iℏ∂t (T ψ ∗ ) = T H∗ T † (T ψ ∗ ) . (1.6.1)

Then the time reversal operation is given by the following substitutions:

ψ → T ψ∗ (1.6.2)
∗ †
H → TH T . (1.6.3)

A given theory is time symmetric if T H∗ T † = H.

1.6.2 Particle-Hole Symmetry

Particle-hole or charge conjugation symmetry is defined by the condition: if there exists a unitary matrix
C such that Cψ ∗ satisfies the Schrödinger equation, then CH∗ C † = −H.

1.6.3 Chiral Symmetry

Chiral or sub-lattice symmetry is defined by:

∃Π : ΠHΠ† = −H (1.6.4)

This definition implies that {H, Π} = 0 so that if |ψn ⟩ is an eigenstate of H with energy En , then Π |ψn ⟩
is also an eigenstate of H with energy −En . So, any Hamiltonian that preserves chiral symmetry has

6
1 Introduction 7

a symmetric spectrum. Any Hamiltonian that preserves chiral symmetry can be written as an anti-
diagonal matrix. The chiral operator is then Σz = σz ⊗ IN/2 , however it’s possible that for a given basis
the Hamiltonian does not present this structure despite being chiral symmetric. In these cases, a unitary
transformation U can be considered that rotates the Hamiltonian to anti-diagonal form:
!
0 h
U HU † = (1.6.5)
h† 0

where h is a given matrix with half the dimension of the Hilbert Space, (N/2). This can be achieved in
real space by identifying two decoupled sub-lattices and proceeding to properly order the real-space basis.
This said, Π can be constructed from U in the following manner:

Σz U HU † Σz = −U HU † ↔
↔ U † Σz U HU † Σz U = −H. (1.6.6)

From this we make the identification Π = U † Σz U = Π† concluding that Π2 = ΠΠ† = 1.

Let us consider a system with time reversal symmetry (T H∗ T † = H), and charge conjugation
(CH∗ C † = −H). From the charge conjugation symmetry and using C unitarity one can invert the
relation and obtain H∗ = −C † HC. We can then insert this relation into the time reversal condition:
   
T H∗ T † = −T C † HCT † = H ↔ T C † H CT † = −H. (1.6.7)

It is clear that a system with charge conjugation and time reversal symmetry also preserves chiral sym-
metry with Π = T C † = CT † . A system can preserve chiral symmetry even if it does not preserve the
other two.
Before concluding this section, we explore some of the properties of this symmetry. Assuming a chiral
symmetric Hamiltonian in anti-diagonal form 1.6.5, its eigenvectors can be written as |ψ⟩ = √1 (ψA , ψB )
2
such that {ψA , ψB } are normalized to 1, thus Π |ψ⟩ = √1 (ψA , −ψB ). Either of the |ψA,B ⟩ obey an
2
eigenvalue equation that involves diagonalizing some lower-dimensional operator. For this, we consider
the following eigenvalue problem:

H2 |ψn ⟩ = En2 |ψn ⟩ . (1.6.8)

Assuming H to be in the form of eq. 1.6.5 then if follows:


 hh†
ψ n = E 2 ψ n
A n A
. (1.6.9)
 h† h
ψ n = E 2 ψ n
B n B

Out of the three discrete symmetries we have introduced, chiral symmetry will play an important role in
the work that follows. In this manner, it is only fitting that we have made a more detailed characterization
of this symmetry. In the chapter that follows, we move away from the introduction and focus on essential
methods to study the vast properties of topological electronic systems.

7
Chapter 2

Methods

2.1 Wannier Functions, Wannier Centers and Boundary Hamiltonian

through Wilson Loops
Wilson loops are a powerful tool when tackling topological properties. They allow us to compute several
quantities, such as bulk polarizations, Chern numbers, etc. They are closely connected to topology and
define the Wannier functions, which are a localized basis for the Hamiltonian interpreted as the tight
binding atomic orbitals. More recently, Wilson loop operators were shown to be adiabatically connected
to the boundary Hamiltonian [64] allowing for the study of boundary topology.

2.1.1 Polarization in 1D crystals

In insulators, the electronic contribution to polarization comes from the displacements of electrons relative
to the atomic positive charges inside the unit cells of the system. These shifts are the so-called Wannier
centers, which can be computed via diagonalization of the projected position operators.
The position operator in a 1D crystal with Lx unit cells and Norb orbitals per UC may be defined
following the work in Ref. [65]:

c†R,α |0⟩ e−i∆k(R+rα ) ⟨0| cR,α ,

X
x̂ = (2.1.1)
R,α

2π
where ∆k = Lx and rα are the relative positions of the orbitals inside the unit cell. Making use of a
discrete Fourier transform, one can pass from real space to reciprocal space and thus x̂ takes the form:

c†k+∆k,α |0⟩ ⟨0| ck,α .

X
x̂ = (2.1.2)
k,α

In reciprocal space, the system Hamiltonian is defined by

c†k,α [hk ]αβ ck,β ,

X
H= (2.1.3)
k

where the sum is over repeated indexes. Making use of [hk ]αβ = n α n ∗ β
P
n [uk ] ϵk,n [(uk ) ] , one can write H
in the band representation:

8
2 Methods 9

c†k,α [unk ]α ϵk,n [(unk )∗ ] ck,β

β
X
H=
k,n
!  

[unk ]α c†k,α
β
[(unk )∗ ] ck,β 
X X X
= ϵk,n 
k,n α β
†
X
= γk,n ϵk,n γk,n , (2.1.4)
k,n

† P n α † †
where γk,n = α [uk ] ck,α . The operator γk,n physically represents the creation of an electron in a band
n with crystalline momentum k, furthermore, this basis fully diagonalizes the Hamiltonian of the system
with spectra ϵk,n . The projector onto occupied bands P occ is defined by:

N occ X
†
X
P occ = γk,n |0⟩ ⟨0| γk,n . (2.1.5)
n=1 k

To obtain the Wannier functions, we need to project the position operator x̂ onto the occupied bands
(P occ x̂P occ ) and proceed with its diagonalizaiton. For this, we start by explicitly writing P occ x̂P occ :

X Nocc X
†
|0⟩ ⟨0| γk,n c†q+∆k,α |0⟩ ⟨0| cq,α γk†′ ,m |0⟩ ⟨0| γk′ ,m .
X
occ occ
P x̂P = γk,n (2.1.6)
q,α n,m=1 k,k′
 α
Substituting ⟨0| γk,n c†q+∆k,α |0⟩ = [(unk )∗ ] δk,q+∆k and ⟨0| cq,α γk†′ ,m |0⟩ = um
α
k′ δk′ ,q (see A.1) in equation
2.1.6 we have

X N occ X
† α
n ∗
X α
P occ x̂P occ = γk,n |0⟩ [um
k′ ] [(uk ) ] δk,q+∆k δk′ ,q ⟨0| γk′ ,m
q,α n,m=1 k,k′

X N
X occ
†  α  n
∗ α
= γq+∆k,n |0⟩ um
q uq+∆k ⟨0| γq,m
q,α n,m=1
N occ
†
X X
= γk+∆k,n |0⟩ unk+∆k |um
k ⟨0| γk,m
n,m=1 k
N occ
†
X X
= γk+∆k,n |0⟩ [Gx,k ]n,m ⟨0| γk,m . (2.1.7)
n,m=1 k

P  m α n ∗ α
To reach this result, the following relation was used unk |um
q = α uq [(uk ) ] ̸= δk,q δn,m which is
different from the usual orthogonality relation ⟨unk |um
k ⟩ = δn,m . P
occ x̂P occ is a single particle operator

that is fully defined by the matrix [Gx,k ]n,m = unk+∆k |um

k which is only unitary in the thermodynamic
limit, limN →∞ Gx,k = Fx,k . For discrete systems, one can make use of singular value decomposition
(SV D) to write Gx,k = U DV † so that we can define a unitary matrix Fx,k ≈ U V † =⇒ D → I, as in
the supplementary material of Ref. [49]. However, the matrix Gx,k quickly becomes unitary, with N ≥ 50
yielding a practically unitary matrix. In this manner, there is no need to perform an SV D decomposition

9
2 Methods 10

in practical cases. From this point on, it is assumed that Lx is big enough such that Gx,k is unitary or
that SV D was applied. The final form for the projected position operator is:

Nocc
†
X X
P occ x̂P occ = γk+∆k,n |0⟩ [Fx,k ]n,m ⟨0| γk,m . (2.1.8)
n,m=1 k

To diagonalize this operator, consider the following equation (P occ x̂P occ ) ψ j = E j ψ j which can be
written in matrix form:
  j  j
0 0 0 . . . FkN vk 1 vk1
    
 Fk1 0 0 ... 0   vk 2   vk2 
    
 = E j  vk3
 0 Fk2 0 ... 0  v
 , (2.1.9)
  
   k3
 . .. .. . . ..  .  .
 ..   ..  ..
 
 . . . . 

 


0 0 0 ... 0 vkN vkN
where Fki are Nocc × Nocc matrices and νki are vectors. The previous matrix equation can be rewritten
as a system of N -coupled equations:

j j j
FkN vkN = E vk1




F v j = E j v j

k1 k1 k2
. (2.1.10)


...


F j j j
kN −1 vkN −1 = E vkN


Iterating this set of equations one obtains:

N
FkN FkN −1 (...)Fk2 Fk1 vkj 1 = E j vkj 1

N
Wx,k vkj = E j vkj . (2.1.11)

Thus, a new unitary operator is defined, the Wilson loop operator (Wx,k ) which sweeps the FBZ along
the x direction. Since the matrices Fx,k are unitary, so is the Wilson loop operator, implying that its
eigenvalue problem can be written as:

j
E E
j j
Wx,k vx,k = e2πiνx vx,k , (2.1.12)

where the phases νxj are defined mod 1. It is also clear that these phases do not depend on the starting
j
k. However, the same cannot generally be stated about the corresponding eigenvectors vx,k . From
the Wilson loop operator phases, one can obtain the eigenvalues of the projected position operator,

N j j
noting the following correspondence: E j = e2πi(νx +m) =⇒ E j = ei∆k(νx +R)1 . The phases νxj are
called the Wannier centers. Since these arguments have been derived using reciprocal space, assuming
PBC, this is valid for each unit cell and is related to a bulk property of the system. In this manner,
1
N is the number of crystalline momenta in the considered direction N = Lx and since our objective is to apply this to
2D systems we make the substitution of m by R where R is a lattice vector R = ma with a = 1.

10
2 Methods 11

E j
E E
j j j
the operator P occ x̂P occ diagonalizes as (P occ x̂P occ ) ψR = ei∆k(R+νx ) ψR , where ψR are called the
Wannier functions:

Nocc X
j
E 1 X †
h
j
in
ψR =√ γn,k |0⟩ vx,k e−ikR (2.1.13)
Lx n=1 k

with j = 1, ..., Nocc . The Wannier

D functions
E are localized functions in real space that form an orthonormal
basis of the Hamiltonian, i.e. i |ψ j
ψR = δR1 ,R2 δi,j . As the Wannier centers νxj represent the shifts of
1 R2
electrons with regard to the background positive charges inside the unit cells, the electronic contribution
to polarization is proportional to the sum of the Wannier centers. Each individual electron that is shifted
by νxj provides a contribution proportional to νxj thus the bulk polarization is:

N
Xocc

p=e νxj mod e. (2.1.14)

j=1

Since νxj are the phases of the Wilson loop operator, one can write:

ie
p=− log (det [Wx,k ]) mod e. (2.1.15)
2π
Furthermore, in the thermodynamic limit, defining the Berry connection [A(k)]n,m = −i ⟨unk |∂k um
k ⟩ yields:

ie  h ´ k+2π i
−i k A(k)dk
p = − log det e mod e (2.1.16)
2π
ˆ k+2π
e
=− Tr {A(k)} dk mod e, (2.1.17)
2π k

which is the expression of polarization in the modern theory of polarization [66].

2.1.2 Polarization in 2D crystals

The formalism presented in 2.1.1 can be generalized for 2D crystals. Most of the algebraic and analytical
arguments presented previously follow in the 2D case. However, one starts by noticing that the projected
position operator now has an additional index ky :

Nocc
†
X X
P occ x̂P occ = γk+∆k x ,n
|0⟩ unk+∆kx |um
k ⟨0| γk,m . (2.1.18)
n,m=1 k

This operator is diagonal in ky and therefore their eigenstates can be labeled by ky . Each block of this
matrix diagonalizes as in the 1D case. Thus, the Wannier functions are:

Nocc X
j
E 1 X †
h
j
in
ψR x ,ky
=√ γn,k |0⟩ vx,k e−ikx Rx , (2.1.19)
Lx n=1 k
x
 
i j
where ψR x ,ky
|ψR ′ ′ = δRx ,R′ δi,j δky ,k′ . The Wilson loop operator is defined exactly in the same way.
x ,ky x y

However, it also depends on ky . Its eigenvalue equation becomes:

11
2 Methods 12

2πiνxj (ky )
E E
j j
Wx,k νx,k =e νx,k . (2.1.20)
E
j
Just as in the 1D case, the Wilson loop eigenvectors νx,k depend on the starting value of k. However,
its phases νxj (ky ) now depend on ky . The contribution to polarization follows in the same manner as in
the 1D case, yielding:

X
px (ky ) = e νxj (ky ) (2.1.21)
j
ie
px (ky ) = − log (det [Wx,k ]) mod e. (2.1.22)
2π

To obtain the bulk polarization, we need to sum over all allowed ky such that:

1 X
px = px (ky ) (2.1.23)
Ly
ky

2.1.3 Wannier Bands and Nested Wilson loop

Now that we know how to compute the bulk polarization of a 2D crystal, we turn our attention to
higher-order bulk multipole moments, such as the quadrupole moment density. It has been shown that
Wx,k = eiHWx (ky ) where HWx is adiabatically connected to the physical boundary Hamiltonian [64]. It
follows that the topological properties of HWx are the same as the true boundary Hamiltonian. This is
where the nested Wilson loop approach reveals its potential. We can, for instance, start by performing
a regular Wilson loop along x and then perform a second Wilson loop over the Wannier bands. In
this manner, the second Wilson loop characterizes boundary topology. As stated, a system with a bulk
quadrupole moment qxy has edge polarizations with qxy = px = py . Classically, this can be done projecting
the tensor qij onto each boundary.
All the calculations necessary to reach the end of this section are in A.2, following the work presented
in [49, 67].The phases νxj (ky ) of the first Wilson loop define a new band structure adiabatically connected to
that of the edge Hamiltonian, being defined in the BZ with ky = [−π, π[. To have a nontrivial topological
phase defined by a bulk quadrupole moment, one needs a boundary Hamiltonian with a gapped band
structure, this way one defines the Wannier sectors νx− and νx+ that are the set of Wannier bands such
that:

νx− = νxj (ky ) : ∀j νxj (ky ) below the gap



νx+ = νxj (ky ) : ∀j νxj (ky ) above the gap .



With the band structure of the edge defined, the objective is to perform a second Wilson loop in the y
direction. The reasoning is analogous to what was previously presented. We define the projector that
spans the occupied Wannier sectors as:

12
2 Methods 13

NW X ED NW Nocc X h in h ∗ i m
j j † j j
X X X
Pνx = ψR x ,ky
ψR x ,ky
= γ n,k |0⟩ v x,k v x,k ⟨0| γm,k , (2.1.24)
j=1 Rx ,ky j=1 n,m=1 k

where NW is the number of occupied Wannier sectors. To study the topological properties of the boundary
Hamiltonian, one needs to project the position operator ŷ into the occupied Wannier sector (Pνx ŷPνx ).
The position operator ŷ has an analogous expression to eq. 2.1.1. Thus, it takes the form:

c†k+∆ky ,α |0⟩ ⟨0| ck,α .

X
ŷ = (2.1.25)
k,α

To simplify notation and calculations, the following definitions are used:

E N occ h in
j j
X
ωx,k = vx,k |unk ⟩ (2.1.26)
n=1
h in,m NW h in D E h ∗ i m
j j
X
νx i i
Fy,k = vx,k+∆k y
ωx,k+∆ky
|ωx,k vx,k . (2.1.27)
i,j=1

Then, Pνx ŷPνx has the form:

Nocc h in,m
†
X X
νx
Pνx ŷPνx = γn,k+∆k y
|0⟩ Fy,k ⟨0| γm,k . (2.1.28)
n,m=1 k

Note that this operator is diagonal in kx . To diagonalize it, one proceeds exactly the same way as
presented in the previous section, which leads to the definition of the second Wilson loop:

h im,n NW ii,j h ∗ i n
νx
X  i
m h νx j
Wy,k = vx,k W̃y,k vx,k , (2.1.29)
i,j=1

νx QLy νx
where W̃y,k = j=0 F̃y,k+j∆kx is the Wilson loop performed in the y direction over the Wannier basis and
h ii,j D E
νx i j νx
F̃y,k = ωx,k+∆k y
|ωx,k . In fact, we only need to work with W̃y,k since it is connected by a unitary
νx
transformation to Wy,k (we further discuss this in A.2). To compute the Wannier sector polarizations,
νx
i.e., the boundary polarization, we diagonalize the Wilson loop W̃y,k such that it obeys the following
eigenvalue equation:

νx ,j
E E
νx νx ,j νx ,j
W̃y,k ν̃y,k = e2πiν̃y (kx ) ν̃y,k . (2.1.30)

Then the boundary polarization contribution from each kx is:

ei h
νx
i
pνyx (kx ) = − log det W̃y,k mod e, (2.1.31)
2π
and the total polarization of the Wannier sector νx :

13
2 Methods 14

1 X νx
pνyx = py (kx ). (2.1.32)
Nx
kx
h in,m
Furthermore, in the thermodynamic limit, defining the Berry connection over the Wannier sector Ãνy,k
x
=
D E
n |∂ ω m
−i ωx,k ky x,k yields:

ˆ
e n o
pνyx =− Tr Ãνy,k
x
d2 k mod e. (2.1.33)
(2π)2 BZ

Final Remarks
Until now, we have explored how Wilson loops emerge from the projected position operator and, con-
sequently, how polarization is connected to these operators. For the reader not as versed in topology,
the connection between Wilson loops and topological indexes might not be clear. The last Wilson loop
defines a topological Z index of the boundary, the winding number (ν).

i h
νx
i
ν = − log det W̃y,k (2.1.34)
π
The winding number effectively fixes the polarization (p = ν/2 mod 1). Furthermore, topologically, it
counts the number of zero-energy boundary states per boundary that appear when the boundaries of the
system are opened. In this manner, computing the winding number for each direction (νx and νy ) tells
us the total number of zero-energy boundary states2 and thus defines bulk topology.

2.2 Real Space Multipole Operators

Wilson loops are an efficient way of calculating the topological properties of a system. However, its
validity holds if certain conditions are met, such as translation invariance or the presence of a spectral
gap. Moreover, Wilson loops are computationally more complex in the absence of translational symmetry
and fail when reflection symmetries commute since there is no Wannier gap [49]. Another way to compute
multipole moments in crystals is through their classical definition and, using Resta’s argument, construct
the quantum mechanical equivalent. Resta in Ref. [65] starts off with the intent of computing ⟨X⟩. As it
turns out, ⟨X⟩ equals the total dipole moment of the system in units of e (electric charge):
D E L 2π P
X̂ = Im log ⟨Ψ0 | ei L x xn̂(x) |Ψ0 ⟩ mod L, (2.2.1)
2π
D E ´
where n̂(x) = c†x cx . One might be tempted to argue that X̂ = dx x n(x) is the correct definition.
However, in a PBC system, the wave function obeys:

ψ (x1 , ..., xi , ..., xN ) = ψ (x1 , ..., xi + L, ..., xN ) (2.2.2)

D E D E D E
and thus X̂ is not invariant under translations. Under a translation of L, X̂ → X̂ + L. Thus, in
systems with Hilbert space defined by eq. 2.2.2 this operator is ill-defined. Using the proposed average
2
2 (νx + νy ) in 2D

14
2 Methods 15

D E
value with the exponential makes X̂ invariant since the ground states are invariant under a translation
of L and the exponential is defined mod L. In Refs. [68, 69] these arguments are generalized to obtain
higher-order multipole moments such as the quadrupole moment density per unit cell. These are defined
as:

1 P
p̃x = Im log ⟨Ψ0 | e2πi x p̂x |Ψ0 ⟩ mod 1 (2.2.3)
2π
1 P
q̃xy = Im log ⟨Ψ0 | e2πi r q̂xy (r) |Ψ0 ⟩ mod 1, (2.2.4)
2π
xn̂(x) xyn̂(r)
where p̂x = Lx , q̂xy (r) = Lx Ly and |Ψ0 ⟩ is the many-body ground state of the system. This approach
full knowledge of the many body ground state of the system, since the exponential operators are many-
body operators. The question that remains
  is how do we implement this method in a crystalline system.
†
It is well known that ⟨Ψ0 |Ψ0 ⟩ = det Uo Uo is a Slater determinant in the absence of interactions, where
U0 in real space takes the form:
 (1) (2) (N ) 
φr1 φr1 . . . φr1 occ
 (1) (2) (N )
 φr2 φr1 . . . φr2 occ


Uo =  .
 .. .. .. , (2.2.5)
 .. . . .


(1) (2) (N )
φrN φrN . . . φrNocc
P
with each column a single particle state. In the real space basis, e2πi x p̂x is very straightforward since it
is diagonal, thus:

2π 2π
P E
⟨x1 , x2 , . . . , xN | ei Lx x xn̂(x)
|Ψ⟩ = e Lx i(x1 +x2 +···+xN ) ⟨x1 , x2 , . . . , xN | I |Ψ⟩ = ⟨x1 , x2 , . . . , xN | I Ψ̃ ,
(2.2.6)
2π P 2π P
E
iL xn̂(x) i xn̂(x)
with Ψ̃ = e x x |Ψ⟩. In this manner, e Lx x can be interpreted as the matrix D1 =
 2πix /L N D E  
†
P
diag e , which leads to ⟨Ψ | e 2πi x p̂x |Ψ ⟩ = Ψ |Ψ̃ = det Uo o , where Ũo is defined by
Ũ
j x
j=1 0 0 0 0
D1 Uo . The same argument can be used for the quadruple moment in 2D. The only change that is required
N
is the that D2 = diag e2πixj yj /(Lx ly ) j=1 . Following this, the multipole moments can be computed via:


1  
p̃x = Im log det Uo† D1 Uo mod 1 (2.2.7)
2π
1  
q̃xy = Im log det Uo† D2 Uo mod 1. (2.2.8)
2π

For the quadrupole moment, it is hard to reach large system sizes, since this method implies diagonalizing
the single particle H that has the dimension of the Hilbert space norb Lx Ly . In Refs. [69, 70, 71] the
authors argue that these definitions need corrections. Thus, the polarization and quadrupole moment are
defined as:

15
2 Methods 16

 
px = p̃x − p(0)
x mod 1 (2.2.9)
 
(0)
qxy = q̃xy − qxy mod 1m (2.2.10)

(0) (0)
where px and qxy correspond to the background positive charge contributions:

N
X xj
p(0)
x = mod 1 (2.2.11)
L
j=1
N
(0) 1 X xj yj
qxy = mod 1. (2.2.12)
2 Lx Ly
j=1

For the present case we will be focusing our attention on the BBH model, which is a 2D model with 4
orbitals per unit cell, and the SSH model, which has 2 orbitals per unit cell both at half filling N = 2Nocc .
Thus, in these particular cases the background contributions are:

(0) Lx +1 (0) (Lx +1)(Ly +1)

px = 2 mod 1 qxy = 2 mod 1 , (2.2.13)

which yield an additional phase factor depending on the parity of Lx , Ly . An insulator at half filling
that preserves chiral symmetry obeys 2qxy = 0 mod 1. In Appendix B, we show that chiral symmetry is
responsible for this quantization.

2.3 Localization Methods

2.3.1 Inverse Participation Ratio (IPR) and Scaling
A detailed characterization of localization properties is fundamental when studying condensed matter. A
simple way of tackling this problem is through an analysis of the inverse participation ratio (IPR) and its
behavior with system size. Consider a state |ψ⟩, for a given basis {|ϕn ⟩} the IPR is defined as:

|⟨ϕn |ψ⟩|4
P
IPR = P n 2 . (2.3.1)
2
n |⟨ϕn |ψ⟩|

Any basis can be considered in IPR calculations. Nonetheless, in the present context, we are more
interested in the real-space and k-space localization properties. States are distinguished according to
three types of localization: extended, critical and localized. Consider a system with volume N . An
extended state with a homogenous distribution has |⟨r|ψ⟩|2 = 1
N, such that IPR = 1
N. On the other
2
hand, for a fully localized state |⟨r|ψ⟩| = δr,0 , which yields IPR = 1. So it is clear that the IPR has a
dependency on system size, and thus in the thermodynamic limit, delocalized states have an IPR going
to zero and localized states have their IPR remaining constant. With this in mind, we introduce the idea
of scaling. The best way of concluding if a state is localized or extended with IPR is through the analysis
of its dependency on system size, N ∝ LD , where D is the dimension of the system. In this manner, we

16
2 Methods 17

propose that IPR = aL−D2 , such that:


D =D extended
 2


D ∈]0, D[ critical , (2.3.2)
 2

D = 0

localized
2

where D2 is the fractal dimension. Numerically, in order to estimate D2 the IP R is computed for a set of
states on a given energy window for various L. Then a fit to the data is performed, estimating a and D2 .
As stated, the calculation of the IPR can be done on any basis of our choice, always keeping in mind
that the localization properties inferred from said IPR calculation are relative to the chosen basis. With
this in mind, it is usually relevant to compute the IPR in reciprocal space, IPRk . Here, one usually finds
two different regimes. For localized states in k-space we are in a ballistic regime, while for extended
states in k-space and in real space the system is in a diffusive regime. Due to the Heisenberg uncertainty
principle, a given state |ψ⟩ cannot be simultaneously localized in real and reciprocal space. Furthermore,
since in the clean limit the solutions to the Schrödinger equation are Bloch waves with well-defined k
(which are localized functions in reciprocal space), their IPR is compatible with that of an extended state.

2.3.2 Transfer Matrix Method (TMM)

Here TMM is presented [72, 73], which allows us to obtain the localization length of the electrons wave
functions at a given energy E. Its implementation on 2D systems enables the classification of electronic
states as localized, extended or critical. In this thesis, it is applied in the context of the study of the
localization properties of the BBH model under disordered potentials, either to study gap edge properties
or to cross-check gap and topological results. This method is of general implementation. Nonetheless,
here we stick with the implementation in a square lattice geometry, that will be employed afterward.
For implementing this model in a 2D system, we start by defining a system with NL and M unit
cells along the longitudinal and transverse directions, respectively, keeping in mind that NL >> M . Its
implementation on a square lattice with nearest neighbors (NN) hoppings is straight-forward. We choose
the x direction as the transversal one and the y direction as the longitudinal one. A tight binding model
with hoppings between NN unit cells has a tridiagonal form:

NL X
M  
ij ij
X
H= |n, i⟩ hij
n ⟨n, j| + |n + 1, i⟩ V n+1,n ⟨n, j| + |n − 1, i⟩ V n−1,n ⟨n, j| , (2.3.3)
n=1 i,j=1

where the state |n, i⟩ belongs to the local site basis with i being the orbital index in the n-th strip and hij
n
is the n-th stripe internal matrix elements including NN hoppings and onsite energies. To further simplify
†
notation, notice that hn represents the Hamiltonian of the nth stripe and Vn+1,n = Vn−1,n = Vn represent
the couplings between consecutive stripes. All of these matrices have dimensions norb M × norb M . In this
manner, the Hamiltonian in matrix form is:

17
2 Methods 18

V1†
 
h1 0 0 0 ··· 0
V2† ···
 

 V1 h2 0 0 0 


 0 V2 h3 V3† 0 ··· 0 

V4†
 
H= 0 0 V3 h4 ··· 0 . (2.3.4)
.. .. .. ..
 
.. .. ..
. . .
 

 . . . . 

†

 0 0 0 ··· VNL −2 hNL −1 VN L −1


0 0 0 ··· 0 VNL −1 hNL

Consider an eigenstate |ψ⟩ of H with energy E. The eigenstates |ψ⟩can be expanded on a local site basis
|ψ⟩ = n,i ψni |n, i⟩, and thus:
P

X ij j
Xh ij j
i
Eδij − hij

j
⟨n, i| H |ψ⟩ = E ⟨n, i|ψ⟩ ↔ Vn,n+1 ψn+1 = n ψn − V ψ
n,n−1 n−1
j j

↔ Vn,n+1 ψ n+1 = (EI − hn ) ψ n − Vn,n−1 ψ n−1 , (2.3.5)

where in the last step we have written the equation in matrix form. We can rewrite this equation:

" #" # " #" #

Vn,n+1 0 ψ n+1 (EI − hn ) −Vn,n−1 ψn
= ↔
0 Vn,n+1 ψn I 0 ψ n−1
" # " #" #
−1 −1
ψ n+1 Vn,n+1 (EI − hn ) −Vn,n+1 Vn,n−1 ψn
↔ =
ψn I 0 ψ n−1
" # " #
ψ n+1 ψn
↔ = Tn
ψn ψ n−1
" # " #
ψ n+1 ψ1
↔ = Mn , (2.3.6)
ψn ψ0

where Tn is the a 2norb M × 2norb M matrix known as the transfer matrix of the system and Mn is the
product of all the transfer matrices from Tn to T1 . Mn satisfies the Oseledets theorem [74] which proves
the existence of the limiting matrix Γ, such that:

  1

MNL M†NL
2NL
Γ = lim . (2.3.7)
NL →∞

This limiting matrix has eigenvalues eγi , with i ∈ {1, 2, ..., 2norb M } and γi being the Lyapunov exponents
of the matrix MNL . If |νi ⟩ is an eigenvector of Γ, then Oseledets theorem implies that:

1
lim ln (|⟨νi | MNL |νi ⟩|) = γi (2.3.8)
NL →∞ NL

These exponents control the exponential increase (γi > 0) or decrease (γi < 0) of the wave function along
the longitudinal direction. The negative Lyapunov exponents should not contribute in the thermodynamic
limit since they should quickly “kill” the wave function amplitudes. Thus, we are interested on studying

18
 2 Methods 19

the positive exponents. For a given system size M , the Lyapunov positive exponents are the inverse of
the localization lengths λM . The smallest γi corresponds to the biggest localization length, and thus
the measured localization length of any given state is λM = min {γi }. As usual, these types of studies
are accompanied by scaling analysis, and TMM is no exception, such that the quantity of interest is
ΛM = λM /M . If ΛM goes to zero as M increases, the states at the considered energy are localized; if ΛM
grows with M , the states are extended; and if ΛM does not scale with M then the states are critical.
It appears we have all the tools to implement this method. However the nature of the Lyapunov
exponents presents some caveats. If we take the product of Tn matrices NL times, the ratio between the
biggest and smallest Lyapunov exponents would quickly be of the order of machine precision (10−16 ), and
thus we would only be capable of computing the largest exponent. To surpass this, we can perform a QR
decomposition of Mn each k products of Tn , where Q is an unitary matrix Q−1 = Q† =⇒ Q† Q = I

and R is an upper triangular matrix. Taking these successive decompositions, one arrives at:

NL NL k NL NL 2k
! ! ! !
Y Y Y Y Y Y
(1) (1) (1)
MNL = Ti = Ti Ti = Ti Q R = Ti Ti Q R(1)
i=1 i=k+1 i=1 i=k+1 i=2k+1 i=k+1
NL n
!
Y Y
= Ti Q(2) R(2) R(1) = ... = Q(n) R(i) = Q(n) R̃(n) , (2.3.9)
i=2k+1 i=1

NL Qn
where n = k and R̃(n) = i=1 R
(i) . Assuming that the Lyapunov exponents are in ascending order
γ1 < γ2 < ... < γ2norb M and through equation 2.3.8 it can be shown:

n
1 
(n)
 1 X  (n) 
lim ln R̃ll = lim ln Rll = γ2norb M −l−1 . (2.3.10)
n→∞ nk n→∞ nk
i=1

In this manner, we only need to store the diagonal entries of R(n) every time a QR decomposition is
made to compute all the Lyapunov exponents and then select the smaller positive one. The choice of k
needs to be tuned to each particular problem. For clean systems k = 10 is usually enough to not lose
information on min {γi }. However, for disordered systems, a smaller k might be needed3 . The choice of
NL can be made taking into account the convergence of λM . For this, we introduce the coefficients ci and
di :

(i) (1)
c1 = ln Rii
(i) (i) (j)
cj = cj−1 + ln Rii , j ∈ {2, ..., n}
(1) 2
 
(1)
d1 = ln Rii
(j) 2
 
(i) (i)
dj = dj−1 + ln Rii , j ∈ {2, ..., n} . (2.3.11)
n o
(1) (n)
The variance of random variables ln Rii , ..., ln Rii is given by
3
Usually k = 2 is enough.

19
2 Methods 20

v !2
u (i) (i)
u dn cn
∆n(i) =t − . (2.3.12)
n n

Since we are interested in studying the convergence of λ−1

M = min {γi } =
cn
we can compute the relative
nk
√
error of λ−1
M by using that the standard deviation of the average cn /n is ∆n / n, thus the relative error
of λ−1
M is:
√
∆n n
ϵ= . (2.3.13)
cn
Throughout this work, the the value for NL is chosen to be a minimum, such that ϵ < 1%.

2.4 Spectral Methods

2.4.1 Kernel Polynomial Expansion (KPM) and the Density of States (DOS)
Here we explore a well known method initially presented and implemented in [75] to compute the time
evolution operator U (t, t0 ). As it turns out, this method has applications in the computation of several
important quantities, such as the DOS of a system. It is of particular interest since it is far more efficient
than exact diagonalization, which allows us to consider systems closer to the thermodynamic limit, thus
removing finite-size effects.

Chebyshev Polinomials

The main trick of this method is that one can expand any analytical function in terms a of an power
series. We start by defining the Chebyshev polynomials Tn (x):

Tn (x) = cos (n arccos (x)) . (2.4.1)

At first glance, it is evident that these polynomials are defined in a given domain x ∈ [−1, 1]. These
polynomials obey the recursive relation:

Tn+1 (x) = 2xTn (x) − Tn−1 (x), (2.4.2)

and follow the orthogonality condition

ˆ 1
Tn (x)Tm (x) 1+δn,0
√ dx = δn,m . (2.4.3)
−1 π 1 − x2 2
In this manner, one can define any integrable function with f : ]−1, 1[ → R, as a series of Chebyshev
poynomials:

∞
X
f (x) = an Tn (x). (2.4.4)
n=0

To compute the coefficients an , we make use of 2.4.3 which yields:

20
2 Methods 21

ˆ 1
2 dx
an = √ Tn (x)f (x). (2.4.5)
1 + δn,0 −1 π 1 − x2
For practicality, the coefficients an can be rewritten at the cost of a new set of coefficients, yielding the
following expression:

∞
X 2∆n
f (x) = Tn (x) (2.4.6)
1 + δn,0
n=0
ˆ 1
dx
∆n = √ Tn (x)f (x). (2.4.7)
−1 π 1 − x2
With this in mind, for a given function f : [−1, 1] → R it is enough to compute ∆n to fully obtain its
Chebyshev polynomial expansion.

Expansion of Dirac Delta in Chebyshev Series

Since our goal is to compute the DOS using KPM, we need to know the form of δ(x) in Chebyshev
polynomials. For the Dirac delta function δ(x − ϵ̃), where ϵ̃ ∈ [−1, 1], it follows:

ˆ 1
dx
∆n = √ Tn (x)δ(x − ϵ̃)
2
−1 π 1 − x
Tn (ϵ̃)
= √ . (2.4.8)
π 1 − ϵ̃2

With 2.4.6 and 2.4.8 is trivial to conclude:

∞
!
1 X
δ(x − ϵ̃) = √ 1+2 Tn (ϵ̃)Tn (x) . (2.4.9)
π 1 − ϵ̃2 n=1

Chebyshev Polynomials and Operators

Functions of operators f (A) are well-defined when the respective functions have a power series expansion.
Using the recursive relation 2.4.2 it can be shown that Tn (x) is indeed a polynomial, thus the operator
Tn (A) is well-defined and is trivial in the basis that diagonalizes A. Keeping in mind that the Chebyshev
polynomials are only defined for x ∈] − 1, 1[, the eigenvalues of the operator A also need to fall within
this interval. In this manner, the objective is to compute moments of the type:

µm = ⟨ψ |Tm (A)| ϕ⟩ . (2.4.10)

The first two moments are trivially defined using T0 (x) = 1 and T1 (x) = x

µ0 = ⟨ψ |ϕ ⟩ (2.4.11)
µ1 = ⟨ψ |A| ϕ⟩ . (2.4.12)

21
2 Methods 22

We can then use the recursion relation 2.4.2 to compute the higher-order moments. Making use of the
notation |ϕm ⟩ = Tm (A) |ϕ⟩ , we can write:

|ϕm+1 ⟩ = 2A |ϕm ⟩ − |ϕm−1 ⟩ . (2.4.13)

Taking the bra with the previous equation, one gets the recursion relation for the moments:

µm+1 = ⟨ψ |2A| ϕm ⟩ − µm−1 . (2.4.14)

The same approach is valid if we want to compute moments of the form µB

m = ⟨ψ |BTm (A)| ϕ⟩. This
can be tackled if we define ψ B = ⟨ψ| B since µB B
m = ψ |Tm (A)| ϕ , which has the same form of the
previously considered moments.

Stochastic Evaluation of the Trace

Before we are able to use this formalism to compute the DOS of a system, we still need an efficient way
of calculating objects of the type Tr [f (A)]. As we know, we can expand the function f at the cost of a
Chebyshev series and so using the linear property of the trace, we only need to know how to calculate
P
Tr [Tm (A)]. Let {|ψi ⟩} be the base in which A is represented. Then Tr [Tm (A)] = i ⟨ψi |Tm (A)| ψi ⟩.
This is a terrible approach to compute the trace since for each term in the Chebyshev series we will
have to take the previous sum over all states {|ψi ⟩}, if A has dimension D, then this sum involves D
moments ⟨ψi |Tm (A)| ψi ⟩. There is a more efficient way of computing this trace, which starts by defining
the following random vector:

X
|r⟩ = ξir |ψi ⟩ , (2.4.15)
i

such that the coefficients ξir have the following properties:

D ′
E
⟨ξir ⟩ = 0 ξir ξjr = δr,r′ δi,j (2.4.16)

The easiest way of defining ξir is by considering a phase that is uniformly distributed in the interval
θ ∈ [0, 2π[, and then taking the complex exponential ξjr = eiθ . The trace of the operator can be estimated
by taking averages over different initial random vectors:

R−1
1 X r
Tr [Tm (A)] ≈ µm , (2.4.17)
R
r=0

where µrm = ⟨r|Tm (A)|r⟩. Thus, one just needs to define a random starting vector and then apply the
method explored previously to recursively compute the moments µrm . The advantage of this method is
that the required number of averages (R) is much smaller than the dimension (D) of the operator A,
R << D. Often, setting R = 10 is more than enough to estimate the trace. Furthermore, the bigger the
operator dimension D, the smaller the number of averages that are needed.

22
2 Methods 23

Density of States (DOS)

The DOS is defined as:

N
1 X
ρ(E) = δ(E − ϵn ) (2.4.18)
N
n=0

Where N is the dimension of the Hilbert space. We substitute the Dirac Delta by the Chebyshev series
expansion 2.4.9. Since we need E ∈] − 1, 1[, a rescale of the system Hamiltonian is performed.

(H−b) (E−b)
H̃ = λ Ẽ = λ
, (2.4.19)

where we choose b and λ such that Ẽ ∈] − 1, 1[:

(Emax +Emin ) (Emax −Emin )

b= 2 λ= 2−ϵ
. (2.4.20)
1
Scaling the energies and using δ(λx) = |λ| δ(x):

N
1 X  
ρ(Ẽ) = δ λẼ + b − λẼn − b
N
n=0
1 h  i
= Tr δ Ẽ − H̃
|λ|N
M
!
1 ∆0 (Ẽ)µ0 X
≈ + ∆m (Ẽ)µm , (2.4.21)
|λ|N 2
m=1

h  i
where ∆m (Ẽ) = 2T
√m (Ẽ) ωm
2
J and µ = Tr T
m m H̃ are computed using stochastic evaluation.
π 1−Ẽ
Numerically, we need to truncate the series to a finite number of M moments, in doing so, the so-called
Gibbs oscillations appear around where the expanded function is discontinuous. To prevent these types
J ) is introduced in the definition of ∆ . The Jackson Kernel is the common
of oscillations, a kernel (ωm m
choice when computing the DOS, which broadens the Dirac deltas as Gaussian curves. Its definition is
given below:
     
πm πm π
(M − m + 1) cos M +1 + sin M +1 cot M +1
J
ωm = . (2.4.22)
M +1

2.4.2 The Recursive method for computing the Local DOS (LDOS)
We have seen an efficient method to compute the DOS of a system using KPM. However there are other
quantities of interest closely related to the DOS. An example is the LDOS. Here we present an efficient
method to compute this quantity known as the recursive method [76, 77]. Let ϵn be an eigenvalue of H
with eigenstate |ϕn ⟩, then the LDOS ρ(R, E) in the local site basis |R, α⟩4 is:
4
where α represents internal degrees of freedom such as orbital index

23
2 Methods 24

X
ρ (R, α, E) = |⟨R, α|ϕn ⟩|2 δ (E − ϵn ) . (2.4.23)
n

The Green’s function is defined as usual by:

G(E) = (EI − H)−1 , (2.4.24)

and it contains information about the spectral properties of the system. A connection between the LDOS
2.4.23 and the local Green’s function exists. We can compute the latter projecting the full Green’s function
2.4.24 onto the local site basis |R, α⟩, thus:

G (R, α, E) = ⟨R, α |G(E)| R, α⟩ . (2.4.25)

This quantity relates to the LDOS through

1
ρ (R, α, E) = − lim Im {G (R, α, E + iη)} . (2.4.26)
π η→0
One needs to compute G (R, α, E) however, full inversion algorithms are inefficient for large enough
systems. To obtain G (R, α, E) we first need to define the Green’s function matrix. To achieve this more
efficiently, we employ a trick that will simplify our calculations further ahead. We start by computing a
basis where H is tridiagonal, this can be done via Lanczos decomposition, as presented in appendix C.1,
by setting the starting vector |u0 ⟩ = |R, α⟩, where |R, α⟩ is the site at which we want to compute the
LDOS. In the Jacobi basis (|un ⟩) the Green’s function is:

 −1
E − a0 −b1 0 ... 0 0
−b1 E − a1 −b2 ... 0 0
 
 
 
 0 −b2 E − a2 ... 0 0 
G(E) =  . (2.4.27)
 
.. .. .. .. .. .. 

 . . . . . . 

0 0 0 ... E − aN −2 −bN −1
 
 
0 0 0 ... −bN −1 E − aN −1

Since we chose |u0 ⟩ = |R, α⟩ the local Green’s function is G (R, α, E) = ⟨u0 |G(E)| u0 ⟩ = G0 which
clarifies the choice |u0 ⟩ = |R, α⟩. For each site, one needs to perform a Lanczos decomposition with a new
starting vector. In appendix C.2 a recursive form of computing the matrix element G00 when the inverse
matrix EI − H is in Jacobi form is presented. This allows us to reach a continued fraction to compute
G0 = G (R, α, E):

1
G0 = 1 , (2.4.28)
E − a0 − b21 E−a −b2 1
1 2
..
.
..
. − b2N −1 E−a1
N −1

where an and bn are the the coefficients obtained from Lanczos decomposition of the systems Hamiltonian
H. In practice, Lanczos decomposition cannot generally generate the full Hilbert space of our problem,

24
2 Methods 25

and for this reason, we can only compute a reduced amount of coefficients 0 < k < N . In this manner, we
truncate the continued fraction to the calculated coefficients. The LDOS in a UC |R⟩ is simply ρ (R, E) =
P
α ρ (R, α, E + iη) where the complex part approximates the delta Dirac function as a Lorentzian with
width defined by η. The recursive method provides an efficient and fast way of computing the LDOS of
a system.

2.5 Boundary Green’s Function

Previously, we presented a way of computing bulk multipole moments, such as the bulk dipole and
quadrupole moments. Although these are bulk properties, if we open the boundaries of the system,
boundary states appear in the gap. Some of these states are topologically protected and are directly
linked to these bulk quantities via bulk-boundary correspondence. There is interest in studying boundary
properties, however an effective boundary Hamiltonian is needed HBound . The boundary Green’s function
method [78] allows us to obtain HBound at a given energy.
The method yields a dimensional reduction approach of the system Hamiltonian. Just as in the case
with TMM, one considers a transversal direction with PBC and size M and a longitudinal direction with
OBC and size NL . The transversal direction can be converted to reciprocal space via Fourier transform
or can be left in real space representation which yields higher dimensional matrices. Let us assume that
our TB Hamiltonian can be written in tridiagonal form.

N h i
†
X
H= ψn† hn ψn + ψn+1 Vn ψn + ψn† Vn† ψn+1 (2.5.1)
n=1

The Hilbert space dimension is norb NL M × norb NL M . The matrix Vn is the matrix that couples con-
secutive transversal stripes of the system, while hn is the Hamiltonian of the nth stripe. In matrix form,
the Hamiltonian is tridiagonal:

V1†
 
h1 0 0 0 ··· 0
V2† ···
 

 V1 h2 0 0 0 


 0 V2 h3 V3† 0 ··· 0 

V4†
 
H= 0 0 V3 h4 ··· 0 . (2.5.2)
.. .. .. ..
 
.. .. ..
. . .
 

 . . . . 

†

 0 0 0 ··· VN −2 hN −1 VN −1


0 0 0 ··· 0 VN −1 hN
The Green function of the system is :

G(E) = (EI − H)−1 . (2.5.3)

The boundary Green’s function of the system is GN (E) = GN N (E) which is the N th block of the full
Green function of the system. Due to the simple Jacobi form of the Hamiltonian, one can take a blockwise
matrix inversion analogous to C.2 to reach a Dyson equation along the longitudinal direction:

25
2 Methods 26

 −1
†
Gn (E) = gn−1 − Vn−1 Gn−1 Vn−1 , (2.5.4)

where gn−1 = EI − hn is the Green’s function of the nth stripe. This recursion can also be written as
a continued fraction, as in eq. 2.4.28. Numerically, this equation converges to GN (E), and so we can
iterate it for a given size NL . In this manner, equation 2.5.4 should reach a fixed point defined by

−1
G = g −1 − V GV † . This fixed point can be computed analytically for simple systems such as the
SSH model by defining the β function, β(Gn (E)) = Gn (E) − Gn−1 (E) as in Ref. [78]. The zeros of the β
function yield the fixed points and thus the boundary Green function. The derivative of the β function
yields the stability of the given fixed point. After computing GN (E), the effective Hamiltonian of the
boundary is trivially defined by:

HBound = − (GN (E = ϵF ))−1 , (2.5.5)

where we have set the E = ϵF since we are interested in topological states that live at Fermi level.
Nonetheless one can take E to be any energy.

2.6 Numerical Simulations on Quasi-Periodic Systems

Simulating incommensurate systems is more complex than simulating disordered ones. To this end, we
explore some of the tricks and details relevant to having coherent and accurate results. For the sake of
simplicity, throughout this chapter, we consider a 1D system with a = 1 and with the quasi-periodic
potential of the form:

W
VW = cos (2πβm + ϕx ) , (2.6.1)
2
where W is the strength of the potential, β the spatial frequency, m the unit cell index, and ϕx the phase
shift. In general, for more complex quasi-periodic potentials, the arguments presented below should hold.
These types of potentials can be taken to be onsite energies or hopping’s modulation.

Choosing System Size (L)

The main purpose of a quasi-periodic modulation is to break translational invariance, which can only be
achieved by choosing β to be an irrational number. For β ∈ Q the desired results is not achieved since,
M
for any rational number z = N, a new unit cell of size N can be chosen, thus preserving translational
invariance. These types of systems are labeled commensurate or periodic systems. We can achieve an
incommensurate modulation by taking β to be any irrational number,
√
and in this thesis one of the most
1+ 5
common values in literature is used, the golden ratio, ϕGR = 2 . However, tackling this problem in
finite systems goes beyond setting β = ϕGR . To guarantee that the system remains incommensurate and
in order to implement periodic boundary conditions without any boundary defect for any system size L,
it is common to approximate ϕGR by a ratio of two co-prime numbers q, p. Luckily, there is a famous
mathematical sequence whose ratio of consecutive numbers Fn and Fn+1 approaches ϕGR monotonously5
5
If this condition is not met, one might have defects when scaling analysis is performed, since when L is increased to
reduce finite-size effects, the overall approximation of our incommensurate systems might worsen due to the ratio drifting

26
2 Methods 27

and whose any two consecutive pair of numbers, Fn and Fn+1 , are co-prime. This is the Fibonacci sequence
2.1. To this end, we set L = Fn and β = Fn+1 /Fn which avoids boundary defects and guarantees that
the system remains incommensurate as we increase system size.

F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 ...

1 1 2 3 5 8 13 21 34 55 89 144 233 377 610 987 1597 ...

Table 2.1: First 17 numbers of the Fibonacci sequence, each new number is obtained from the sum of the
two that precede it.

Averaging over Phase Shifts

Nothing has been said regarding the potential phase shifts (ϕx ). At first glance, the results should be
dependent on any particular choice of ϕx , however, in the thermodynamic limit, the system should not
depend on the choice of the phase shift. To motivate this statement. let us consider an infinite system with
an observer in the middle of the lattice (R = 0). For a given value of ϕ0x , the observer sees a particular
configuration of the potential. Now, changing ϕ0x → ϕ1x the observer sees another configuration. We can
translate the system (or the observer) to a new site, where he will see the initial configuration. In this
manner, for a finite system with size L we average over phase shifts, but as the system approaches the
thermodynamic limit, the dependence on ϕx vanishes. Thus, we choose ϕx to be uniformly distributed
in the range [0, 2π[. Furthermore, for finite systems, extended states are not usually dependent on the
choice of ϕx , while critical and localized states are more susceptible to local changes in the potential.

Averaging over Phase Twists

Beyond phase shifts, there is another quantity over which we can take averages: phase twists. To apply
phase twists, the boundaries are periodically closed (as for periodic boundary conditions), but with an
additional twist, such that:

ψα (R + Lai ) = eiθi ψα (R), (2.6.2)

where ψα (R) = c†R,α |0⟩. To implement this in a tight binding system, we add the term eiθi to the matrix
elements at the boundary (|n = 0⟩ ⟨n = L − 1|) which is equivalent to spreading this contribution over
all intra-UC hopping as follows t → teiθi /Li . As it turns out, having the first or the latter is nothing
more than choosing a particular gauge, with the latter being more intuitive. As the system approaches
the thermodynamic limit, any dependence on phase twists should vanish due to limLi →∞ eiθi /Li → 1. In
contrast to phase shifts and for finite system sizes, extended and critical states are usually dependent on
the choice of θi , while localized states remain unaffected. For these reasons, as in the previous case, we
can also perform averages over phase twists by sampling θi from a uniform distribution defined in the
interval [0, 2π[.
away from ϕGR .

27
2 Methods 28

Phase Twists and Spectral Gap

Not always taking averages over phase twists is to our advantage. Phase twists are slightly more tricky
than phase shifts, especially when dealing with small system sizes and gap-dependent results. For a finite
periodic system of size L, the allowed k’s are ki = − Lπ + 2π


L n bi with n ∈ Z. Here, we have assumed
that the FBZ is centered in the Γ point. Introducing
 phase twists leads to shifts in the allowed k’s, such
π 2π θi
that if a twist θi is introduced then ki = − L + L n + L bi .
For the sake of simplicity, consider a 1D TB system with a gap that closes at kx = 0 (maximum
of the valence band), and that our target computation is the calculation of the spectral gap via exact
L
diagonalization (ED) of the Hamiltonian for a system with size L. If we set kn = 0, then it follows n = 2.
L
Since n ∈ Z, n = 2 is valid only if L is even. However, for odd L, kn = 0 is not an allowed crystalline
momentum, such that performing ED of the Hamiltonian for and odd L, will lead to an overestimation of
the spectral gap. This effect is particularly relevant for small L and for |∂kx ϵkx | ≫ 1. One way of tackling
2π
this problem is to consider bigger system sizes, since eventually L will be small enough that considering
even or odd L is no longer relevant.
Due to the high computational cost of performing ED for large system sizes, another approach is
required to tackle gap scaling in finite crystalline systems. We can add a twist that shifts all the allowed
kx ’s in such a way that, for odd L, kn = 0 is a possible value of crystalline momenta. This is achieved by
considering θx = π for odd L while for even L we use regular periodic boundary conditions6 . This trick is
also relevant even if a direct calculation of the spectral gap is not being performed. Gap-dependent results
as topological phases and TPT are sensible to gap closures, and thus for small systems, it is important
to make an appropriate choice of phase twists even when evaluating system topology. With this in mind,
throughout the main results of the thesis, it is always specified if averages over phase twists or a fixed
twist were considered.
The previous argument holds when translational invariance is preserved, although it is not clear
that in the absence of translational symmetry, using phase twists is advantageous. In fact, even for
some disordered systems, phase twists solve spectral gap finite-size effects. Quasi-periodic systems in
the low W regime maintain ballistic states at Fermi level, and so setting a non-zero twist can improve
gap calculations, especially when the gap closes in a ballistic-ballistic transition. In contrast, regular
uncorrelated disorder does not benefit from phase twists, since all of the electronic states are Anderson
localized (for low-dimensional systems).

6
which is the same as considering twists with θx = 0

28
Chapter 3

Quasi-Periodic modulation and

first-order Topology

Before tackling HOTIs it is logical that we first build up our intuition with a first-order topological
insulator. To this end, we focus this chapter on the analysis of a Chern insulator under quasi-periodic
modulation while maintaining brevity. This is original work that is available at arXiv [37].

3.1 Model
We consider a QBCP model in a square lattice with two orbitals per unit cell (UC), as in Ref. [79]. This
model is defined by the following clean limit k-space Hamiltonian:

Ψ†k Hk Ψk ,
X
H0 = (3.1.1)
k
 
where Ψ†k = c†k,A , c†k,B , with c†k,α the creation operator of a state with crystal momentum k in the α
orbital and

Hk = h(k) · σ, (3.1.2)

with σ the vector of Pauli matrices and the vector h(k) given by

B+1
hx = 2tx sin (kx ) sin (ky ) hy = 1 + 2 (cos (kx ) + cos (ky )) hz = 2tz (cos (kx ) − cos (ky )) . (3.1.3)

Throughout this chapter, the hopping amplitudes are fixed to unity (tx = ty = t = 1), so that every
physical observable with energy units comes in units of t.
For hy = 0, the system is gapless with quadratic band crossing points at the Fermi level. In contrast
with Dirac semimetals, QBCP systems exhibit a finite DOS at Fermi level (ϵF = 0). Moreover, these
new types of crossings display an interesting topological character since QBCP transport a total Chern
number of 1, unlike Dirac cones, which transport a total Chern number of 1/2. The introduction of
a non-zero hy perturbation opens a gap and breaks time reversal symmetry, which are two ingredients

29
3 Quasi-Periodic modulation and first-order Topology 30

2.5 2.5
E 0.0 E 0.0

−2.5 −2.5
−π −π
−π 0 −π 0
0 kx 0 kx
ky π π ky π π
(a) B = −1 and C = 0 (b) B = −3 and C = 2

Figure 3.1: Band structure for different values of topological mass B.

needed for the emergence of Chern insulating phases. For this model, a single QBCP occurs at M for
B = 0, and for B = −2 at Γ. These are the only values of B for which the system becomes gapless and
where topological phase transitions between phases with C = 2 ↔ C = 0 (B = −2) and C = 0 ↔ C = −2
(B = 0) take place. For trivial phases (C = 0), quadratic band crossing points (QBCP in Fig. 3.1a) with
symmetric berry curvatures (Fig. 3.2a) occur at Γ = (0, 0) and M = (±π, ±π) points of the FBZ. When
integrating over the entire FBZ, the symmetric contributions cancel each other out yielding C = 0, as
shown in Fig. 3.2a1 . For topological phases such as B = −3, the QBCPs only appear at either the Γ or
M points of the FBZ, as seen in Fig. 3.1b, which leads to C = 2 phases. This is justified by the fact
that the Berry curvature is either positive or negative throughout the entire FBZ, with most of its weight
concentrated around the QBCP, as seen in Fig. 3.2b.
To study the effect of quasi-disorder, which is our main focus, we consider the Aubry-André unidi-
rectional potential given by the real-space Hamiltonian,

W X
HW = cos (2πβn) c†R,α cR,α , (3.1.4)
2
R,α

where R = maêx + naêy is a lattice vector and β is the potential frequency. In the thermodynamic limit,
we choose β to be an irrational number in order to break translational invariance, in particular the golden
√
ratio (β = ϕGR = (1 + 5)/2) as previously discussed. The potential in Eq. 3.1.4 maintains translation
invariance along the x direction, and so one can perform the Fourier transform along this direction to
obtain Hkx , an Hamiltonian diagonal in Bloch momentum kx (see Appendix D).
1
B = −1 is used as an example, in fact for B ∈] − 2, 0[ the system is in a trivial phase, however the band structure and
the Berry curvature follows the same behavior.

30
3 Quasi-Periodic modulation and first-order Topology 31

×10−5 ×10−5
π π
2 5
π/2 π/2
1 4

Ω(k)

Ω(k)
3
ky

ky
0 0 0

2
−1
−π/2 −π/2
1
−2
−π −π 0
−π −π/2 0 π/2 π −π −π/2 0 π/2 π
kx kx

(a) B = −1 and C = 0 (b) B = −3 and C = 2

Figure 3.2: Berry curvature (Ω(k)) for different values of topological mass B and L = 1000. The scale of
the color maps differs between Fig. (a) and (b). In (a) the scale is symmetric, while in (b) the scale is
definite and positive.

3.2 Results and Discussion

We start by characterizing the topological phase diagram, shown in Fig. 3.3(left) in the plane of topological
mass B and intensity of the quasi-periodic potential W . The phase diagram was obtained for a system
with 21 × 21 unit cells averaged over 20 random phase twists with twisted boundary conditions.
The clean limit presents topological phase transitions from C = ±2 to C = 0, with the gap closing
at a QBCP (B = {−2, 0}). These transitions are robust for small W , still occurring at B = {−2, 0}.
The observed behavior contrasts with the case of uncorrelated disorder, studied in Ref. [79], where it
was shown that for infinitesimal disorder strength W , C = ±1 phases appear. In the present case of an
incommensurate potential, C = ±1 phases only appear at large enough potential strengths, indicating
a rather different phenomenology. Interestingly, along the lines of fixed B = −2, 0, where the QBCP
occurs, we see the appearance of very small C = ±1 bubbles, indicated by the arrows in Fig. 3.3. If we
keep increasing W , topological phase transitions occur into C = ±1 phases. These phases are not present
in the clean limit, which shows that a quasi-periodic potential can induce topological phases of entirely
new character. As we will see in the following chapters, the same is true about HOTIs. For large enough
potential strength, all topological phases are suppressed, and a trivial phase is eventually reached. Unlike
in the disordered case, where all the C = ±1 phases are gapless, there are both gapped and gapless
C = ±1 phases separated by the black lines in Fig. 3.3.
Regarding spectral properties, we can see in Fig. 3.3(right) that the gap closes and reopens in most
of the topological phase transitions. A small gapless region for strong disorder is present (wider blackish
regions at high W ). This greatly contrasts with the case of Anderson disorder, where the novel C = ±1
phases are always gapless.
Besides relevant topological properties, quasi-periodicity is far richer than Anderson disorder in regard

31
3 Quasi-Periodic modulation and first-order Topology 32

10 10

8 8
C=1 C = -1
6 6
W

W
4 4
C=0
C=2 C = -2 3
2 Bubbles 2 2

Gap
1
0 0
−4 −3 −2 −1 0 1 2 −4 −3 −2 −1 0 1 2
B B

Figure 3.3: (left) Chern number as a function of B and W for a system size of L = 21 and 20 different
twist realizations. The grayed-out areas are gapless. (right) Energy gap as a function of B and W for
Ly = 1597. No twist realizations were considered. The gray lines represent the lines at which a topological
phase transition occurs.

to localization phenomena. To this end, TMM is a powerful tool to characterize each of the observed
phases. By sweeping the system at Fermi level (E = 0), we are able to catch gap opening and closing
transitions since TMM captures a diverging localization length due to the delocalized bulk states whenever
the gap closes, i.e., at TPT, as is evident from Fig. 3.4.

10

6
W

4
log10 (ξ)

2 5.0
2.5
0
−4 −3 −2 −1 0 1 2
B

Figure 3.4: TMM results at Fermi level (E = 0) at the kx where the gap closes. Obtained for a 1D system
size of N = 30000 such that ϵ ≈ 1%. The grid discretization is 300 × 300.

The localization properties were also found to be very distinct from the disordered case at small W .
In the presence of disorder, all the eigenstates are always localized, except for extended states that live
on very narrow energy windows. This contrasts with the quasi-periodic case, where considerably large
clusters of bulk states remain extended (top plots of Fig. 3.5). Furthermore, at large potential strengths, a

32
3 Quasi-Periodic modulation and first-order Topology 33

−2

−4
E

7.7 7.2 5.7

log10 (ξ)

log10 (ξ)

log10 (ξ)
−6
W: 3.5 W: 4.5 W: 5.5
-0.4 -0.4 -0.4
0
0.0 0.0 0.0
−2
−0.5 −0.5 −0.5
−4 4 4 4
E

2 2 2
5.0 5.0 5.0
−1.0 −1.0 −1.0
log10 (ξ)

log10 (ξ)

log10 (ξ)
−6 −π −π + 0.1 −π −π + 0.1 −π −π + 0.1
W: 8.0 W: 8.5 W: 9.0
-0.4 -0.4 -0.4
−π −π/2 0 π/2 π −π −π/2 0 π/2 π −π −π/2 0 π/2 π
kx kx kx

Figure 3.5: TMM bulk phase diagram at B = −3 for the entire range of the valence band, E ∈ [−7, 0]
and kx ∈ [−π, π]. The insets follow the same procedure for a smaller region (E ∈ [−1, 0] and kx ∈
[−π, −π + 0.2]) with a grid discretization of 600 × 600.

mechanism similar to the “levitation and annihilation” mechanism can be observed close to the transitions
from the C = ±1 phases into the trivial phase. In fact, in this large potential limit, the C = ±1 phases
can become gapless, with extended states only existing at very narrow energy windows (bottom plots of
Fig. 3.5). After the topological transition into the trivial phase, all states become localized, as in the
disordered case.
The topological and localization properties therefore show a mixture of features present in the ho-
mogeneous and disordered cases, as was also previously observed for quasi-periodic Chern insulators [?
]. The small potential results are also in agreement with Ref. [80], where a different QBCP system was
found to be robust to the application of a (in this case, fully two-dimensional) quasi-periodic potential.

33
Chapter 4

Benalcazar-Bernevig-Hughes (BBH)
Model in 2 dimensions

In this chapter, we turn to the study of HOTIs by introducing the paradigmatic BBH model, that will
be used extensively hereafter. Initially, we focus our attention on the clean limit as presented in Ref. [49]
elaborating on its topological properties. Lastly, we dive into the disordered limit and study how a chiral
symmetry-preserving disorder can induce novel topological quadrupole phases, as in Ref. [50].

4.1 Clean Limit

The BBH model is a 2D tight binding model with four orbitals per UC that describes spineless electrons
in a square lattice, as seen in Fig. 4.1.

Figure 4.1: Tight binding scheme of the BBH model. The dotted “connections” are negative in relation
to γ and λ.

The BBH model is defined by the following Bloch Hamiltonian:

Hk = [γ + λ cos(kx )] Γ4 + λ sin(kx )Γ3 + [γ + λ cos(ky )] Γ2 + λ sin(ky )Γ1 + δΓ0 , (4.1.1)

where Γ are 4 × 4 matrices (Γ0 = τ3 ⊗ σ0 , Γk = −τ2 ⊗ σk , Γ4 = τ1 ⊗ σ0 ) that define the internal degrees of
freedom within a unit cell. The matrixes τi and σi are Paulli matrixes. Throughout the rest of this thesis,

34
4 Benalcazar-Bernevig-Hughes (BBH) Model in 2 dimensions 35

we set the lattice constants as the length scale of our problem (ax , ay = 1). The system has 4 energy bands
p
(E(k) = δ ± 2λ2 + 2γ 2 + 2λγ (cos (kx ) + cos (ky ))) each of which is twofold degenerate and preserves all
the three discrete symmetries, C4 and mirror symmetries mx = τ1 ⊗σ3 , my = τ1 ⊗σ1 . The BBH model has
its topological phase protected by several distinct symmetries, such as C4 and mirror symmetries mx , my ,
as well as chiral symmetry. In this manner, as long as one of these symmetries is enforced, topological
quadrupole insulating phases can be observed. In Fig 4.2 the band structure is represented for different

2
2 2

E 0 E 0 E 0
−2 −2
−π −π −2 −π
−π 0 −π 0 −π 0
0 kx 0 kx 0 kx
ky π π ky π π ky π π

(a) γ/λ = −1.5 (b) γ/λ = −1 (c) γ/λ = −0.5

Figure 4.2: Band Structure of the 2D BBH model for different values of the hopping ratio γ/λ and with
δ=0

values of the hopping ratio γ/λ. At γ/λ = 1(−1) the system is a Dirac semimetal, with the gap closing
at the M (Γ) points of the FBZ. For |γ/λ| > 1 the system is in a gapped trivial insulating phase (qxy = 0)
and for |γ/λ| < 1 the system is in a gapped topological quadrupole insulating phase (qxy = 0.5).
By bulk-boundary correspondence, opening the boundaries of the system leads to the appearance
of boundary properties, thus, at the topological phase (|γ/λ| < 1), the BBH model displays zeroth-
dimensional boundary states entitled corner states as depicted in Fig. 4.3(left) in red. These corner states
give rise to localized fractional corner charges Qi = ±0.5 where i ∈ {1, 2, 3, 4} labels each corner and Qi
comes in units of elementary charge e, Fig. 4.3(right). Away from the topological regime, the trivial phase
(|γ/λ| > 1) is defined by qxy = 0, with an absence of zero-energy corner modes.
Regarding the electronic charge density, it displays a bulk homogeneous distribution ρe (r) = 2 that
is canceled by the contribution from the positive atomic charges. Therefore, integrating the total charge
density over one of the corner regions yields a fractional corner charge Qi = ±0.5. A small δ is considered
to break the hybridization of the fourfold zero-energy corner states, and the inter unit cell hopping is chosen
to be small (γ = 10−3 ) in order to resolve the corner states in Dirac deltas. Furthermore, approaching
the TPT by increasing γ leads to increasing localization length of the corner modes. For example, for
γ ≈ 0.5, the corner states span a few UC’s around the corner. From this point on, we shall set λ = 1.

Topology and bulk-boundary correspondence on the BBH model

From a classical perspective, a canonical quadrupole moment is composed of two canceling dipoles and
four corner charges. In fact, this perspective should hold whether the quadrupole moment has its origins
in a perfect canonical structure or if it emerges from quantum behavior. These properties are observed
in the quadrupole insulating (QI) topological phase of the BBH model (|qxy | = |pi | = |Qi | = 0.5).

35
4 Benalcazar-Bernevig-Hughes (BBH) Model in 2 dimensions 36

Electron Density
1 2.4
2.2
2.0
E/λ

0
1.8
1.6
−1
0
5
−2
15 10 y
−1.0 −0.5 0.0 0.5 1.0 10 15
x 5
γ/λ 0

Figure 4.3: (left) Energy Spectrum as a function of γ/λ obtained by ED for a system with open boundaries
and L = 30. The gap does not close properly at |γ/λ| = 1 due to finite-size effects. In red, the energy
corner modes. (right) Electron charge density for γ = δ = 10−3 .

This said, how can we characterize the topology of such types of topological insulators? For the
particular case of the BBH model, the simplest way of computing the bulk quadrupole moment is via
real-space methods, as in Sec. 2.2. However, there are more efficient ways of achieving the same result. The
Nested Wilson loop (A.2) is one of these approaches, allowing us to compute both boundary polarizations
and thus define a boundary invariant P = 4 |px py |. We can take advantage of the hierarchy of nested
topological insulators to characterize the topological properties of the bulk via the topological properties
of the boundary. Apart from the nested Wilson loops that require the existence of translational invariance,
there is an additional method capable of reaching the boundary. The boundary Green’s function method
tackled in Sec. 2.5 is a valid approach in reciprocal-space and in real-space, which makes it ideal to
implement in disordered systems. With this method, the authors in Ref. [78] obtained the boundary
Bloch Hamiltonian of the BBH model, given by:
!
0 γ + λe−ik λ(λ+αγ)
HBound = fk , fk = γ 2 +λ2 +2λγ cos(k)
. (4.1.2)
γ+ λeik 0
Aside from the pre-factor fk , the Hamiltonian is identical to that of the SSH model and is defined for
̸ 1.1 Thus, it can be shown that HBound has the same topological properties as the SSH model. Let
|γ| =
us prove the previous statement. Let uSSH
k be a Bloch eigenstate of HSSH (k) with energy ϵSSH
k that
obeys the following eigenvalue equation:

HSSH (k) uSSH

k = ϵk uSSH
k . (4.1.3)
1
For |γ| = 1 the system is a metal, and the Green’s function at zero-energy is not defined, which in turn leads to a
non-defined boundary Green’s function.

36
4 Benalcazar-Bernevig-Hughes (BBH) Model in 2 dimensions 37

0.5 1.0

0.4 0.8

0.3 0.6
qxy

P
0.2 0.4

0.1 0.2
(a) (b)
0.0 0.0
−2 −1 0 1 2 −2 −1 0 1 2
γ γ

Figure 4.4: (a) qxy as a function of the hopping γ computed via real-space methods for a system size of
L = 10. (b) P = 4 |px py | invariant computed via nested Wilson loops for a system size of L = 10.

The boundary Hamiltonian obtained analytically from the boundary Green’s function is HBound (k) =
fk HSSH (k) implying that [HBound , HSSH ] = 0. In this manner, uSSH
k is also an eigenvector of HBound (k)
with energy fk ϵSSH
k . Since the Bloch wavefunctions remain unchanged and following the definition of
topological indexes via Wilson loops as in Sec. 2.1.1, it is clear that the two models have the same
topology. Going from reciprocal-space into real-space, the factor fk , which is k dependent, introduces
long range hopping’s.
With either of those methods, we can compute the boundary polarizations px and py , thus defining
the boundary invariant: 
1 qxy = 0.5
P = 4 |px py | = . (4.1.4)
0 qxy = 0

This characterization holds for the BBH model but care is needed since there is indication that topological
phases can occur with px = 0, py = 0.5 and vice versa [81, 82], albeit we do not dwell such regimes.
To conclude this section, we make some finishing remarks. If our focus is on quadruple phases
protected by crystalline symmetries, the nested Wilson loop approach and the traditional Z2 invariants
computed by constructing the Pfaffian over high symmetry k-points (as in the Fu-Kane invariant, Ref. [83])
are possible invariants that define qxy . However, since we are interested in disordered chiral QIs, these
two approaches are not viable. The BBH model falls within the BDI (chiral orthogonal) symmetry class
[13] which has no Z topological index in 2 dimensions. However, such classification might not hold for
HOTIs since the tenfold way can only properly classify first-order topological insulators. More recently, it
has been shown that chiral-symmetric HOTIs have well-defined topological invariants entitled multipole
chiral numbers (MCNs), which are a generalization of the winding number to higher dimensions [84].
Furthermore, the authors show that phases with qxy = 0.5 are always topological, while phases with
qxy = 0 might be topological or trivial. This said, a complete picture of the topology of chiral QIs
can be accomplished not only by the characterization of the quadrupole moment using PBC but also
by opening the boundaries, studying the number of corner states in each regime, and examining the

37
4 Benalcazar-Bernevig-Hughes (BBH) Model in 2 dimensions 38

emergent boundary signatures. Overall, a chiral symmetric quadrupole topological phase is defined by
a quadrupole moment quantized to 0 or 0.5 as long as zero-energy corner modes are present, and their
number is divisible by the number of boundaries in the system (4 in 2D). Regarding the BBH model and
the disordered variants considered in this thesis, the phases with qxy = 0 are always trivial, and so we
refrain from computing multipole chiral numbers. It follows that in such cases, the quadrupole moment
acts as a topological invariant.

4.2 Chiral Disorder

In this section, we explore the work published in Ref. [50], where the authors study the effect of chiral
disorder in the BBH model. We replicate some results from the paper in question, which allow us to
test much of the methods we have introduced in Chapter 2, building up our intuition to later tackle
quasi-periodicity. In the original paper, the authors study a model of their creation that is connected via
a local transformation to the BBH model. The results that follow are for chiral disorder, and we use the
BBH model as the underlying TB system. As we will see, applying chiral disorder to the BBH model
yields identical results to the ones obtained by the authors of Ref. [50]. Before proceeding, let us note
that most of the graphs display vertical dotted lines that mark phase transitions, with dark blue lines
marking gap closures or openings and gray lines representing TPTs or localization transitions. These act
as guidelines throughout the discussion that follows.

Disordered hopping
In order to preserve chiral symmetry, onsite Anderson disorder cannot be introduced into the system.
With this in mind, the internal UC hoppings (γ) can be perturbed with Anderson disorder, ensuring that
chiral symmetry is preserved. In this manner, we take:

W
γ→γ+ αi , (4.2.1)
2
where W is the disorder strength and αi is sampled from a uniform distribution such that αi ∈ [−1, 1]
with i ∈ {1, 2, 3, 4}.

Main results
In Fig. 4.5(a), we show the quadrupole moment and the boundary invariant as a function of W and in
Fig. 4.5(b), we show the spectral gap (∆E) as a function of disorder strength W . For W = 0, the system
is in a trivial phase that is robust to increasing disorder strength (I). Eventually, the gap closes and
reopens at WcI→II ≈ 2.1. This behavior stands in stark contrast to Anderson disorder, which leads to
the permanent closure of the spectral gap. Into phase II, the boundary invariant plateaus at 1, which is
accompanied by the gap closing and reopening. Although the boundary invariant (P ) quantizes to 1 in the
topological phase (II), even for L = 200, it still suffers from finite-size effects. Regarding the quadruple
moment, it never quantizes, despite the clear scaling with increasing system size. As it turns out, the
absence of quantization is a consequence of small system sizes. To tackle this, the authors perform a
scaling analysis in these regions and conclude that, for big enough L, qxy → 0.5. Phase III is gapless but

38
4 Benalcazar-Bernevig-Hughes (BBH) Model in 2 dimensions 39

topological. To this end, we refer to phase II and III as higher-order topological Anderson insulators
(HOTAI).

I II III IV V I II III IV V
1.0
0.25
P
0.8 L: 10
L: 20
0.20
0.6 L: 30

∆E
qxy

0.15

0.4 0.10

0.2 0.05
(a) (b)
0.0 0.00
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
W W

Figure 4.5: (a) qxy and P as a function of W . The P invariant was computed for a system size of
L = 200 and averaged over 200 disorder configurations. qxy was obtained via real space methods, with
1000 disorder configurations for γ = 1.1, L = 10 and 20 while for L = 30 500 configurations were
considered. (b) Spectral gap (∆E) as a function of W for γ = 1.1, L = 200 and averaged over 150
disorder configurations.

Further increasing disorder induces a TPT at WcIII→IV ≈ 3.5, and the system reaches a Griffiths
regime (phase IV ), which displays “mixed topology”. Although for each disorder configuration the
quadrupole moment and boundary invariant remain quantized, their average yields an intermediary value.
At high W (phase V ) the system exhibits the same phenomenology of an Anderson insulating phase.

Spectral Properties

0.20 W 0.20
2.4
2.8
0.15 0.15
3.2
3.6
ρ(E)

ρ(E)

4.0
0.10 0.10
9.0

0.05 0.05

(a) (b)
0.00 0.00
−10 −5 0 5 10 −1.0 −0.5 0.0 0.5 1.0
E E

Figure 4.6: DOS as a function of E calculated through KPM for a system of size L = 1000, M = 5000
moments, R = 10 disorder and random vectors configurations for several values of disorder strength. (a)
Full DOS. (b) Zoomed-in view around Fermi level.

39
4 Benalcazar-Bernevig-Hughes (BBH) Model in 2 dimensions 40

Now we turn to the spectral properties under increasing W . In Fig. 4.6 we show the density of states
(ρ(E)) as a function of E obtained via KPM for several values of disorder strength. Just as Anderson
disorder, chiral disorder flattens and increases the bandwidth of the system. However, for certain values of
W a singularity at E = 0 appears, which is later killed, further strengthening the disorder, as in Ref. [85].
To confirm that II and III are topological, alongside a quantized quadrupole moment, the system
also needs to display zero-energy corner modes when the boundaries are opened. To this end, in Fig. 4.7
the local DOS at E = 0 computed for an open boundary system is plotted as a function of r for a W in
each of the topological phases. For either of the phases, there is a clear spectral weight concentrated at
the corner. Overall, the quantized quadrupole moment, boundary invariant, and the existence of corner
modes at these particular phases are enough to conclude that phase II and III correspond to gapped
and gapless HOTAIs respectively.

40 40 0.30
0.25
0.25
30 30
0.20
0.20
LDOS

LDOS
20 0.15 20
y

0.15
0.10
10 10 0.10
0.05
0 0 0.05
0 10 20 30 40 0 10 20 30 40
x x

Figure 4.7: LDOS for E = 0 as a function of position r = (x, y) calculated using the recursive method for
a system size L = 40. (left) W = 2.6 (phase II) and 500 disorder configurations. (right) W = 3.4 (phase
III) and 1000 disorder configurations.

Localization Properties
We now turn to the study of the localization properties of each phase. Implementing TMM at the Fermi
level is a powerful tool to tackle topological phase transitions and compute the localization length of
eigenstates at the Fermi level, in gapless regimes. In gapped regions, TMM generates localized wave
functions corresponding to evanescent waves that live in the gap. In Fig. 4.8, we plot the normalized
localization length (ξ/M ) at Fermi level (E = 0) as a function of W .
In the gapped phases I and II, TMM gives localized states. However, at the TPT, the normalized
localization length diverges, indicating the existence of extended states at Fermi level. In phase III, which
is gapless, the bulk Fermi level states are localized. This is relevant to ensure that a TPT does not occur,
since localized states at the Fermi level can hybridize with the corner modes without inducing a TPT.
Phase IV is a critical metal, which can be distinguished by the absence of scaling of ξ/M as M increases,
implying that the localization length is always of the order of the system size. These types of critical
metal regimes have been previously observed in systems with disordered hoppings [85, 86]. Eventually,

40
4 Benalcazar-Bernevig-Hughes (BBH) Model in 2 dimensions 41

for high enough disorder, the typical AI phase is reached (V ).

101 I II III IV V
M: 20
M: 40
M: 60
M: 80
100

ξ/M

10−1

0 1 2 3 4 5 6 7 8
W

Figure 4.8: Normalized localization length (ξ/M ) as a function of W at Fermi level, obtained with TMM
for γ = 1.1 for different transversal sizes M . The size of the longitudinal direction N was chosen such
that ϵ < 1%.

41
Chapter 5

Chiral Quasi-Periodic Quadrupole

Insulator (QPQI)

Finally, we reach the main focus of this thesis. Until this point, we explored numerical and analytical
methods we can employ to study single particle systems on a lattice, first-order and higher-order topolo-
gies, and introduced the BBH model and its disordered cousin. In this chapter, we perform a detailed
study of a chiral QPQI model, starting with the topological properties, followed by spectral and local-
ization properties, and finishing with boundary phenomenon and its connection to the observed bulk
properties.

5.1 Model
At this point in the thesis, it should be clear that chiral symmetry is important to have a quantized
quadrupole moment. With this in mind, an appropriate form of the potential needs to be chosen in order
to break translational invariance while preserving chiral symmetry. This can be achieved by perturbing
the internal UC hopping’s (γ) with a quasi-periodic (QP) modulation while maintaining the intra-UC
hopping’s (λ) unchanged. Thus, in analogy to what was done in Chapter 3, we consider the BBH model
with the following hopping modulation:

W
γ→γ+ cos (2πβm + ϕx ) cos (2πβn + ϕy ) , (5.1.1)
2
where β is an approximant of ϕGR (the golden ratio), R = mêx + nêx is a lattice vector and ϕi are the
phase shifts. Unlike the case of the QBCP Chern insulator, the QP potential is chosen to vary in both
directions to fully break translational invariance. For generic ϕ = (ϕx , ϕy ), the QP modulation breaks
the mirror and C4 crystalline symmetries. Therefore, this symmetry is broken for the finite-size systems
that we study. Nonetheless, it should be recovered in the thermodynamic limit, when there should be no
ϕ dependence of the bulk properties.

42
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 43

5.2 Main Results

In this section, we unveil the main results of this thesis, namely the phase diagram of the system on the
(γ, W ) plane, as shown in Fig. 5.1, where 5.1(a) is for the topological properties and 5.1(b) for the spectral
properties. These plots allow for a general perspective of the phase diagram that enable us to focus on
specific regions of interest. In this manner, in Fig. 5.1(a) the quadrupole moment and in Fig. 5.1(b) the
bulk spectral gap phase diagrams.

6 6

5 5

4 QPQI 4 QPQI
W

W
3 3

2 2 0.0
QI T 0.4 QI T

∆E
qxy

1 0.2 1 -2.0
(a) 0.0 (b)
0 0
0 0.5 1.0 1.5 2.0 0 0.5 1.0 1.5 2.0
γ γ

Figure 5.1: (a) Quadrupole moment (qxy ) in the (γ, W ) plane, computed for a system size L = 13,
with θi = π and averaged over 100 random phase shifts realizations. (b) The logarithm spectral gap
(log10 (∆E)) in the (γ, W ) plane for a system size L = 34, with θi = 0 and averaged over 50 realizations
of phase shifts. A minimum value of the spectral gap was chosen (∆E < 10−3 ) to highlight all the gapped
and gapless phases appearing in the phase diagram. The dotted white lines are constant γ and W cuts
for γ = 0.5, 1.1 and W = 3, respectively.

The clean limit quadrupole insulating (QI) and trivial (T) phases are robust to QP modulations. In
the trivial regime, these modulations can induce a TPT into a QI phase adiabatically connected to the
clean limit phase or remain trivial, while in the topological case, the QP modulations eventually induce
a TPT into a gapless critical metal, blueish region in Fig. 5.1 between QI and QPQI phase. Further
increasing the strength of the QP modulation leads to a novel reentrant topological regime, which we
entitle quasi-periodic QI (QPQI). These phases exhibit the same boundary signatures as their clean limit
counterparts, such as boundary polarizations (px = py = 0.5) and four-fold zero-energy corner modes
that give rise to fractional corner charges. In these regimes, an intricate interplay between phase shift-
dependent edge states and corner modes emerges, leading to behavior not observed in the QI phases, such
as edge-corner hybridization. This behavior is unique to the QPQI phase. Overall, the observed TPT
between various regimes occurs in gap closing or opening as seen in Fig. 5.1(b).
To thoroughly characterize each of the observed phases, two cuts along constant γ were characterized
(depicted as vertical dotted white lines), with the intent of analyzing various electronic properties as W
increased. In this manner, two distinct values of the hopping amplitude were selected such that multiple
regions of the phase diagram were incorporated into the cuts. Furthermore, to study the robustness of
each clean limit phase, we opted for γ = 0.5(1.1), which for W = 0 corresponds to the topological (trivial)

43
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 44

insulating phase. Additionally, we also consider a constant W cut at W = 3 seen in Fig. 5.1 (depicted by
a horizontal dotted white line). The results associated with this cut are in Appendix F and despite the
gapless QPQI regime with localized states at Fermi level, no relevant regimes have been found.

5.3 Topological and Spectral Properties

As we know, topological phase transitions are usually associated with some gap mechanism, whether it
be a gap closure as usually seen in clean systems or gap-edge localization transitions that induce a TPT
as observed in topological Anderson insulators. Although gap closures are not a requirement for a TPT,
they are still a good indicator that one has occurred, especially when the system has delocalized gap-edge
states. To this end, in Fig. 5.2, the spectral gap (∆E) is plotted as a function of W for each of the
different starting phases. The spectral gap was obtained via ED and estimated as ∆E = 2ϵ+ , where ϵ+ is

1.50 I II III IV I II III

0.4
1.25
0.05
0.3 0.005
1.00 a b c a b c
∆E

∆E

0.75 0.2
0.50 0 0
3.1 3.5 3.4 3.5
0.1
0.25
(a) (b)
0.00 0.0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
W W

Figure 5.2: Spectral gap (∆E) as a function of W computed for an even system size L = 144 =⇒ θi = 0
and averaged over 50 phase shifts realizations. (a) γ = 0.5. (b) γ = 1.1.

the positive eigenvalue closest to zero. The dotted dark blue lines mark gap closures or openings and thus
a change of region per example, I → II. As per the dotted gray lines, they mark level crossings, which
will be clearer when we tackle localization properties. From start to finish of our analysis, the dotted
lines act as a visual guide throughout the discussion that follows. The calculation of ∆E was performed
for an even system size. However, the results can be identically duplicated for odd L if we add a phase
twist θi = π in both directions. Furthermore, no error bars are plotted since the curves ∆E ± 2σ are
over-imposed with the main plot, indicating that the bulk spectral gap has no dependence on phase shifts.
In Fig. 5.3 the quadrupole moment (qxy ) and the boundary invariant (P ) are plotted as a function of
the quasi-periodic potential strength (W ). All the results discussed in this chapter revolve around these
two cuts, and so they will be mentioned throughout this discussion.
The quadrupole moment was computed for several system sizes via the multipole operators. Since
some of the sizes are odd, a twist of π was fixed in each direction in order to have a well-defined gap
closure1 . The boundary invariant (P ) was obtained via the boundary Green’s function method followed
1
If we perform the calculation of qxy for a big enough odd value of L the twist will play no part in resolving the TPT.

44
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 45

I II III IV I II III P
1.0 1.0
L:13
L:21
0.8 0.8 Peak L:34
Peak
0.6 0.6
qxy

qxy
0.4 0.4

0.2 0.2
(a) (b)
0.0 0.0
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
W W

Figure 5.3: qxy and P as a function of W . The P invariant was computed for a system size of L = 610
with 115 averages over phase shifts. qxy was obtained via real-space methods with 100 averages over phase
shifts. For even L, θx = θy = 0; for odd L θx = θy = 0. (a) γ = 0.5. (b) γ = 1.1.

by the multipole operator for polarization, with θi = 0 in both directions2 since we take L to be even.

Discussion - Starting Topological The results obtained by increasing W , starting at a topological

phase, are shown in Fig. 5.3(a) and 5.2(a). For γ = 0.5 and W = 0 the system is in a topological phase.
Both the quadrupole moment and the boundary invariant are quantized, respectively, to 0.5 or 1 for each
realization of phase shifts. This quantization remains with increasing W, leading us to conclude that the
QI phase (I) is robust to QP modulations. Eventually, at WcI→II ≈ 3.05, the gap closes and a TPT occurs
into a critical metal phase (IIa, see inset of Fig. 5.2) which we motivate below after tackling localization
L=34
properties. Reaching IIb, the gap reopens into a topological regime (IIb) with both the P and qxy
invariants exhibiting a sharp peak in the region, as indicated in Fig. 5.3(a). The fact that these peaks
L=34 in IIa and IIc go to zero, hint


approach 1(0.5) with increasing system size while the value of P qxy
that IIb is a quadrupole insulating phase. Additionally, because this is a very narrow W window and the
gap is small (∆E ≈ 10−3 ), we attribute the lack of quantization of either invariants to finite-size effects.
Further increasing W induces a trivial insulating phase (IIc) before another quadrupole insulating regime
is reached (III) that suffers from the same finite-size effects of IIb. Since both IIb and III are gapped,
the proof presented in Appendix B holds, and in the thermodynamic limit qxy and P should quantize
to 0.5 and 1 respectively. In this manner, quasi-periodic modulations can induce a novel type of SOTI
phases which we entitle QPQI phases3 . Towards the strong W limit, the gap closes (WcIII→IV ≈ 4.3)
and the system reaches a gapless trivial phase (IV ) with a finite, although not quantized, quadrupole
moment (qxy ̸= 0). For each individual phase shift realization qxy (P ) remain quantized to either 0 or
0.5(1), however, its average yields a finite intermediary value, indicating a trivial regime. One might be
tempted to use chiral symmetry to argue that this region is either trivial or topological. However, due to
2
The open boundary is only considered in the longitudinal direction. Along the transversal direction, the boundaries are
periodically closed since the considered system size is even. Although a twist can be added, for such large system sizes, they
should play no part in resolving gap closures and thus TPTs.
3
Quasi-Periodic Quadrupole Insulator

45
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 46

the gapless nature of the system, its ground state is ill-defined due to the high degeneracy of zero-energy
states, which in turn leads to a non-quantized quadrupole moment. In spite of this, since both invariants
scale down to zero as the system size increased, phase IV is trivial.

Discussion - Starting Trivial For γ = 1.1 and W = 0 the system is in a gapped trivial phase, which
is stable to the application of a QP modulation (I). Increasing W induces a TPT into a gapped QI phase
(II) with the gap closing and reopening at WcI→II ≈ 1.73. Furthermore, the gap reaches its maximum,
with a level crossing between energy levels in the gap edge occurring at W ≈ 2.64 (dotted gray line).
Away from the QI phase, the gap closes (WcII→III ≈ 3.4) and reopens in IIIb with a sharp peak in P
that almost quantizes to 1 as seen in Fig. 5.3(b). Just as in the previous case (γ = 0.5), we attribute
the lack of quantization of the boundary invariant to finite-size effects, classifying IIIb as QPQI phase.
Eventually, at the high W regime, the system undergoes a TPT into a gapless trivial regime (IIIc).
Around W ≈ 4, peaks can be observed in the P invariant and qxy , with the latter approaching 0.5
with increasing system size. Bulk gap calculations (Fig. 5.2) indicate a gapless regime with localized
states at the gap-edge (D2 → 0 as seen in Fig. 5.9), however, we cannot conclude that these regions are
topological due to finite-size effects.

0.25
I II III IV I II III
0.4
0.20
0.3
0.15
ρ(0)

ρ(0)

0.2
0.10
10−4
0.1 0.05
10−7
(a) (b) 1.50 1.75
0.0 0.00
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
W W

Figure 5.4: The DOS at zero-energy (ρ(0)) as a function of W . The following KPM parameters were
used: L = 987, M = 5000, R = 10. (a) γ = 0.5. (b) γ = 1.1.

Density of States In Fig. 5.4, we plot the DOS at E = 0 as a function of W with the intent of verifying
the results of the spectral gap presented in Fig. 5.2. In gapped regions ρ(0) ≈ 0,4 while in gapless regions,
the system should have a finite DOS (excluding the possibility of a semimetal). All gapless phases (II and
IV for γ = 0.5 and III for γ = 1.1) display a finite DOS at Fermi level. Furthermore, at the transition
I → II (γ = 1.1) the gap closes and reopens at a critical QP potential strength (WcI→II ≈ 1.73), which is
signaled by a small increase of ρ(0) around it, as shown in Fig. 5.4(b-inset). The Jackson kernel broadens
the Dirac delta as a Gaussian with spread σ ∝ 1/M . For this reason, it was not possible to reach enough
resolution to capture the gapped IIb and IIc phases that are filled by the tails of the Gaussian. The same
4
It might not be zero since, in gapped regions, the DOS scales to zero with the number of Chebyshev moments M .

46
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 47

occurs for γ = 1.1 in phase IIIb. These narrow phases are only expected to be resolved for larger system
sizes and number of moments.
Away from the Fermi level, the effects of QP modulations on the energy spectrum contrast with
the effects of onsite or chiral disorder. As per the latter, disorder tends to smooth the DOS, quickly
killing singularities, increasing bandwidth and closing gaps. Furthermore, for high enough W , regular
disorder transforms the DOS into a homogeneous distribution with width W . To explore the effects of
quasi-periodicity, we compute the DOS with KPM using the Jackson kernel. In Fig. 5.5 the clean limit
DOS is plotted as a function of E for the two cases (γ = 0.5 and γ = 1.1).

1.0
0.6
0.8

0.6 0.4
ρ(E)

ρ(E)
0.4
0.2
0.2
(a) (b)
0.0 0.0
−2 −1 0 1 2 −2 0 2
E E

Figure 5.5: Clean limit DOS as a function of E. The following KPM parameters were used: L = 987,
M = 4000, R = 10 (a) γ = 0.5. (b) γ = 1.1.

To complete our study, we show in Fig. 5.6 the evolution of ρ(E) under increasing W for γ = 1.1.
The QP potential induces mid-bulk gaps even for small W . For higher W , all the gaps are killed, with the
bandwidth of the system growing monotonously with W . The same behavior can be observed for γ = 0.5
(Fig. F.1).

W =1.0 W =3.0 W =6.0

0.75 0.3 0.2
ρ(E)

0.50 0.2
0.1
0.25 0.1

0.00 0.0 0.0

−2.5 0.0 2.5 −5 0 5 −5 0 5
E E E

Figure 5.6: DOS as a function of E for γ = 1.1 and different values of quasi-periodic potential strength
(W ). The following KPM parameters were used: L = 987, M = 2000, R = 10.

Overall, the effects of a QP modulation on the energy spectra of the system is independent of the

47
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 48

starting phase. Furthermore, these types of modulations (whether they are chiral or onsite) tend to induce
bulk gaps and sometimes even spectral gaps that remain well into a strong W regime. This phenomenology
is common in quasi-periodic systems and can be observed in the notorious Aubry-André model [27] and
in the QBCP Chern insulator with QP modulations discussed in Chapter 3.

5.4 Localization properties

In this section, we turn our attention to localization properties. In Fig. 5.7, we plot the normalized
localization length (ξ/M ) at Fermi level (E = 0), obtained using TMM. In regions with a larger gap, it
is clear that ξ/M scales down to zero, since TMM captures localized evanescent wave solutions of the
Schrödinger equation. At gapless regions, or regions with small gaps, the results are noisier, and in some
cases, sufficiently larger system sizes would be needed to understand whether ξ/M converges with M ,
showing critical behavior, or if it decreases with M signaling gapless localized states. This will become
more clear from the eigenstate analysis based in exact diagonalization that we carry out below.

I II III IV I II III
101 M: 13
101 M: 21
M: 34
100 M: 55
ξ/M

ξ/M

100
10−1

10−1
10−2 (a) (b)
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
W W

Figure 5.7: TMM at Fermi level (E = 0) for several transversal system sizes (M ). For odd M we add a
twits of θM = π. The size of the longitudinal direction was chosen such that the relative error ϵ < 1%.
(a) γ = 0.5. (b) γ = 1.1.

Below, we characterize the localization properties of the eigenstates with energies closer to E = 0. In
the gapped phases, these correspond to eigenstates at the gap edge that are not captured by the TMM
at Fermi level. In Fig. 5.8, we plot the average IP R and IP Rk of the 8 eigenstates closer to the Fermi
level as a function of W . Whenever a gap closure occurs, TMM and IP R calculations are in agreement,
as we discuss below.
We also plot the fractal dimensions D2 and D2k in Fig. 5.9 estimated with the calculations of IP R
and IP Rk respectively.

Discussion - Starting Topological For γ = 0.5, in the gapped phase I, the eigenstates at the gap
edge start off ballistic, as seen in Fig. 5.9, D2 → 2 and D2k → 0. They undergo a ballistic→localized
(D2 → 0 and D2k → 2) transition at W ≈ 1.19 (gray line) further ascertaining that a level crossing between

48
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 49

100
I II III IV I II III

10−1

10−2
IP R

10−3 (a) (b)

10−4

10−5

100
L: 34
−1 L: 55
10
L: 89
10−2 L: 144
L: 233
IP Rk

10−3 L: 377

10−4

10−5
(c) (d)
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
W W

Figure 5.8: IP R (a,b) and IP Rk (c,d) calculations for different system sizes obtained with exact diago-
nalization methods. The IP R and IP Rk are averaged over 100 phase twists, shifts and over the first 8
eigenstates with energies closest to E = 0. (a,c) γ = 0.5. (b,d) γ = 1.1.

gap edge states occurs. This transition is not captured by the TMM since the system remains gapped.
At the TPT (I → II) the gap-edge states are ballistic, however, these states quickly turn critical, as seen
by D2 ≈ 1, indicating that the gapless critical metal regime is reached (IIa). The normalized localization
length (ξ/M ) obtained via TMM, does not diverge at the transition. Instead, it goes to a constant value
(see Fig. 5.7(a)), in agreement with the critical metal regime. In the QPQI phase (IIb), the gap-edge
states remain critical, localizing for the IIc phase. In these latter regimes, TMM fails to capture such
narrow behavior, due to the lack of discretization in TMM calculations. At the transition from II → III,
IP R, IP Rk (Fig. 5.8) and D2 (Fig. 5.9) indicate a ballistic regime, with the gap-edge states undergoing
a ballistic-localized transition halfway through phase III.
Phase IV is of difficult classification with any of the methods at our disposal. At the transition
III → IV the IP R results suggest that the eigenstates are critical.
After this transition, TMM and IP R results are noisy, however, considering the fractal dimension
computed with the last three system sizes, D2 approaches zero, indicating that the system as reached an
Anderson insulating phase (D2 → 0).

49
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 50

I II III IV I II III
3

2
D2

0
(a) (b)

I II III IV I II III
3

2
D2k

0
(c) (d)
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
W W

Figure 5.9: D2 (a,b) and D2k (c,d) calculations for different system sizes obtained from the IP R and IP Rk
results respectively. (a,c) γ = 0.5. (b,d) γ = 1.1. The red curves were estimated making a fit in log − log
scale for all L = {34, 55, 89, 144, 233, 377}, while the dotted gray lines were obtained only considering the
biggest three system sizes L = {144, 233, 377}.

Discussion - Starting Trivial Starting at the trivial phase with γ = 1.1, the gap edge eigenstates are
ballistic within phase I and remain ballistic even after the transition into phase II (D2 → 2 and D2k → 0,
see Fig. 5.9). At the TPT (I → II), a gap closing and reopening topological transition with ballistic gap
edge states occurs, as in the absence of quasi-periodicity, which explains why this particular gap closure
is highly dependent on the choice of twists, especially for small system sizes [87]. As in the previous
case, the ballistic regime is maintained until the level crossing between ballistic and localized gap edge
levels occurs (gray line) where we observe the gap-edge states sharply localize (D2 → 0). For higher W
sightly before the gap closing at WcII→III ≈ 3.4, gap edge states delocalize and then become critical at
the transition (D2 ≈ 1.2). The results for the final phase (III) are similar to phase IV of the previous
cut, leading us to conclude that an Anderson insulating regime is reached.

5.5 Opening the Boundaries

Until now, we have spent much of our time on calculations involving periodic or twisted boundary con-
ditions. However, a full characterization of these phases is not complete without considering an open

50
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 51

boundary system.
As it was discussed, an important hallmark of topological quadrupole insulators is that quadrupole
phases defined by qxy = 0.5 have zero-energy corner modes that give rise to fractional corner charges in
open boundary systems. In this manner, a study of boundary signatures for each of the observed phases
is important in order to have a complete and detailed picture of the chiral QPQI. Before proceeding, we
introduce a simple yet powerful quantity, the corner occupation probability (Pocc ) defined as

X 4
X
Pocc = |ψnα (R)|2 , (5.5.1)
R∈corner α=1

where ψnα (R) represent the αth orbital at site R of the eigenstate with energy En . For the calculations
that follow, we average out Pocc of the four eigenstates closest to the Fermi level. On one hand, in a QI
phase zero-energy corner modes occur thus Pocc ≈ 1, on the other hand, for a trivial phase, no corner
modes exist such that Pocc < 1. Since the corner states display a localization length dependent on W
and γ we compute Pocc over an l × l region which we assume to be the corner, i.e. we restrict the sum
in Eq. 5.5.1 to that region. Therefore, l is effectively an estimation of the localization length of the
corner modes. In Fig. 5.10(a-b) the Pocc as a function of W is shown for several sizes of the corner (l).
P
In Fig. 5.10(c-d) we show the averaged corner charge Q̄ = i |Qi | as a function of W , obtained from
Qi = R∈corner ρ (R), where ρ(R) = 2 − n∈occ. 4α=1 |ψnα (R)|2 is the charge density. The first and
P P P

second terms arise from the atomic positive charges and from the electronic density, respectively. Unlike
the computations for Pocc the system is partitioned into four quadrants, each of the size of a quarter of
the system. Then, the corner charge is computed by integrating ρ (R) over each of the quadrants. In
quadrupole insulating phases Q̄ = 0.5.

Discussion - Starting Topological For γ = 0.5, Pocc has perfect plateaus at 1 for every considered l,
indicating the existence of highly localized corner states (ξl < 10 where ξl is the localization length of the
corner modes). Similarly, the corner charge also quantizes at Q̄ = 0.5 throughout the topological phase.
In phase II, the critical metal (IIa) and trivial insulator (IIc) display a finite, yet significantly smaller
Pocc , that arises from contributions of edge or bulk states (below we will provide a detailed discussion
on edge states). The same is observed regarding the corner charge. In the narrow topological regime
observed (IIb), Pocc sharply peaks, scaling closer to 1 as seen in Fig. 5.10(a). Despite this, the peak does
not properly quantize to 1 for the studied l, that we attribute to the large localization length of the corner
modes ξl > 30, which is expected due to the very small gap in this phase. A peak can also be observed in
the corner charge for L = 55 as seen in Fig. 5.10(c), albeit smaller due to the considerably smaller system
size. Regarding phase III, Pocc has a noisy plateau that scales to 1 as l increases, indicating again that
the corner states display a larger localization length than the maximum considered corner (l = 30). The
corner charge also scales to 0.5 with increasing size, further ascertaining that III is indeed topological,
with corner modes characterized by a large localization length.
Finally, as expected, phase IV has a small Pocc . However, we also see a growth with increasing
corner/system size. This increase is explained by the gapless nature of this phase. As l increases, more
and more bulk spectral weight falls within the considered l × l corner, leading to a visible increase in
Pocc . Regarding the corner charge, although there are no quantized corner charges, for some realizations

51
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 52

I II III IV I II III
1.0 1.0 l: 5
l: 10
l: 15
0.8 0.8 Peak
Peak l: 20
l: 25
0.6 0.6
Pocc

Pocc
l: 30

0.4 0.4

0.2 0.2
(a) (b)
0.0 0.0
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
W W

I II III IV I II III
0.5 0.5
L: 21
L: 34
0.4 0.4 L: 55

0.3 0.3
Q̄

Q̄

0.2 0.2

0.1 0.1
(c) (d)
0.0 0.0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
W W

Figure 5.10: (a-b) Corner occupation probability (Pocc ) as a function of W for the first 2 zero-energy
states. The results were obtained for a system size of L = 377 and averaged over 200 phase shifts. (c-d)
Corner charge (Q̄) as a function of W for several system sizes and averaged over 100 phase shifts and over
the 4 corners. A small δ = 10−5 was used in both cases to split the degeneracy of the four-fold corner
modes. (a,c) γ = 0.5. (b,d) γ = 1.1.

of phase shifts the charge density displays fluctuations which leads to finite contributions to the corner
charge. This occurs due to the breaking of chiral symmetry by the small δ introduced to lift the degeneracy
of the corner modes. Overall, the Pocc and Q̄ of each phase agrees well with the topological phase diagram.

Discussion - Starting Trivial For γ = 1.1 both trivial phases I and IIIc exhibit Pocc = 0, as seen in
Fig. 5.10(b). For phase II the corner occupation probability, though initially suffers from finite-size effects,
eventually reaches 1. Its dependence on l after the TPT (I → II) indicates a diverging localization length
of the corner states as the critical point is approached. Deeper into the QI phase, the corner modes become
more localized. This is exactly the same behavior observed around the TPT of the clean BBH model.
Regarding IIIb, Pocc has a small peak in this region as indicated in Fig. 5.10(b), however, the corner
charge calculations do not have enough resolution to catch these narrow topological regimes. Regarding
the corner charge, it displays scaling with increasing system size. Nonetheless, a plateau cannot be seen

52
 5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 53

even for the largest considered size L = 55. Looking at Pocc in phase II, it is clear that for a system size
L = 377 considering the corner to be a 30×30 square is not enough to measure all the weight of the corner
state since they display a localization length ξl > 30. Therefore, simulating a system of size L = 55,5
leads to the hybridization of the corner modes, which in turn leads to a non-quantized corner charge. In
Appendix E some remarks are made regarding corner mode hybridization, its relation to finite-size effects,
and consequently to non-quantized corner charges, as seen in phase II with γ = 1.1.

Density of States Another quantity worth exploring is the ratio of open boundary DOS (ρOBC ) and
periodic boundary DOS (ρP BC ). Specifically, for the same system, we compute the DOS with open
boundary conditions (ρOBC ), then with periodic boundary conditions (ρP BC ) and proceed to define the
ratio ρOBC /ρP BC . In regions where there are no boundary states, we expect ρOBC /ρP BC ≈ 1,6 however,
in gapped regions where boundary states can live, we expect ρOBC /ρP BC ≫ 1. In Fig. 5.11 we show the
ratio ρOBC /ρP BC for γ = 1.1 as a function of E. As expected, in gapped regions (grayed-out zones) we
have ρOBC /ρP BC ≫ 1.
Overall, ρOBC /ρP BC ≫ 1 in gapped regions (grayed-out zones), furthermore, for values of W in the
topological phases (W = 2.0 and W = 3.0 in phase II) spectral weight at Fermi level (E = 0) is observed,
indicating the existence of topological corner modes. For high W , all the bulk gaps eventually close,
leading to ρOBC /ρP BC → 1. The exact same behavior can be seen for γ = 0.5 in Fig. F.2 in Appendix F.

W=1.0 W=2.0 W=3.0 W=4.0

2
10
102 102 102
ρOBC
ρPBC

101 101 101

101

100 100 100 100

−2.5 0.0 2.5 −2.5 0.0 2.5 −2.5 0.0 2.5 −2.5 0.0 2.5
E E E E

Figure 5.11: Ratio of open boundary DOS (ρOBC ) and periodic boundaries DOS (ρP BC ) as a function of
energy for W in different phases and γ = 1.1. The grayed-out regions are bulk gaps opened by the QP
modulation and were estimated by setting a threshold for the PBC DOS (ρP BC < 10−3 ). The following
KPM parameters were used: L = 987, M = 5000, R = 10.

Properties of QPQI phases Until now, a proper characterization of the QPQI phases has not been
provided. In the paragraphs that follow, we aim to fully define these quasi-periodic-induced regime.
We start with their definition: a QPQI phase is a topological quadrupole insulating phase that is not
adiabatically connected to any of the clean limit phases7 while displaying a quasi-periodic induced bulk-
spectral gap8 .
5
Notice that for ξl > 30, the localization length of the corner state is of the order of the linear size of the entire system.
6
For big enough system sizes.
7
Two phases are adiabatically connected if there is at least one continuous path connecting both phases, throughout which
the bulk spectral gap remains open.
8
Or is gapless but displays localized states at the Fermi level, as discussed in Appendix F in relation to the constant W
cut).

53
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 54

σE=0 (10−5 ) log10 (Eedge )

0 5 10 −4 −3 −2

2π 2π

3π/2 3π/2
φy

π π

π/2 π/2

(a) (b)
0 0
0 π/2 π 3π/2 2π 0 π/2 π 3π/2 2π
φx φx

Figure 5.12: (a) Standard deviation (σE=0 ) of the energies of the corner modes as a function of the phase
shifts (ϕx , ϕy ). (b) Logarithm of edge state energy (log10 (Eedge )) as a function of the phase shifts (ϕx , ϕy ).
Results obtained for W = 3, γ = 0.5 for a system size L = 233.

We have found that quasi-periodic-induced gapped phases usually display edge states that disperse
with the phase shifts. This phenomenon is common in QP systems and can be observed in the Aubry-
André model. As ϕ changes, the energy of the edge states can move closer or even cross the Fermi level.
This phenomenon is particularly relevant in topological regimes since the edge states can hybridize with
zero-energy topological modes, thus pulling them away from zero-energy. We expect that when such
crossing occurs, the system behaves as in the trivial regime, displaying P = 0. Furthermore, Pocc is also
impacted since the corner modes are hybridized with edge states that exhibit spectral weight outside the
corners.
In order to check the existence of edge states and study edge-corner hybridization, we resort to exact
diagonalization of the open boundary system Hamiltonian, calculating the 6 eigenvalues closest to Fermi
level. Assuming we choose a point of the phase diagram that resides in a SOTI phase, out of the 6 energies,
4 correspond to zero-energy corner modes, with the remaining two corresponding to the edge states with
energy 
closest toq
Fermi level.
 We then proceed to compute the standard deviation of the corner modes
1 P 2
energy σE=0 = N i Ei and the energy of the edge state closest to the Fermi level (Eedge ). In this
manner, when the edge states energy approaches the Fermi level (Eedge → 0) a peak should be seen
in σE=0 , indicating that the corner modes have hybridized with the edge states. Before tackling QPQI
phases that display complex dynamics, we consider a simpler case. In Fig. 5.3(a), just before the TPT
from I → II, a small dip can be seen in the P invariant (W ≈ 3 and γ ≈ 0.5), suggesting that some
dependence on the phase shifts might exist. In this manner, in Fig. 5.12 we show the density plots of σE=0
(a) and Eedge (b) in the (ϕx , ϕy ) plane. Four lines of constant ϕx can be observed where σE=0 > 0 for
Eedge → 0, indicating that the corner modes hybridized away from zero energy (see Fig. 5.14). Moreover,
not all four corner modes hybridize with the edge states. In fact, only two of them move away from
zero-energy while also demonstrating signs of hybridization, prompting us to distinguish the four corner

54
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 55

states. We attribute the term “Inner Corner Modes” to the two corner modes that remain at E = 0 and
“Outer Corner Modes” to the remaining corner states that move away from zero energy.

σE=0 (10−5 ) log10 (Eedge )

0 1 2 −6 −4 −2

2π 2π

3π/2 3π/2
φy

π π

(a) (b)
π/2 π/2

0 0
0 π/2 π 3π/2 2π 0 π/2 π 3π/2 2π
φx φx

Figure 5.13: (a) Standard deviation (σE=0 ) of the energies of the corner modes as a function of the phase
shifts (ϕx , ϕy ). (b) Logarithm of edge state energy (log10 (Eedge )) as a function of the phase shifts (ϕx , ϕy ).
Results obtained for W = 3.8 and γ = 0.2 in the QPQI phase for a system size L = 377.

A distinct feature of QPQI phases is their much stronger dependence on phase shifts. In Fig. 5.13
we show σE=0 and Eedge → 0 for the choice of parameters γ = 0.2 and W = 3.8, well inside the QPQI
phase. Just as before, regions with σE=0 > 0 are correlated to regions where Eedge → 0. Furthermore,
increasing system size confines these regions to singular points (for smaller system sizes, see Appendix F).
This behavior suggests that as these regions collapse to singular points, all the quantities of relevance
(P ,Pocc and qxy ) quantize to their respective values, since these points will have no statistical importance.

log10 (IP Redge ) log10 (IP Router ) log10 (IP Rinner )

−2 −1 −2 −1 −2 −1

2π 2π 2π

3π/2 3π/2 3π/2

φy

φy

φy

π π π

π/2 π/2 π/2

0 0 0
0 π/2 π 3π/2 2π 0 π/2 π 3π/2 2π 0 π/2 π 3π/2 2π
φx φx φx

Figure 5.14: IP R of the edge, outer and inner states plotted from left to right respectively for L = 377.

As Eedge → 0 we expect the localization length of the corner modes to increase as the states hybridize.

55
5 Chiral Quasi-Periodic Quadrupole Insulator (QPQI) 56

In Fig. 5.14, we plot the IP R of the edge and corner states for L = 377. As discussed previously, only
one of the pairs of corner modes hybridizes, which can be seen by the clear pattern emerging in the
middle plot of Fig. 5.14 corresponding to the IP R of the outer corner states. As the edge states cross
the Fermi energy, the IP R of the outer corner modes decreases, indicating that the localization length
increases. Although, IP R alone is not enough to conclude about the localization length of a state, for
fully delocalized states IP R ≈ 1
N, thus for the considered system size (L = 377), IP R ≈ 10−5 . Therefore,
an IP R of the order of 10−2 is a good indication that the corner modes remain localized throughout the
(ϕx , ϕy ) plane.
Overall, QPQI phases exhibit a strong correlation between the localization length of the corner modes
and the edge gap (which is ϕ dependent). As the edge states energy approach the Fermi level, they localize,
moving towards the corners, thus hybridizing with the corner modes. The hybridization “pulls” the outer
corner modes away from zero-energy while increasing their localization length. In the QI regime, the
corner modes display oscillating localization length with varying phase shifts, however, to a far lesser
degree than in the QPQI phases. Furthermore, unlike the QPQI regimes, there is no apparent correlation
between the localization length of the corner modes and the energy of the edge states, suggesting that
the oscillations on the localization length have a different origin than corner-edge hybridization.

56
Chapter 6

Conclusion and Outlook

Throughout this thesis, we sought to study and characterize the effects of a quasi-periodic modulation on
the prototypical BBH model for higher-order topology.
In Chapter 2 we have explored several methods to tackle topological systems, starting with the
introduction of the nested Wilson loop approach by demonstrating its connection with the diagonalization
of the projected position operator. Furthermore, we related their spectra to bulk and boundary topology
with the introduction of the winding number, which in turn defines the bulk electronic polarization. We
have also explored generalizations of Resta’s formula to higher-order multipole moments that are needed
for the characterization of quadrupole insulators, showing that these multipole moments are quantized by
chiral symmetry. Since our main goal was to study systems lacking translational invariance, we introduced
several methods that allow us to study spectral (KPM and recursive method), localization (TMM and
IP R) and boundary properties (Green’s functions) via real-space approaches.
Before tackling higher-order topological insulators, we study a Chern insulator under the effect of
a 1D quasi-periodic onsite potential in Chapter 3, where it was demonstrated that Chern insulating
phases (C = ±2) are robust to quasi-periodic modulations while eventually a gap closing and opening
in Dirac cone transition occurs. This induces a topological phase transition into a gapped C = ±1
phase characterized by several extended bulk states. These results greatly contrast to those obtained for
Anderson disorder, where it has been shown that the gap closes before the topological phase transitions
and that the entire spectra is Anderson localized, but for a few punctual critical states that carry all the
topology [79]. Overall, QP topological systems are phenomenologically different from their disordered
cousins. With this analysis, we built up our intuition to tackle higher-order quasi-periodic topological
systems.
In Chapter 4, we started off with the introduction of the BBH model and its overall properties.
Later the effects of chiral symmetry preserving disorder was considered on the BBH model, where it was
shown that chiral disorder induces a topological phase transition into a phase with quantized quadrupole
moment (qxy = 0.5), quantized boundary polarizations (px = py = 0.5) and zero-energy corner modes.
This SOTI phase is initially gapped, with a smaller gapless region appearing. Thus, two distinct phases
were found which the authors in Ref. [50] entitled gapped and gapless HOTAI. This behavior is identical
to the one observed in first-order TAIs. Eventually the system undergoes a transition into a critical metal
with mixed topological properties, the Griffiths regime. For high enough disorder, the entire spectrum

57
 Conclusion and Outlook 58

localizes, and the system reaches an Anderson insulating phase.

In Chapter 5 we reached the main focus of this thesis, where the properties of the BBH model
under chiral quasi-periodic modulations were studied. We performed a detailed study of the system,
characterizing most of the relevant phases observed in the phase diagram. Overall, clean limit topological
phases (QI) are shown to be robust to the application of such type of modulations. Furthermore, quasi-
periodic modulations can induce transitions into QPQI phases. Although these phases are defined by a
quantized quadrupole moment (|qxy | = 0.5), quantized boundary polarizations (|px | = |py | = 0.5) and
zero-energy corner modes, just like the regular QI regimes, these newly found QPQI phases display far
richer localization properties and dynamics, like corner-edge hybridization, gapped and gapless regimes,
and are not adiabatically connected to regular QI phases. For low modulation strength, ballistic-ballistic
transitions occur and even transitions with critical states at Fermi level are observed. In the range of W
comparable to where the Griffiths regime is observed for the disordered case, the same mixed topology
emerges, however, with localized states at Fermi level. For high modulation strength, the system reaches
an Anderson Insulating phase. Furthermore, all the observed SOTI phases exhibit zero-energy corner
modes that give rise to quantized fractional corner charges.
Unfortunately, this thesis does not answer all the questions, with some work left to be done. Reaching
bigger system sizes will be an instrumental tool in better defining some of the observed phases, especially
the QPQI regimes. Furthermore, a characterization of each phase with the multipole chiral numbers
might paint a more illuminating picture in the mixed regimes. Regarding boundary properties, we have
encountered an interesting dynamics involving the phase shift and the energy of the edge states that
separates the QI phases from the QPQI regimes. Certainly, further studying this interplay between
quasi-periodic topology and chiral SOTI phases is of interest in the present state of quasi-periodicity and
HOTIs. Moreover, a perturbative approach via self-consistent Born approximation would be welcomed
to better understand the mechanism behind the TPT from I → II at γ = 1.1, which we suspect to be
in the common hopping renormalization mechanism. Last but not least, it would be interesting to find
signatures of the corner-edge dynamics that could be used in possible cold atom experiments to detect
QPQI phases.

58
Appendix A

Wilson Loops - Supplementary

Calculations

A.1 ⟨0| γk,n c†q+∆k,α |0⟩ and ⟨0| cq,α γk†′ ,m |0⟩ correlators
Here we calculate the correlators:

⟨0| γk,n c†q+∆k,α |0⟩ (A.1.1)

⟨0| cq,α γk†′ ,m |0⟩ . (A.1.2)

For the first one (A.1.1):

⟨0| γk,n c†q+∆k,α |0⟩ =

[(unk )∗ ] ⟨0| ck,β c†q+∆k,α |0⟩
β
X
=
β
β
[(unk )∗ ] δα,β δk,q+∆k
X
=
β
α
= [(unk )∗ ] δk,q+∆k . (A.1.3)

For the second one (A.1.2) the calculation is analogous:

⟨0| cq,α γk†′ ,m |0⟩ =

†
X β
= [um
k′ ] ⟨0| cq,α ck′ ,β |0⟩
β
X β
= [um
k′ ] δα,β δk′ ,q
β
α
= [um
k′ ] δk′ ,q . (A.1.4)

59
A Wilson Loops - Supplementary Calculations 60

A.2 Polarization of Wannier Sectors

The phases νxj (ky ) of the first Wilson loop define a new band structure of the edge, they are defined in
the BZ with ky = [−π, π[. To have a non-trivial topological phase defined by a bulk quadrupole moment,
one needs a boundary Hamiltonian with a gapped band structure. This way, one defines the Wannier
sectors νx− and νx+ that are the set of Wannier bands.

νx− = νxj (ky ) : ∀j νxj (ky ) below the gap



νx+ = νxj (ky ) : ∀j νxj (ky ) above the gap



With the band structure of the edge defined, the objective is to perform a second Wilson loop in the y
direction. The projector that spans the occupied Wannier sectors is given by:

NW X ED
j j
X
Pνx = ψR x ,ky
ψR x ,ky
j=1 Rx ,ky
NW X N occ X
1 X X †
h
j
in
−ikx Rx
h
j
∗ i m ′
= γn,[kx ,ky ]
|0⟩ v x,[kx ,ky ] e ⟨0| γ m,[k ′
,k ] v ′ eikx Rx
Nx ′
x y x,[kx ,ky ]
j=1 Rx ,ky n,m=1 kx ,k
x
NW X Nocc h in h ∗ i m 1 X  ′

† j j −i kx −kx Rx
X X X
= γn,[kx ,ky ]
|0⟩ vx,[kx ,ky ] ⟨0| γm,[k′ ,ky ] vx,[k′ ,k ] e
x x y Nx
j=1 ky n,m=1 kx ,k′ Rx
x
NW X Nocc h in h ∗ i m
† j j
X X X
= γn,[k x ,ky ]
|0⟩ v x,[kx ,ky ] ⟨0| γ ′
m,[k ,ky ] v ′
x,[k ,k ]
δkx ,k′
x x y x
j=1 ky n,m=1 kx ,k′
x
NW Nocc X h in h ∗ im
† j j
X X
= γn,k |0⟩ vx,k vx,k ⟨0| γm,k . (A.2.1)
j=1 n,m=1 k

Now one needs to project the position operator ŷ into the occupied Wannier sector (Pνx ŷPνx ). The
position operator ŷ has an analogous expression to eq. 2.1.1 in reciprocal-space, its form is:

c†k+∆ky ,α |0⟩ ⟨0| ck,α .

X
ŷ =
k,α

The projected position operator takes the form:

NW Nocc
†  i n h i
∗ i m
⟨0| γm,k c†q+∆ky ,α |0⟩
X X XX
Pνx ŷPνx = γn,k |0⟩ vx,k vx,k
i,j=1 n,m=1 k,k′ α
n′ ,m′ =1 q
h in′ h ∗ im′
⟨0| cq,α γn† ′ ,k′ |0⟩ vx,k
j
′ v j
x,k′ ⟨0| γm′ ,k′ . (A.2.2)

Substituting the results deduced previously for the correlators A.1:

60
A Wilson Loops - Supplementary Calculations 61

NW Nocc
X X XX †  i n h i
∗ im m ∗ α h n′ iα
Pνx ŷPνx = γn,k |0⟩ vx,k vx,k [(uk ) ] uk′ δk,q+∆ky
i,j=1 n,m=1 k,k′ α
n′ ,m′ =1 q
h in′ h ∗ im′
j j
δk′ ,q vx,k ′ vx,k ′ ⟨0| γm′ ,k′ ,

NW Nocc h in h ∗ im h ∗ i α h ′ i α
†
X X X
i i m
Pνx ŷPνx = γn,q+∆ky
|0⟩ v x,q+∆ky v x,q+∆ky uq+∆ky unq
i,j=1 n,m=1 q,α
n′ ,m′ =1
 j n′ h j
∗ im′
vx,q vx,q ⟨0| γm′ ,q ,

 
NW Nocc X in Nocc ∗ i m 
h D h
n′ ′
E
†
X X X
i
um i j
unq 

Pνx ŷPνx = γn,q+∆k y
|0⟩ vx,q+∆k y
 q+∆ky vx,q+∆k y
vx,q
i,j=1 n,m′ =1 q n′ ,m=1
h
j

∗ im′
vx,q ⟨0| γm′ ,q . (A.2.3)

To simplify notation and calculations the following definitions are used:

E N occ h in
j j
X
ωx,k = vx,k |unk ⟩ (A.2.4)
n=1
h in,m NW h in D E h ∗ i m
j j
X
νx i i
Fy,k = vx,k+∆k y
ωx,k+∆ky
|ωx,k vx,k . (A.2.5)
i,j=1

Substituting these definitions in eq. A.2.3 we get:

Nocc h in,m
†
X X
νx
Pνx ŷPνx = γn,k+∆k y
|0⟩ Fy,k ⟨0| γm,k . (A.2.6)
n,m=1 k

Note that this operator is diagonal in kx . To diagonalize the operator let us consider the following
eigenvectors equation Pνx ŷPνx ψ j = E j ψ j which in matrix form for each kx reduces to (for simplicity
in notation the kx index is hidden):
  j  j
νx
0 0 0 . . . Fy,kN
νy,k1 νy,k1
 νx    
 Fy,k 0 0 ... 0   νy,k2   νy,k2 
 1    
 0 νx  ν j ν
 Fy,k 2
0 ... 0   y,k3

 = E  y,k3  ,

(A.2.7)
 . .. .. . . ..  .  .
  ..  ..
 
 .. . . . .  
    
0 0 0 ... 0 νy,kN νy,kN

61
A Wilson Loops - Supplementary Calculations 62

νx
where Fy,k i
are Nocc × Nocc matrices. This eigenvalue problem can be rewritten has a set of N equations:




νx
Fy,k νj
N y,kN
j
= E j νy,k 1


νx j j j
F ν = E νy,k2


 y,k1 y,k1


νx
Fy,k ν j = E j νy,k
2 y,k2
j
3
, (A.2.8)


...






F νx j j j
y,kN −1 νy,kN −1 = E νy,kN


which fully defines

N
νx
Fy,k F νx
N y,kN −1
νx
(...)Fy,k F νx ν j = E j
2 y,k1 y,k1
j
νy,k1

νx j

N j
Wy,k νy,k = E j νy,k . (A.2.9)

νx
One can now write explicitly the Wilson loop operator as a product of matrices Fy,k and then expand
them using eq. A.2.5.

Ny
Y
νx νx
Wy,k = Fy,k+i∆ky
i=0
 
Ny
Y
νx  F νx F νx
= Fy,k+i∆ky y,k2 y,k1
i=2
h im,n X h i h i
νx νx νx
Wy,k = Um,l Fy,k 2 ′
Fy,k1 ′
l,l l ,n
l,l′
N occ NW ∗ il′ h ′ il′ D ′ E h ′ ∗ in
 i l D i j′
E h
j j j
X X
i i
= Um,l vx,k 3
ω x,k 3
|ωx,k2 v x,k2 v x,k 2
ωx,k 2
|ω x,k1 vx,k 1
l,l′ =1 i,j=1
i′ ,j ′ =1
N NW
occ X E h ′ ∗ in
 i l D i j′
E D ′
j j
X
i
= Um,l vx,k 3
ω |ω
x,k3 x,k2 δj,i′ ω |ω
x,k2 x,k1 vx,k 1
l=1 i,j=1
i′ ,j ′ =1
N NW
occ X E h ′ ∗ in
 i l D i j′
ED
j j j
X
= Um,l vx,k 3
ω |ω
x,k3 x,k2 ω |ω
x,k2 x,k1 vx,k 1
(A.2.10)
l=1 i,j=1
j ′ =1

νx
Iterating this procedure until there are no Fy,k left in the product yields:

h im,n NW h ii,j h ∗ i n
νx
X  i
m νx j
Wy,k = vx,k W̃y,k vx,k , (A.2.11)
i,j=1

νx QLy νx
where W̃y,k = j=0 F̃y,k+j∆kx is the Wilson loop performed in the y direction over the Wannier basis and
h ii,j D E
νx i j νx
F̃y,k = ωx,k+∆k y
|ωx,k . As it turns out, it suffices to work with W̃y,k . To diagonalize the operator

62
A Wilson Loops - Supplementary Calculations 63

νx
Pνx ŷPνx we need to diagonalize the Wilson loop operator defined in the occupied bands subspace Wy,k .
νx
The relevant information should come from the Wilson loop defined in the Wannier sector subspace W̃y,k .
In fact these two Wilson loop operators are connected via a unitary transformation U defined as an
Nocc × NW matrix:
 
| | | |
 | | | |
 

NW
 
 1 2
U =  vx,k vx,k ... vx,k  (A.2.12)

 | | | |
 

| | | |
νx
The transformation U maps the Wilson loop W̃y,k onto the occupied bands subspace, and obeys:

U U † = INocc ×Nocc (A.2.13)

U † U = INW ×NW . (A.2.14)

νx νx † νx νx
Thus, one can write Wy,k = U W̃y,k U and W̃y,k = U † Wy,k U , which follows from the unitary conditions of
E νx ,j
E
νx νx ,j νx ,j
U . Assuming that W̃y,k ṽy,k = e2πiν̃y (kx ) ṽy,k , it can be shown that from the NW eigenvectors and
E E
νx νx νx ,j νx ,j
eigenvalues of W̃y,k , half of the eigenvalues and eigenvectors of Wy,k are defined with vy,k = U ṽy,k .
With the Wilson loop phases, one can now compute the boundary polarization:

X
pyνx (kx ) = e ν̃yνx ,j (kx ), (A.2.15)
j

and the total polarization of the Wannier sector νx is:

1 X νx
pνyx = py (kx ). (A.2.16)
Nx
kx

Under certain symmetries constrains pνyx (kx ) does not depend on kx and thus the sum is redundant.

63
Appendix B

Chiral Symmetry and Bulk Multipole

Moments
 
(0)
Here we present a proof that shows that the quadrupole moment qxy = q̃xy − qxy mod 1 is quantized
to 0 or 0.5 when a half-filled system preserves chiral symmetry. This proof is available in [50]. The
quadrupole moment is:

1  
q̃xy = Im det Uo† DUo , (B.0.1)
2π
where Uo is the matrix whose columns are the occupied eigenstates, which corresponds to half the states
at half filling. Now considering a model with chiral symmetry such that ΠHΠ† = −H =⇒ {H, Π} = 0,
one can construct the unoccupied eigenstates via the chiral symmetry operator Π, such that if |ψno ⟩ is
an occupied eigenstate with energy En then Π |ψno ⟩ is an unoccupied eigenstate with energy −En . The
unoccupied states matrix is Uu = ΠUo .

1  
u
q̃xy = Im det Uu† DUu
2π
1  
= Im det Uo† Π† DΠUo (B.0.2)
2π

Clearly D commutes with the chiral symmetry transformation Π since both matrixes are diagonal, i.e.,
[D, Π] = 0, and so

1  
u
q̃xy = Im det Uo† DUo = q̃xy . (B.0.3)
2π
The bulk quadrupole moment is defined by
 
(0)
qxy = q̃xy − qxy mod 1, (B.0.4)

(0)
where qxy is the contribution from the background static positive charges. We want to prove that
u = 0 mod 1 which is equivalent to q̃ + q̃ u − 2q (0)
qxy + qxy xy xy xy = 0 mod 1. We start by defining Ut = (Uo , Uu )
which is the unitary transformation that diagonalizes the Hamiltonian of the system. It is clear that:

64
B Chiral Symmetry and Bulk Multipole Moments 65

na
(0)
X x j yj
2qxy =
Lx Ly
j=1
1
=
Im log det (D) mod 1
2π
1  
= Im log det Ut† DUt mod 1, (B.0.5)
2π
   
where the last step follows from the unitarity of Ut , so det Ut† DUt = det Ut† det (D) det (Ut ) = det(D).
(0)
Now the objective is to write 2qxy at the cost of the electronic contributions to the quadrupole moment:
!
  Uo† DUo Uo† DUu
Im log det Ut† DUt = Im log det . (B.0.6)
Uu† DUo Uu† DUu
!
A B
Let M be a 2 × 2 block matrix of the form M = . If A is invertible, the following algebraic
C D
identity holds:
!
A B
= det (A) det D − CA−1 B .


det (B.0.7)
C D
 −1  −1
Making the due correspondences yields CA−1 = Uu† DUo Uo† DUo = Uu† DUo Uo−1 D−1 Uo† . Al-
though Uo is a na × nc matrix, its inverse is still-defined, Uo−1 is the pseudo-inverse of Uo such that the
definition of inverse holds, i.e. Uo−1 Uo = Inc ×nc , in fact, Uo−1 = Uo† , which follows from the orthogonality
relations of the eigenbasis of H. The pseudo-inverse is tricky in the sense that Uo−1 Uo ̸= Uo Uo−1 = Ina ×na .
Keeping this in mind, it follows:

 −1
Uu† DUo Uo−1 D−1 Uo† = Uu† DD−1 Uo

= Uu† Uo
= 0, (B.0.8)

where we have used the orthogonality of the occupied and unoccupied subspaces. We then obtain:

!
Uo† DUo Uo† DUu n    o
Im log det mod 2π = Im log det Uo† DUo det Uu† DUu mod 2π
Uu† DUo Uu† DUu
   
= Im log det Uo† DUo + Im log det Uu† DUu mod 2π
1   1  
= Im log det Uo† DUo + Im log det Uu† DUu mod 1,
2π 2π
(B.0.9)

which leads to:

65
B Chiral Symmetry and Bulk Multipole Moments 66

     
u
2π q̃xy + q̃xy (0)
− 2qxy = Im log det Uo† DUo + Im log det Uu† DUu −
   
− Im log det Uo† DUo − Im log det Uu† DUu mod 2π = 0 mod 2π. (B.0.10)

u = 0 mod 1 using q
We have qxy + qxy u
xy = qxy one concludes 2qxy = 0 mod 1, namely, qxy is quantized to
0 or 0.5. In this manner, the quadrupole moment is quantized by chiral symmetry in half-filling insulators.
An analogous approach can be made regarding any of the remaining multipole operators, concluding each
time that chiral symmetry also quantizes the bulk polarization and bulk octupole moment density at
half-filling insulators.

66
Appendix C

Tridiagonal Matrixes

When considering TB systems, it is common to have a Hamiltonian that is tridiagonal, in fact, for several
of the employed methods, the tridiagonal form has been exploited to our advantage. It is only natural
that we delve into some properties of such matrices.

C.1 Lanczos Decomposition

We start by presenting an algorithm that, from a general hermitian matrix A in some basis {|ϕn ⟩}, can
generate a new basis {|un ⟩} in which A is tridiagonal (Lanczos algorithm [88]). In Jacobi form, A can be
written as:
 
a0 b1 0 0 0 ...
 
 b1 a1 b2 0 0 ... 
 
A= 0 b2 a2 b3 0 ... . (C.1.1)

 
 0 0 b3 a3 b4 ... 
.. .. .. .. .. ..
 
. . . . . .

The first big advantage of an hermitian matrix in Jacobi form is that we can store the entire matrix with
two vectors, one that stores the coefficients an and another that stores the coefficients bn . Knowing these
coefficients fully defines A. In the tridiagonal orthonormal basis {|un ⟩}, the following equation holds:

A |un ⟩ = an |un ⟩ + bn |un−1 ⟩ + bn+1 |un+1 ⟩ . (C.1.2)

The way to generate the new basis and the coefficients is as follows:

1. We choose an arbitrary normalized state |u0 ⟩ which defines a0 = ⟨u0 |A| u0 ⟩

2. We obtain |ũ1 ⟩ by acting with A on |u0 ⟩ and removing the leftover projection |u0 ⟩ i.e. |ũ1 ⟩ =
A |u0 ⟩ − a0 |u0 ⟩

3. Now we normalize |ũ1 ⟩. With C.1.2 we get ⟨ũ1 |ũ1 ⟩ = |b1 |2 thus, |u1 ⟩ = 1
b1 |ũ1 ⟩

4. We keep iterating this method

67
C Tridiagonal Matrixes 68

(a) an = ⟨un |A| un ⟩

(b) |ũn+1 ⟩ = H |un ⟩ − an |un ⟩ − bn−1 |un−1 ⟩
(c) |bn+1 |2 = ⟨ũn+1 |ũn+1 ⟩
1
(d) |un+1 ⟩ = bn+1 |ũn+1 ⟩

Generally, we choose |u0 ⟩ which has at least one caveat: we might not be able to generate the entirety of
the Hilbert space of A, and thus, for a general |u0 ⟩, we might be confined to a subspace of A. Luckily, for
most applications, we can truncate the Lanczos method to a reduced number of terms.

C.2 Inverting Tridiagonal Matrices

Another relevant operation where Jacobi form simplifies the calculations is when we want to obtain a given
entrance of the inverse of a tridiagonal matrix. Let us assume A has the form in C.1.1. The objective is
to compute A−1 00 , but first let us see how to invert a matrix. The inverse of A is defined as:

Adj {A}
A−1 = , (C.2.1)
det {A}
where Adj {A} is the adjoint matrix of A, which is simply the transpose of the cofactor matrix C :
Adj {A} = C T . In turn, the cofactor matrix is defined by

Cij = (−1)i+j Mij , (C.2.2)

where Mij is the determinant of the matrix obtained by removing the ith row and j th column of A. In
this manner, setting G = A−1 , we have that

M00
G00 = G0 = . (C.2.3)
det {A}
To compute det {A}, we make a Laplace expansion of A along the zeroth row, arriving at:

det {A} = a0 M0,0 − b1 M0,1 (C.2.4)

Defining the following determinants of sub-matrices of A

 
an bn+1 0 0 0 ...
 
 bn+1 an+1 bn+2 0 0 ... 
 
 0 bn+2 an+2 bn+3 0 ... 
Dn = det  , (C.2.5)
 
 0 0 bn+3 an+3 bn+4 ... 
.. .. .. .. .. ..
 
. . . . . .

it is clear that M00 = D1 and M01 = b1 D2 , where we employed another Laplace expansion along the first
column of M01 . We then reach

1
G00 = . (C.2.6)
a0 − b21 D
D1
2

68
C Tridiagonal Matrixes 69

D2

−1 (1)
The ratio D1 is nothing more than than A(1) 00
= G00 = G1 , which is the first element of the inverse
of the sub-matrix of A
 
a1 b2 0 0 0 ...
 
 b2 a2 b3 0 0 ... 
 
A(1) = 0 b3 a3 b4 0 ... . (C.2.7)

 
 0 0 b4 a4 b5 ... 
.. .. .. .. .. ..
 
. . . . . .

I hope that now is clear that the tridiagonal form leads to an iterative equation that allows us to compute
G0 at the cost of sub-matrices of A which yields a continued fraction equation

1
G0 = 1 , (C.2.8)
a0 − b21 a −b2 1
1 2
..
.
.. 1
. − b2N −1 a
N −1

or in recursive form

1
Gn = (C.2.9)
an − b2n+1 Gn+1
1
GN −1 = . (C.2.10)
aN −1

This iterative approach is particularly relevant when we are dealing with Green’s functions and need to
compute a specific entry or block of a Green’s function. We can usually truncate this method to a number
of coefficients smaller than N − 1. This greatly improves the numerical efficiency since we do not need
(and generally cannot) generate the full basis {|un ⟩}. This approach is valid if the objective is to compute
A−1 N −1,N −1 . In this case, the procedure is just applied backwards. For block tridiagonal matrices, the

generalization is analogous. One just needs to trade the coefficients for the respective matrices and be
careful with the order of matrix multiplication.

69
Appendix D

kx-dependent Model

In this section, we present the kx -dependent Hamiltonian for the quasi-periodic QBCP model. We start
from the Hamiltonian written in the real space basis and take plane wave solutions along the invariant
direction x. Since we are reducing our problem to an effective 1D model H(kx ), we can reach sizes far
closer to the thermodynamic limit. This transformation comes with some caveats, since to have the full
picture of the 2D system, we need to consider the entire FBZ along the invariant direction, and so when
computing gaps, for example, one needs to know exactly at which kx the gap closes. The real-space
Hamiltonian is:

H = H0 + HB + H1D (D.0.1)

Where each term is defined by:

X
H0 = {tz (|R ± aêx , A ⟩⟨ R, A| − |R ± aêy , A ⟩⟨ R, A|) −
R

− tz (|R ± aêx , B ⟩⟨ R, B| + |R ± aêy , B ⟩⟨ R, B|) +

tx
+ (|R ± aêx ± aêy , A ⟩⟨ R, B| − |R ± aêx ∓ aêy , A ⟩⟨ R, B|) +
2 
tx
+ (|R ± aêx ± aêy , B ⟩⟨ R, A| − |R ± aêx ∓ aêy , B ⟩⟨ R, A|) , (D.0.2)
2

X
HB = {i (|R, B ⟩⟨ R, A| − |R, A ⟩⟨ R, B|)
R
B+1
−i (|R ± aêx , A ⟩⟨ R, B| + |R ± aêy , A ⟩⟨ R, B|)
4 
B+1
+i (|R, B ⟩⟨ R ± aêx , A| + |R, B ⟩⟨ R ± aêy , A|) , (D.0.3)
4

W X
H1D = cos (2πβna) {|R, A ⟩⟨ R, A| + |R, B ⟩⟨ R, B|} . (D.0.4)
2
R

70
D kx -dependent Model 71

√1 e−ikx ma |kx , α⟩,

P
We can now make a change of basis along the invariant direction, using |m, α⟩ = Lx kx
P
and obtain the 1D hamiltonian H = kx H(kx ) |kx ⟩⟨ kx |:

H(kx ) = H0 (kx ) + HB (kx ) + H1D (kx ), (D.0.5)

where:

X
H0 (kx ) = {tz (2 cos (kx a) |n, A ⟩⟨ n, A| − |n ± 1, A ⟩⟨ n, A|) +
n

+ tz (−2 cos (kx a) |n, B ⟩⟨ n, B| + |n ± 1, B ⟩⟨ n, B|) +

+ itx sin (kx a) (|n + 1, A ⟩⟨ n, B| − |n − 1, A ⟩⟨ n, B|) +
+ itx sin (kx a) (|n + 1, B ⟩⟨ n, A| − |n − 1, B ⟩⟨ n, A|)} , (D.0.6)

X
HB (kx ) = {i (|n, B ⟩⟨ n, A| − |n, A ⟩⟨ n, B|) −
n
B+1
−i (2 cos (kx a) |n, A ⟩⟨ n, B| + |n ± 1, A ⟩⟨ n, B|) +
4 
B+1
+i (2 cos (kx a) |n, B ⟩⟨ n, A| + |n, B ⟩⟨ n ± 1, A|) , (D.0.7)
4

W X
H1D (kx ) = cos (2πaβn) {|n, A ⟩⟨ n, A| + |n, B ⟩⟨ n, B|} . (D.0.8)
2 n

71
Appendix E

Corner Mode Hybridization

As we have seen, computations of the corner charge for phase II at γ = 1.1 do not yield a quantized
value, despite its scaling with system size. In this appendix, we argue that the lack of quantization of
the corner charge is due to corner mode hybridization, brought up by finite-size effects. In Fig. E.1 the
probability density (|ψ(R)|2 ) of the 6 eigenstates closer to Fermi level for γ = 1.1 and W = 2.2 are shown.
We consider δ = 10−7 to split the degeneracy of the fourfold corner modes, thus, for ED calculations, two
corner modes are shifted towards negative energies E = −δ and the remaining two are shifted towards
positive energies E = δ. Moreover, each of the pairs have their corner modes living in opposite corners
to each other. Therefore, in a half-filled system the two opposite occupied corner modes give rise to an
excess of charge in the corresponding corners ( Q̄ = 0.5) while the unoccupied corner modes lead to a
deficit of charge in the corresponding corners (Q̄ = −0.5).
In Fig. E.1(a) (L = 89) the four middle states are localized in the corners, however, their energies
do not come in pairs and the two modes with negative energy do not live in opposite corners. In fact,
each eigenstate has spectral weight in three of the four corners, suggesting that they hybridize with each
other. This is also seen in the energies of each of the corner modes. Although the spectrum remains
symmetric due to particle hole symmetry, as indicated in Fig. E.1, there are “outer” corner modes with
E ≈ 10−4 and “inner” corner modes with E ≈ 10−5 , which indicates that they hybridize away from
zero-energy (in this case E = ±δ). Furthermore, in Refs. [84, 89], the authors argue that for chiral
QI, each corner mode has a well-defined chiral charge (Π |ψ⟩ = ± |ψ⟩)1 and so when two corner modes
with different corner charges hybridize, their energy moves away from zero-energy, which is the observed
behavior. This hybridization occurs when the localization length of the corner modes is of the order of the
system size (ξl ≈ L), leading to the corner modes decaying into the bulk thus “feeling the influence” of
their neighbors. This is exactly the observed behavior. Considering larger system sizes, as in Fig. E.1(b),
further localizes each mode in its respective corner, impeding corner mode hybridization and consequently
leading to a quantized corner charge. The observed hybridization is dependent on the random phase shifts
realization, with some realizations leading to lower hybridization and so to a well-defined corner charge,
while other realizations lead to higher hybridization between the corner modes and thus to non-quantized
corner charges.

1
Which is the same as stating that each corner mode lives in only one sub-lattice.

72
E Corner Mode Hybridization 73

E = -0.06680891 E = -0.000105 E = -2.351e-05

75 75 75

50 50 50

25 25 25

0 0 0
0 50 0 50 0 50
E = 2.351e-05 E = 0.000105 E = 0.06680891

75 75 75

50 50 50

25 25 25

0 0 0
0 50 0 50 0 50
(a) L = 89

E = -0.06069273 E = -1e-07 E = -1e-07

200 200 200

100 100 100

0 0 0
0 100 200 0 100 200 0 100 200
E = 1e-07 E = 1e-07 E = 0.06069273

200 200 200

100 100 100

0 0 0
0 100 200 0 100 200 0 100 200
(b) L = 233

Figure E.1: Probability density (|ψ(R)|2 ) of the 6 eigenstates around Fermi level for γ = 1.1,W = 2.2
(phase II in Fig. 5.3) for one realization of random shifts and with δ = 10−7 to split the degeneracy of
the four-fold corner modes. Each sub-figure is a different system size, increasing from top to bottom.

73
Appendix F

Complementary Results

Here we complement the results of Chapter 5 with additional plots. In Fig. F.1 and Fig. F.2 we provide
similar plots as those in Fig. 5.6 and Fig. 5.11, respectively, but for γ = 0.5.

0.4 0.3
1.0 W =0.8 W =2.0 W =3.2 W =3.6
0.4
0.2
ρ(E)

0.5 0.2
0.2 0.1

0.0 0.0 0.0 0.0

−2 0 2 −2.5 0.0 2.5 −2.5 0.0 2.5 −2.5 0.0 2.5
0.2
W =4.1 W =5.0 W =6.0 W =8.0
0.2 0.15
0.2
ρ(E)

0.1 0.10
0.1 0.1
0.05

0.0 0.0 0.0 0.00

−5 0 5 −5 0 5 −5 0 5 −5 0 5
E E E E

Figure F.1: DOS as a function of energy (E) for γ = 0.5 and different values of quasi-periodic potential
strength (W ). The following parameters were used in the KPM: system size L = 987, number of Chebyshev
momenta M = 2000 and number of averages over traces and phase shifts R = 10.

W=0.8 W=3.2 W=3.6 W=4.1

3 102
10 101
ρOBC
ρPBC

101
1
10 100
0 0
10 10
−2 0 2 −2.5 0.0 2.5 −2.5 0.0 2.5 −2.5 0.0 2.5
E E E E

Figure F.2: Ratio of open boundary DOS and periodic boundaries DOS as a function of energy for W in
different phases and γ = 0.5. The grayed out regions are bulk gaps opened by the QP modulation. The
following KPM parameters were utilized: L = 987, M = 5000, R = 10.

74
F Complementary Results 75

Constant W Cut
In the main text, we have focused our analysis in two constant γ cuts. However, a constant W cut was
also performed, as seen in Fig. 5.1. We plot several quantities as a function of hopping strength (γ),
such as the quadrupole moment and boundary invariant (a), the bulk spectral gap (b), fractal dimensions
(c-d), corner occupation probability (e), and the corner charge (f).
The most interesting result is regarding phase I which is a gapless QPQI phase. Although this regime
is gapless, the states at Fermi level are localized, which allows for the existence of localized topological
corner modes at zero energy. Despite the topological nature of this regime, Pocc does not quantize since
the numerical ED of the system Hamiltonian yields linear combinations of localized corner modes and
localized bulk states that live at zero energy. Phase II displays several trivial gapped regimes with either
a localized or critical gap edge. From phase II to III the gap closes in a ballistic gap-edge into a gapped
QI phase with well quantized qxy and P . Regarding the corner occupation probability, it plateaus for
most of phase III, however, at around γ ≈ 0.5, it displays two dips that are associated with corner edge
hybridization, which we further discuss in Appendix. F. Furthermore, if we consider the constant γ = 0.5
cut and set W = 3 (end of phase I, Fig. 5.3(a)), a small dip can also be observed in the P invariant, in
Pocc and in Q̄. Phase IV is a band gap trivial insulator. Overall, the results of this analysis agree with
the previous results. However, we have found the existence of a gapless QPQI phase which has not been
seen thus far. This topological gapless regime is phenomenologically similar to the gapless HOTAI phase
found in Ref. [50].

75
F Complementary Results 76

1.0
I II III IV 1.50 I II III IV

P
0.01
1.25
0.8 L:13
L:21 1.00
0.6 L:34

∆E
qxy

0.75
0
0.4 0 0.5
0.50
0.2 0.25
(a) (b)
0.0 0.00
3 3
I II III IV I II III IV

2 2

D2k
D2

1 1

0 (c) 0 (d)

I II III IV I II III IV
1.0 l: 5 0.5
l: 10
L: 21
0.8 l: 15 0.4 L: 34
l: 20 L: 55
0.6 l: 25 0.3
Pocc

Q̄i

l: 30
0.4 0.2

0.2 0.1
(e) (f)
0.0 0.0
0 0.5 1.0 1.5 2.0 0 0.5 1.0 1.5 2.0
γ γ

Figure F.3: (a) qxy and P as a function of W . The P invariant was computed for a system size of L = 610
with 100 averages over phase shifts. qxy was obtained via real space methods with 100 averages over phase
shifts. For even L, θx = θy = 0; for odd L θx = θy = 0. (b) Spectral gap as a function of W computed
for an even system size L = 144 =⇒ θi = 0 and averaged over 50 phase shifts realizations. (c-d) Fractal
dimension averaged out over 100 phase twists, shifts and over the first 8 eigenstates with energies closest
to E = 0. (e) Corner occupation probability (Pocc ) as a function of W for the first 2 zero-energy states.
The results were obtained for a system size of L = 377 and averaged over 100 phase shifts. (f) Corner
charge (Q̄) as a function of W for several system sizes and averaged over 100 phase shifts and over the 4
corners. (e-f) l δ = 10−5 .

76
F Complementary Results 77

Corner-Edge Hybridization

σE=0 (10−5 ) log10 (Eedge )

0 10 20 −6 −4 −2

2π 2π

3π/2 3π/2
φy

φy
π π

L=89
π/2 π/2

0 0
0 π/2 π 3π/2 2π 0 π/2 π 3π/2 2π
φx φx

σE=0 (10−5 ) log10 (Eedge )

0 5 10 −6 −4 −2

2π 2π

3π/2 3π/2
φy

φy

π π

L=144
π/2 π/2

0 0
0 π/2 π 3π/2 2π 0 π/2 π 3π/2 2π
φx φx

σE=0 (10−5 ) log10 (Eedge )

0 2 4 −6 −4 −2

2π 2π

3π/2 3π/2
φy

φy

π π

L=233
π/2 π/2

0 0
0 π/2 π 3π/2 2π 0 π/2 π 3π/2 2π
φx φx

Figure F.4: Standard deviation (σE=0 ) of the energies of the corner modes as a function of the phase
shifts (ϕx , ϕy ) and logarithm of edge state energy (log10 (Eedge )) as a function of the phase shifts (ϕx , ϕy ).
Results obtained for W = 3.8 and γ = 0.2 in the QPQI phase for system sizes L = {89, 144, 233} as
indicated. 77
Bibliography

[1] Klaus von Klitzing. The quantized hall effect. Reviews of Modern Physics, 58:519–531, 7 1986.

[2] K. v. Klitzing, G. Dorda, and M. Pepper. New method for high-accuracy determination of the
fine-structure constant based on quantized hall resistance. Physical Review Letters, 45:494–497, 8
1980.

[3] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs. Quantized hall conductance in a
two-dimensional periodic potential. Physical Review Letters, 49:405–408, 8 1982.

[4] F. D. M. Haldane. Model for a quantum hall effect without landau levels: Condensed-matter real-
ization of the ”parity anomaly”. Physical Review Letters, 61:2015–2018, 10 1988.

[5] B. I. Halperin. Quantized hall conductance, current-carrying edge states, and the existence of ex-
tended states in a two-dimensional disordered potential. Physical Review B, 25:2185–2190, 2 1982.

[6] M. Z. Hasan and C. L. Kane. Colloquium : Topological insulators. Reviews of Modern Physics,
82:3045–3067, 11 2010.

[7] Xiao-Liang Qi and Shou-Cheng Zhang. Topological insulators and superconductors. Reviews of
Modern Physics, 83:1057–1110, 10 2011.

[8] W. P. Su, J. R. Schrieffer, and A. J. Heeger. Solitons in polyacetylene. Physical Review Letters,
42:1698–1701, 6 1979.

[9] Xiao-Liang Qi and Shou-Cheng Zhang. Topological insulators and superconductors. Rev. Mod. Phys.,
83:1057–1110, 2011.

[10] B. Andrei Bernevig with Taylor L. Hughes. Topological Insulators and Topological Superconductors.
Princeton University Press, 2013.

[11] Ching-Kai Chiu, Jeffrey C. Y. Teo, Andreas P. Schnyder, and Shinsei Ryu. Classification of topo-
logical quantum matter with symmetries. Rev. Mod. Phys., 88:035005, Aug 2016.

[12] Alexander Altland and Martin R. Zirnbauer. Nonstandard symmetry classes in mesoscopic normal-
superconducting hybrid structures. Physical Review B, 55:1142–1161, 1 1997.

[13] Andreas P. Schnyder, Shinsei Ryu, Akira Furusaki, and Andreas W. W. Ludwig. Classification of
topological insulators and superconductors in three spatial dimensions. Physical Review B, 78:195125,
11 2008.

78
F Bibliography 79

[14] Liang Fu. Topological crystalline insulators. Physical Review Letters, 106:106802, 3 2011.

[15] Timothy Hsieh, Hsin Lin, Wenhui Duan, Arun Bansil, and Liang Fu. Topological crystalline insulators
in the snte material class. Nature Communications, 7 2012.

[16] M. V. Berry. Quantal phase factors accompanying adiabatic changes. Royal Society, 392:45–57, 3
1984.

[17] P. W. Anderson. Absence of diffusion in certain random lattices. Physical Review, 109:1492–1505, 3
1958.

[18] N.F. Mott and W.D. Twose. The theory of impurity conduction. Advances in Physics, 10:107–163,
4 1961.

[19] Patrick A. Lee and T. V. Ramakrishnan. Disordered electronic systems. Reviews of Modern Physics,
57:287–337, 4 1985.

[20] E. Abrahams, P. W. Anderson, D. C. Licciardello, and T. V. Ramakrishnan. Scaling theory of

localization: Absence of quantum diffusion in two dimensions. Physical Review Letters, 42:673–676,
3 1979.

[21] Binglan Wu, Juntao Song, Jiaojiao Zhou, and Hua Jiang. Disorder effects in topological states: Brief
review of the recent developments. Chinese Physics B, 25:117311, 11 2016.

[22] Emil Prodan, Taylor L Hughes, and B Andrei Bernevig. Entanglement Spectrum of a Disordered
Topological Chern Insulator. Phys. Rev. Lett., 105:115501, 2010.

[23] Emil Prodan. Disordered topological insulators: A non-commutative geometry perspective. Journal
of Physics A: Mathematical and Theoretical, 44(11):113001, mar 2011.

[24] Masaru Onoda, Yshai Avishai, and Naoto Nagaosa. Localization in a Quantum Spin Hall System.
Phys. Rev. Lett., 98:76802, 2007.

[25] Eduardo V. Castro, M. Pilar López-Sancho, and Marı́a A.H. Vozmediano. Anderson localization
and topological transition in Chern insulators. Physical Review B - Condensed Matter and Materials
Physics, 92(8), 2015.

[26] Jian Li, Rui-Lin Chu, J. K. Jain, and Shun-Qing Shen. Topological anderson insulator. Physical
Review Letters, 102:136806, 4 2009.

[27] Serge Aubry and Gilles André. Analyticity breaking and anderson localization in incommensurate
lattices. Proceedings, VIII International Colloquium on Group-Theoretical Methods in Physics, 3, 1
1980.

[28] Magnus Johansson and Rolf Riklund. Self-dual model for one-dimensional incommensurate crystals
including next-nearest-neighbor hopping, and its relation to the Hofstadter model. Phys. Rev. B,
43(16):13468–13475, jun 1991.

79
F Bibliography 80

[29] Miguel Gonçalves, B. Amorim, Eduardo V. Castro, and Pedro Ribeiro. Renormalization group
theory of one-dimensional quasiperiodic lattice models with commensurate approximants. Phys.
Rev. B, 108:L100201, Sep 2023.

[30] Miguel Gonçalves, Bruno Amorim, Eduardo V. Castro, and Pedro Ribeiro. Critical phase dualities
in 1d exactly solvable quasiperiodic models. Phys. Rev. Lett., 131:186303, Nov 2023.

[31] J. H. Pixley, Justin H. Wilson, David A. Huse, and Sarang Gopalakrishnan. Weyl semimetal to metal
phase transitions driven by quasiperiodic potentials. Phys. Rev. Lett., 120:207604, May 2018.

[32] Yixing Fu, Elio J König, Justin H Wilson, Yang-Zhi Chou, and Jedediah H Pixley. Magic-angle
semimetals. npj Quantum Materials, 5(1):71, 2020.

[33] Miguel Gonçalves, Hadi Z Olyaei, Bruno Amorim, Rubem Mondaini, Pedro Ribeiro, and Eduardo V
Castro. Incommensurability-induced sub-ballistic narrow-band-states in twisted bilayer graphene.
2D Materials, 9(1):011001, nov 2021.

[34] Attila Szabo and Ulrich Schneider. Mixed spectra and partially extended states in a two-dimensional
quasiperiodic model. Physical Review B, 101:014205, 1 2020.

[35] Shujie Cheng, Reza Asgari, and Gao Xianlong. From topological phase to transverse anderson
localization in a two-dimensional quasiperiodic system. Phys. Rev. B, 108:024204, Jul 2023.

[36] M. F. Madeira and P. D. Sacramento. Quasidisorder-induced topology. Phys. Rev. B, 106:224505,

Dec 2022.

[37] Raul Liquito, Miguel Gonçalves, and Eduardo V Castro. Fate of quadratic band crossing under
quasiperiodic modulation. arXiv:2309.03896, 2023.

[38] Tiago S Gonçalves, Miguel Gonçalves, Pedro Ribeiro, and Bruno Amorim. Topological phase tran-
sitions for any taste in 2d quasiperiodic systems. arXiv:2212.08024, 2022.

[39] Pranjal Bordia, Henrik Lüschen, Sebastian Scherg, Sarang Gopalakrishnan, Michael Knap, Ulrich
Schneider, and Immanuel Bloch. Probing slow relaxation and many-body localization in two-
dimensional quasiperiodic systems. Physical Review X, 7:041047, 11 2017.

[40] Theorore Corcovilos and Jahnavee Mittal. Two-dimensional optical quasicrystal potentials for ultra-
cold atom experiments. Appl. Opt., 58, 3 2019.

[41] Michael Schreiber, Sean S. Hodgman, Pranjal Bordia, Henrik P. Lüschen, Mark H. Fischer, Ronen
Vosk, Ehud Altman, Ulrich Schneider, and Immanuel Bloch. Observation of many-body localization
of interacting fermions in a quasirandom optical lattice. Science, 349(6250):842–845, 2015.

[42] Fangzhao Alex An, Karmela Padavić, Eric J. Meier, Suraj Hegde, Sriram Ganeshan, J. H. Pixley,
Smitha Vishveshwara, and Bryce Gadway. Interactions and mobility edges: Observing the generalized
aubry-andré model. Phys. Rev. Lett., 126:040603, Jan 2021.

80
F Bibliography 81

[43] Y. Lahini, R. Pugatch, F. Pozzi, M. Sorel, R. Morandotti, N. Davidson, and Y. Silberberg. Observa-
tion of a localization transition in quasiperiodic photonic lattices. Phys. Rev. Lett., 103:013901, Jun
2009.

[44] Yaacov E. Kraus and Oded Zilberberg. Topological equivalence between the fibonacci quasicrystal
and the harper model. Physical Review Letters, 109:116404, 9 2012.

[45] Wenting Cheng, Emil Prodan, and Camelia Prodan. Experimental demonstration of dynamic
topological pumping across incommensurate bilayered acoustic metamaterials. Phys. Rev. Lett.,
125:224301, Nov 2020.

[46] Ze-Guo Chen, Weiwei Zhu, Yang Tan, Licheng Wang, and Guancong Ma. Acoustic realization of
a four-dimensional higher-order chern insulator and boundary-modes engineering. Phys. Rev. X,
11:011016, Jan 2021.

[47] Leon Balents, Cory R Dean, Dmitri K Efetov, and Andrea F Young. Superconductivity and strong
correlations in moiré flat bands. Nat. Phys., 16(7):725–733, jul 2020.

[48] Eva Y. Andrei and Allan H. MacDonald. Graphene bilayers with a twist. Nature Materials,
19(12):1265–1275, Dec 2020.

[49] Wladimir A. Benalcazar, B. Andrei Bernevig, and Taylor L. Hughes. Quantized electric multipole
insulators. Science, 357:61–66, 7 2017.

[50] Yan-Bin Yang, Kai Li, L.-M. Duan, and Yong Xu. Higher-order topological anderson insulators.
Physical Review B, 103:085408, 2 2021.

[51] Motohiko Ezawa. Higher-order topological insulators and semimetals on the breathing kagome and
pyrochlore lattices. Physical Review Letters, 120:026801, 1 2018.

[52] Yichen Xu, Ruolan Xue, and Shaolong Wan. Topological corner states on kagome lattice based chiral
higher-order topological insulator. arXiv:1711.09202, 2017.

[53] Bing Liu, Lede Xian, Haimen Mu, Gan Zhao, Zhao Liu, Angel Rubio, and Z.F. Wang. Higher-order
band topology in twisted moiruperlattice. Physical Review Letters, 126:066401, 2 2021.

[54] Baokai Wang, Xiaoting Zhou, Hsin Lin, and Arun Bansil. Higher-order topological insulator phase
in a modified haldane model. Physical Review B, 104:L121108, 9 2021.

[55] Serra-Garcia, Marc, Peri, Valerio, Süsstrunk, Roman, Bilal, Osama R., Larsen, Tom, Villanueva, Luis
Guillermo, Huber, and Sebastian D. Observation of a phononic quadrupole topological insulator.
Nature, 555(7696):342–345, 2018.

[56] Jia Bao, Deyuan Zou, Weixuan Zhang, Wenjing He, Houjun Sun, and Xiangdong Zhang. Topoelec-
trical circuit octupole insulator with topologically protected corner states. Phys. Rev. B, 100:201406,
Nov 2019.

81
F Bibliography 82

[57] Weixuan Zhang, Deyuan Zou, Qingsong Pei, Wenjing He, Jiacheng Bao, Houjun Sun, and Xiangdong
Zhang. Experimental observation of higher-order topological anderson insulators. Physical Review
Letters, 126:146802, 4 2021.

[58] Frank Schindler, Ashley M. Cook, Maia G. Vergniory, Zhijun Wang, Stuart S. P. Parkin, B. Andrei
Bernevig, and Titus Neupert. Higher-order topological insulators. Science Advances, 4, 6 2018.

[59] Hugo Lóio, Miguel Gonçalves, Pedro Ribeiro, and Eduardo V. Castro. Third-order topological
insulator induced by disorder. Phys. Rev. B, 109:014204, Jan 2024.

[60] C. Wang and X. R. Wang. Disorder-induced quantum phase transitions in three-dimensional second-
order topological insulators. Phys. Rev. Res., 2:033521, Sep 2020.

[61] Rui Chen, Chui-Zhen Chen, Jin-Hua Gao, Bin Zhou, and Dong-Hui Xu. Higher-order topological
insulators in quasicrystals. Phys. Rev. Lett., 124:036803, Jan 2020.

[62] Dániel Varjas, Alexander Lau, Kim Pöyhönen, Anton R. Akhmerov, Dmitry I. Pikulin, and
Ion Cosma Fulga. Topological phases without crystalline counterparts. Phys. Rev. Lett., 123:196401,
Nov 2019.

[63] Miguel A N Araújo and Pedro Sacramento. Topology In Condensed Matter: An Introduction. World
Scientific, 1 edition, 2021.

[64] Lukasz Fidkowski, T. S. Jackson, and Israel Klich. Model characterization of gapless edge modes
of topological insulators using intermediate brillouin-zone functions. Physical Review Letters,
107:036601, 7 2011.

[65] Raffaele Resta. Quantum-mechanical position operator in extended systems. Physical Review Letters,
80:1800–1803, 3 1998.

[66] R. D. King-Smith and David Vanderbilt. Theory of polarization of crystalline solids. Physical Review
B, 47:1651–1654, 1 1993.

[67] Wladimir A. Benalcazar, B. Andrei Bernevig, and Taylor L. Hughes. Electric multipole moments,
topological multipole moment pumping, and chiral hinge states in crystalline insulators. Physical
Review B, 96:245115, 12 2017.

[68] Byungmin Kang, Ken Shiozaki, and Gil Young Cho. Many-body order parameters for multipoles in
solids. Physical Review B, 100:245134, 12 2019.

[69] William A. Wheeler, Lucas K. Wagner, and Taylor L. Hughes. Many-body electric multipole operators
in extended systems. Physical Review B, 100:245135, 12 2019.

[70] Seishiro Ono, Luka Trifunovic, and Haruki Watanabe. Difficulties in operator-based formulation of
the bulk quadrupole moment. Phys. Rev. B, 100:245133, Dec 2019.

[71] Yan-Bin Yang, Jiong-Hao Wang, Kai Li, and Yong Xu. Higher-order topological phases in crystalline
and non-crystalline systems: a review. arXiv: 2309.03688, 2023.

82
F Bibliography 83

[72] A. MacKinnon and B. Kramer. One-parameter scaling of localization length and conductance in
disordered systems. Physical Review Letters, 47:1546–1549, 11 1981.

[73] A. MacKinnon and B. Kramer. The scaling theory of electrons in disordered solids: Additional
numerical results. Zeitschrift für Physik B Condensed Matter, 53:1–13, 3 1983.

[74] Valery Iustinovich Oseledets. A multiplicative ergodic theorem. characteristic ljapunov, exponents
of dynamical systems. Trudy Moskovskogo Matematicheskogo Obshchestva, 19:179–210, 1968.

[75] Alexander Weiße, Gerhard Wellein, Andreas Alvermann, and Holger Fehske. The kernel polynomial
method. Reviews of Modern Physics, 78:275–306, 3 2006.

[76] Henry Ehrenreich, Frederick Seitz, and David Turnbull. Solid State Physics, volume 35. Academic
Press, 1 edition, 6 1950.

[77] R Haydock and C M M Nex. Comparison of quadrature and termination for estimating the density
of states within the recursion method. Journal of Physics C: Solid State Physics, 17:4783–4789, 9
1984.

[78] Yang Peng, Yimu Bao, and Felix von Oppen. Boundary green functions of topological insulators and
superconductors. Physical Review B, 95:235143, 6 2017.

[79] Nicolau Sobrosa, Miguel Gonçalves, and Eduardo V. Castro. Instability of qbc systems to topological
anderson insulating phases. 10 2021.

[80] Junhyun Lee and J. H. Pixley. Emulating twisted double bilayer graphene with a multiorbital optical
lattice. SciPost Phys., 13:033, 2022.

[81] Yu-Liang Tao, Jiong-Hao Wang, and Yong Xu. Quadrupole insulator without corner states in the
energy spectrum. arXiv:2307.00486, 2023.

[82] Jinbing Hu, Songlin Zhuang, and Yi Yang. Synthetic gauge fields enable high-order topology on
brillouin real projective plane. arXiv:2306.15477, 2023.

[83] Liang Fu and C. L. Kane. Topological insulators with inversion symmetry. Phys. Rev. B, 76:045302,
Jul 2007.

[84] Wladimir A. Benalcazar and Alexander Cerjan. Chiral-symmetric higher-order topological phases of
matter. Physical Review Letters, 128:127601, 3 2022.

[85] T. P. Eggarter and R. Riedinger. Singular behavior of tight-binding chains with off-diagonal disorder.
Phys. Rev. B, 18:569–575, Jul 1978.

[86] Ian Mondragon-Shem, Taylor L. Hughes, Juntao Song, and Emil Prodan. Topological criticality in
the chiral-symmetric aiii class at strong disorder. Phys. Rev. Lett., 113:046802, Jul 2014.

[87] Miguel Gonçalves, Bruno Amorim, Eduardo V. Castro, and Pedro Ribeiro. Hidden dualities in 1D
quasiperiodic lattice models. SciPost Phys., 13:046, 2022.

83
F Bibliography 84

[88] C. Lanczos. An iteration method for the solution of the eigenvalue problem of linear differential and
integral operators. Journal of Research of the National Bureau of Standards, 45:255, 10 1950.

[89] Eslam Khalaf, Wladimir A. Benalcazar, Taylor L. Hughes, and Raquel Queiroz. Boundary-obstructed
topological phases. Phys. Rev. Res., 3:013239, Mar 2021.

84