Resta Lecture Notes

Geometry and Topology
in Electronic Structure Theory
Raffaele Resta
Notes subject to ongoing editing

This version run through LATEX on 11–Mar–13 at 14:39
Contents
1 Introduction 1
1.1 About the present Notes . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Notice to the reader . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Why writing these Notes? . . . . . . . . . . . . . . . . . . 1
1.2 What topology is about . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 Gauss-Bonnet theorem . . . . . . . . . . . . . . . . . . . . 2
1.2.2 Euler characteristic . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Electronic wavefunctions . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Gauge and flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6.1 Classical mechanics . . . . . . . . . . . . . . . . . . . . . . 6
1.6.2 Quantum mechanics, open boundary conditions . . . . . . 6
1.6.3 Quantum mechanics, periodic boundary conditions . . . . 7
1.6.4 Example: Free particle in 1d . . . . . . . . . . . . . . . . 7
1.6.5 Flux and flux quantum . . . . . . . . . . . . . . . . . . . 8
2 Early discoveries 10
2.1 The Aharonov-Bohm effect: A paradox? . . . . . . . . . . . . . . 10
2.2 Conical intersections in molecules . . . . . . . . . . . . . . . . . . 11
2.3 Quantization of the surface charge . . . . . . . . . . . . . . . . . 14
2.4 Integer quantum Hall effect . . . . . . . . . . . . . . . . . . . . . 15
2.4.1 Classical theory (Drude-Zener) . . . . . . . . . . . . . . . 15
2.4.2 Landau levels . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4.3 The experiment . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.4 Early theoretical interpretation . . . . . . . . . . . . . . . 18
3 Berryology 21
3.1 Phases and distances . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Berry phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Connection and curvature . . . . . . . . . . . . . . . . . . . . . . 24
3.4 Chern number . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.5 Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.6 Sum over states . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.7 Time-reversal and inversion symmetries . . . . . . . . . . . . . . 27
3.8 NonAbelian case . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.9 Bloch orbitals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
i
4 Manifestations of the Berry phase 32
4.1 A toy-model Hamiltonian . . . . . . . . . . . . . . . . . . . . . . 32
4.1.1 Connection and curvature . . . . . . . . . . . . . . . . . . 32
4.1.2 Chern number . . . . . . . . . . . . . . . . . . . . . . . . 33
4.1.3 Berry phase . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.1.4 Numerical considerations . . . . . . . . . . . . . . . . . . 33
4.2 Early discoveries reinterpreted . . . . . . . . . . . . . . . . . . . . 34
4.2.1 Aharonov-Bohm effect . . . . . . . . . . . . . . . . . . . . 34
4.2.2 Molecular Aharonov-Bohm effect . . . . . . . . . . . . . . 36
4.2.3 Digression: the Z2 invariant . . . . . . . . . . . . . . . . . 37
4.2.4 Integer quantum Hall effect . . . . . . . . . . . . . . . . . 38
4.3 Adiabatic approximation in a magnetic field . . . . . . . . . . . . 39
4.4 Anomalous Hall effect . . . . . . . . . . . . . . . . . . . . . . . . 41
4.5 Semiclassical transport . . . . . . . . . . . . . . . . . . . . . . . . 42
4.5.1 Textbook equations of motion . . . . . . . . . . . . . . . . 42
4.5.2 Modern equations of motion . . . . . . . . . . . . . . . . . 42
4.5.3 Equations of motion in symplectic form . . . . . . . . . . 43
4.5.4 Geometrical correction to the density of states . . . . . . 44
4.5.5 Outstanding consequences of the modified density of states 45
4.6 Quantum transport . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.6.1 Transport by a single state . . . . . . . . . . . . . . . . . 46
4.6.2 Current carried by filled bands . . . . . . . . . . . . . . . 47
4.6.3 Quantization of charge transport . . . . . . . . . . . . . . 47
5 Macroscopic polarization 49
5.1 Polarization and electric field . . . . . . . . . . . . . . . . . . . . 49
5.2 Polarization “itself” vs. polarization difference . . . . . . . . . . 50
5.3 Independent electrons . . . . . . . . . . . . . . . . . . . . . . . . 52
5.3.1 The King-Smith and Vanderbilt formula . . . . . . . . . . 52
5.3.2 The quantum of polarization . . . . . . . . . . . . . . . . 53
5.3.3 Wannier functions . . . . . . . . . . . . . . . . . . . . . . 54
5.3.4 The surface charge theorem . . . . . . . . . . . . . . . . . 55
5.3.5 Noncrystalline systems: The single-point Berry phase . . 57
5.4 Correlated wavefunctions . . . . . . . . . . . . . . . . . . . . . . 58
5.4.1 Single-point Berry phase again . . . . . . . . . . . . . . . 58
5.4.2 Kohn-Sham polarization vs. real polarization . . . . . . . 59
6 Quantum metric and the theory of the insulating state 61

6.1 Nongeometrical theories of the insulating state . . . . . . . . . . 61
6.2 Metric-curvature tensor . . . . . . . . . . . . . . . . . . . . . . . 62
6.2.1 Open boundary conditions . . . . . . . . . . . . . . . . . . 62
6.2.2 Periodic boundary conditions . . . . . . . . . . . . . . . . 63
6.2.3 Sum over states again . . . . . . . . . . . . . . . . . . . . 64
6.3 Geometrical theory of the insulating state . . . . . . . . . . . . . 64
6.3.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.3.2 Linear response . . . . . . . . . . . . . . . . . . . . . . . . 65
6.3.3 Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.3.4 Sum rules . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.3.5 Screened vs. unscreened field . . . . . . . . . . . . . . . . 67
6.4 Localization in the insulating state . . . . . . . . . . . . . . . . . 68
ii
6.4.1 Independent electrons . . . . . . . . . . . . . . . . . . . . 69
6.4.2 Band insulators and band metals . . . . . . . . . . . . . . 71
6.5 Wannier and hermaphrodite orbitals . . . . . . . . . . . . . . . . 72
6.5.1 Wannier orbitals . . . . . . . . . . . . . . . . . . . . . . . 72
6.5.2 Hermaphrodite orbitals . . . . . . . . . . . . . . . . . . . 72
6.5.3 Maximally localized Wannier orbitals . . . . . . . . . . . 74
6.6 Localization in different kinds of insulators . . . . . . . . . . . . 75
6.6.1 Small molecules . . . . . . . . . . . . . . . . . . . . . . . . 75
6.6.2 Band insulators . . . . . . . . . . . . . . . . . . . . . . . . 75
6.6.3 Correlated (Mott) insulators . . . . . . . . . . . . . . . . 76
6.6.4 Disordered (Anderson) insulators . . . . . . . . . . . . . . 77
7 Topological invariants and topological insulators 79

7.1 Chern invariant . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.1.1 Computing the Chern number . . . . . . . . . . . . . . . 80
7.1.2 Chern number as the curvature for unit area . . . . . . . 80
7.1.3 Chern number and hermaphrodite orbitals . . . . . . . . . 82
7.1.4 Lowest Landau level . . . . . . . . . . . . . . . . . . . . . 83
7.2 Quantum Hall insulators: Correlated wavefunction . . . . . . . . 84
7.3 Topological insulators . . . . . . . . . . . . . . . . . . . . . . . . 85
7.3.1 Haldanium . . . . . . . . . . . . . . . . . . . . . . . . . . 86
7.3.2 Time-reversal symmetric topological insulators . . . . . . 87
7.4 Digression: Bulk-boundary correspondence . . . . . . . . . . . . 88
7.5 Topological order in r space . . . . . . . . . . . . . . . . . . . . . 88
8 Orbital magnetization 90
8.1 Magnetization and magnetic field . . . . . . . . . . . . . . . . . . 91
8.2 Magnetization “itself” . . . . . . . . . . . . . . . . . . . . . . . . 92
8.2.1 The operator P r Q . . . . . . . . . . . . . . . . . . . . . . 93
8.2.2 Magnetization in k space . . . . . . . . . . . . . . . . . . 94
8.2.3 Magnetization in r space . . . . . . . . . . . . . . . . . . 95
Bibliography 99
iii
iv
Chapter 1
Introduction
1.1 About the present Notes

1.1.1 Notice to the reader
This version of the Notes, posted at http://www-dft.ts.infn.it/~resta/ is
not final, and probably will never be such. On the contrary, these Notes are
likely to be continuously edited, updated, and augmented. Subsequent versions
will replace the present one at the same URL.
It is easy to guess these Notes to be plagued by many typos and even more
serious errors. I kindly ask any reader spotting such occurrences to notify me
at resta@democritos.it.
1.1.2 Why writing these Notes?

The occasion, or I should better say the pretext, for starting to these Notes
is a graduate course delivered at SISSA (International School for Advanced
Studies, Trieste) in 2012 and covering a part—and a part only—of the topics
dealt with here. Further editing continues towards an advanced undergraduate
course delivered at the University of Trieste from March 2013 onwards.
The need for some support material for such courses is hardly compelling,
since many very good review papers exist, whose union covers about the whole
of the present topics, all of them state-of-the-art at the time of their publication.
Some of these reviews are authored or coauthored by me [1, 2, 3, 4, 5, 6, 7, 8, 9],
and many more by outstanding colleagues. Of these, I quote only the most
recent ones; a non exhaustive list is [10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20].
The writing of a set of Lecture Notes in due form appears as a heavy burden,
diverting for a significant amount of time the author from his day-to-day research
work, already also diverted by administration and teaching. Instead, I used
above the word “pretext”, because these Notes are mostly written for my own
sake. In fact I regard such burden as an important occasion for a pause of
reflection, devoted to critical rethinking about the subject in its wholeness. In
the present case, the pause of reflection was made possible by an extended stay
(four months) in 2011-12 at CECAM in Lausanne, and another (three months)
in 2012-13 at DIPC in San Sebastian.
I wrote Lecture Notes in four other occasions, last time in 2000 [21]. I
know by experience that the task of writing in uniform notations and in a
1
Figure 1.1: The hallmark of topology,
as in many popular presentations. Hun-
dreds of figures like this, and even some
very perspicuous videos, can be down-
loaded from the internet. The two
closed surfaces (“two-dimensional com-
pact manifolds”) have the same topo-
logical invariant g = 1, which measures
the number of handles.
logical sequence many results scattered in the literature leads me to scrutinize

from a novel viewpoint not only other people’s work, but even my own. Here
LATEX plays a major role; soon after its appearance in the early 1980s my way
of thinking has changed dramatically. While previously I—like everybody—
reasoned by jotting formulas on a sheet of paper, nowadays I cannot reason
clearly unless the formulas are neatly typeset in LATEX.
The very fact of writing many different known results altogether has the
beneficial effect of making quite evident several links, hitherto unsuspected.
The perspective view gained while writing the present Notes (and typesetting
them in LATEX) will doubtless influence the future course of my research activity,
as in fact already happened in the previous occasions.
1.2 What topology is about

Topology is defined as a branch of mathematics that describes properties which
remain unchanged under smooth deformations; such properties are usually la-
belled by integer numbers, named topological invariants. The concepts and
tools belonging to topology are continuity and connectivity, open and closed
sets, neighborhoods, and the like.
Differentiability, or even a metric structure, are not needed; theorems are
proved under very general hypotheses, and are therefore very powerful, being
applicable to very diverse frameworks. The tradeoff is that proofs, and even def-
initions, look clumsy and obscure to readers with the mathematical background
of a typical condensed matter physicist. The good news is that the topological
properties most relevant for electronic structure theory can be formulated in the
more familiar language of differential geometry.
Many introductions to topology start with the statement that, to a topolo-
gist, a coffee cup and a doughnut are the same thing, as in Fig. 1.1. Intuitively,
the common feature of the two objects is the presence of one, and only one,
handle. The mathematical definition of “handle” is coming soon.
1.2.1 Gauss-Bonnet theorem

We start with the simplest example, a sphere, and a tangent plane at a given
point. In a local system of Cartesian coordinates on the plane the equation of
2
Figure 1.2: A sphere of radius R, and
its tangent plane in a generic point. The
Gaussian curvature in this trivial case is
Ω = 1/R2 .
the sphere is
p x2 + y 2
z =R− R2 − x2 − y 2 ≃ , (1.1)
2R
and the Hessian matrix is
∂2 z ∂2z
! !
1/R 0
∂x2 ∂x∂y
H= ∂2 z ∂2z
= . (1.2)
∂y∂x ∂y 2 0 1/R
The Gaussian curvature Ω is by definition the determinant of the Hessian at the

tangency point. It is obviously constant and equal to 1/R2 at any point of the
sphere; notice that the orientation of the z axis (either inwards or outwards) is
irrelevant. The integral of Ω over the whole closed surface is 4π.
Next we consider a smooth (i.e. twice differentiable) closed surface of arbi-
trary shape: the Gaussian curvature is defined as the determinant of the Hessian
at the tangent plane, similarly to what we did for the sphere:
2 2
!
∂ z ∂ z
∂x2 ∂x∂y
Ω = det ∂2 z ∂2z
. (1.3)
∂y∂x ∂y 2
In general, Ω can be positive, negative (at a saddle point), or zero (e.g. for
a plane or a cylinder). The Gauss-Bonnet theorem states that for any closed
smooth surface
1
Z
dσ Ω = 2(1 − g), (1.4)
2π S
where g is a nonnegative integer, called the “genus” of the surface. Surfaces
which can be continuously deformed into each other (i.e. “homeomorphic”)
have the same genus. For the sphere and any surface homeomorphic to it g = 0;
both the coffee cup and the doughnut, Fig. 1.1 have g = 1; a double-handle
cup has g = 2. The genus is thus the mathematical definition for the number
of handles.
1.2.2 Euler characteristic

We have considered smooth surfaces so far, but topological invariants are
based on the more general condition of continuity, and—to the delight of
mathematicians—unsurprisingly can be defined even for pathological surfaces
3
(“manifolds” in topology-speak). The simplest non smooth case addresses poly-
hedra, where the Gaussian curvature is either zero (on the faces) or singular (at
vertices and edges).
The Euler characteristic is defined as χ = V − E + F , where V is the num-
ber of vertices, E is the number of edges, and F is the number of faces. If we
address the set of regular polyhedra (tetrahedron, cube, octahedron, dodeca-
hedron, icosahedron) it is easily verified that χ = 2. All these surfaces can be
continuously deformed into (“are homeomeorphic to”) each other, and into a
sphere. In fact there is a one-to-one relationship between the Euler character-
istic and the genus: χ = 2(1 − g). Polyhedra can also have χ 6= 2, like the
doughnut-shaped one shown in Fig. 1.4.
1.3 Electronic wavefunctions

In the domain of electronic structure, the typical object addressed via geomet-
rical and/or topological concepts is the electronic ground state of some system.
Whenever an observable effect has the nature of a topological invariant, i.e.
it is an integer number, two remarkable features occur. (1) The observable
is measurable in principle with infinite precision (10−9 is actually attained for
the quantum Hall effect). (2) The observable is very robust under even strong
variations of the sample conditions; a very disruptive perturbation is needed to
switch from one integer to another. Topology concerns mostly insulators: in
this case the disruptive perturbation amounts to crossing a metallic state.
These Notes are entirely devoted to physical properties having a topological
and/or geometrical character. I am not sure of always using the right seman-
tics. Loosely speaking, I would use the term “topological” for something which
is quantized, and “geometrical” for something which is not. The framework
and the mathematical tools are often the same for quantized and nonquantized
quantities, the former frequently occurring as special cases of the latter.
The Berry phase is the typical geometrical quantity which is not quantized,
although it can be quantized in high-symmetry cases. The macroscopic po-
larization of a solid is a Berry phase, and is obviously (from an experimental
viewpoint) a nonquantized observable. Nonetheless, there are aspects of the
modern theory of polarization that I would define topological. The same ap-
plies to other geometrical properties considered in this Notes. It is reassuring
that even other authors often use “geometrical” and “topological” as synony-
Figure 1.3: Some surfaces and their genus
4
Figure 1.4: A doughnut shaped polyhe-
dron. This surface has Euler character-
istic χ = 0 or, equivalently, genus g = 1.
mous, and that an illustrious author like Michael Berry confesses an original
mistake about the semantics [22].
Finally, a few words about the many calculations cited and sometimes briefly
outlined here. Unless otherwise stated, the term “first-principle calculations”,
when referred to a condensed matter system, means density functional calcula-
tions; independent-electron eigenfunctions and eigenvalues are the Kohn-Sham
(KS) ones. Despite these Notes mostly address a computational physics reader-
ship, no technical details are given (basis sets, pseudopotentials, functionals...);
they are obviously detailed in the original literature, while the focus here is on
the physical properties.
1.4 Units
We use Gaussian electromagnetic units throughout: these have the advantage
(at variance with SI units) that electric and magnetic fields have the same dimen-
sions. Furthermore, the nasty ε0 and µ0 disappear; SI formulas are converted
by setting 4πε0 = 1 and 4π/µ0 = 1.
For a single particle, the Newton equation of motion and the Hamiltonian
read, respectively
dv 1
M =f =Q E+ v×B , (1.5)
dt c
2
1 Q
H= p − A(r) + QΦ(r). (1.6)
2M c
Generally, Gaussian electromagnetic units are associated to mechanical cgs
units, but this is no means necessary. In electronic structure theory, it is ex-
pedient to associate Gaussian electromagnetic units with atomic units (a.u.),
defined as e2 = 1, me = 1, ~ = 1. The unit of energy is the hartree (1 Ha = 2
Ry = 27.21 eV). In the present Notes the electron charge is −e, with e > 0; this
sign choice agrees with most (but not all) the recent literature. For instance, the
very popular review of Ref [1] adopts the symbol e for the “algebraic” electron
charge (e < 0).
The speed of light in a.u. is c = 137. This immediately hints at why the
largest atomic number Z in the periodic table is Z ≃ 100: in fact the core
electrons have (in a.u.) energies of the order of Z 2 , hence velocities of the order
of Z.
1.5 Symbols
I am faced here with two contrasting issues: adopting the symbols most currently
used in the literature, and adopting different symbols for different objects. This
5
proved to be near to impossible in a work of the present kind, if baroque symbols
are to be ruled out. For instance along the present Notes I do use A, A, A, A, A ,
all with a different meaning. Similarly, I use P, P̂ , P, P, P. Despite this, I
found unavoidable to use—in different Chapters—the same symbol for different
objects. For instance, depending on the context, the symbol P may indicate
a projector or, otherwise, a one-dimensional electrical polarization. Therefore
caution is in order when extrapolating a given symbol from its own context.
1.6 Gauge and flux

We consider here a simple exercise which plays the role of a very important
paradigm; it illustrates basic concepts and results which are going to reappear
several times all along the present Notes.
We address the single-particle Hamiltonian
1 e
H= (p + A)2 + V (r), (1.7)
2m c
where the vector potential A is independent of space and time. It is usually
said that A is a pure gauge, meaning with this that it does not affect the fields:
1 ∂A
B = ∇ × A ≡ 0, E=− ≡ 0. (1.8)
c ∂t
1.6.1 Classical mechanics

Let us first adopt a classical viewpoint. The Hamilton equation of motions are
∂H
ṗ = − = −∇V (r) (1.9)
∂r
∂H 1 e
ṙ = = (p + A). (1.10)
∂p m c
From these we get
e
p = mṙ − A, (1.11)
c
which leads to the Newton equation of motion
mr̈ = −∇V (r). (1.12)
The bottom line looks quite obvious: a pure gauge has no effect. A basic tenet
of classical mechanics is that the equations of motion can always be directly
expressed in terms of the forces (i.e. the fields), while the potentials—scalar
and vector—are auxiliary quantities, devoid of physical meaning.
1.6.2 Quantum mechanics, open boundary conditions

Next we switch to quantum mechanics. It is expedient to rewrite Eq. (1.7) as
1 e
H(κ) = (p + ~κ)2 + V (r), κ= A, (1.13)
2m c~
6
where κ, having the dimensions of an inverse length, will be referred to as
“twist” in the following. The Schrödinger equation is
H(κ)|ψn (κ)i = ǫn (κ)|ψn (κ)i. (1.14)
The eigenvectors and eigenvalues of the Schrödinger equation depend on the

boundary conditions assumed.
The so-called open boundary conditions (OBCs) require that the bound
eigenstates are square-integrable over R3 . Let |ψn (0)i be a nondegenerate
eigenstate of the “untwisted” Hamiltonian within OBCs. Then the state
e−iκ·r |ψn (0)i obviously obeys OBCs as well, and also obeys Eq. (1.14) with
a κ-independent eigenvalue. Therefore it coincides with the n-th eigenstate
|ψn (κ)i of the twisted Hamiltonian; notice that this eigenstate is arbitrary by
a κ-dependent phase factor.
We conclude that a pure gauge within OBCs affects the wavefunction, but
does not affect any of the observable quantities, such as expectation values,
density, and current. We spell this, in jargon, by saying that the “twist” is
easily “gauged away” within OBCs.
1.6.3 Quantum mechanics, periodic boundary conditions

We assume periodic boundary conditions (PBCs) over a cubic box of side L, i.e.
we require the eigenstates of Eq. (1.14) to be Born-von-Kàrmàn periodic with
period L over x, y, and z at any given κ. Each Cartesian coordinate is therefore
equivalent to an angle, e.g. ϕx = 2πx/L.
If |ψn (0)i is an eigenstate of the untwisted Hamiltonian within PBCs, then
the state e−iκ·r |ψn (0)i obeys Eq. (1.14) with a κ-independent eigenvalue, but
for a general κ it does not obey PBCs, and therefore in general does not coincide
with the genuine eigenstate |ψn (κ)i. Within PBCs the spectrum of Eq. (1.14)
depends on the twist κ in a nontrivial way.
If |ψn (κ)i is an eigenstate of Eq. (1.14) within PBCs with eigenvalue ǫn (κ),
then the auxiliary state |ψ̃n (κ)i = eiκ·r |ψn (κ)i obeys the untwisted (κ = 0)
Schrödinger equation, and quasi-periodical (a.k.a. “twisted” or “skewed”)
boundary conditions: at any two opposite faces of the cube the wavefunction
differs by a κ-dependent phase factor.
In other words the problem can be formulated in two equivalent ways: either
the Hamiltonian is κ-dependent, as in Eq. (1.14), and the boundary conditions
are κ-independent; or the Hamiltonian is κ-independent but the boundary con-
ditions are “twisted” in a κ-dependent way.
1.6.4 Example: Free particle in 1d

For the sake of simplicity, we consider Eq. (1.14) in 1d, and with V ≡ 0:
2
~2

d
−i +κ |ψn (κ)i = ǫn (κ)|ψn (κ)i. (1.15)
2m dx
The eigenfunctions within PBCs and the spectrum are

2πn
hx|ψn (κ)i ∝ ei L x , n ∈ Z, (1.16)
7
2
~2

2πn
ǫn (κ) = +κ , (1.17)
2m L
where the nontrivial κ-dependence is perspicuous. The velocity operator can be
written as
1 ∂H
v= , (1.18)
~ ∂κ
and the Hellmann-Feynman theorem yields
1 dǫn (κ)
hψn |v|ψn i = . (1.19)
~ dκ
We have introduced PBCs as a basic framework of condensed matter physics.
Many concepts (like the Bloch vector or the Fermi surface) make sense only
within PBCs. But we also may regard this problem as if the electrons were
confined to a circular rail of circumference L, as in Fig. 1.5. There is no field
(electric or magnetic) on the rail, but a constant vector potential of intensity
A = c~κ/e is present along the rail; eigenvectors and eigenfunctions depend on
its value.
1.6.5 Flux and flux quantum

The constant vector potential A on the circular rail corresponds to a magnetic
flux φ = LA threading the surface encircled by the rail, in a region not vis-
ited by the electronic system; it has been appropriately called by some authors
“inaccessible flux”.
We further observe that the spectrum, Eq. (1.17), is periodic in κ with period
2π/L; alternatively, it is periodic in the flux φ with period φ0 = 2π~c/e = hc/e,
the elementary flux quantum. In cgs units hc/e = 4.135×10−7 gauss cm2 , while
in SI units φ0 = h/e = 4.136 × 10−15 Wb. Notice also that, in the framework
of superconductivity, the same symbol φ0 indicates one half of this (it refers to
electron pairs).
We stress that only the fractional part of the flux affects the results in a
nontrivial way. This is perspicuous if we recast Eq. (1.17) as
2 2
~2

2π φ
ǫn (φ) = n+ . (1.20)
2m L φ0
The flux breaks time-reversal symmetry (κ → −κ), and the spectrum is non-
degenerate, except when φ = 0 or φ = φ0 /2, the latter also called “π flux”. In
Figure 1.5: The electron motion is con-

fined to a circular rail. A constant vec-
tor potential A = c~κ/e along the rail,
as in Eq. (1.15), corresponds to vanish-
ing fields (electric and magnetic), yet
the spectrum depends on the “inaccessi-
ble flux” threading the surface encircled
by the rail.
8
these two cases (and in these cases only) the eigenfunctions can be chosen as
real.
When the flux is varied with time, an emf is induced along the loop. Using
Eq. (1.19), the current is
dǫn
I = −ev = −c . (1.21)
dφ
This result is remarkable: it holds even in presence of a potential V (x), and
generalizes straightforwardly to N noninteracting electrons. It will be used in
the discussion of the quantum Hall effect: see Eq. (2.17) below.
9
Chapter 2
Early discoveries
2.1 The Aharonov-Bohm effect: A paradox?

The Aharonov-Bohm effect is the paradigm for a measurable effect induced by
an inaccessible flux. We anticipate that in many other phenomena such flux may
be purely “geometrical” or “topological”, without any relationship to a genuine
magnetic field: this is e.g., the case considered in the next Section. It is only
in the Aharonov-Bohm effect that one addresses indeed the inaccessible flux of
a magnetic field, as present e.g. inside a solenoid. An interference experiment
detects the presence of the flux even when the electronic motion is confined in
the region outside the solenoid, where the magnetic field is zero. This seems
paradoxical: something which “happens” in a region not visited by the quantum
particle may affect some observable properties. Indeed, the founding fathers of
quantum mechanics (in the 1920s) failed to notice such peculiar feature. It only
surfaced more than 30 years afterwards in the milestone paper by Aharonov and
Bohm [23], appeared in 1959, whose abstract states verbatim “...contrary to the
conclusions of classical mechanics, there exist effects of potential on charged
particles, even in the regions where all the fields (and therefore the forces on the
particles) vanish”.
The paper was shocking, and its conclusions were challenged by several au-
thors; nonetheless experimental validations appeared as early as 1960 [25, 26].
The main message of Ref. [23] is at the basis of many subsequent developments
Figure 2.1: The Aharonov-Bohm interference experiment (From Ref. [24])
10
in electronic structure theory, many of them illustrated below in the present
Notes. The Aharonov-Bohm effect is also at the root of the commercial SQUID
technology [27].
It is remarkable that R. P. Feynman included the Aharonov-Bohm effect
in his legendary lectures, delivered to the sophomore class at Caltech during
the 1962-63 academic year [24]. In the final sentence about this topic, Feynman
says: “...E and B are slowly disappearing from the modern expression of physical
laws; they are being replaced by A and Φ”.
It is also remarkable and shameful that a paper bearing the title “Nonex-
istence of the Aharonov-Bohm effect” [28] was published as late as 1978. All
challenges disappeared with the publication in 1984 of the celebrated paper by
Michael Berry (now Sir Michael Berry [29]), where the eponymous phase made
its first appearance [30].
2.2 Conical intersections in molecules

At the time Berry wrote his famous paper, only two occurrences of a geometrical
phase (called Berry phase soon afterwards) in quantum mechanics were known to
him: the Aharonov-Bohm effect and a somewhat exotic phenomenon occurring
in molecular physics. Even the latter was known since the late 1950s [31, 32],
and appropriately rebaptized in the late 1970s as “molecular Aharonov-Bohm
effect” [33, 34]. In the subsequent years Berry phases were discovered in many
branches of physics.
The smallest molecular system where the molecular Aharonov-Bohm effect is
possible is a trimer, having three internal coordinates (e.g. the three internuclear
distances), and the simplest trimers are of course the homonuclear ones, where
symmetry plays a major role. I give a simple outline for this particular system:
a dynamical Jahn-Teller effect, bearing the conventional symmetry label E ⊗ ε.
We focus on a trimer of monovalent atoms, e.g. H3 or Na3 , and we assume
an independent-electron picture in the Born-Oppenheimer approximation. We
start with the molecule in the equilateral configurations, Fig. 2.2. Two of the
valence electrons occupy a totalsymmetric orbital, while the unpaired electron
occupies the next available one, which has E symmetry and is doubly degener-
ate. In a simple tight-binding (alias minimal-basis LCAO) scheme, a possible
basis in the two-dimensional manifold is:
1 1
|1i = √ ( |Bi−|Ci ) ; |2i = √ ( 2|Ai−|Bi−|Ci ) , (2.1)
2 6
where A,B,C are atomic labels (as in the figure). This choice deserves an impor-
tant comment. We are adopting OBCs, as appropriate for an isolated molecule,
and the Hamiltonian is invariant under time reversal (no magnetic field, no
spin-orbit interaction). These two conditions guarantee that the orbitals may
always be chosen as real. They may, but they don’t need: it may instead be
convenient to choose a complex basis in the same two-dimensional degenerate
manifold.
Figure 2.2: A homonuclear trimer in its equilateral configu-

ration.
11
Figure 2.3: A schematic representation of a (counterclockwise) pseudorota-
tion, where subsequent snapshots differ by 2π/3. The corresponding electronic
ground states in the tight-binding approximation, are also shown.
When we distort the molecule from its equilateral configuration, the doublet
is linearly split: one of the two components is energetically favored, the molecule
undergoes Jahn-Teller distortion, and the electronic ground state in the Born-
Oppenheimer approximation becomes nondegenerate.
Next we analyze the motion of the nuclei. There are three linearly inde-
pendent normal modes for the small oscillations of the internal coordinates. Of
course, in absence of a Jahn-Teller effect, the equilateral configuration is the
equilibrium one. One mode is totalsymmetric, and cannot split the electronic
levels. The remaining modes are degenerate, having in fact E symmetry, and
couple to the electronic doublet, originating in fact the dynamical Jahn-Teller
effect. The notation E ⊗ ε means indeed that an E vibrational mode is coupled
to an E electronic state: conventionally, one uses upper case letters as symmetry
labels for the vibrational states, and lower case ones for the electronic states.
The adiabatic electronic ground state follows the nuclear motion. For a
cyclic pseudorotation, shown in Fig. 2.3, the Hamiltonian is periodical, but the
electronic wavefunction is antiperiodical. The total wavefunction in the Born-
Oppenheimer approximation factors into the electronic one times the nuclear
one. Given that the total wavefunction must be single-valued, even the nuclear
wavefunction must be quantized using antiperiodical boundary conditions, and
this affects the pseudorotation spectrum in a measurable way.
This feature has to do with the peculiar shape of the Born-Oppenheimer
surface, shown in Fig. 2.4. If we adopt a two-dimensional Cartesian normal
coordinate ξ = (ξ1 , ξ2 ), the ionic displacements are parametrized as:
xA = ξ1 √
yA = ξ2 √
xB = − 21 ξ1 + 3
√2 2
ξ yB = − 3
√ 2 1
ξ − 21 ξ2
xC = − 12 ξ1 − 3
2 ξ2 yC = 3 1
2 ξ1 − 2 ξ2
Figure 2.4: The Born-Oppenheimer sur-

face of the Jahn-Teller split doublet: a
double-valued function with a conical
intersection (a.k.a. diabolical point).
The potential minimum is a circle of
radius ξmin centered at the degeneracy
point.
12
The meaning of this coordinate choice is transparent with reference to Fig 2.3:
when the atom A is displaced by ξ, the displacements of B and C are of equal
magnitude |ξ|, but pointing in directions rotated by −2π/3 and −4π/3, re-
spectively. If we neglect Jahn-Teller coupling beyond linear order, no potential
energy is associated to a motion at constant |ξ|, which is indeed a free pseudoro-
tation (or a “rotation wave”), also schematized in the succession of snapshots
in Fig. 2.3.
In absence of Jahn-Teller coupling, the surface would simply be a parabola,
everywhere doubly degenerate. The linear Jahn-Teller splitting is function of
|ξ|, hence to linear order the electronic eigenvalues are:
E± (ξ) ∝ |ξ|2 ± const |ξ|. (2.2)
This double-valued function is displayed in Fig. 2.4 and has a conical intersection
at the origin. The double cone is also called a diabolo (after a spinning toy of
the same shape), so the degeneracy points are also called “diabolical”. The
lowest sheet E− (ξ) has a circular valley of radius ξmin , where a classical particle
travels freely (if nonlinear Jahn-Teller coupling is neglected). Nothing exotic
happens if the nuclear motion can be considered as classic; but when we quantize
the nuclear degrees of freedom, antiperiodical boundary conditions have to be
imposed for the cyclic motion, as said above.
A simple approximation for the rotovibrational levels is thus:
1
E(u, j) = (u + ) ω0 + Aj 2 , (2.3)
2
corresponding to an oscillation of frequency ω0 and quantum number u, and
a two-dimensional internal rotation with rotor constant A. The antiperiodical
boundary conditions imply half-odd-integer values for the quantum number j;
notice that the ground state (j = ±1/2) is twofold degenerate.
The pseudorotation term in the spectrum can be compared to Eq. (1.20);
the moment of inertia in the prefactor becomes obviously a nuclear rotor, but
the spectrum is the same if we identify the inaccessible flux φ with half a flux
quantum φ0 (a.k.a. π flux).
There is no magnetic field in this problem; the flux is purely topological
and can be regarded as an obstruction: the nuclear path cannot be contracted
without crossing a degeneracy point. It is remarkable that the topological nature
of this problem was clearly stated as early as 1963—much earlier than topology
became fashionable in electronic structure—by Herzberg and Longuet-Higgins
[32], who say verbatim: “...a conically self-intersecting potential surface has a
different topological character from a pair of distinct surfaces which happen to
meet at a point. Indeed, if an electronic wave function changes sign when we
move round a closed loop in configuration space, we can conclude that somewhere
inside the loop there must be a singular point at which the wave function is
degenerate”.
In modern jargon, we would say that the cases φ = 0 and φ = φ0 /2 are topo-
logically distinct; owing to time-reversal invariance, other flux values are ruled
out. We anticipate that the present problem is revisited in order to introduce
the Z2 topological invariant in Sec. 4.2.3.
The present paradigm also illustrates the robustness of topological proper-
ties against smooth deformations. For instance, here we have addressed the
13
ultrasimple tight-binding model, but the ground wavefunction can be “contin-
uously deformed” to the exact correlated wavefunction: topology-wise, the two
wavefunctions are essentially the same object, insofar as the conical intersection
is present. Notice also that at the conical intersection the Born-Oppenheimer
approximation breaks down.
One could also address more general closed paths, according to their winding
number round the obstruction. Only paths having the same winding number
can be continuously deformed into each other: they are “homotopic”.
2.3 Quantization of the surface charge

The pioneering selfconsistent calculations of the electronic structure of surfaces,
performed at IBM (Yorktown Heights) and at Bell Labs in the mid 1970s,
pointed out the occurrence of quantization of charge at insulating surfaces.
After an early paper by V. Heine in 1966 [35], the theorem made its appearance
in a 1974 paper by Appelbaum and Hamann [36]. Other papers addressed the
issue in the 1970s [37, 38], but the topological explanation came much later; it
will be discussed below, Sec. 5.3.4.
The quantization of surface charge may appear counterintuitive, if one sticks
at the idea that a solid is an array of classical charges (ions), as many people
still do. Possibly because of its counterintuitive content, this important theorem
is surprisingly ignored even by well known specialists in surface physics. The
extreme of such ignorance occurs in a recently published invited review paper—
which I abstain from quoting—about polar surfaces. The theorem is even more
ignored in quantum chemistry, where it addresses end charges in linear polymers.
Electrons are quantum particles, and classical ideas may prove wrong. Solids
are not assemblies of ions; they are assemblies of atoms, having ionic character
only because neighboring atoms have a different electronegativity [39]. At the
surface, one has to look at what happens to the bonds.
A simple statement of the theorem is the following. If the bulk of the crystal
is centrosymmetric, and if the surface is insulating, then the charge per surface
cell may only be an integer or half integer; the surface charge can be nonquan-
tized only if the bulk is noncentrosymmetric, or if the surface is metallic.
Quite often, the actual quantized value is zero because of energy consider-
ations; therefore even polar surfaces are (counterintuitively) neutral under the
above two essential hypotheses, which I stress again: the bulk is centrosymmet-
ric, and the surface is insulating. The microscopic mechanism can be understood
as an intrinsic surface-state neutralization [39]; however, topology guarantees
quantization independently of microscopic details.
We nowadays regard bulk-surface correspondence in many phenomena as
one of the hallmarks of geometry and topology in condensed matter physics. In
modern jargon, I would say that the surface charge of insulators is “topologically
protected”. More about the bulk-boundary correspondence will be said in Sec.
7.4.
14
2.4 Integer quantum Hall effect
2.4.1 Classical theory (Drude-Zener)
We consider any 2d system, in the setup shown in Fig. 2.5. If dissipation is
accounted for by a single relaxation time τ , the Newton equation of motion for
a single carrier of mass m and charge −e, is

dv 1 1
m + v = −e E + v × B . (2.4)
dt τ c
Setting dv/dt = 0 we get the steady-state solution:

eτ 1
v=− E+ v×B . (2.5)
m c
eB
In terms of the cyclotron frequency ωc = mc the solution with Ey = 0 is
eτ
vx = − Ex − ωc τ vy
m
vy = ωc τ vx . (2.6)
If n is the carrier density, the current is j = −ne v:
ne2 τ
jx = Ex − ωc τ jy
m
jy = ωc τ jx . (2.7)
in zero B field we retrieve the standard Drude (diagonal) conductivity:
ne2 τ
jx = σ0 Ex , σ0 = , (2.8)
m
while for B 6= 0 the conductivity tensor is:
σ0
jx = Ex = σxx Ex
1 + (ωc τ )2
ωc τ σ0
jy = Ex = σyx Ex . (2.9)
1 + (ωc τ )2
Inversion of the conductivity tensor

σxx σyx
ρxx = 2 + σ2
ρxy = 2 + σ2
(2.10)
σxx yx σxx yx
Figure 2.5: Hall effect in a 2d system.

The E field is applied along x, while the
B field is along z. The system is shorted
in the y direction; the current j has both
longitudinal (x) and transverse, a.k.a.
Hall (y) components.
15
provides a remarkably simple expression for the longitudinal and transverse
resistivity
m mωc 1
ρxx = 1/σ0 = 2 , ρxy = = B. (2.11)
ne τ ne2 nec
The Hall resistivity is therefore linear in B and independent of both mass and
relaxation time; more accurately, since we may consider even carriers of positive
charge e (“holes”), its sign does depend on the carrier charge. Notice also that
in the nondissipative regime (τ ≫ 1/ωc ) both ρxx and σxx vanish.
If we write n as N/A (number of carriers per unit area), then
1 φ
ρxy = B= , (2.12)
nec N ec
where φ = AB is the magnetic flux through area A. Although we are still at a
purely classical level, it is instructive to multiply and divide ρxy by h. We may
thus identically write
1 h N φ0
ρxy = , ν= . (2.13)
ν e2 φ
Here φ0 = hc/e is the flux quantum, as defined above. The dimensionless
quantity ν, called the filling factor, equals the ratio between the number of
electrons N and the number of flux quanta φ/φ0 . Eq. (2.13) expresses the
transverse resistivity in terms of the natural resistance unit h/e2 . Since 1990
this is a new metrology standard, accurate to more than nine figures: 1 klitzing
= h/e2 = 25812.807557(18) ohm.
In 2d resistance and resistivity have the same dimensions, and coincide in
the transverse case. We write therefore the Hall resistance as
1 h
Vy /Ix = RH ≡ Rxy = . (2.14)
ν e2
Upon obvious dimensionality arguments, even in the quantum case the Hall
resistance can be written in this way; but then the concentration- and B-
dependence of ν are expected to be very different from the simple monothonical
form of Eq. (2.13).
2.4.2 Landau levels

In quantum mechanics, the Schrödinger equation for an electron in 2d subject
to a perpendicular B field (and in a flat potential) can be exactly dealt with,
both in the Landau gauge and in the central gauge. The spectrum is discrete
εn = (n + 12 )~ωc . We define the magnetic length as ℓ = (~c/eB)1/2 ; it diverges
in the zero-field limit, and is of the order of 100 angstrom in a typical quantum
Hall experiment. In the Landau gauge (Ax = By, Ay = 0) the eigenfunctions
with energy n are
ψnk (x, y) ∝ eikx χn (y − ℓ2 k), (2.15)
where χn (y) are harmonic oscillator eigenfunctions with frequency ωc . Each LL
is infinitely degenerate (one eigenfunction for each k). For a system of area A,
the number of states in each level is N = A/(2πℓ2 ); this has a simple form in
terms of the magnetic flux φ through area A: N = φ/φ0 .
If we now consider N noninteracting electrons, the lowest LL is completely
filled when N = N ; more generally, one expects a periodicity in the filling factor
ν = N/N = N φ/φ0 , whenever ν crosses an integer value, in most physical
properties.
16
Figure 2.6: The original figure from von
Klitzing et al. Ref. [40]. The gate volt-
age Vg was supposed to control the car-
rier density. Instead, the Hall resistance
is quantized and insensitive to Vg over
a large interval; over the same interval,
the longitudinal resistance vanishes.
2.4.3 The experiment

The Hall resistance of a noninteracting 2d electron gas has been computed
quantum-mechanically by Ando in 1974 [41]. The result, when expressed as in
Eq. (2.14) showed indeed oscillations in ν with integer period. The experiment,
performed by von Klitzing and collaborators in 1980 [40], provided qualitatively
different and very surprising results, shown in Fig 2.6. The discovery of the
quantum Hall effect triggered a revolution with far reaching consequences in
electronic structure theory at large; Klaus von Klitzing was awarded the Nobel
prize in 1985.
In the original experiment, the 2d electrons were confined by a MOSFET
(metal-oxide-semi-conductor field-effect transistor); later, higher mobilities were
obtained at semiconductor heterojunctions. Fig. 2.6 shows a very robust
Figure 2.7: A modern realization of the integer quantum Hall effect
17
plateau, where Rxx = 0 and Rxy = 6453.3 ± 0.1, corresponding to the fill-
ing factor ν = 4. The accuracy in the quantized Rxy value is clearly far beyond
the experimental control of the carrier concentration and of the B field uni-
formity over the sample. A novel, qualitatively different, state of matter was
discovered. In modern jargon, the plateaus are “topologically protected”.
A modern realization of the integer quantum Hall effect is shown in Fig.
2.7, where ρxx and ρxy are plotted as a function of the magnetic field. The
plateau quantization is accurate to nine figures. The 2d electron gas is typically
confined at a GaAs/GaAlAs heterojunction. The ν = 1 value is achieved above
≃ 10 tesla; at low field (high ν) the system becomes dissipative (ρxx > 0), while
the classical linear behavior of ρxy is recovered; the slope depends on electron
concentration n.
2.4.4 Early theoretical interpretation

The breakthrough paper, by Laughlin, appeared as early as 1981 [42]. This
is a remarkably concise paper (two pages) which, in retrospect, is based on
topological arguments. One key ingredient of the theory is disorder: in fact, the
quantum Hall effect becomes less spectacular for very “clean” samples, while
some “dirtyness” enhances the effect.
Laughlin devised a Gedankenexperiment based on the setup shown in
Fig. 2.8. The corresponding 2d Schrödinger Hamiltonian in the Landau gauge
is
1 e 2 2
H(ϕ) = px + Ax + py + eEy + V (x, y), (2.16)
2m c
where E is the field across the ribbon, and V is an arbitrary substrate potential.
The addition of a constant vector potential along x, Ax → Ax +∆A, in Eq. (2.16)
corresponds to threading a flux ϕ = L ∆A through the cylinder; we use the
symbol ϕ, not to be confused with the real magnetic flux φ normal to the
surface.
Similarly to what discussed in Sec. 1.6, the eigenvalues acquire a ϕ de-
pendence. According to Eq. (1.21), if εn (ϕ) is the n-th eigenvalue the current
transported by the corresponding eigenstate is Ix = −c ∂εn (ϕ)/∂ϕ. For an
independent-particle system with N carriers the current is thus
∂U (ϕ)
Ix = −c , (2.17)
∂ϕ
Figure 2.8: Geometry for Laughlin’s

Gedankenexperiment. A 2d channel is
bent into a loop of circumference L, and
a magnetic B field of constant magni-
tude pierces the cylinder normal to the
surface. A current I ≡ Ix circles the
loop; VH ≡ Vy is the Hall voltage. The
loop may be threaded by the inaccessi-
ble flux ϕ.
18
Figure 2.9: The density of states for a 2d system of noninteracting carriers. (a)
Clean system, with zero substrate potential. (b) Actual sample, in presence of
substrate disorder and impurities.
where U (ϕ) is the total energy of the system. Implicitly, we are assuming a
dissipationless system.
The expression in Eq. (2.17) for the current is remarkably simple, general,
and robust: it does not depend on the substrate potential V (x, y), nor the
number N of carriers, and not even on their mass m. But for a disordered
potential the eigenstates come in two kinds: localized and extended. The latter
ones are phase-coherent round the loop, while the former are exponentially
localized for L → ∞. The localized states are insensitive to the flux insertion
(like the OBCs eigenstates in Sec. 1.6), and the whole current is carried by
the delocalized ones. Therefore Eq. (2.17) provides a nonzero result insofar
at least one of the occupied eigenstates in the disordered sample is extended,
i.e. phase-coherent round the loop; besides this, the number and nature of the
current-carrying stated is irrelevant. It is therefore crucial to address the nature
of the single-particle eigenstates in a quantum Hall sample. For a clean sample
(flat substrate) the LLs are sharp, all eigenstates are extended (in the Landau
gauge), and the density of states is a series of delta functions, shown in Fig. 2.9
(a); the weight of each delta is φ/φ0 . In presence of disorder, the deltas broaden
into alternating bands of localized and extended states, as sketched in Fig. 2.9
(b).
The electron fluid is in the quantum Hall regime whenever the Fermi level
falls in a region of localized states. Therefore σxx = 0 (the fluid is a “quantum
Hall insulator”), and ρxx = 0 (transport is dissipationless).
We now imagine to adiabatically increase the vector potential by an amount
∆A = φ0 /L, where φ0 is a flux quantum: all of the current-carrying states are
mapped back into themselves, while the localized ones are unaffected. Hence the
ground state has the same energy; nonetheless Eq. (2.17) implies U (ϕ + φ0 ) −
U (ϕ) ≃ −φ0 Ix /c 6= 0. This is only possible if an integer number of electrons is
transferred from one cylinder edge to the other, each of them contributing the
energy eVy . If we call −ν such integer number, the relationship is then
φ0 1 h
φ0 Ix /c = νeVy ; RH = Vy /Ix = = . (2.18)
νce ν e2
The flux ϕ acts therefore as a charge pump; the pump cycle is one flux quantum
φ0 .
Ideally, the sample ground state can be continuously “deformed” from dirty
to clean. Insofar as the Fermi level stays is in a region of nonconducting states,
19
the (topological) integer ν cannot change, even if the number of current carry-
ing states does obviously change. The identification of ν with the number of
filled LLs comes from the clean-sample limit, which is exactly soluble. Setting
V (x, y) = 0 the eigenfunctions of Eq. (2.16) are
cE
ψnk (x, y) = eikx χn (y − y0 ), y0 = ℓ2 k − . (2.19)
ωc B
For a finite L, the allowed k’s are integer multiples of 2π/L and the correspond-
ing centers y0 are spaced by 2πℓ2 /L = Lφ0 /φ. Threading a flux ϕ shifts y0
linearly in ϕ; when ϕ equals one flux quantum each eigenfunction goes over to
the next. Therefore one carrier is shifted for each n; the integer index ν measures
therefore the number of occupied LLs. Similar arguments can be reformulated
in different gauges and in different geometries [43, 44].
20
Chapter 3
Berryology
Berryology is used as synonymous for geometry in nonrelativistic quantum me-

chanics. Topological (i.e. quantized) quantities are defined via geometrical
quantities, analogously to what we did above in Sec. 1.2. But many important
quantities (notably the Berry phase) are merely geometrical [22].
Let us assume that a generic time-independent quantum Hamiltonian has a
parametric dependence. The Schrödinger equation is
H(ξ)|Ψ(ξ)i = E(ξ)|Ψ(ξ)i, (3.1)
where the d-dimensional real parameter ξ is defined in a suitable domain of Rd :

a 2d ξ has been chosen for display in Fig. 3.1. In most of this Section we discuss
the most general case, and therefore we do not specify which quantum system is
described by this Hamiltonian, nor what the physical meaning of the parameter
ξ is.
In the subsequent Chapters |Ψ(ξ)i will be identified with either a single-
particle wavefunction (a.k.a. orbital) or a many-electron wavefunction. As for
the parameter ξ, it may be a nuclear coordinate, a phase angle, a magnetic
flux, a Bloch vector, a momentum, and more: it could therefore have various
dimensions. Sometimes, the parameter ξ is called the “slow variable”, while the
electronic coordinates are the “fast variable”. At the end of this Chapter, Sec.
3.9, we address the special case where the parameter ξ is identified with a Bloch
vector.
The state vectors |Ψ(ξ)i are all supposed to be normalized and to reside
in the same Hilbert space: this amounts to saying that the wavefunctions are
supposed to obey ξ-independent boundary conditions. We focus on the ground
state |Ψ0 (ξ)i, and we assume it to be nondegenerate for ξ in some domain of
Rd .
Any quantum mechanical state vector is arbitrary by a constant phase fac-
tor. Here we refer to choosing this phase as to the choice of the gauge. The
semantic is a bit ambiguous. In presence of magnetic fields, we may change the
magnetic gauge: this changes the Hamiltonian and the eigenfunctions. Once
the magnetic gauge—hence the Hamiltonian—is fixed, we still remain with the
phase arbitrariness referred to above. All measurable quantities (e.g. expecta-
tion values) are obviously gauge-invariant (in both senses), but the reverse is
also true: all gauge-invariant properties are—at least in principle—measurable.
It is expedient to define the ground-state projector (a.k.a. density matrix)
21
and its complement, i.e.
P̂ (ξ) = |Ψ0 (ξ)ihΨ0 (ξ)|; Q̂(ξ) = 1̂ − P̂ (ξ). (3.2)
Both P̂ (ξ) and Q̂(ξ) are gauge-invariant (for a fixed Hamiltonian).
3.1 Phases and distances

We define the phase difference between the ground eigenstates at two different
ξ points in the most natural way:
hΨ0 (ξ 1 )|Ψ0 (ξ 2 )i
e−i∆ϕ12 = ; (3.3)
|hΨ0 (ξ 1 )|Ψ0 (ξ 2 )i|
∆ϕ12 = − Im log hΨ0 (ξ 1 )|Ψ0 (ξ 2 )i . (3.4)
For any given choice of the two states, Eqs. (3.3) and (3.4) provide a ∆ϕ12
which is unique modulo 2π, except in the very special case that the states are
orthogonal. However, it is also clear that such ∆ϕ12 is gauge-dependent and
cannot have, by itself, any physical meaning.
The distance between quantum states has been defined by Bures [45] as:
2
D12 = 1 − |hΨ0 (ξ 1 )|Ψ0 (ξ 2 )i|2 . (3.5)
Such distance fulfills the familiar axioms from calculus textbooks; it vanishes
when the two states physically coincide (i.e. independently of the phase factor),
and is maximum (equal to one) when the states are orthogonal. At variance
with ∆ϕ12 defined above, the Bures distance is gauge-invariant, and can be
explicitly expressed in terms of ground-state projectors
2
D12 = 1 − Tr {P̂ (ξ 1 )P̂ (ξ 2 )}, (3.6)
where “Tr” is the trace over the Hilbert space.
3.2 Berry phase

We have already observed that the phase difference ∆ϕ12 between any two states
is gauge-dependent and cannot have any physical meaning. Matters are quite
different when we consider the total phase difference along a closed path which
joins several points in a given order, as shown in Fig. 3.2:
γ = ∆ϕ12 + ∆ϕ23 + ∆ϕ34 + ∆ϕ41
= − Im log hΨ0 (ξ 1 )|Ψ0 (ξ 2 )ihΨ0 (ξ 2 )|Ψ0 (ξ 3 )i ×
× hΨ0 (ξ 3 )|Ψ0 (ξ 4 )ihΨ0 (ξ 4 )|Ψ0 (ξ 1 )i. (3.7)
Figure 3.1: State vectors in the two-

dimensional ξ-space. The phase dif-
ference between two of them is de-
hΨ (ξ )|Ψ (ξ )i
fined as e−i∆ϕ12 = |hΨ0 (ξ1 )|Ψ0 (ξ2 )i| ,
0 1 0 2
2
and their distance as D12 = 1 −
2
|hΨ0 (ξ 1 )|Ψ0 (ξ 2 )i| .
22
Figure 3.2: A closed path joining four
states in ξ-space.
It is now clear that all the gauge-arbitrary phases cancel in pairs, such as to
make the overall phase γ a gauge–invariant quantity. The above simple–minded
algebra leads to a result of overwhelming physical importance: in fact, a gauge–
invariant quantity is potentially a physical observable. In essence, this is the
revolutionary message of Berry’s celebrated paper, appeared in 1984 [30, 46].
Next we consider a smooth closed curve C in the parameter domain, such
as in Fig. 3.3, and we discretize it with a set of points on it. Using Eq. (3.3),
we write the phase difference between any two contiguous points as
hΨ0 (ξ)|Ψ0 (ξ+∆ξ)i

e−i∆ϕ = . (3.8)
|hΨ0 (ξ)|Ψ0 (ξ+∆ξ)i|
If we further assume that the gauge is so chosen that the phase varies in a
differentiable way along the path, then from Eq. (3.8) we get to leading order
in ∆ξ:
−i∆ϕ ≃ hΨ0 (ξ)|∇ξ Ψ0 (ξ)i · ∆ξ. (3.9)
In the limiting case of a set of points which becomes dense on the continuous
path, the total phase difference γ converges to a circuit integral:
M
X I
γ= ∆ϕs,s+1 −→ A(ξ) · dξ, (3.10)
s=1 C
where A(ξ) is called the Berry connection:
A(ξ) = i hΨ0 (ξ)|∇ξ Ψ0 (ξ)i. (3.11)
Figure 3.3: A smooth closed curve C in

ξ-space, and its discretization.
23
Since the state vectors are assumed to be normalized at any ξ, the connection
is real; we can therefore equivalently write
A(ξ) = −Im hΨ0 (ξ)|∇ξ Ψ0 (ξ)i. (3.12)
A number of manifestations of the Berry phase occurring in molecular and

condensed matter phenomena will be discussed in the present Notes. Many
reviews papers; here we quote Refs. [46, 47, 48, 3, 13].
At this point it is also worth emphasizing that in computational physics there
are no derivatives. The ground state Ψ0 (ξ) is generally found by diagonalizing
a matrix on a finite set of ξ points, and the phase (i.e. the gauge) is chosen by
the diagonalization routine; the phase is therefore nonsmooth and possibly even
random. The discrete form in Eq. (3.11) at finite M is the one universally used
in numerical work; it is clearly unaffected by any erratic phase factor.
3.3 Connection and curvature

The loop integral of the Berry connection (i.e. the Berry phase γ) is non trivial
in two cases: either the curl of A(ξ) is nonzero, or the curl is zero but the curve
C is not in a simply connected domain. In the former case, we can invoke Stokes
theorem; the formulation is very simple when ξ is a 3d parameter. If C is the
boundary of a surface Σ (i.e. C ≡ ∂Σ), and the curl of A(ξ) is regular on Σ,
then Stokes’ theorem reads
I Z
γ= A(ξ) · dξ = Ω(ξ) · n dσ, (3.13)
∂Σ Σ
where Ω is the Berry curvature, defined as
Ω(ξ) = ∇ξ × A(ξ) = −Im h∇ξ Ψ0 (ξ)| × |∇ξ Ψ0 (ξ)i

= ih∇ξ Ψ0 (ξ)| × |∇ξ Ψ0 (ξ)i, (3.14)
with the usual meaning of the cross product between three-component bra and
ket states. Equation (3.13) may be spelled out by saying that the curvature is
the Berry phase per unit area of Σ.
For d 6= 3 the Berry curvature is conveniently written as the d × d antisym-
metric matrix
Ωαβ (ξ) = −2 Im h∂α Ψ0 (ξ)|∂β Ψ0 (ξ)i; (3.15)
Greek subscripts are Cartesian coordinates throughout, and ∂α = ∂/∂ξα . The
Stokes theorem can still be applied, generalizing Eq. (3.13) to
1
Z
γ= dξ α ∧ dξ β Ωαβ (ξ). (3.16)
2 Σ
The Berry connection is also known as “gauge potential”, and the Berry
curvature as “gauge field” [48]. It is worth pointing out that the former is
gauge-dependent, while the latter is gauge-invariant and therefore corresponds
in general to a measurable quantity, even before any integration. The two quan-
tities play (in ξ-space) a similar role as the vector potential and the magnetic
field in elementary magnetostatics: A(r) is gauge-dependent, nonmeasurable;
B(r) = ∇r × A(r) is gauge-invariant, measurable.
24
The Berry phase γ, defined as the integral over a closed curve C of the
connection, is gauge invariant only modulo 2π. This indeterminacy is resolved
by Eqs. (3.13) and (3.16) whenever the curve C is the boundary ∂Σ of a surface
Σ where the curvature is regular. In fact, the curvature is gauge-invariant and
has no modulo 2π indeterminacy.
3.4 Chern number

The rhs of Eqs. (3.13) and (3.16) is the flux of the Berry curvature on the surface
Σ; such flux remains meaningful even on a closed surface (e.g. a sphere or a
torus), in which case ∂Σ is the empty set. The key result is that such an integral
is quantized. Here we limit ourselves to 3d, where we identify Σ with the sphere
S 2 (Fig. 3.4):
1
Z
Ω(ξ) · n dσ = C1 ; (3.17)
2π S 2
C1 is an integer ∈ Z, called Chern number of the first class.
The proof is based on a similar algebra as for Dirac’s theory of the magnetic
monopole [47, 49]. The theorem goes sometimes under the name of Gauss-
Bonnet-Chern theorem; the analogy with Eq. (1.4) is perspicuous. A specific
example is dealt with in detail in Sec. 4.1.
The curvature is regular (and divergence-free) on the closed surface S 2 ; the
lhs of Eq. (3.17) is the flux of Ω(ξ) across S 2 . The integrand Ω(ξ) is the curl of
the connection A(ξ); the latter in general cannot be defined as a single-valued
function globally on S 2 , but only on patches of it [47, 49]. To fix the ideas,
suppose that Ω(ξ) is singular at ξ = 0, and that S 2 is the spherical surface
centered at the origin (Fig. 3.4). We cut this surface at the equator ξz = 0 and
we consider the flux across the two open surfaces:
Z Z Z
Ω(ξ) · n dσ = Ω(ξ) · n dσ + Ω(ξ) · n dσ. (3.18)
S2 S+ S−
We notice that ∂S+ = ∂S− = C, but the surface normals n have opposite
orientations. From Stokes theorem, Eq. (3.13), we get:
Z I
Ω(ξ) · n dσ = ± A± (ξ) · dξ (3.19)
S± C
Z I I
Ω(ξ) · n dσ = A+ (ξ) · dξ − A− (ξ) · dξ . (3.20)
S2 C C
The two upper and lower Berry connections A± (ξ) may only differ by a gauge
transformation; the rhs of Eq. (3.20) is the difference of two Berry phases on
Figure 3.4: A sphere cut at the equator

in two hemispheres.
25
the same path and is necessarily a multiple of 2π. This concludes the proof of
Eq. (3.17).
We emphasize that the Chern number is a robust topological invariant of
the wavefunction, and is at the origin of observable effects. The Chern number
made its first appearance in electronic structure in 1982, in the famous TKNN
paper about the quantum Hall effect [50] (Sec. 4.2.4). Nowadays more complex
topological invariants are in fashion, and they characterize a completely novel
class of insulators, called “topological insulators” [10, 11, 15, 51, 52, 53, 54, 55,
56].
3.5 Metric
Starting from Eq. (3.5), the infinitesimal distance is
d
X
2
Dξ ,ξ+dξ = gαβ (ξ)dξα dξβ , (3.21)
α,β=1
where the metric tensor is easily shown to be
gαβ (ξ) = Re h∂α Ψ0 (ξ)|∂β Ψ0 (ξ)i

− h∂α Ψ0 (ξ)|Ψ0 (ξ)ihΨ0 (ξ)|∂β Ψ0 (ξ)i
= Re h∂α Ψ0 (ξ)|Q̂(ξ)|∂β Ψ0 (ξ)i; (3.22)
the projector Q̂(ξ) is the same as defined in Eq. (3.2). This quantum metric
tensor was first proposed by Provost and Vallee in 1980 [57].
At this point we may compare Eq. (3.22) to Eq. (3.15), noticing that the
insertion of Q̂(ξ) is irrelevant in the latter, i.e.
Ωαβ (ξ) = −2 Im h∂α Ψ0 (ξ)|Q̂(ξ)|∂β Ψ0 (ξ)i. (3.23)
It is therefore clear that gαβ and Ωαβ are, apart for a trivial −2 factor, the real
(symmetric) and the imaginary (antisymmetric) parts of the same tensor, which
we are going to call Fαβ in the following:
Fαβ (ξ) = h∂α Ψ0 (ξ)|Q̂(ξ)|∂β Ψ0 (ξ)i. (3.24)
The metric-curvature tensor Fαβ is gauge-invariant. A compact equivalent ex-

pression is
Fαβ (ξ) = Tr {∂α P̂ (ξ)Q̂(ξ)∂β P̂ (ξ)}, (3.25)
manifestly gauge-invariant and Hermitian.
3.6 Sum over states

The ξ-derivatives entering many of the previous equations, e.g. Eq. (3.24), can
be expressed starting from perturbation theory:
|Ψ0 (ξ + ∆ξ)i − |Ψ0 (ξ)i (3.26)

X′ hΨn (ξ)| [ H(ξ + ∆ξ) − H(ξ) ] |Ψ0 (ξ)i
≃ |Ψn (ξ)i ;
E0 (ξ) − En (ξ)
n6=0
26
X′ hΨn (ξ)|∂α H(ξ)|Ψ0 (ξ)i
|∂α Ψ0 (ξ)i = |Ψn (ξ)i . (3.27)
E0 (ξ) − En (ξ)
n6=0
These seemingly obvious and innocent formulae need some caveat. It is clear
that inserting Eq. (3.27) into the Berry connection, Eqs. (3.11) and (3.12), we
get a vanishing result for any ξ. This happens because the simple expression
of Eq. (3.26) corresponds to a very specific gauge choice (called the parallel-
transport gauge [3]); multiplying the rhs by a ξ-dependent phase factor is le-
gitimate, and must not modify any physical result, while the Berry connection
is instead affected. Nonetheless, since our Fαβ (ξ) is a gauge-invariant quantity,
we may safely evaluate it in any gauge, including the parallel-transport gauge,
implicit in Eq. (3.27). The result is
Fαβ (ξ) (3.28)
X′ hΨ0 (ξ)|∂α H(ξ)|Ψn (ξ)ihΨn (ξ)|∂β H(ξ)|Ψ0 (ξ)i
= .
[E0 (ξ) − En (ξ)]2
n6=0
This expression shows explicitly that both the curvature and the metric are
ill defined and singular wherever the ground state is degenerate with the first
excited state. Indeed, this is the main reason why the domain may happen not
to be simply connected.
3.7 Time-reversal and inversion symmetries

According to Eq. (3.25) the ground-state projector uniquely determines the
curvature
Ωαβ (ξ) = −2 Im Tr {∂α P̂ (ξ)Q̂(ξ)∂β P̂ (ξ)}. (3.29)
It is therefore expedient to analyze the symmetries of the ground-state projector
P̂ (ξ), which coincide with the symmetries of the Hamiltonian; these in turn
depend on the nature of the parameter ξ. We only address spinless electrons
(no spin-orbit coupling).
When ξ is even under time reversal (like e.g. a nuclear coordinate), then
time-reversal invariance implies that both H(ξ) and P̂ (ξ) are real for any ξ, and
Eq. (3.29) warrants that the curvature is everywhere vanishing. The Berry phase
γ can be nonzero (modulo 2π) only if the curve C loops around a singularity or,
more generally, it does not lie in a simply connected domain; the only allowed
values are γ = 0 or γ = π.
When instead ξ is odd under time-reversal (like e.g. a momentum) then
time-reversal symmetry requires P̂ (−ξ) = P̂ ∗ (ξ), therefore
Ωαβ (−ξ) = −Ωαβ (ξ). (3.30)
The Berry phase along an inversion-symmetric path vanishes; the Chern number
vanishes as well.
Next we switch to inversion symmetry. When evaluating any ξ-dependent
matrix-element (or trace), inversion of the coordinates at fixed ξ is equivalent to
keeping the coordinates fixed and inverting ξ. This statement holds whether ξ
is a coordinate or a momentum; both are in fact odd under inversion. Therefore
inversion symmetry implies P̂ (−ξ) = P̂ (ξ), and
Ωαβ (−ξ) = Ωαβ (ξ). (3.31)
27
If both time-reversal and inversion symmetry are present, then the Berry curva-
ture is everywhere vanishing. The Berry phase can be only zero or π; the latter
case requires a domain which is not simply connected, as above.
Crucial to the above arguments is the fact that the double derivative appear-
ing in Eq. (3.29) are even under either time-reversal or inversion. Summarizing
the symmetry results, for the case where ξ is a momentum: a non vanishing
Chern number can only occur in absence of time-reversal symmetry, but may
occur even in inversion-symmetric cases.
3.8 NonAbelian case

Very soon after the appearance of Berry’s milestone paper, Wilczek and Zee [58]
introduced a many-state generalization. Suppose we do not focus on a single
state in the Hilbert space (say the ground state), while we are interested instead
in the behavior of n different states altogether, as a function of ξ. The original
motivation was the possibility of degeneracy amongst the n lowest eigenstates
|ψj (ξ)i of a given Hamiltonian at some points of the path, and they formulated
the theory under the hypothesis that these lowest states are never degenerate
with the n+1-th. Strictly speaking, an Hamiltonian is unnecessary: one only
needs to unambiguously identify an n-dimensional (n fixed) manifold of states,
as function of the parameter ξ. A gauge transformation is then a n × n unitary
matrix; in the Abelian case, this matrix is a diagonal one, with unitary elements.
The gauge invariant quantity which defines the manifold is the n-dimensional
projector
Xn
P (ξ) = |ψj (ξ)ihψj (ξ)|. (3.32)
j=1
Within loss of generality, we may address the case which is most interesting in
electronic structure. We assume that the n states |ψj (ξ)i are spin orbitals, and
that |Ψ(ξ)i is the many-body wavefunction built as their Slater determinant:
1
Ψ(ξ)i = √ |ψ1 (ξ)ψ2 (ξ) . . . ψN (ξ)|. (3.33)
N!
We can therefore apply some of the results of the previous Sections. The many-
body Berry phase is given in Eqs. (3.10) and (3.11), which we prefer to rewrite
here in the form I
e−iγ = exp A(ξ) · dξ, (3.34)
C
where A is the connection of the many-body wavefunction.
In the nonAbelian case the connection generalizes to a vector of n × n Her-
mitian matrices
Ajj ′ (ξ) = −Im hψj (ξ)|∇ξ ψj ′ (ξ)i, (3.35)
and the Berry phase factor of Eq. (3.34) generalizes to the unitary matrix
I
e−iΓ = P exp A (ξ) · dξ. (3.36)
C
Here P is a path-ordering operator, owing to the noncommuting nature of

the A matrices at different ξ in the nonAbelian case. The P operator has a
28
precise meaning when the series expansion of the exponential is considered. The
discretization of Eq. (3.36) is rather straightforward and will not be discussed
here [21].
While the single-state phase factor e−iγ is gauge-invariant, the Wilczek-Zee
unitary matrix e−iΓ is only gauge-covariant, i.e. a gauge transformation yields
a matrix unitarily equivalent to e−iΓ ; the gauge can be fixed by choosing the
vectors |ψj (ξ)i at one point of the path. The eigenvalues γj of the Hermitian
matrix Γ, defined modulo 2π, are gauge-invariant and in principle individually
observable. It is easily proved that their sum, i.e. the trace of Γ, coincides with
the Berry phase γ of the many-body wavefunction, Eq. (3.34) [21].
In order to write the metric-curvature tensor in the nonAbelian case, we
start writing the many-body Abelian metric-curvature tensor of the Slater-
determinant wavefunction, Eq. (3.24), as
Fαβ (ξ) = h∂α Ψ0 (ξ)|∂β Ψ0 (ξ)i − h∂α Ψ0 (ξ)|P̂ (ξ)|∂β Ψ0 (ξ)i. (3.37)
Next, we need now to explicitate the Cartesian indices α, β at the same time as
the matrix indices j, j ′ . The nonAbelian metric-curvature tensor is the general-
ized form of Eq. (3.25), i.e.
Fαβ,jj′ (ξ) = h∂α ψj (ξ)|∂β ψj ′ (ξ)i − h∂α ψj (ξ)|P (ξ)|∂β ψj ′ (ξ)i, (3.38)
where now P (ξ) is the single-particle projector; even this matrix is gauge-
covariant. If we rewrite the Cartesian components of Eq. (3.35) as Aα,jj ′ (ξ) =
−Im hψj (ξ)|∂α ψj ′ (ξ)i, the nonAbelian curvature becomes
Ωαβ,jj′ (ξ) = ∂α Aβ,jj ′ (ξ) − ∂β Aα,jj′ (ξ) − i[Aα (ξ), Aβ (ξ)]jj ′ . (3.39)
With respect to Eq. (3.15) notice the presence of an extra term, which vanishes
in the Abelian case. At fixed jj ′ Eq. (3.39) is clearly antisymmetric in the αβ
Cartesian indices, while at fixed αβ it is an Hermitian n × n matrix.
The trace over the j index of the (nonAbelian) metric-curvature tensor equals
the corresponding (Abelian) metric-curvature tensor of the many-body ground
state, Eq. (3.25):
n
X
Fαβ,jj (ξ)
j=1
Xn n
X
= h∂α ψj (ξ)|∂β ψj (ξ)i − h∂α ψj (ξ)|ψj ′ (ξ)ihψj ′ (ξ)|∂β ψj (ξ)i
j=1 jj ′ =1
= Fαβ (ξ). (3.40)
The proof of the last line of Eq. (3.40) is tedious, but straightforward.
3.9 Bloch orbitals

We have remained very general so far. The case where the parameter ξ coincides
with the Bloch vector k bears a particular relevance in the context of the present
Notes. In the framework of first-principle calculations for crystalline systems,
the Bloch states |ψjk i are the KS orbitals.
29
The domain where the k parameter varies (the reciprocal cell or, equivalently,
the Brillouin zone, BZ) has the geometry of a torus in 1d, 2d, 3d. The whole
BZ is a closed surface; we denote as G a reciprocal vector.
The Bloch orbital of the j-th band is |ψjk i = eik·r |ujk i, where |ujk i are
BZ-periodical, and are eigenfunctions of the Hamiltonian Hk = e−ik·r Heik·r .
While the |ψjk i at different k’s are mutually orthogonal, the |ujk i live instead
in the same Hilbert space (they are all BZ-periodical).
The physical meaning of all the mathematical quantities introduced next
will be discussed in the following Chapters, and not anticipated in the present
one.
The Berry connection of the j-th orbital is
Aj (k) = ihujk |∇k ujk i. (3.41)
The relative phases at different k’s are arbitrary. Whenever possible, it is cus-
tomary to enforce the so-called periodic gauge |ψjk+G i = |ψjk i, which implies
|ujk+G i = e−iG·r |ujk i. (3.42)
We stress, however, that in topologically nontrivial crystals it is generally im-
possible to adopt a periodic gauge.
The interesting closed paths C on the torus are lines across the reciprocal
cell, from one face to the opposite one. For an insulator with n occupied bands
the Berry phase is, according to the previous section,
Xn Z n Z
X
γ=i Aj (k) · dk = i dk · hujk |∇k ujk i. (3.43)
j=1 C j=1 C
This Berry phase depends on the choice of the origin in the crystal cell. For
centrosymmetric crystals, if the origin is at a center of inversion symmetry, the
only allowed values are γ = 0 and γ = π (modulo 2π).
In case of band crossings the definition of individual bands is ambiguous,
but the Berry phase in Eq. (3.43) is not affected by such ambiguity. More
generally, the results of the previous Section show that Eq. (3.43) is invariant
by a nonAbelian gauge transformation, i.e. by mixing of the occupied orbitals
between themselves by an arbitrary unitary matrix Ujj ′ (k). The mixed orbitals
are no longer Hamiltonian eigenstates; any gauge where instead the |ujk i are
eigenstates will be called “Hamiltonian gauge”.
The discretization of Eq. (3.43) is nowadays implemented in most electronic
structure codes [59, 60, 61, 62] in order to compute the macroscopic polarization
of crystalline dielectrics (Sec. 5.3). Such discretization is based on the following
result: if |Ψ(k)i is the Slater determinant of the n occupied |ujk i, then
hΨ(k1 )|Ψ(k2 )i = det S(k1 , k2 ), (3.44)
where S(k1 , k2 ) is the n × n overlap matrix
Sjj ′ (k1 , k2 ) = hujk1 |uj ′ k2 i. (3.45)
We discretize with M+1 points on the line, where kM+1 = k1 +G; it is understood
that |ujkM+1 i = e−iG·r |ujk1 i. The discretized formula is then
Z
γ = i dk · hΨ(k)|∇k Ψ(k)i → −Im log ΠM s=1 hΨ(ks )|Ψ(ks+1 )i
C
= −Im log det ΠM
s=1 S(ks , ks+1 ). (3.46)
30
Notice that this is numerically gauge invariant, i.e. it does not depend on
how the diagonalization routine chooses the phases and/or the ordering of the
eigenvectors.
The metric-curvature tensor of n occupied bands obtains straightforwardly
from Eq. (3.40):
n
X n
X
Fαβ (k) = h∂α ujk |∂β ujk i − h∂α ujk |uj ′ k ihuj ′ k |∂β ujk i. (3.47)
j=1 jj ′ =1
This is usually integrated over the BZ; being gauge invariant, it carries in prin-
ciple physical meaning even before any integration. Indeed, the k-dependent
(single-band) curvature enters the theory of semiclassical transport in crystalline
solids [63, 64, 13], Sec. 4.5.
Eq. (3.47) is the trace of the nonAbelian metric-curvature whose expression
is
n
X
Fαβ,jj′ (k) = h∂α ujk |∂β uj ′ k i − h∂α ujk |uj”k ihuj”k |∂β uj ′ k i
j”=1
∞
X
= h∂α ujk |usk ihusk |∂β uj ′ k i. (3.48)
s=n+1
The nonAbelian metric and curvature are

1
gαβ,jj′ = [ Fαβ,jj′ (k) + Fβα,jj′ (k) ]
2
Ωαβ,jj′ = i[ Fαβ,jj′ (k) − Fβα,jj′ (k) ]. (3.49)
The many-band curvature obtains from either Eq. (3.47) or Eq. (3.49) as
n
X
Ωαβ = −2 Im h∂α ujk |∂β ujk i. (3.50)
j=1
31
Chapter 4
Manifestations of the Berry phase
4.1 A toy-model Hamiltonian

Our toy model here is a two-level spinless Hamiltonian, of the form
H(ξ) = ξ · ~σ
= ξ (sin ϑ cos ϕ σx + sin ϑ sin ϕ σy + cos ϑ σz ), (4.1)
where σα are the three Pauli matrices. The spectrum is non degenerate for
ξ 6= 0, and the lowest eigenvalue is −ξ. Upon symmetry arguments, we can
already guess the curvature to be isotropic.
4.1.1 Connection and curvature

The lowest eigenvector is
sin ϑ2 e−iϕ

|ψ(ϑ, ϕ)i = .
− cos ϑ2
This corresponds to a specific gauge choice; the eigenvector can be multiplied by

an overall (ϑ, ϕ)-dependent phase factor. The Berry connection and curvature
are
Aϑ = ihψ|∂ϑ ψi = 0
ϑ
Aϕ = ihψ|∂ϕ ψi = sin2
2
1
Ω = ∂ϑ Aϕ − ∂ϕ Aϑ = sin ϑ. (4.2)
2
The curvature is gauge-invariant, while the connection is gauge-dependent.
Within our gauge choice the connection displays a vortex at the south pole
(ϑ = π); other gauges yield the singularity at a different point, but a singu-
larity is unavoidable. It is impossible to find a gauge which is smooth on the
whole closed surface, and a nonsingular connection; the singularity—often called
“obstruction”—can be moved but not removed. The algebra is the same as for
Dirac’s theory of the magnetic monopole [47, 49]: the degeneracy at the origin
is the monopole, and the singularity is the “Dirac string”.
32
Figure 4.1: A closed curve C on the surface of the
sphere, and the solid angle spanned by it.
4.1.2 Chern number

The domain of the parameters (ϑ, ϕ) is a rectangle, which indeed has the topol-
ogy of a torus: a closed surface. Integrating the Berry curvature therein we
get
1 1
Z
dϑdϕ sin ϑ = 1, (4.3)
2π S 2 2
i.e. the Chern number of the lowest eigenstate in this problem. This integer
measures the strength of the singularity (magnetic monopole), which resides in
a site inaccessible to the quantum system. The highest eigenstate has C1 = −1,
since the total Chern number is zero.
This simple example illustrates well the meaning of a topological invariant of
the quantum mechanical ground state. The Chern number is in fact very robust
under continuous deformations of the surface and of the Hamiltonian: its value
is always one insofar as one (and only one) degeneracy point is included in the
closed surface.
4.1.3 Berry phase

Suppose now we evaluate the Berry phase over any closed curve C on the sphere
(Fig. 4.1) I
γ= A(ξ) · dξ. (4.4)
C
Owing to Stokes theorem, the Berry phase for this toy model problem clearly
equals the solid angle spanned by the curve, divided by 2. The inherent 4π
arbitrariness in the solid angle leads to the well known 2π arbitrariness in the
Berry phase: for instance at the equator γ = π modulo 2π. If we cut the sphere
in two hemispheres, as in Sec. 4.1.2 (see Fig. 3.4), the difference of the boundary
Berry phases equals 0 modulo 2π, i.e. 2π times an integer, as it must be. But in
order to tell which integer (the actual Chern number) the two connections are
useless: one has to integrate the curvature, as in Eq. (4.3). Despite this feature,
in numerical work Chern numbers are typically computed via Berry phases, as
described in the next Section.
4.1.4 Numerical considerations

This exactly soluble example also provides the occasion for illustrating the stan-
dard computational approach to Chern numbers. Suppose we discretize the
(ϑ, ϕ) domain with a rectangular mesh, and that we diagonalize the Hamil-
tonian at the points of the mesh. The gauge at any point is chosen by the
33
Figure 4.2: Discretization of the domain [0, π] ×
[0, 2π]; the Hamiltonian is not diagonalized at the
empty circles, only at the black ones, thus enforcing
toroidal topology.
diagonalization routine and is thus erratic; we only enforce the toroidal topol-
ogy by requiring that the phases at the opposite edges of the rectangle are the
same: Fig. 4.2.
Then for each small rectangle we compute the discrete Berry phase as in
Eq. (3.7), i.e.
γ = − Im log hψ(ϑ, ϕ)|ψ(ϑ + ∆ϑ, ϕ)ihψ(ϑ + ∆ϑ, ϕ)|ψ(ϑ + ∆ϑ, ϕ + ∆ϕ)i

× hψ(ϑ + ∆ϑ, ϕ + ∆ϕ)|ψ(ϑ, ϕ + ∆ϕ)ihψ(ϑ, ϕ + ∆ϕ)|ψ(ϑ, ϕ)i. (4.5)
The Berry curvature is the Berry phase per unit (ϑ, ϕ) area. In this simple,
analytically soluble, case we know the exact value; Eq. (4.2) implies for Eq. (4.5)
γ = 12 sin ϑ ∆ϑ∆ϕ modulo 2π. The Chern number is the integral over the
domain, and is therefore equal to the sum of all the phases computed as in
Eq. (4.5) and covering the whole domain.
How do we then get rid of the modulo 2π indeterminacy in Eq. (4.5)? The
size of ∆ϑ∆ϕ is very small, and each contribution γ to the sum is also small
(proportional to ∆ϑ∆ϕ), although Eq. (4.5) is in principle arbitrary modulo 2π.
It should be now pretty clear that the right solution is in choosing the “Im log”
branch with values in [−π, π].
Last but not least: where is the obstruction? In the continuous formulation,
any gauge choice yields a singularity at a single (ϑ, ϕ) point. In the discrete for-
mulation, there is no way to locate the singularity: in some sense, the singularity
is everywhere since the gauge is in principle erratic at all mesh points.
4.2 Early discoveries reinterpreted

4.2.1 Aharonov-Bohm effect
Here we reformulate the Aharonov-Bohm effect as a special case of a Berry
phase. Suppose we have an electron in a box (infinite potential well) centered
at the origin. We take the ground wavefunction as real, and we write it as χ(r).
The time-independent Schrödinger equation is:
2
p
+ V (r) χ(r) = Eχ(r). (4.6)
2m
Displacing the center of the box at position R changes the Hamiltonian to
p2
H(R) = + V (r − R) : (4.7)
2m
34
we will identify the ξ parameter with the box position R. Because of transla-
tional invariance, the R-dependence of the state vectors is
hr|ψ(R)i = χ(r − R), (4.8)
while the eigenvalue is R-independent.

Suppose now that a magnetic field is switched on somewhere in space. Then
the Hamiltonian becomes
1 h e i2
H(R) = p + A(r) + V (r − R) , (4.9)
2m c
where A is the vector potential and e is the electron charge. It can be easily
verified that a solution of the Schrödinger equation can be formally written in
the form:
ie r
Z
′ ′
hr|ψ(R)i = exp − A(r ) · dr χ(r − R). (4.10)
~c R
But such a solution in general is not a single-valued function of r, since the
phase factor depends on the path. Therefore we restrict ourselves to a less
general case, where the magnetic field is generated by a solenoid: the B field
is nonzero only within a given cylinder, and we don’t allow our box to overlap
this cylinder by suitably restricting the domain of R. This situation is sketched
in Fig. 4.3. With such a choice the wavefunction, Eq. (4.10), is a single valued
function of r for any fixed R, and is therefore an honest ground wavefunction.
As for the dependence on R, Eq. (4.10) only guarantees local single-valuedness,
since the domain is not simply connected: when the system is transported on a
closed path winding once round the solenoid, the electron wavefunction picks up
a Berry’s phase. This phase difference can be actually detected in interference
experiments.
The Berry connection of the problem is
e
Z
A(R) = ihψ(R)|∇R ψ(R)i = − A(R) − i dr χ(r − R)∇R χ(r − R), (4.11)
~c
where the last term vanishes. Therefore in the present case the Berry connection
is proportional to the ordinary vector potential. A gauge transformation in
the quantum-mechanical sense also coincides with an electromagnetic gauge
quantum
system
Figure 4.3: A particle in a box, trans-

ported round a solenoid
35
transformation, which changes A while leaving B invariant. In fact, in this
example B is essentially the Berry curvature. The Berry phase is
e 2π
I I
γ=− A(R) · dR = − A(R) · dR, (4.12)
~c C φ0 C
where φ0 is the flux quantum. Therefore γ measures the flux of the magnetic field
across the interior of the solenoid, a space region not accessed by the quantum
system: above, we have called it “inaccessible flux”. Only the fractional part of
the flux has physical meaning.
4.2.2 Molecular Aharonov-Bohm effect

Here we identify the “slow coordinate” ξ with a d-dimensional nuclear coordi-
nate, and the state vector |Ψ(ξ)i with the electronic ground-state wavefunction
in the Born-Oppenheimer approximation.
We start from the complete Hamiltonian H of an isolated molecular system,
and we explicitly separate the nuclear kinetic energy:
d
1 X
H(ξ, [x]) = M−1
αβ pα pβ + H(ξ, [x]), (4.13)
2
α,β=1
where [x] indicates the electronic degrees of freedom collectively, pα =

−i~ ∂/∂ξα is the canonical momentum conjugated to ξα , and the inverse mass
matrix M−1 in general may be a function of ξ, but not of the momenta.
The Born-Oppenheimer approximation starts by writing the eigenfunctions
of Eq. (4.13) in the Schrödinger representation as the product h [x] |Ψ(ξ)i Φ(ξ).
Our aim is obtaining an effective Schrödinger equation for the nuclear wave-
function Φ(ξ), where the electronic degrees of freedom have been integrated
out. We start considering the effect of the canonical nuclear momentum p on
the product ansatz:
p |Ψ(ξ)i Φ(ξ) = −i~ |Ψ(ξ)i ∇ξ Φ(ξ) − i~ |∇ξ Ψ(ξ)i Φ(ξ). (4.14)
We then multiply by the electronic eigenbra hΨ(ξ)| on the left, thus integrating
over the electronic degrees of freedom. We get the effective nuclear momentum
π acting on Φ as:
π Φ(ξ) = [ p − i~hΨ(ξ)|∇ξ Ψ(ξ)i ] Φ(ξ) = [ p − ~A(ξ) ] Φ(ξ), (4.15)
where we easily recognize the Berry connection. The momentum π is the kine-
matical (also called covariant, or mechanical) momentum, to be distinguished
from p = −i~∇ξ , which is the canonical momentum.
Whenever the time scales of nuclear and electronic motions are well sepa-
rated the coupling between different electronic states can be neglected, and the
adiabatic approximations allows to treat the slow variable ξ in H(ξ, [x]) as a
classical parameter. The electronic eigenvalue E(ξ) of a given state (e.g. the
ground state) plays therefore the role of a (scalar) potential for nuclear motion,
whose effective Hamiltonian acting on Φ(ξ) is then:
d
1 X
Heff = M−1
αβ πα πβ + E(ξ). (4.16)
2
α,β=1
36
In the molecular physics literature the extra term in Eq. (4.15) is seldom
mentioned, and π is identified with p. The reason is that for a time-reversal-
invariant Hamiltonian, and in absence of spin-orbit interaction, the wave func-
tion can always be taken as real. This corresponds to the parallel transport
gauge, and the Berry connection vanishes at all ξ; the tradeoff is that—in some
cases—the electronic wave function is not single valued along a closed path:
see Fig. 2.3. The alternative approach, due to Mead and Truhlar [33, 65], is
to choose a different gauge, where the electronic wave function is single valued
and complex. The Berry phase is gauge invariant (modulo 2π); the values al-
lowed by time-reversal symmetry are 0 and π; the two cases are experimentally
distinguishable, as previously shown in Sec. 2.2.
We stress that, whenever the ionic motion is purely classical and governed
by Newton’s equation, the vector-potential-like term in Eq. (4.15) is irrelevant:
the corresponding curvature (magnetic-field-like) is in fact identically vanishing
along the nuclear trajectory on the Born-Oppenheimer surface. We anticipate
that the case where a genuine magnetic field is present—and the Hamiltonian
is no longer time-reversal-invariant—is qualitatively different in this respect, see
Sec. 4.3 below.
4.2.3 Digression: the Z2 invariant

The topological nature of the molecular Aharonov-Bohm effect has already been
pointed out in Sec. 2.2. Here we want to recast the finding in modern topological
language. The group Z2 is the additive group of the integers modulo 2: it has
only two elements, which can be labeled as 0 and 1, or as “even” and “odd”.
Owing to time-reversal invariance, the electronic Berry phase in a molecule—
on any closed path in configuration space—can only be either 0 or π mod 2π,
hence the Z2 nature of this invariant is obvious. We focus on the simplest
possible case, dealt with in Sec. 2.2: a two-level system with (or without)
a conical intersection. We identify our closed “surface” with the closed loop:
a one-dimensional “torus” or “sphere”. In absence of spin-orbit interaction,
a gauge which respects time-reversal symmetry requires the electronic wave
function to be everywhere real. In the nontrivial case (odd, i.e. γ = π) the
connection cannot be smooth on the whole “surface”, and necessarily has an
“obstruction” at a point on the path: see Fig. 2.3. The obstruction can be
moved but not removed, and is due to the “monopole” at the conical intersection,
in a region not visited by the molecular system.
However, it is a simple exercise to devise a smooth gauge throughout the
path: it is enough to adopt complex electronic wavefunctions. In modern jargon,
the obstruction is removed at the price of breaking time-reversal symmetry.
The topological invariant is easily retrieved from the discrete approach: only
three Hamiltonian diagonalizations on the path are needed. With reference to
Fig. 2.3 we define
1 1 1
|ψ1 i = √ (|Bi − |Ci), |ψ2 i = √ (|Bi − |Ai), |ψ3 i = √ (|Ci − |Ai). (4.17)
2 2 2
Independently of any arbitrary gauge factor, whether it respects time-reversal
symmetry or otherwise, we always have
1
hψ1 |ψ2 ihψ2 |ψ3 ihψ3 |ψ1 i = − , (4.18)
8
37
i.e. γ = π mod 2π: the ground state is Z2 odd.
The Z2 topological invariant is extremely robust against smooth deforma-
tions of the Hamiltonian. For instance, here we have addressed the ultrasim-
ple tight-binding model, but the ground wavefunction can be “continuously
deformed” to the exact correlated wavefunction: topology-wise, the two wave-
functions are essentially the same object: they can be regarded as analogous to
the proverbial doughnut and coffee cup, Figs. 1.1 and 1.3.
Only two conditions are essential to conserve the Z2 invariant: (1) the defor-
mation of the Hamiltonian must conserve time-reversal symmetry, and (2) the
number of conical intersections inside the path must remain the same. Notice
that at the conical intersection the Born-Oppenheimer approximation breaks
down. If (2) is violated, then a singularity has crossed the path during the con-
tinuous transformation: the gap between the ground state and the first excited
state did not remain open.
4.2.4 Integer quantum Hall effect

The famous TKNN (Thouless, Kohmoto, Nightingale, and den Nijs) paper ap-
peared in 1982 [50] and marks the very first occurrence of a topological invariant
(the first Chern number) in electronic structure. The outline provided here is
inspired by Kohmoto [66].
We consider a two-dimensional independent-electron system in a lattice-
periodical potential, and subject to a perpendicular B field. The Hamiltonian
is not translationally invariant, but one can address the magnetic translation
group. We choose a large enough “supercell”, such that the magnetic flux is
commensurate (i.e. an integer number of flux quanta φ0 thread the supercell):
in this case a continuous k vector can be defined in the magnetic Brillouin zone.
As in Sec. 3.9,we define |ψjk i = eik·r |ujk i and Hk = e−ik·r Heik·r ; the latter
takes here the form
1 h e i2
Hk = p + ~k + A(r) + V (r), (4.19)
m c
where V is the substrate potential. The velocity can be expressed as
1
v= ∇k Hk , (4.20)
~
a formula often recurring in the present Notes in various forms, see e.g.
Eqs. (1.18) and (2.17).
The Kubo formula for transverse conductivity is [67]
X 1 Z hujk |vx |uj ′ k ihuj ′ k |vx |ujk i
σxy = 2~e2 Im 2
dk (4.21)
(2π) ǫjk <ǫF <ǫj′ k (ǫjk − ǫj ′ k )2
jj ′
e2 X 1 Z hujk |∂x Hk |uj ′ k ihuj ′ k |∂y Hk |ujk i

= 2 Im 2
dk .
~ ′
jj
If we now consider the case where the Fermi level lies in a gap, with n filled
bands (Landau levels in a flat potential), Eq. (4.21) becomes the BZ integral
n ∞
e2 1 hujk |∂x Hk |uj ′ k ihuj ′ k |∂y Hk |ujk i
Z X X
σxy = 2 Im dk . (4.22)
~ (2π)2 BZ j=1 ′ (ǫjk − ǫj ′ k )2
j =n+1
38
The integrand is just a simple generalization of the sum-over-states formula of
Eq. (3.28). Using the same arguments as in Ch. 3 it is rather straightforward
to arrive at the identity
n ∞
X X hujk |∂x Hk |uj ′ k ihuj ′ k |∂y Hk |ujk i
Im
j=1 ′
(ǫjk − ǫj ′ k )2
j =n+1
n
X 1
= Im h∂x ujk |∂x ujk i = − Ωxy (k), (4.23)
j=1
2
where the many-band Berry curvature, Eq. (3.50), appears. Since the BZ is a
torus, the BZ integral of the curvature equals 2π times an integer, the (first)
Chern number C. The milestone TKNN discovery is that Hall conductivity is
a Chern number when expressed in klitzing−1 :
e2
σxy = − C. (4.24)
h
Notice that the sign choices are not uniform across the literature.
Conductivity is a property of the excitations of the system, as it is perspic-
uous in the Kubo formula above. The Chern number, instead, is a ground state
property. The identity relating them belongs to the general class of fluctuation-
dissipation theorems, although this looks like an oxymoron, the Hall conduc-
tivity being here dissipationless. The interpretation of the Chern number as
a ground-state quantum fluctuation will be elaborated somewhere else in the
present Notes.
The topological nature of the observable explains its extreme robustness un-
der variations of magnetic field, carrier density, substrate disorder, and more.
The topological invariant C is identified with the filling ν using the same argu-
ments as in Sec. 2.4.4; the integer can only be varied by crosssing a conducting
state.
4.3 Adiabatic approximation in a magnetic field

The general problem of the nuclear motion—both classical and quantum—
in presence of an external magnetic field has been first solved in 1988 by
Schmelcher, Cederbaum, and Meyer [68]. It is remarkable that such a funda-
mental problem was solved so late, and that even today the relevant literature
is ignored by textbooks and little cited. The solution is a manifestation of geo-
metrical effects in electronic wavefunctions [69], which appears in a spectacular
way even when the nuclear motion is addressed at the classical level.
When a genuine magnetic field, generated by some external source, acts on
the molecular system, the Hamiltonian of Eq. (4.13) is modified by the addition
of a vector potential term in the kinetic energies of both the nuclei and the
electrons. Proceeding as in the zero field case, one writes an ansatz wavefunction
and arrives at the effective Hamiltonian for the ionic motion, Eq. (4.16), where
an extra term must be added to the kinematical momentum π of Eq. (4.15).
There are thus two vector potentials in the effective nuclear Hamiltonian: a
geometric one, and a genuinely magnetic one.
However, with respect to the zero-field case, there is a qualitative differ-
ence whose importance is overwhelming. Since the electronic Hamiltonian is no
39
longer invariant under time-reversal, the electronic wavefunction is necessarily
complex, and the curvature is in general nonzero. No singularity is needed to
produce geometrical effects on the nuclear motion; the Berry phase will be in
general nonzero on any path in the space of nuclear coordinates.
Suppose we are interested into the nuclear motion at the purely classical
level. The Hamiltonian of Eq. (4.16)—whose kinematical momentum π includes
now the two different vector potentials—yields the Hamilton equations of mo-
tion, which can be transformed into the Newton equations of motion: within the
latter, the effects of the vector potentials appear in terms of fields, in the form
of Lorentz forces. The curl of the magnetic vector potential obviously yields
the magnetic field due to the external source; the curl of the geometric vector
potential (Berry curvature) yields an additional “magnetic-like” field which is
nonzero even on the classical trajectory of the nuclei. We stress that this is at
variance with the zero-field case, where the Berry phase had no effect on the
ionic motion at the classical level, and could only be detected when quantizing
the ionic degrees of freedom.
Within a naı̈ve Born–Oppenheimer approximation—where Berry phases are
neglected—the magnetic field acts on the nuclei as it they were “naked” charges:
a proper treatment must instead account for electronic screening: this is pro-
vided by the geometric vector potential. Surprisingly, there are very few calcu-
lations of the effect: it is pretty clear, however, that the geometric term is no
small correction.
For pedagogical purposes we consider the case of a hydrogen atom, hence we
identify the electronic degrees of freedom [x], used in Sec. 4.2.2, with a single
coordinate r, and the parameter ξ with the nuclear coordinate R. If the atom is
subject only to a magnetic field, the complete Hamiltonian H and the electronic
Hamiltonian H are
1 h e i2
H(R, r) = p − A(R) + H(R, r); (4.25)
2M c
1 h e i2 e2
H(R, r) = −i~∇r + A(r) − . (4.26)
2m c |r − R|
As explained above, the nuclear kinematical momentum of Eq. (4.15) becomes
e
π = p − ~A(R) − A(R) (4.27)
c
The case of a constant B field can be dealt with analytically. We choose the
central gauge A(r) = 21 B × r. If φ(r) is the exact ground eigenfunction when
the proton sits at R = 0, the eigenfunction at a generic R is:
ie
hr|ψ(R)i = e− 2~c r·B×R φ(r − R), (4.28)
with an R-independent eigenvalue. The Berry connection is clearly
e e
A(R) = ihψ(R)|∇R ψ(R)i = − hψ(R)|B × r|ψ(R)i = − B×R
2~c 2~c
e
= − A(R), (4.29)
~c
since the R-derivative of φ(r − R) does not contribute. Replacing Eq. (4.29)
into Eq. (4.27) we find π = p, as it must be: the nucleus travels at constant
speed, and is not deflected by a Lorentz force.
40
Remarkably, the “magnetic-like” field due to the Berry phase is—in this sim-
ple example—exactly opposite to the external magnetic field, thus providing the
complete screening which is physically expected. In less trivial situations, the
screening affects significantly the molecular vibrations and the classical nuclear
motion in general. The case of H2 has been investigated in 2007 by Ceresoli et
al. [70].
4.4 Anomalous Hall effect

Edwin H. Hall discovered the eponymous effect, nowadays in all solid state
textbooks, in 1879. Shortly later, in 1881, he discovered that the effect in
ferromagnetic materials is “anomalous”. While in nonmagnetic materials the
Hall resistivity is linear in the magnetic field, in ferromagnetic ones it saturates
to a value roughly proportional to the macroscopic magnetization.
In 1954 Karplus and Luttinger [71] provided a theory of the anomalous Hall
effect (AHE) which, with hindsight, was indeed geometrical. The understanding
of AHE, however, remained controversial for another 40 years and more. One of
the reasons for this state of affairs is that the intrinsic geometrical contribution
is partly obscured by extrinsic contributions (jargon: skew scattering and side
jumps), not easily disentangled in the experimental data. A comprehensive
review appeared in 2010 [12]. Here we only outline the intrinsic theory, based
on geometrical concepts, which owes to a couple of papers appeared in 2002, by
Jungwith, Niu, and MacDonald [72], and by Onoda and Nagaosa [73].
The starting point is the Kubo formula of Eq. (4.21), which we rewrite in
dimension d as
2
X 1 Z hujk |vx |uj ′ k ihuj ′ k |vx |ujk i
σxy = 2~e Im d
dk . (4.30)
′
jj
While in Sec. 4.2.4 this was integrated in the BZ, for a metal the k integral is
limited to the volume inside the Fermi surface. Nonetheless, the transformation
of the integrand, from a sum over states into a curvature, proceeds in the same
way. The geometrical contribution to the AHE is therefore proportional to the
integral of the Berry curvature within the Fermi surface. To allow for band
crossings, we write the intrinsic AHE conductivity in dimension d as
∞ Z
e2 1 X
σxy = 2 Im dk h∂x ujk |∂x ujk i; (4.31)
~ (2π)d j=1 ǫjk <ǫF
the analogy to Eqs. (4.22) and (4.23) is self evident. But this analogy is partly
misleading: at variance with the quantum Hall case, no macroscopic B field is
present here, and the u orbitals in Eq. (4.31) are the (periodic parts of) genuine
Bloch orbitals.
Given that the Fermi surface is symmetrical under k → −k, the symmetry
considerations of Sec. 3.7 show that Eq. (4.31) can be nonzero only if time-
reversal symmetry is broken, while inversion symmetry is irrelevant. The typical
case studies are the ferromagnetic metals, whose ground state breaks indeed
time-reversal symmetry in absence of a macroscopic B field.
First-principle calculations were performed for Ni, Cu, and Fe, as well as
or for some oxides. The intrinsic geometric contribution appears to be the
41
dominant one. These calculations also pointed out the crucial role played by
avoided crossings of the bands near the Fermi surface, which induce a very spiky
behavior of the Berry curvature in the BZ. More than 106 k points where used
in Ref. [74] in order to perform the integration in Eq. (4.31); a more efficient
strategy was devised later [75].
A noninteracting (e.g. KS) many-electron system is a trivial example of
a Fermi liquid. Haldane [76] pointed out that the very basic tenet of Lan-
dau’s Fermi-liquid theory is that charge transport involves only quasiparticles
with energies within kB T of the Fermi level. This is apparently at odds with
Eq. (4.31), which is an integration over the whole occupied Fermi sea. The two
viewpoints can be reconciled, essentially via an integration by parts [76]. Even
this alternative form has been implemented in first-principle calculations [77].
4.5 Semiclassical transport

The semiclassical theory of Bloch electron dynamics plays a fundamental role in
the physics of metals and semiconductors, and is a typical textbook topic [78].
The theory addresses the motion of a wave packet built as a superposition of
Bloch states from the n-th band
Z
|W i = dk a(k, t)|ψnk i, (4.32)
BZ
where the envelope function is well localized in k-space. Because of this, it

is delocalized in r space; we assume, however, that its center of mass is well
defined. Owing to this, we may define the wave vector k and the center r of the
wave packet as
Z
k = dk′ k′ |a(k′ , t)|2 ; r = hW |r|W i. (4.33)
BZ
4.5.1 Textbook equations of motion

In absence of collisions, the equations of motion (EoM) reported in textbooks
and routinely used in device engineering are
1 ∂ǫk
ṙ =
~ ∂k
1
~k̇ = −e (E + ṙ × B), (4.34)
c
where ǫk is the band structure of the relevant band, and E and B are the
perturbing fields, assumed weak and slowly varying in space and time.
As emphasized in Ref. [78], the derivation of Eq. (4.34), despite the formal
simplicity of the result, is “a formidable task”. The early derivations date from
the 1930s; the problem was reconsidered several times in the literature, by Slater
[79], Luttinger [80], and Zak [81] among others.
4.5.2 Modern equations of motion

The modern analysis of semiclassical transport owes to a couple of papers by
Q. Niu and coworkers [63, 64] (see also Ref. [13]). They start observing that a
42
wave packet generally possesses two kinds of motion: the center-of-mass motion
and the self-rotation around its center. Owing to the latter, the wave packet is
endowed with an orbital magnetic moment, whose expression is
ie
m(k) = − h∇k uk | × (Hk − ǫk ) |∇k uk i. (4.35)
2~c
Because of this, the band structure acquires a B-dependent term
ǫk → ǫ̃k − m(k) · B; (4.36)
Furthermore, the canonical momentum ~k has to be replaced with the kinetic

momentum, which includes the (geometrical) vector potential. In the Newton-
like EoM, its contribution is reminiscent of a Lorentz force in reciprocal space.
Eq. (4.34) must then be replaced with
1 ∂ǫ̃k
ṙ = − k̇ × Ω(k)
~ ∂k
1
~k̇ = −e (E + ṙ × B), (4.37)
c
where Ω(k) is the Berry curvature of the relevant band, having the dimensions
of a squared length. Notice that the curvature is nonzero even in presence of
time-reversal symmetry, provided that the crystal is noncentrosymmetric.
4.5.3 Equations of motion in symplectic form

A more accurate analysis performed in 2003 by Panati, Spohn, and Teufel [82]
proves that Eq. (4.37) is not yet the ultimate form for the semiclassical EoM;
it is only correct if B is constant in space. Furthermore, the equations which
supersede Eq. (4.37) are more symmetrical and elegant.
One starts defining
H (r, k) = ǫk − eφ(r) − m(k) · B(r), (4.38)
where φ(r) is the potential of the electric field E(r), and m(k) is the wave
packet’s moment, Eq. (4.35). Then the EoM are
1 ∂H (r, k)
ṙ = − k̇ × Ω(k)
~ ∂k
1 ∂H (r, k) e
k̇ = − − ṙ × B(r), . (4.39)
~ ∂r ~c
In order to realize the relationship of Eq. (4.39) to the standard symplectic form
of Hamilton’s equations, it is expedient to adopt atomic units, and furthermore
Figure 4.4: A wave packet car-

ries quasi momentum ~k, but
also an orbital moment and
a corresponding magnetic mo-
ment m(k)
43
to redefine B by including the 1/c factor: we get
ṙ = ∇p H (r, p) − ṗ × Ω(p)
ṗ = −∇r H (r, p) − ṙ × B(r). (4.40)
Notice that the translational momentum is coupled to B via the explicit Lorentz
force ṙ × B(r), unlike in the standard Hamiltonian formulation where such
coupling would appear via a vector-potential-dependent H . It is easy to prove
that Eq. (4.40) conserves the energy:
dH
= ∇r H · ṙ + ∇p H · ṗ = 0 (4.41)
dt
In the simple case where Ω and B are identically vanishing, H becomes the
standard classical Hamiltonian and Eq. (4.40) are the corresponding Hamilton’s
equations. If I is the 3 × 3 identity, the standard 6 × 6 symplectic matrix is

0 −I
Θ= , (4.42)
I 0
and the symplectic form of Eq. (4.40) is

ṙ ∇r H
Θ = . (4.43)
ṗ ∇p H
When either B or Ω (or both) are nonzero, one recasts identically Eq. (4.40) in
the similar form
ṙ ∇r H
Θ̃ = , (4.44)
ṗ ∇p H
where Θ̃ is a deformed symplectic matrix defined as
~

B(r) −I
Θ̃ = ~ , (4.45)
I Ω(p)
~
and the 3 × 3 antisymmetric matrices B(r) ~
and Ω(p) are
   
0 −Bz By 0 −Ωz Ωy
B~ =  Bz 0 −Bx  , Ω ~ =  Ωz 0 −Ωx  . (4.46)
−By Bx 0 −Ωy Ωx 0
~ and the
Finally, we notice that if we set to zero both the Berry curvature Ω
moment m(k) in Eq. (4.38), then Eq. (4.44) provides an alternative form for
Hamilton’s equations, where the B field appears in the symplectic matrix and
no vector potential appears in the Hamiltonian H . This form is well known in
classical mechanics, and goes under the name of gauge-invariant Hamiltonian
formulation.
4.5.4 Geometrical correction to the density of states

In a remarkable 2005 paper by Xiao, Shi, and Niu [83] it was pointed out
that Eq. (4.37), in presence of a nonzero B field, violates Liouville’s theorem.
This means that the volume element ∆V = ∆r∆p changes in time during the
44
evolution of the system; it is possible, however to remedy this shortcoming in
an elegant way.
We are not following the original paper; instead, we start from the more
elegant symplectic formulation above. The time evolution of the volume element
is
1 ∂V
= ∇r · ṙ + ∇p · ṗ
∆V ∂t
= ∇r · [ ∇p H − ṗ × Ω(p) ] + ∇p · [ −∇r H − ṙ × B(r) ]
= −∇r · [ ṗ × Ω(p) ] − ∇p · [ ṙ × B(r) ]. (4.47)
After a somewhat lengthy calculation, one arrives at a very simple result,

which—most notably—is a total derivative
1 ∂V d
= − log [ 1 + B(r) · Ω(p) ], (4.48)
∆V ∂t dt
and is clearly nonzero whenever B and Ω are non constant in phase space. It is
worth mentioning that Eq. (4.48) bears a simple relationship to the deformed
symplectic matrix, Eq. (4.45): in fact
q
1 + B(r) · Ω(p) = det Θ̃(r, p). (4.49)
In order to arrive at Eqs. (4.48) and (4.49) two special features of the matrix
Θ̃(r, p) seem to be essential: its elements are either functions of r or p, but not
of both; the divergences of B(r) and Ω(p) are both zero.
Within ordinary statistical mechanics, the density of states in phase space is
1/hd (in dimension d). The number of states in volume ∆V = ∆r∆p is ∆V /h3 ,
and—owing to Liouville’s theorem—this number remains constant during time
evolution. Here instead, such number remains constant only if we appropriately
modify the density of states. For the sake of clarity we restore Gauss units; the
modification needed is
1 1 2π
−→ d [ 1 + B(r) · Ω(p/~) ], (4.50)
hd h φ0
where φ0 is the flux quantum.
4.5.5 Outstanding consequences of the modified density

of states
After Ref. [83] appeared, it become immediately clear that the geometrical
correction to the density of states, Eq. (4.50), goes well beyond the scope of
semiclassical approximation. To my present taste, a convincing proof of the
universality of Eq. (4.50) has not yet appeared; nonetheless its consequences are
unavoidable. Statistical mechanics paraphernalia, such as partition functions
and the like, must adopt Eq. (4.50).
As a consequence of the modified density of states, the Fermi volume of a
metal changes when a macroscopic B field is switched on at constant electron
density. If instead we keep the chemical potential µ constant, then the electron
45
density n depends on B. At zero temperature (for each spin channel)

1 2π
Z
n = dk 1 + B · Ω(k) ϑ(µ − ǫk )
(2π)d BZ φ0

∂n 1
Z
= dk Ω(k) ϑ(µ − ǫk ) (4.51)
∂B µ (2π)d−1 φ0 BZ
The latter assumes a perspicuous meaning in 2d, when µ is in a gap:

∂n 1 C1 1
Z
= dk Ω(k) = = − σxy . (4.52)
∂B µ 2πφ0 BZ φ0 ec
For a quantum Hall system, this goes under the name of Streda formula [84],
and had been first derived in 1982 in a very different way.
4.6 Quantum transport

4.6.1 Transport by a single state
We are going to study here the current induced by an adiabatic change of the
potential, or more generally of the Hamiltonian, in the single-particle case. We
indicate as ψn (t) the adiabatic instantaneous eigenstates, and with ψ(t) the time
evolution of the ground state. In order to get rid of the dynamical phase, it is
better to deal with density matrices
ρ(t) = |ψ(t)ihψ(t)| = |ψ0 (t)ihψ0 (t)| + ∆ρ(t). (4.53)
The velocity of this state is
v(t) = Tr {ρ(t) v} (4.54)
X
= hψ0 (t)| v |ψ0 (t)i + hψ0 (t)| ∆ρ(t) |ψn (t)ihψn (t)| v |ψ0 (t)i.
n
Since the adiabatic density matrix commutes with H(t), the time evolution
is
d
[H(t), ∆ρ(t)] = i~ρ̇(t) ≃ i~ |ψ0 (t)ihψ0 (t)|
dt
= i~ ( |ψ̇0 (t)ihψ0 (t)| − |ψ0 (t)ihψ̇0 (t)| ), (4.55)
where we are neglecting a term of higher order in the adiabaticity parameter.
We now take the matrix elements between hψ0 (t)| and |ψn (t)i:
(E0 − En )hψ0 (t)| ∆ρ(t) |ψn (t)i = i~ ( hψn (t)|ψ̇0 (t)i − hψ̇0 (t)|ψn (t)i ). (4.56)
The term with n = 0 in the rhs vanishes because of norm conservation; replace-
ment into Eq. (4.54) yields
v(t) = hψ0 (t)| v |ψ0 (t)i
" #
X hψ̇0 (t)|ψn (t)ihψn (t)| v |ψ0 (t)i
+i~ − c.c. . (4.57)
E0 − En
n6=0
The first term on the rhs is zero in the special case where—at all times—the
Hamiltonian H(t) is time-reversal invariant and the state is nondegenerate.
46
4.6.2 Current carried by filled bands
We now exploit the previous result for a system of noninteracting electrons in
the case where the Hamiltonian H(t) is lattice periodical and the ground state
is insulating; this means that the gap remains finite at all t. It will be enough
to consider the simple case of just one filled band, with band index zero; the
current carried by each Bloch orbital |ψ0k i = eik·r |u0k i is
" #
X hψ̇0k |ψjk ihψjk | v |ψ0k i
vk (t) = hψ0k | v |ψ0k i + i~ − c.c.
ǫ0k − ǫjk
j6=0
X hu̇0k |ujk ihujk | v |u0k i
= hu0k | v |u0k i + i~ − c.c. , (4.58)
ǫ0k − ǫjk
j6=0
where the t dependence of the rhs is now implicit. We then adopt the usual
formula for the velocity, Eqs. (1.18) and (4.20), and the analog of Eq. (3.27):
1 1X hujk | v |u0k i
v= ∇k Hk , |∇k u0k i = |ujk i (4.59)
~ ~ ǫ0k − ǫjk
j6=0
1
vk (t) = ∇k ǫ0k + i ( hu̇0k |∇k u0k i − h∇k u0k |u̇0k i ). (4.60)
~
The first term on the rhs integrates to zero over the BZ, while the second is
clearly a Berry curvature component in the four-dimensional k, t domain. The
current density carried by a filled band in dimension d is
ie
Z
j(t) = − dk ( hu̇0k |∇k u0k i − h∇k u0k |u̇0k i ); (4.61)
(2π)d BZ
the Bloch states are normalized to one in the crystal cell (as everywhere in the
present Notes).
4.6.3 Quantization of charge transport

Let us consider the special case of a simple cubic crystal with lattice constant
a. The transported charge in time T in the z direction across one cell is
T T π/a π/a
ea2
Z Z Z Z
Q = dt Iz (t) = a2 dt jz (t) = − dkx dky
0 0 (2π)2 −π/a −π/a
T π/a
i
Z Z
× dt dkz ( hu̇0k |∂z u0k i − h∂z u0k |u̇0k i ). (4.62)
2π 0 −π/a
If the time evolution of the Hamiltonian is cyclic H(T ) = H(0), then the second
line in Eq. (4.62) is clearly a Chern number C (in the kz , t variables) and is
integer. Notice also that C is dimensionless, and therefore does not depend on
how fast the Hamiltonian varies with time; ideally, the adiabatic regime means
T → ∞.
We arrive therefore at the outstanding result
Z T
Q= dt Iz (t) = e × integer (4.63)
0
47
first proved by Thouless in 1983 [85]; it holds of course for any dimension d.
Let me restate the theorem: if the Hamiltonian is changed adiabatically in such
a way that it returns to its starting value in time T , the transported charge
in an infinite periodic system is quantized provided that the system remains
insulating at all times. A cycle pumps an integer number of elementary charges
across the system.
Among the examples which realize a “Thouless pump”, the original paper
suggests a sliding charge-density wave. A more outstanding manifestation of
quantized charge transport was pointed out shortly afterwards by Pendry and
Hodges [86]: Faradays’ laws of of electrolysis (1832). The mass/charge transfer
ratio shows that charge is always transported in units of e per ion, to the extent
that electrolytic cells are used as standards of current. If a given ion sits at one
electrode at t = 0, and if it drifts to the other one at t = T , the Hamiltonian
can be considered as cyclic, whence charge quantization follows. However, at
intermediate t values the charge “belonging” to a given ion is definitely non
quantized, and arbitrarily defined: for a review of the possible definitions, see
Ref. [87].
Thouless quantization of charge transport [85], discussed above, also has
profound relationships to later advances: namely, to the topological explanation
of the quantization of surface charge (discussed in Sec. 5.3.4), and to the modern
theory of polarization (discussed in Sec. 5.3).
48
Chapter 5
Macroscopic polarization
The macroscopic polarization P is a fundamental concept that all undergradu-

ates learn about in elementary courses [88, 89]. In view of this, it is truly ex-
traordinary that until the early 1990s there was no generally accepted formula
for P in condensed matter, even as a matter of principles. P is an intensive
vector quantities that intuitively carries the meaning of dipole per unit volume.
Most textbooks [78, 90] provide a flawed definition of P, not implementable in
practical computations [91].
A genuine change of paradigm was initiated by a couple of important pa-
pers [92, 93], after which the major development was introduced by King-Smith
and Vanderbilt in 1992 (paper published in 1993 [94]). Other important ad-
vances occurred during the 1990s [95, 96] and the so-called “modern theory of
polarization”it is at a mature stage since more than a decade; several reviews
have appeared in the literature over the years [1, 2, 3, 5, 6, 7, 8, 20].
Aiming at a computational physics readership, it is worth emphasizing that
most ab-initio electronic-structure codes on the market, for dealing with ei-
ther crystalline or noncrystalline materials, implement the modern theory of
polarization as a standard option. A nonexaustive list includes abinit [59],
crystal [60], quantum-espresso [62], siesta [97], vasp [61], and cpmd [98];
the code tutorials often include an outline of polarization theory. Its imple-
mentations have been instrumental since more than a decade in the study of
ferroelectric and piezoelectric materials [99, 100, 101].
The basic concepts of the modern theory of polarization also start reaching
a few textbooks [102, 103], though very slowly; most of them are still plagued
with erroneous concepts and statements.
5.1 Polarization and electric field

The modern theory of polarization, at least in its original form, only addresses
the polarization P in a null macroscopic E field; in this case P can be nonzero
only if the medium breaks inversion symmetry. This applies to noncentrosym-
metric crystals, and more generally to cases where inversion symmetry is broken
by some perturbation (as e.g. a frozen phonon, or piezoelectric strain).
It must be realized that, insofar as we address an infinite system with no
boundaries, the E field is quite arbitrary. The microscopic charge density is
neutral in average and lattice periodical; the value of E is just an arbitrary
49
z
+ + + + + +
− − − − − −
x
Figure 5.1: Macroscopic polarization P in a slab normal to z, for a vanishing

external field E(ext) . Left: When P is normal to the slab, a depolarizing field
E = −4πP is present inside the slab, and charges at its surface, with areal
density σsurface = P · n Right: When P is parallel to the slab, no depolarizing
field and no surface charge is present.
boundary condition for the integration of Poisson’s equation. The usual choice
(performed within all electronic-structure codes) is to impose a lattice-periodical
Coulomb potential, i.e. E = 0. Imposing a given nonzero value of E is equally
legitimate (in insulators), although technically more difficult [104, 105]).
When addressing a finite sample with boundaries, the E field is in princi-
ple measurable inside the material, without reference to what happens at the
sample boundary; this is not the case of D. In fact, E obtains by averaging
over a macroscopic length scale the microscopic electric field E(micro) (r), which
fluctuates at the atomic scale [89]. In a macroscopically homogeneous system
the macroscopic field E is constant, and in crystalline materials it coincides with
the cell average of E(micro) (r). A lattice-periodical potential enforces E = 0; for
a supercell calculation, this applies to the field average over the supercell, while
in different regions there can be a nonzero macroscopic field.
As explained so far, there is no need of addressing finite samples and external
vs. internal fields from a theoretician’s viewpoint. Nonetheless a brief digression
is in order, given that experiments are performed over finite samples, often in
external fields. Suppose a finite macroscopic sample is inserted in a constant
external field E(ext) : the microscopic field E(micro) (r) coincides with E(ext) far
away from the sample, while it is different inside because of screening effects.
If we choose an homogeneous sample of ellipsoidal shape, then the macroscopic
average of E(micro)(r), i.e. the macroscopic screened field E, is constant in
the bulk of the sample. The shape effects are embedded in the depolarization
coefficients [88]: the simplest case is the extremely oblate ellipsoid, i.e. a slab of
a macroscopically homogeneous dielectric; more details are given in Ref. [8]. For
the slab geometry in a vanishing external field E(ext) the internal field E vanishes
when P is parallel to the slab (transverse polarization), while E = −4πP is the
depolarization field when P is normal to the slab (longitudinal polarization):
see Fig. 5.1.
5.2 Polarization “itself” vs. polarization differ-

ence
Novel ideas about macroscopic polarization emerged in the early 1990s [92, 93];
these led to the modern theory, based on a Berry phase, which was founded
by King-Smith and Vanderbilt soon afterwards [94]. At its foundation, the
50
Figure 5.2: Top panel: The 14-
atom BeO supercell in a ver-
tical plane through the BeO
bonds; the wurtzite (W) and
zincblende (ZB) stackings are
perspicuous. Bottom panel:
Macroscopic averages of the va-
lence electron density (solid)
and of the electrostatic poten-
tial (dotted).
modern theory was limited to a crystalline system in an independent-electron

framework (either KS or Hartree-Fock). Later, the theory was extended to
correlated and/or disordered systems [95, 96].
The first calculation ever of spontaneous polarization was published in
1990 [92]. The case study was BeO: it has the simplest structure where in-
version symmetry is absent (i.e. wurtzite), and furthermore its constituents
are first-row atoms. The idea was to address the macroscopic polarization of
a slab of finite thickness, with faces normal to the c axis, embedding it in an
ad hoc medium which (i) has no bulk polarization for symmetry reasons, and
(2) does not produce any geometrical or chemical perturbation at the interface.
The optimal choice is a fictitious BeO in the zincblende structure. Because of
obvious reasons, the system is periodically replicated in a supercell geometry
(Fig. 5.2, top panel). The selfconsistent calculation shows well localized interface
charges, of opposite sign and equal magnitudes at the two nonequivalent inter-
faces (Fig. 5.2, bottom panel). The interface charge is related to the difference
in polarization between the two materials: σinterface = ∆P · n. The computer
experiment provides the value of σinterface , and since P vanishes by symmetry
in the zincblende slab, one thus obtains the bulk value of P in the wurtzite
material. Notice that here P is a longitudinal polarization, in a depolarizing
field.
It must be emphasized that the quantity really “measured” in this computer
experiment is ∆P, not the polarization P itself. After Ref [92] was published,
a study of the experimental literature showed that—contrary to an incorrect
widespread belief—no experimental value of P in any wurtzite material ex-
ists: only estimates are available. Ref. [92] marks, as said above, a change of
paradigm: polarization must be defined by means of differences, and the con-
cept of polarization “itself” must be abandoned. With hindsight, it is nowadays
pretty clear that the problem, as well as its solution, exists already at the clas-
sical level: this is sketched in Fig. 5.3. Most textbooks are missing this very
basic fact.
The modern theory avoids addressing the “absolute” polarization of a given
equilibrium state, quite in agreement with the experiments, which invariably
measure polarization differences. Instead, the theory addresses differences in
polarization between two states of the material that can be connected by an
51
Figure 5.3: A 1d solid with infinite
length. Different choices of the unit cell
give different P values: (a), (b). On the
other hand, the change of polarization
∆P does not depend on the choice of
the unit cell (c).
adiabatic switching process. The time-dependent Hamiltonian is assumed to

remain insulating at all times, and the polarization difference is then defined
[93] as the time-integrated transient macroscopic current that flows through the
insulating sample during the switching process:
Z ∆t
∆P = P(∆t) − P(0) = dt j(t). (5.1)
0
In the adiabatic limit ∆t → ∞ and j(t) → 0, while ∆P stays finite. Ad-

dressing currents (instead of charges) explains the occurrence of phases of the
wavefunctions (instead of square moduli) in the modern theory. Eventually the
time integration in Eq. (5.1) will be eliminated, leading to a two-point formula
involving only the initial and final states.
5.3 Independent electrons

5.3.1 The King-Smith and Vanderbilt formula
For a crystalline system of independent electrons the expression for the tran-
sient current occurring in Eq. (5.1) is precisely the same as previously derived
for quantum transport, Eq. (4.61). Therefore for one band (index 0) and sin-
gle occupancy in dimension d the electronic contribution to the polarization
difference is
Z ∆t Z
ie
∆P = − dt dk ( hu̇0k |∇k u0k i − h∇k u0k |u̇0k i ); (5.2)
(2π)d 0 BZ
the (classical) nuclear contribution must be added separately. We remind that

crystal-cell neutrality is essential. Notice also that, given the occurrence of
Bloch states, the Hamiltonian is lattice periodical at all t: this implicitly means
that in Eq. (5.2) ∆P is evaluated at E = 0.
It is now expedient to introduce a dimensionless adiabatic time λ, with
|ujk (t)i = |ujk (λ(t))i, λ(0) = 0, λ(1) = ∆t. Eq. (5.2) becomes then, for n
doubly occupied bands (index j = 1, n) in 3d:
Z 1
∆P = P(1) − P(0) = dλ ∂λ P(λ)
0
n Z
2ie X
∂λ P = − dk ( h∂λ ujk |∇k ujk i − h∇k ujk |∂λ ujk i ). (5.3)
(2π)3 j=1 BZ
52
It is essential that the gap does not close, i.e. the system remain insulating, for
all λ values.
The expression in Eq. (5.3) can be integrated with respect to λ to obtain
n Z
2ie X
P(λ) = − dk hujk |∇k ujk i : (5.4)
(2π)3 j=1 BZ
this is the (by now famous) King-Smith and Vanderbilt formula [94], yielding
the polarization of the final state minus the polarization of the initial state,
Eq. (5.3). To understand the meaning of the k integral in 3d we take the simple
example of a simple cubic lattice of constant a, similarly to Eq. (4.62):
n Z
" Z #
2e X π/a
Z π/a π/a
Pz (λ) = − dkx dky i dkz hujk |∂kz ujk i , (5.5)
(2π)3 j=1 −π/a −π/a −π/a
where the square parenthesis highlights the Berry-phase, to be compared to

Eq. (3.43).
In first-principle implementations, the Berry phase is discretized as in
Eqs. (3.45) and (3.46), and the remaining 2d k-integral is discretized in the
trivial way. The first calculation ever of the “spontaneous” polarization of a
ferroelectric material (KNbO3 ) appeared in 1993 [106], and agreed within 10%
with the measured values. As said above, the Berry-phase formula is nowadays
implemented in most first-principle codes.
5.3.2 The quantum of polarization

Given that every phase is defined modulo 2π, all of the two-point formulas for
∆P in terms of Berry phases are arbitrary modulo a polarization “quantum”.
This is the tradeoff one has to pay when switching from the curvature formula,
Eq. (5.3)—where no such arbitrariness exists—to the two-point King-Smith and
Vanderbilt formula, where only the connection occurs. The actual arbitrariness
of ∆P in 3d is 2eR/Vcell, where R is a lattice vector and Vcell is the cell volume
(the 2 factor owes to double band occupancy). A similar arbitrariness of an
integer times eR/Vcell occurs for the classical nuclear contribution to polariza-
tion.
The quantum arbitrariness is rarely a problem in practice. In most cases, the
change in P that can be induced by a perturbation, such as a small sublattice
displacement, is insufficient to cause P to change by a large fraction of the
quantum. Where exceptions exist the ambiguity can be resolved by subdividing
the adiabatic path into several shorter intervals, for each of which the change
in P is unambiguous for practical purposes. Additional problems may occur in
the discretized version of the Berry-phase formula; this is discussed e.g. in Ref.
[8] and, in more detail, in Ref. [7].
Here we stress that the quantum ambiguity is an essential aspect of the
theory. For example, for the case of a closed cyclic adiabatic evolution of the
system, in which the parameter values λ = 0 and λ = 1 label the same physical
state of the system, we retrieve the quantization of charge transport, discussed
in Sec. 4.6, and governed by a Chern number.
53
5.3.3 Wannier functions
The KS (or Hartee-Fock) ground state is a Slater determinant of doubly occupied
orbitals; any unitary transformation of the occupied states among themselves
leaves the determinantal wavefunction invariant (apart for an irrelevant phase
factor), and hence it leaves invariant any KS ground-state property.
For an insulating crystal the Bloch KS orbitals of completely occupied bands
can be transformed to localized Wannier orbitals (or functions) WFs. This is
known since 1937 [107], but for many years the WFs have been mostly used as a
formal tool; they became a popular topic in computational electronic structure
only after the 1997 seminal work of Marzari and Vanderbilt [108]. A compre-
hensive review appeared as Ref. [19], and a public-domain implementation is in
wannier90 [109]. If the crystal is metallic, the WFs can still be technically
useful [110], but it must be emphasized that the ground state cannot be writ-
ten as a Slater determinant of localized orbitals of any kind, as a matter of
principle [111].
The transformation of the Berry phase formula in terms of WFs provides an
alternative, and perhaps more intuitive, viewpoint. The formal transformation
was known since the 1950s [112], although the physical meaning of the formalism
was not understood until the advent of the modern theory of polarization.
The unitary transformation which defines the WF wjR (r), labeled by band
j and unit cell R, within our normalization is
Vcell
Z
|wjR i = dk eik·R |ψjk i . (5.6)
(2π)3 BZ
If one then defines the “Wannier centers” as rjR = hwjR |r|wjR i, it is rather
straightforward to prove that Eq. (5.4) is equivalent to
n
2e X
P(el) (λ) = − rj0 . (5.7)
Vcell j=1
This means that the electronic term in the macroscopic polarization P is (twice)
the dipole of the Wannier charge distributions in the central cell, divided by the
cell volume. The nuclear term is obviously similar in form to Eq. (5.7); the sum
of both terms is charge neutral.
WFs are severely gauge-dependent, since the phases of the |ψjk i appearing in
Eq. (5.6) can be chosen arbitrarily. However, their centers are gauge-invariant
modulo a lattice vector. Therefore P(el) in Eq. (5.7) is affected by the same
“quantum” indeterminacy discussed above. We also stress, once more, that
Eq. (5.7)—as well as Eq. (5.3)—is to be used in polarization differences, and
does not define polarization itself.
The modern theory, when formulated in terms of WFs, becomes much
more intuitive, and in a sense vindicates the venerable Clausius-Mossotti view-
point [113]: in fact, the charge distribution is partitioned into localized contri-
butions, each providing an electric dipole, and these dipoles yield the electronic
term in P. However, it is clear from Eq. (5.6) that the phase of the Bloch or-
bitals is essential to arrive at the right partitioning. Any decomposition based
on charge only is severely nonunique and does not provide in general the right
P, with the notable exception of the extreme case of molecular crystals.
54
In the latter case, in fact, we may consider the set of WFs centered on a
given molecule; their total charge distribution coincides—in the weakly inter-
acting limit—with the electron density of the isolated molecule (possibly in a
local field). This justifies the elementary Clausius-Mossotti viewpoint. It is
worth mentioning that the dipole of a polar molecule is routinely computed in a
supercell geometry via the single-point Berry phase discussed below [114]. The
dipole value coincides with the one computed in the trivial way in the large
supercell limit. Finite-size corrections, due to the local field (different in the
two cases), can also be applied [115].
The case of alkali halides—where the model is often phenomenologically
used—deserves a different comment [8]. The electron densities of isolated
ions (with or without fields) are quite different from the corresponding WFs
charge distributions, for instance because of orthogonality constraints: hence
the Clausius-Mossotti model is not justified in its elementary form, despite con-
trary statements in the literature. For a detailed analysis, see Ref. [116].
5.3.4 The surface charge theorem

The early occurrences of the theorem of quantization of the surface charge [35,
36, 37, 38] are discussed above, Sec. 2.3. The topological nature of this theorem
was first realized by Niu [117] in 1986; here we follow the treatment of Vanderbilt
and King-Smith [118] (see also Ref. [5]).
According to elementary electrostatics the macroscopic bound surface charge
density σsurface residing on the surface of a sample is related to the polarization
in the interior by σsurface = n̂ · P, where n̂ is the surface normal. One defines
the bound charge σsurface by saying that no free charge is present, but what,
precisely, does this mean? The surface must be insulating, with the electron
chemical potential lying in a gap that is common to both bulk and surface.
But this is not a unique prescription, since there can be a surface band which
is entirely occupied or entirely empty. The two cases differ by a polarization
quantum in the corresponding P value. In fact, given that the bulk polarization
P is arbitrary modulo eR/Vcell, it follows that the charge per surface area is
defined modulo a quantum
e
σsurface = n̂ · P modulo . (5.8)
Asurface
An equivalent formulation of the surface charge theorem can be arrived at by
means of WFs. The WF approach is most perspicuous for quasi-1d systems (e.g.
insulating polymers); a pedagogical presentation is in Ref. [119]. Notice that all
terminations are “insulating” in a polymer, there are no “surface” conducting
states.
The bulk-surface correspondence encoded in Eq. (5.8) is an outstanding man-
ifestation of topology in condensed matter physics: the surface charge of an in-
sulating surface is “topologically protected”. The actual value of σsurface among
the discrete allowed values is then determined by energy considerations.
For a centrosymmetric crystal it is tempting to guess that P(λ) in the Berry-
phase formula, Eq. (5.4), vanishes for any λ. Instead, this is not the case:
centrosymmetry only dictates that P = −P modulo a quantum, and therefore
Eq. (5.4) yields P equal to an integer or half integer multiple of the quantum
eR/Vcell (for single band occupancy). Then from Eq. (5.8) it follows that the
55
Figure 5.4: A cen-
trosymmetric insulating
quasi-1d “crystal” with
two different termina-
tion: alternant trans-
polyacetylene. Here the
”bulk” is five-monomer
long.
charge per surface cell may only be an integer or half integer, as first discov-
ered many years ago, and previously discussed in Sec. 2.3. Therein, it was
observed that this important theorem is often ignored even by specialists in sur-
face physics. A thorough analysis of polar surfaces, in the light of the present
theorem, has been published in 2011 by Stengel [120].
Among the values dictated by topology, Nature chooses the minimum-energy
one. If the electric field vanishes outside the solid, a charged surface implies a
nonvanishing field inside. This has an extensive energy cost (proportional to the
square of the field times the volume). If the bulk of the solid is centrosymmetric,
the surface charge is zero; or otherwise the surface is metallic.
Quasi-1d “crystalline” systems (i.e. stereoregular polymers) are more in-
teresting. Therein: (1) the energy cost of the field is nonextensive and (2) the
“surface” (i.e. the termination) cannot be “metallic”. Therefore different values
of the surface quantum can be actually realized for the same bulk: we illustrate
this with Figs. 5.4 and 5.5, taken from Ref. [119].
Alternant trans-polyacetylene is a centrosymmetric quasi-1d insulating crys-
tal: its end charges are quantized in units of half an election charge. We con-
sider two different terminations, shown in Fig. 5.4: notice that in both cases
the molecule as a whole is not centrosymmetric, although the bulk is. The end
charges are trivially related to the dipole per unit length (or per monomer).
Fig. 5.5 shows that, in the long system limit, the end charges are either zero
or one, depending on the termination. The structure is either “neutral” or
“charge-transfer”; the end groups are “donor” (NH2 ) and “acceptor” (COOH).
The figure also shows that at finite sizes both molecules are polar; the quanti-
zation is exact in principle in the infinite system limit; in the present case it is
attained at about 10-20 monomers. This length is essentially the exponential
decay length of the one-body density matrix of bulk polyacetylene.
In the polyacetylene case considered so far topology mandates the end
charges to be zero modulo 1 (in units of e). Other cases where instead the
end charge is 1/2 modulo 1 are also considered in Ref. [119]. The actual value
of the end charges depends on the relative ionicity of the end groups vs. the
bulk. Counterintuitively, while the ionicity varies on a continuous scale, the end
charges may only vary by an integer.
56
7
[+-]
[NN]
6
3
Figure 5.5: Quantization of the
2
end charges in polyacetylene.
1 Dipole per monomer as a func-
0 tion of the number of monomers
-1
in the chain, for the two differ-
1 10 100
ent terminations.
5.3.5 Noncrystalline systems: The single-point Berry

phase
The key ingredient for computing the infrared spectrum of amorphous or liquid
systems is the power spectrum of the autocorrelation function of the macro-
scopic polarization hP(t) · P(0)i. Since Car-Parrinello simulations are custom-
arily performed using only k = 0 in the (supercell) BZ, we need to analyze the
single-point version of the Berry-phase formula for polarization, Eq. (5.5).
We consider a simple cubic supercell of side L; in the large L limit the BZ
integral of any function f (k) is approximated as
(2π)3
Z
dk f (k) → f (0). (5.9)
BZ L3
We start with the Berry phase, i.e. with the 1d integral in square parenthesis
in Eq. (5.5):
Z π Z 2π
L L
γz = i dkz hujk |∂kz ujk i = i dkz hujk |∂kz ujk i → −Im log det S(k1 , k2 ),
π
−L 0
(5.10)
where k1 = (0, 0, 0) and k2 = (0, 0, 2π/L). In Eq. (5.10) we have used the
discretized Berry phase, Eq. (3.46), with only one factor in the matrix product.
2πz
Then, as in Sec. 3.9, we notice that |ujk2 i = e−i L |ujk1 i: therefore the overlap
matrix in Eq. (5.10) becomes
2πz
Sjj ′ (k1 , k2 ) = huj |e−i L |uj ′ i, (5.11)
where all the orbitals |uj i = |ψj i are evaluated at k = 0. We then approximate
even the remaining integrals in Eq. (5.5) with a single point. At any time during
the simulation the electronic term in the polarization is thus
e e
Pz(el) (t) = − γz = Im log det S. (5.12)
πL2 πL2
The nuclear (or core) contribution has a very simple form. If zm is the instan-
taneous z coordinate of the m-th nucleus, and eZm the corresponding charge,
the total polarization is
e e X
Pz (t) = 2
Im log det S + 3 Zm z m . (5.13)
πL L m
57
This the expression currently used in computing power spectra and infrared
spectra [121], and, more generally, whenever a single k point is used in the
first-principle simulations [122, 114].
The polarization quantum in Eq. (5.13) is e/L2 , which vanishes in the L → ∞
limit. This does not make any problem, and in fact Eq. (5.12) is routinely used
for evaluating polarization differences in noncrystalline materials. The key point
is that the L → ∞ limit is not actually needed; for an accurate description of
a given material, it is enough to assume a finite L, actually larger than the
relevant correlation lengths in the material. For any given length, the quantum
e/L2 sets an upper limit to the magnitude of a polarization difference accessible
via the Berry phase. The larger are the correlation lengths, the smaller is the
accessible ∆P. This is no problem at all in practice, either when evaluating
static derivatives by numerical differentiation, such as e.g. in Ref. [122, 114],
or when performing Car-Parrinello simulations [121]. In the latter case ∆t is
a Car-Parrinello time step (a few a.u.), during which the polarization varies
by a tiny amount, much smaller than the quantum (the typical size of a large
simulation cell nowadays is L ≃ 50 bohr). Whenever needed, the drawback may
be overcome by splitting ∆t into several smaller intervals.
5.4 Correlated wavefunctions

The treatment given so far assumes an independent-particle scheme, where po-
larization is evaluated as a Berry phase of one-electron orbitals, typically the
KS ones. Shortly after the appearance of the King-Smith and Vanderbilt paper
[94], Ortiz and Martin [95] provided the many-body generalization of the theory,
where polarization is expressed as a Berry phase of the many-body wavefunc-
tion.
A subsequent development [96] provides a unified treatment of macroscopic
polarization, dealing on the same footing with either independent-electron or
correlated systems, and with either crystalline or disordered systems.
5.4.1 Single-point Berry phase again

Suppose we have N electrons in a cubic box of volume L3 , with a many-body
Hamiltonian
N
1 X
Ĥ = |pi |2 + V̂ , (5.14)
2me i=1
and eigenfunctions |Ψn i, normalized in the hypercube of volume L3N . In

Eq. (5.14) the potential V̂ includes one-body and two-body (electron-electron)
contributions. As usual in condensed-matter theory, we adopt periodic Born-
von-Kàrmàn boundary conditions over each electron coordinate ri indepen-
dently, whose Cartesian components ri,α are then equivalent to the angles
2πri,α /L. The potential V̂ enjoys the same periodicity, which implies that the
electric field averages to zero over the sample.
The formula for correlated wavefunctions is quite similar to Eq. (5.12), which
in fact is a special case of the latter. The starting ingredient is the single-point
Berry phase
2π 2π
P P
γz = Im log hΨ0 |ei L i zi
|Ψ0 i = −Im log hΨ0 |e−i L i zi
|Ψ0 i, (5.15)
58
and the polarization formula obtains by replacing γz in Eq. (5.13), or equiva-
lently
2π
(det S)2 → Im log hΨ0 |e−i L i zi |Ψ0 i.
P
(5.16)
e 2π
Pz(el) = Im log hΨ0 |e−i L i zi |Ψ0 i.
P
(5.17)
2πL2
The proof of Eq. (5.17) is provided in the original paper [96], as well as in some
review papers [3, 6, 7, 8].
Here we content ourselves to prove that Eq. (5.17) coincides with Eq. (5.12)
in the special case where |Ψ0 i is the Slater determinant of the (doubly occu-
pied) k = 0 orbitals |uj i = |ψj i. The key observation is that the many-body
wavefunction
2π
|Ψ̃0 i = e−i L i zi |Ψ0 i
P
(5.18)
2πz
is the Slater determinant built from the orbitals |ũj i = e−i L |uj i, hence
2π
P
hΨ0 |e−i L i zi
|Ψ0 i = hΨ0 |Ψ̃0 i = ( dethuj |ũj ′ i )2 = (det S)2 , (5.19)
where the second power owes to double orbital occupancy.
5.4.2 Kohn-Sham polarization vs. real polarization

All of the independent-electron formulas discussed in Sec. 5.3 are exact for
noninteracting electrons, but the obvious aim is to implement them with KS or-
bitals, in a given density-functional theory (DFT) framework. Since macroscopic
polarization applies to insulators only, we stress that we mean “KS insulator”
throughout: that is, we assume that the KS spectrum is gapped. In the class of
“simple” (i.e. computationally friendly) materials a genuine insulator is also a
KS insulator, although pathological cases (computationally unfriendly) do exist.
Having specified this, the key issue is then: Does the KS polarization coincide
with the physical many-body one? The answer is subtle, and is different whether
one chooses either “open” boundary conditions, as appropriate for molecules
and clusters, or periodic boundary conditions (Born-von Kármán), as invariably
done in the present Notes.
Within open boundary conditions the KS orbitals vanish at infinity, as well
as the charge density of the sample. P is then the first moment of the charge
density, divided by the sample volume. The basic tenet of DFT is that the
microscopic density of the fictitious noninteracting KS system coincides with the
density of the interacting system. Therefore the exact P coincides by definition
with the one obtained from the KS orbitals.
Matters are quite different within periodic boundary conditions: we have
seen above that P is not a function of the charge density, hence the value of
P obtained from the KS orbitals, in general, is not the correct many-body P.
This was first shown in 1995 by Gonze, Ghosez, and Godby [123], and later
discussed by several authors. A complete account of the issue can be found in
Ref. [6]. Here we just mention that the exact P is provided by Eq. (5.17), while
the KS P is provided by Eq. (5.12), where the KS orbitals enter Eq. (5.11); both
expressions are to be evaluated in the large-L limit. The two expressions are
clearly different whenever the ground wave function is not a Slater determinant.
Therefore P cannot be exactly expressed within DFT, but the exact func-
tional is obviously inaccessible, and even sometimes pathological. The practical
59
issue is whether the current popular functionals provide an accurate approxi-
mation to the experimental values of P in a large class of materials.
A vast first-principle literature has accumulated over the years by either
linear-response theory [124]—not reviewed here—or by the modern theory. The
errors are typically of the order of 10-20% on permittivity, and much less on most
other properties (infrared spectra, piezoelectricity, ferroelectricity) for many
different materials. It is unclear which part of the error is to be attributed to
DFT per se, and which part is to be attributed to the approximations to DFT.
60
Chapter 6
Quantum metric and the theory of the

insulating state
6.1 Nongeometrical theories of the insulating

state
The standard textbook approach to the insulating state of matter is based on
band theory. Following Bloch’s theorem [125] in 1928, the main result is due
to Wilson in 1931 [126]. The single-particle spectrum of a lattice-periodical
Hamiltonian is in general gapped, and the electron count determines where the
Fermi level lies. If it crosses a band one has a conductor: an applied electric field
induces free acceleration of the electrons (at T = 0 in absence of dissipation).
If the Fermi level lies instead in a gap, one has an insulator: in presence of
a field the electronic system polarizes, but no steady-state current flows for
T → 0. This very successful theory explains the insulating/conducting behavior
of most common materials across the periodic table, for which band structure
calculations became soon available.
At the root of band theory are two basic assumptions: the electrons are
noninteracting (in a mean-field sense), and the solid is crystalline. By the early
1960s, however, it became clear that there are solids where these two assump-
tions are very far from the truth, and where the insulating behavior is due to
completely different mechanisms. The works of Mott in 1949 [127] and of An-
derson in 1958 [128] opened new avenues in condensed matter physics. In the
materials which we now call Mott insulators the insulating behavior is due to
electron correlation [129], while in those called Anderson insulators it is due to
lattice disorder [130].
In a milestone paper appeared in 1964 Kohn [131] provided a more compre-
hensive characterization of the insulating state of matter, which encompasses
band insulators, Mott insulators, Anderson insulators, and eventually any kind
of insulating material. According to Kohn, the electrons in the insulating state
satisfy a many-electron localization condition [132]. This kind of localization
must be defined in a subtle way given that, for instance, the Hamiltonian eigen-
states in a band insulator are obviously not localized. According to the original
Kohn’s formulation, the insulating behavior arises whenever the ground-state
wavefunction of an extended system breaks up into a sum of contributions which
are localized in essentially disconnected regions of the many-electron configura-
61
tion space.
Kohn’s theory remained little visited for many years [133] until the 1990s,
when a breakthrough occurred in electronic structure theory: the modern theory
of polarization. Inspired by the fact that electrical polarization discriminates
qualitatively between insulators and metals, Resta and Sorella [134] in 1999
provided a definition of many-electron localization rather different from Kohn’s,
deeply rooted in the theory of polarization, and therefore based on geometrical
concepts. Their program was completed soon after by Souza, Wilkens and
Martin [135] (hereafter quoted as SWM), thus providing the foundations of the
modern theory of the insulating state. An early review paper appeared in 2002
[4], and a more recent one in 2011 [9]; the latter is at the root of the present
Chapter.
6.2 Metric-curvature tensor

As in Sec. 5.4 we address an interacting N -electron system, whose most general
Hamiltonian we write, in the Schrödinger representation and in Gaussian units,
as
N
1 X e
Ĥ(κ) = |pi + A(ri ) + ~κ|2 + V̂ . (6.1)
2me i=1 c
Equation (6.1) is exact in the nonrelativistic, infinite-nuclear-mass limit. With

respect to Eq. (5.14), the velocity is augmented with two terms: A(r) is a vector
potential of magnetic origin, and κ, having the dimensions of an inverse length,
is a “flux” or “twist” of the same kind as the single-particle flux thoroughly
discussed in Sec. 1.6.
In Sec. 5.4 we adopted periodic Born-von-Kàrmàn boundary conditions
(PBCs) over a cubic supercell of side L, and the eigenfunctions |Ψn i were nor-
malized in the hypercube of volume L3N . In this Chapter, instead, it is expedi-
ent to consider in parallel both “open” boundary conditions (OBCs) and PBCs.
The former are appropriate to molecular physics, and require that the many-
electron wavefunction of a bound state is square-integrable over the whole co-
ordinate space R3N . PBCs are instead appropriate for extended systems, either
crystalline or disordered, either independent-electron or correlated. It is worth
emphasizing that the choice of boundary conditions (either OBCs or PBCs)
corresponds to choosing the Hilbert space, which in turn affects profoundly the
geometrical properties discussed in Ch. 3.
6.2.1 Open boundary conditions

The case of OBCs is by far the simplest. We write for the sake of simplicity
the ground state of Eq. (6.1) at κ = 0 as |Ψ0 i ≡ |Ψ0 (0)i, and we define the
many-body position operator as
N
X
r̂ = ri . (6.2)
i=1
As in Sec. 1.6.2 the flux is easily “gauged away”: the state e−iκ·r̂ |Ψ0 i coincides
with the ground eigenstate |Ψ0 (κ)i of the twisted Hamiltonian, Eq. (6.1). It
62
is legitimate to multiply this eigenvector by any κ-dependent (and position-
independent) phase factor; our choice is then
|Ψ0 (κ)i = e−iκ·(r̂−d) |Ψ0 i, (6.3)
where d = hΨ0 |r̂|Ψ0 i is the electronic dipole of the molecular system. It follows
that the κ-derivative needed in our metric-curvature tensor, Eq. (3.24) is
|∇κ Ψ0 i = −i(r̂ − d)|Ψ0 i = −iQ̂(0) r̂|Ψ0 i, (6.4)
Exploiting the idempotency of the projector, the tensor Fαβ , evaluated at κ = 0,

is
Fαβ (0) = hΨ0 |r̂α Q̂(0)r̂β |Ψ0 i

= hΨ0 |r̂α r̂β |Ψ0 i − hΨ0 |r̂α |Ψ0 ihΨ0 |r̂β |Ψ0 i, (6.5)
where Q̂ is the projector over the unoccupied many-body eigenstates. F (0)

clearly a real symmetric tensor, hence the Berry curvature vanishes within
OBCs, and the metric gαβ (0) coincides with Fαβ (0). It will be shown that
within PBCs, instead, the tensor Fαβ (0) may acquire a nonvanishing imaginary
part.
Clearly, the metric tensor in Eq. (6.5) is the cumulant second moment of the
position operator, or equivalently the ground state fluctuation of the dipole of
the molecular system; this quantity is extensive (scales like N in macroscopically
homogenous systems). We anticipate that Fαβ (0)/N discriminates, in the large
N -limit, between insulators and metals.
6.2.2 Periodic boundary conditions

First of all, a key observation about the position operator is in order. The simple
multiplicative operator r̂, as defined in Eq. (6.2), is “forbidden” within PBCs:
in fact it maps any periodic wavefunction |Ψi into the nonperiodic function
r̂|Ψi. In other words, it maps a function which belongs to the Hilbert space
to a function outside of it [96]. Incidentally, this is the main reason why the
polarization problem has remained unsolved until the early 1990s. In the present
context, formulae like Eq. (6.5) are ill defined and absurd within PBCs.
Within PBC the state e−iκ·r̂ |Ψ0 i is not an eigenstate of Eq. (6.1), except
for the discrete κ values
2π
κ m1 m2 m3 = (m1 e1 + m2 e2 + m3 e3 ), (6.6)
L
where mα ∈ Z, and eα are the Cartesian versors. For fractional values of κ—i.e.
for values different from those in Eq. (6.6)—the κ-dependence of the eigenvalues
and eigenvectors of Eq. (6.1) is nontrivial. We have thoroughly discussed the
single-particle version of this feature in Sec. 1.6.3.
If |Ψ0 (κ)i is the genuine ground eigenstate of Eq. (6.1) within PBCs, then
the auxiliary function |Ψ̃0 (κ)i = eiκ·r̂ |Ψ0 (κ)i is a solution of Ĥ(0), but ful-
fills quasi-periodic boundary conditions: at any two opposite faces of the cube
the wavefunction differs by a κ-dependent phase factor. In other words the
problem can be formulated in two equivalent ways: either the Hamiltonian is
κ-dependent, as in Eq. (6.1), and the boundary conditions are κ-independent;
63
or the Hamiltonian is κ-independent but the boundary conditions are “twisted”
in a κ-dependent way.
Within PBCs the tensor Fαβ (0) cannot be simplified—as e.g. in Eq. (6.5)—
and must be therefore addressed in its original form by actually evaluating the
κ derivatives:
Fαβ (κ) = h∂α Ψ0 (κ)|Q̂(κ)|∂β Ψ0 (κ)i. (6.7)
As within OBCs, even within PBCs this tensor is extensive. In the thermody-
namic limit (i.e N → ∞, keeping N/L3 constant), the well defined quantity is
Fαβ (0)/N . We also observe that for time-reversal symmetric systems Fαβ (0) is
real symmetric, and coincides therefore with the metric tensor gαβ (0).
6.2.3 Sum over states again

We express hrα rβ ic using the sum-over-states formula, Eq. (3.28) :
1 X′ hΨ0 |∂α Ĥ(0)|Ψn ihΨn |∂β Ĥ(0)|Ψ0 i

hrα rβ ic = . (6.8)
N (E0 − En )2
n6=0
The κ-derivative of the Hamiltonian of Eq. (6.1) is

N
~ Xh e i
∇κ Ĥ(0) = pi + A(ri ) , (6.9)
me i=1 c
where the rhs is nothing else than ~ times the velocity operator v̂; Eq. (6.8)
becomes then
1 X′ hΨ0 |v̂α |Ψn ihΨn |v̂β |Ψ0 i
hrα rβ ic =
~2 N (E0 − En )2
n6=0
1 X′ hΨ0 |v̂α |Ψn ihΨn |v̂β |Ψ0 i
= 2 , (6.10)
N ω0n
n6=0
where ω0n = (En − E0 )/~.

The velocity operator is also commonly expressed as v̂ = i[Ĥ(0), r̂]/~, but
it is worth emphasizing that while the position r̂ is well defined within OBCs
and ill defined within PBCs, the velocity v̂ is well defined in both cases.
The basic sum-over states formula, Eq. (6.10), applies therefore to both
OBCs and PBCs. In general, it is not very useful on practical grounds, since it
would require the evaluation of slowly convergent sums. Nonetheless, Eq. (6.10)
is instead essential to gather understanding into the physical meaning of Fαβ ,
as will be shown below.
6.3 Geometrical theory of the insulating state

6.3.1 Fundamentals
The modern formulation of the theory of the insulating state is based on a
localization tensor (squared localization length in 1d), first introduced by Resta
and Sorella in 1999 [134]. This work was followed soon afterwards by SWM
64
[135], where additional results are found, most notably relating localization to
conductivity. Since then, most authors (including the present one) have adopted
the notation hrα rβ ic for the localization tensor, where “c” stays for cumulant.
The formulation within OBCs, using the language and the notations familiar in
quantum chemistry, dates since 2006 [111].
The tensor hrα rβ ic has the dimensions of a squared length; it is an inten-
sive quantity that characterizes the ground-state many-body wavefunction as a
whole. Its key virtue is that it discriminates between insulators and metals: it is
finite in the former case and divergent (in the large-system limit) in the latter.
This is the main message of the present Chapter (and of the modern theory of
the insulating state): we are going to prove it below, Sect. 6.4
Several expressions for the localization tensor have been given in the lit-
erature, all of them equivalent; here we define the localization tensor as the
intensive quantity
hrα rβ ic = Fαβ (0)/N, (6.11)
where the thermodynamic limit is understood. A glance at the OBCs expression,
Eq. (6.5), shows the cumulant nature of F and explains the reason for the
notation, which we adopt within PBCs as well.
Until 2005 the theory of the insulating state implicitly addressed time-
reversal invariant systems only, where the F tensor is real symmetric, and
coincides with the metric: in such systems therefore
hrα rβ ic = gαβ (0)/N. (6.12)
It was found in 2005 [136] that—in absence of time-reversal symmetry and

within PBCs—the tensor hrα rβ ic is naturally endowed with an antisymmetric
imaginary part, whose physical meaning is also outstanding. This is discussed
in Sects. 7.1 and 7.2.
6.3.2 Linear response

So far, we have only discussed ground-state properties of our N -electron sys-
tem. Suppose now that it is subject to a small time-dependent perturbation
contributing to the Hamiltonian the term:
Z ∞
1 1
δ Ĥ(t) = dω f (ω) (Âe−iωt + Â† eiωt ), (6.13)
2π −∞ 2
where Â determines the “shape” of the perturbation and f its amplitude. In

order to get an Hermitian δ Ĥ, we assume f (ω) = f (−ω). We wish to measure
the response to such perturbation by means of the expectation value of some
observable B̂, i.e.:
δhB̂i = hΨ̃|B̂|Ψ̃i − hΨ0 |B̂|Ψ0 i, (6.14)
where Ψ̃ = Ψ0 + δΨ(t) is the perturbed time-evolved ground state. If we limit
ourselves to study terms which are linear in the response, it is enough to consider
the single oscillatory perturbation:
1
Ĥ ′ (ω) = (Âe−iωt + Â† eiωt ), (6.15)
2
65
whose response can be written, using the compact notations due to Zubarev
[67, 137, 138], as:
1
δhB̂i = ( hhB̂|Âiiω e−iωt + hhB̂|Âii−ω eiωt ). (6.16)
2
The quantity hhB̂|Âiiω is by definition the linear response induced by the per-
turbation Â at frequency ω on the expectation value hB̂i. Straightforward
first-order perturbation theory provides its explicit expression as:
1 X′ hΨ0 |B̂|Ψn ihΨn |Â|Ψ0 i

hhB̂|Âiiω = lim
~ η→0+ ω − ω0n + iη
n6=0
!
hΨ0 |Â|Ψn ihΨn |B̂|Ψ0 i
− , (6.17)
ω + ω0n + iη
where ω0n are the excitation frequencies of the unperturbed system, and the
positive infinitesimal η ensures causality. Expressions of the kind of Eq. (6.17)
go under the name of Kubo formulae.
6.3.3 Conductivity
The conductivity tensor σαβ (ω) measures the current linearly induced by an
electric field: jα = σαβ Eβ . We therefore identify Â with the potential of an
electric field along β, i.e. Â = eE rˆβ , and B̂ with the current operator −ev̂α /L3 .
An important detail must be stressed at this point. The macroscopic field
inside the sample includes by definition screening effects due to the electronic
system, while the perturbation δ Ĥ entering Eq. (6.1)—via the Â operator—is
the “bare”, or unscreened one. This point will be discussed below (Sect. 6.3.5);
for the time being we simply identify screened and unscreened fields.
The Kubo formula for conductivity is therefore
e2
σαβ (ω) = − hhv̂α |r̂β iiω ; (6.18)
L3
this is correct within OBCs, but meaningless within PBCs, owing to the explicit
presence of the position operator. This, however, makes no harm, since only
its off-diagonal matrix elements are required: see Eq. (6.17). As usual, we may
exploit the identity hΨ0 |r̂|Ψn i = ihΨ0 |v̂|Ψn i/ω0n . The Kubo formula becomes
then
ie2 X′ 1 hΨ0 |v̂α |Ψn ihΨn |v̂β |Ψ0 i
σαβ (ω) = lim
~L3 η→0+ ω0n ω − ω0n + iη
n6=0

hΨ0 |v̂β |Ψn ihΨn |v̂α |Ψ0 i
+ . (6.19)
ω + ω0n + iη
We introduce a compact notation for the real and imaginary parts of the
numerators in Eq. (6.19), i.e.
Rn,αβ = Re hΨ0 |v̂α |Ψn ihΨn |v̂β |Ψ0 i, (6.20)

In,αβ = Im hΨ0 |v̂α |Ψn ihΨn |v̂β |Ψ0 i, (6.21)
66
which are symmetric and antisymmetric, respectively. Using then
1 1
lim = P − iπδ(x), (6.22)
F →0+ x + iη x
and omitting the principal part, we separate for ω > 0 the symmetric and
antisymmetric parts in the conductivity tensor as
(+) πe2 X′ Rn,αβ

Re σαβ (ω) = δ(ω − ω0n )
~L3 ω0n
n6=0
(−) 2e2 X′ In,αβ

Re σαβ (ω) = = 2 − ω2 . (6.23)
~L3 ω0n
n6=0
6.3.4 Sum rules

At this point, we are ready to compare with the sum-over-states formulae for
the tensor hrα rβ ic . In the present notations, we rewrite Eq. (6.10) as
1 X ′ Rn,αβ
Re hrα rβ ic = 2
N ω0n
n6=0
1 X ′ In,αβ
Im hrα rβ ic = 2 . (6.24)
N ω0n
n6=0
A glance at Eq. (6.23) shows that

Z ∞
~L3 dω (+)
Re hrα rβ ic = 2
Re σαβ (ω) (6.25)
πe N 0 ω
~L3 (−)
Im hrα rβ ic = Re σαβ (0). (6.26)
2e2 N
Eq. (6.25) has been arrived at by SWM in 2000 [135], and Eq. (6.26) by Resta
in 2005 [136]. First of all, these identities show that hrα rβ ic , defined here as a
basic geometric feature, is indeed a measurable quantity (whenever it does not
diverge).
As emphasized throughout this work hrα rβ ic is a ground-state property,
while the rhs of Eqs. (6.25) and (6.26) are properties of the system excitations,
owing to the Kubo formula. Indeed, both Eqs. (6.25) and (6.26) look like the
zero-temperature limit of a fluctuation-dissipation theorem, several forms of
which are known in statistical physics [139, 140]: in the lhs we have a ground-
state fluctuation—see in particular Eq. (6.5)—while the ingredient of the rhs is
conductivity (dissipation).
6.3.5 Screened vs. unscreened field

The Kubo formula for conductivity has been obtained identifying the Â operator
with E rˆβ , where E is the macroscopic field inside the sample. In general this
is not quite correct, since instead Â = eE0 r̂β , where E0 is the “bare” field, i.e.
the field that would be present inside the sample in absence of screening. The
latter originates from the two-body (electron-electron) terms in the potential V̂
67
entering Schrödinger equation. The relationship between E and E0 is not a bulk
property, and depends on the shape of the sample. Alternatively, it depends on
the boundary conditions assumed for integrating Poisson equation: we refer to
Ref. [8] for a thorough discussion. Whenever E = 6 E0 , the sum rule in Eq. (6.25)
must be modified.
The ground-state fluctuations, as e.g. Eq. (6.5), are fluctuations of the
macroscopic polarization, which in a finite sample induce a surface charge at
the boundary. This in turn generates a depolarizing field, which counteracts
polarization. Therefore the localization tensor hrα rβ ic depends on the sample
shape, or equivalently on the boundary conditions assumed when taking the
thermodynamic limit. The choice of PBCs in Eq. (6.1), however, implies E = E0
[141]; hence Eqs. (6.25) and (6.26) are correct as they stand within PBCs. This
no longer holds within OBCs: in this case Eq. (6.25) needs to be modified, while
Im hrα rβ ic = 0.
Ideally the equality E = E0 corresponds to choosing a sample in the form
of a slab, and to addressing the component of the fluctuation tensor hrα rβ ic
parallel to the slab [8]; the thermodynamic limit amounts then to the infinite
slab thickness. Owing to the long range of Coulomb interaction the order of
the limits (first a slab, then its infinite thickness) is crucial. For instance, if
the limit is taken instead by considering spherical clusters of increasing radius,
the SWM fluctuation-dissipation sum rule, Eq. (6.25), assumes a different form:
this is discussed in Refs. [111, 141]. The explicit form of the generalized sum
rule is given therein.
Last but not least, the effect leading to E = 6 E0 within OBCs is a pure
correlation effect. It originates from explicitly correlated wavefunctions, and
does not occur within mean-field theories (Hartree-Fock and Kohn-Sham) [111,
141]. Within such theories, therefore, the sum rules hold in the simple form
of Eqs. (6.25) and (6.26); the conductivity therein is the independent-particle
conductivity (“uncoupled” response in quantum-chemistry jargon).
6.4 Localization in the insulating state

The basic tenet of the modern theory of the insulating state is that the localiza-
tion tensor hrα rβ ic is the ground-state property which sharply discriminates—
in the spirit of Kohn’s seminal work [131, 132]—between insulators and metals.
The real part of hrα rβ ic remains finite in the thermodynamic limit in any insu-
lator, while it diverges in any metal.
The theory is very general, and has found applications to various different
kinds of insulators: band insulators [142, 143, 144, 145]; correlated (i.e. Mott)
insulators, either by means of Hubbard-like model Hamiltonians [134, 146] or
realistic ones [147, 148]; and Anderson insulators [149]. Some of these applica-
tions are reviewed in Sect. 6.6. The cases of Chern insulators, quantum Hall
insulators, and topological insulators are discussed in Chap. 7.
The ultimate proof of the key property of Re hrα rβ ic is based on the SWM
sum rule, Eq. (6.25). Since the tensor is real symmetric, it is enough to consider
the diagonal elements (over its principal axes)
Z ∞
~L3 dω
Re hrα rα ic = 2 Re σαα (ω). (6.27)
πe N 0 ω
68
The f -sum rule yields
∞ ωp2 πe2 N
Z
dω Re σαα (ω) = = , (6.28)
0 8 2me L3
where ωp is the plasma frequency. Therefore the integral in Eq. (6.27) always
converges at ∞; its convergence/divergence is dominated by the small-ω behav-
ior of Re σαα (ω).
Suppose first that the spectrum is gapped, i.e. the spacing between the
ground state and the first excited state stays finite in the thermodynamic limit.
If the gap is Eg the conductivity vanishes for ω < Eg /~, and Eq. (6.28) yields
Z ∞
~L3 dω
Re hrα rα ic = 2
Re σαα (ω)
πe N Eg /~ ω
Z ∞
~2 L 3
< dω Re σαα (ω)
πe2 N Eg 0
~2
= . (6.29)
2me Eg
This inequality is due to SWM and clearly proves that Re hrα rβ ic is finite in
any gapped insulator, as e.g. band insulators (considered in more detail in Sect.
6.4.2).
The main message of Kohn’s 1964 paper, however, is that “insulating char-
acteristics are a strict consequence of electronic localization (in an appropriate
sense) and do not require an energy gap”. For any gapless material, the small-ω
behavior of Re σαα (ω) is the result of a competition between numerators and
denominators in the Kubo formula, Eq. (6.19). Since we aim at a continuous
function of ω, the singularities in Eq. (6.23) must be smoothed: this can be done
by keeping the “dissipation” η finite while performing the thermodynamic limit
first [150]. For a band metal the localization tensor diverges (see below). Ac-
cording to SWM, a gapless material is insulating whenever Re σαα (ω) → 0 like
a positive power of ω, and metallic otherwise. The only example of gapless in-
sulator considered so far is a model Anderson insulator in 1d [149]. Simulations
prove indeed that hx2 ic is finite therein (Sect. 6.6.4).
6.4.1 Independent electrons

For noninteracting electrons the potential V̂ in Eq. (6.1) is the sum of identical
PN
one-body terms: V̂ = i=1 V (ri ). The many-electron Hamiltonian is separa-
ble and the exact ground state |Ψ0 (κ)i is a Slater determinant of one-particle
orbitals (doubly occupied in the singlet case). At a mean-field level, the one-
body potential V (r) includes electron-electron interaction in a selfconsistent
way. In the Hartree-Fock (HF) framework the Slater determinant is regarded
as an approximate many-electron wavefunction. Instead, in the DFT frame-
work the orbitals—called Kohn-Sham (KS) orbitals—are auxiliary quantities,
individually devoid of physical meaning. In particular, their Slater determinant
does not coincide with the many-electron wavefunction |Ψ0 (κ)i, as a matter of
principle. Therefore the exact localization tensor does not coincide, at least in
principle, with the one obtained from the Slater determinant of KS orbitals; the
issue is similar to the one discussed above for macroscopic polarization P (Sec.
69
5.4.2). We stress, however, a difference: the exact P coincides with the KS P
within OBCs, and the problem occurs only within PBCs. Instead, the exact
localization tensor differs in principle from the KS one within both OBCs and
PBCs.
So much for the matters of principle. On practical grounds such difference
is routinely disregarded (e.g. when dealing with polarization, magnetization
[5, 6, 8], and more), given that it is not at all clear what is the relative im-
portance of this “intrinsic” error, compared with the errors due to the choice
of the functional itself. We therefore address here either the HF or the KS
wavefunction |Ψ0 (κ)i, having the form of a Slater determinant.
Whenever the wavefunction is a Slater determinant, all ground-state prop-
erties can be explicitly cast in terms of the one-body density matrix
N/2
X
ρ(r, r′ ) = 2P (r, r′ ) = 2 ϕj (r)ϕ∗j (r′ ), (6.30)
j=1
where a singlet ground state is assumed, and ϕj (r) are the occupied one-particle
orbitals (either HF or KS); P (r, r′ ) is the projector over the occupied manifold.
The expression for the localization tensor is easily found within OBCs start-
ing from Eq. (6.5) [111]:
1
Z
hrα rβ ic = drdr′ (r − r′ )α (r − r′ )β |P (r, r′ )|2 . (6.31)
N
If we define the complementary projector
Q(r, r′ ) = δ(r − r′ ) − P (r, r′ ), (6.32)
an equivalent expression is [142]

2
hrα rβ ic = Tr {rα P rβ Q}, (6.33)
N
where “Tr” is the trace over the single-particle Hilbert space (not on Cartesian
indices).
If we consider a cluster, cut out of a crystalline solid, Eq. (6.31) becomes in
the large-N limit
1
Z Z
hrα rβ ic = dr dr′ (r − r′ )α (r − r′ )β |P (r, r′ )|2 , (6.34)
Nc cell all space
where Nc is the number of electrons per crystal cell. According to the discussion
in Sec. 6.3.5, we need not to worry about shape issues in taking the limit; we
also stress that the density matrix, Eq. (6.30), is independent of the boundary
conditions (either OBCs or PBCs) in the large-N limit.
Finally, we notice that the nasty position operator r is ill defined within
PBCs. But only the relative coordinates appear in Eq. (6.34), while instead in
Eq. (6.33) the operator r is “sandwiched” between a P and a Q. The main
feature making this expression viable in the thermodynamic limit is that P rQ
is well defined even within PBCs. More details about this are given below, Sec.
8.2.1.
70
6.4.2 Band insulators and band metals
As observed, Eq. (6.34) holds for a crystalline solids. Therefore the inner integral
on the rhs must converge for a band insulator, and must diverge for a band metal.
This is confirmed by the well known fact that the asymptotic behavior of P is
qualitatively different in insulators and in metals. In the former materials,
in fact, P (r, r′ ) decays exponentially [151, 152, 153, 154] for large values of
r − r′ : therefore the integral converges and the localization tensor is finite.
In conducting materials, instead, P (r, r′ ) decays only polynominially, and the
inner integral diverges. This divergence can be explicitly verified for the simplest
conductor of all, namely, the noninteracting electron gas, whose density matrix
is exactly known in analytic form [142, 155] for 1d, 2d, and 3d. In all these
cases, the inner integral in Eq. (6.34) diverges like the linear dimension of the
system.
It is therefore clear that the localization tensor, when expressed in the form
of Eq. (6.34), measures in a perspicuous way the “nearsightedness”[156] of the
electron distribution. Such measure is qualitatively different in insulators and
in metals.
The one-particle orbitals (either HF or KS) in a crystalline solid have the
Bloch form. We therefore may wish to replace the orbitals in the expression for
P (r, r′ ), Eq. (6.30),
ϕj (r) → ψjk (r) = eik·r ujk (r), (6.35)
where j is the band index and k is the Bloch vector. We stress that PBCs are
at the very root of Bloch theorem. If the orbitals are normalized to one over
the crystal cell of volume Vcell , the ground-state projector in insulating crystals
is
n Z
Vcell X
P (r, r′ ) = dk ψjk (r)ψjk∗
(r′ ) (6.36)
(2π)3 j=1 BZ
n Z
Vcell X ′
= dk eik·(r−r ) ujk (r)u∗jk (r′ ),
(2π)3 j=1 BZ
where n = Nc /2 is the number of occupied bands, and the integral is taken over
the Brillouin zone (equivalently, over the reciprocal cell).
Using Eq. (6.36), the localization tensor in a band insulator becomes (for
double occupancy) proportional to BZ integral of the Bloch metric-curvature
tensor, Eq. (3.47), i.e.
Vcell
Z
hrα rβ ic = dk Fαβ (k)
(2π)3 n BZ
X n n
X
Fαβ (k) = h∂α ujk |∂β ujk i − h∂α ujk |uj ′ k ihuj ′ k |∂β ujk i. (6.37)
j=1 jj ′ =1
The proof is given in Refs. [4, 142], and will not be repeated here.
Within OBCs the localization tensor is always real: this is perspicuous
in Eqs. (6.31) and (6.34). Instead the imaginary part of the PBC hrα rβ ic ,
Eq. (6.37), is the BZ integral of the Bloch Berry curvature, discussed below,
Sec. 7.1
71
6.5 Wannier and hermaphrodite orbitals
The Wannier and hermaphrodite orbitals are only defined for an independent-
electron insulator; as shown below, their theory makes use of expressions such
as Eq. (6.37) and the like. But these orbitals deserve a discussion of their own;
we limit ourselves to the crystalline case.
6.5.1 Wannier orbitals

The unitary transformation which defines the WFs, Eq. (5.6), labeled by band
j and unit cell R, is
Vcell
Z X
|wjR i = 3
dk eik·R |ψjk i ; |ψjk i = eik·R |wjR i. (6.38)
(2π) BZ
R
Bloch orbitals are normalized over the unit cell, while WFs are square-integrable
and normalized over R3 (in 3d). Any WF centered in R is just the translate of
a central-cell WF, i.e.:
Vcell
Z
hr|wjR i = hr − R|wj0 i, |wj0 i = dk |ψjk i. (6.39)
(2π)3 BZ
The density matrix is clearly invariant by any unitary transformation of the

occupied orbitals among themselves. The transformation of Eq. (6.30) to WFs
yields
n Z
Vcell X X
P = 3
dk |ψjk ihψjk | = |wjR ihwjR |. (6.40)
(2π) j=1 BZ
jR
In several circumstances, Eq. (6.40) is the most straightforward “bridge” be-

tween PBCs and OBCs; localized orbitals have a long history within quantum
chemistry.
The WF definition, Eq. (6.38), leaves a large arbitrariness (a.k.a. gauge
freedom), given that: (1) the Bloch orbitals |ψjk i are arbitrary by a k-dependent
phase factor, and (2) the band labeling j is ambiguous in case of band crossings.
It is expedient to fix a criterion for defining the best localized orbitals. Al-
though the criterion is somewhat arbitrary, the orbitals obtained in this way
often carry some intuitive physical meaning, and are in some sense additive: a
large system can be regarded as the sum of its parts. In quantum chemistry, a
very popular criterion is due to Boys [157]: it amounts to minimize the quadratic
spread of the orbitals, averaged over the molecule. The same criterion has been
adopted and implemented for crystalline systems by Marzari and Vanderbilt in
1997 [108], and has subsequently enjoyed a very large popularity [19].
6.5.2 Hermaphrodite orbitals

Hermaphrodite orbitals were first introduced in Ref. [142]. It is expedient to
start with a finite crystallite within OBCs, with N electrons in a singlet ground
state:
N/2
X
ρ=2P =2 |ϕi ihϕi |. (6.41)
i=1
72
The average quadratic spread in the x direction is by definition
N/2
2 X
λ2xx = (hϕi |x2 |ϕi i − hϕi |x|ϕi i2 ). (6.42)
N i=1
We recast this identically as:

2 X X 2 X
λ2xx = hϕi | x ( 1 − |ϕj ihϕj | ) x |ϕi i + |hϕi |x|ϕj i|2 .
N i j
N
i6=j
(6.43)
The first term in Eq. (6.43) is gauge invariant, since we can identically write:
2 2 X
λ2xx = Tr xP x(1 − P ) + |hϕi |x|ϕj i|2 . (6.44)
N N
i6=j
The gauge–invariant term in Eq. (6.44) coincides with the xx element hx2 ic of
the localization tensor hrα rβ ic , Eq. (6.33).
If we look for the orbitals which minimize the average spread in the x di-
rection, the solution, after Eq. (6.44), is provided by those orbitals which diag-
onalize the position operator x, projected over the occupied manifold, i.e. the
operator P xP . Obviously, a set of orthonormal orbitals which diagonalize it
can always be found, since P xP is a Hermitian operator. The quadratic spread
of these orbitals is the minimum and equals then hx2 ic .
If we now switch to the thermodynamic limit for a crystalline system, trans-
lational symmetry implies P (r, r′ ) = P (r+ R, r′ + R). For the sake of simplicity
we limit ourselves to a rectangular lattice, where R = (X, Y, Z), and each of
the X, Y, Z are one-dimensional lattices. It is then obvious that the eigenstates
of P xP can be labelled with a Bloch vector in the yz directions. If we use the
notation χs ky kz (r) for a generic eigenstate of P xP , then it obeys the Bloch
theorem in the form
χs ky kz (x, y + Y, z) = eiky Y χs ky kz (x, y, z), (6.45)
and analogous in the z direction.

We address the simple case of a single occupied band. If χ0 ky kz (x, y, z)
is eigenstate of P xP with eigenvalue x0 , then its lattice-translate χ0 ky kz (x −
X, y, z) is also eigenstate of P xP , with eigenvalue x0 + X: this is reminiscent
of WFs, see Eq. (6.39). The proof is very simple: P (x − X)P = P xP − XP ,
and XP on the occupied manifold is just a constant multiplication by X.
In the many-band case we need—as for WFs—a band index j and a 1d cell
index X. All hermaphrodite orbitals are obtained by lattice translation from
the central cell ones:
hx, y, z|χj X ky kz i = hx − X, y, z|χj 0 ky kz i; (6.46)
a glance at Eq. (6.39) explains the notations and the meaning. Notice that the
hermaphrodite-orbital center x0 does depend on the (ky , kz ) Bloch vector, as
well as on the band index j.
The name “hermaphrodite orbitals” is therefore pretty clear (see Fig 6.1):
they are Bloch-like in the yz direction, and Wannier-like in the x direction.
73
Figure 6.1: Two-masted vessels. Brig: both masts are square rigged. Schooner:
both masts are fore-and-aft rigged (gaff sails and topsails here). Hermaphrodite
brig: the foremast is square rigged, the mainmast is fore-and-aft rigged.
The present construction guarantees that they are maximally localized in the
x direction and their quadratic spread is equal to the xx component of the
(gauge-invariant) localization tensor. Similarly to standard Bloch and Wannier
orbitals, our hermaphrodite orbitals are an orthonormal basis for the occupied
manifold.
6.5.3 Maximally localized Wannier orbitals

For a one-dimensional system the hermaphrodite orbitals, as defined above, are
those localized orbitals which minimize the quadratic spread, a.k.a. MLWFs
(maximally localized WFs). The minimum spread equals the gauge-invariant
cumulant moment, a.k.a. squared localization length hx2 ic .
For a 3d system we are interested in minimizing the spherical quadratic
spread. In general we cannot diagonalize simultaneously P xP , P yP , and P zP ,
and the spherical spread will be strictly larger than the Cartesian trace of the
localization tensor. This is a key feature in the work of Boys [157], and of
Marzari and Vanderbilt well. The relationship between our localization tensor
hrα rβ ic with the gauge-invariant quadratic spread ΩI common in the MLWF
literature [108, 19] is
d
X 1
hrα rα ic = ΩI , (6.47)
α=1
n
where n is the number of occupied bands. As said above, the trace in Eq. (6.47)
is a lower bound (and not a minimum in dimension d > 1) for the spherical
second (cumulant) moment—a.k.a. quadratic spread—of the WFs, averaged
over the sample.
We stress that Eqs. (6.36) and (6.37) make sense only insofar the Fermi level
falls in a gap, in which case hrα rβ ic is always finite. If we vary the Hamiltonian
continuously, allowing the gap to close, then hrα rβ ic diverges; the quadratic
spread of the Wannier functions diverges as well [111].
Our previous findings bear an important additional message concerning Boys
localization. For a cluster of finite size (or for a crystallite), no matter how large,
one can doubtless build localized Boys orbitals within OBCs. But our results
prove that, in the large N limit, the average quadratic spread of these Boys
orbitals diverges whenever the cluster is metallic.
74
6.6 Localization in different kinds of insulators
6.6.1 Small molecules
The modern theory of the insulating state clearly addresses extended sys-
tems, i.e. the N → ∞ limit; indeed it makes little sense to ask whether a
small molecule is insulating or conducting. Nonetheless the concepts of lo-
calized/delocalized electronic states is of the utmost importance in quantum
chemistry as well, notably in relationship to aromaticity.
The tensor hrα rβ ic within OBCs is always real symmetric. If the ground-
state wavefunction is a Slater determinant, then the trace of the tensor at finite
N has the meaning of a lower bound for the quadratic spread of the Boys
localized orbitals, averaged over all the occupied orbitals [111].
The small-N version of the main concepts of the present review (in their
OBCs flavour [111]) has been adopted in quantum chemistry by Àngyàn
[158, 159]. Besides providing HF calculations of hrα rβ ic for a sample of small
molecules, Àngyàn even provides experimental values drawn from compilations
of the dipole oscillation-strength distributions: basically, from Eq. (6.25).
6.6.2 Band insulators

The theory warrants that the localization tensor is finite in any insulator. How-
ever, quantitative calculations for both model tight-binding Hamiltonians and
realistic solids within DFT have been used to illustrate the theory and to iden-
tify trends. For instance one expects much smaller diagonal elements hrα rα ic
for strong (i.e. large-gap) insulators than for weak (small-gap) ones. This is
also suggested by the SWM inequality, Eq. (6.29).
Let us start with a simple tight-binding (a.k.a. Hückel) Hamiltonian in 1d:
[ (−1)j ∆ c†jσ cjσ − t(c†jσ cj+1σ + H.c.) ]

X
Ĥ = (6.48)
jσ
where t > 0 is the first neighbor hopping (β = −t in most chemistry literature)

and H.c. stays for Hermitian conjugate. This toy model schematizes a binary
ionic crystal; the band structure is
p
ǫ(q) = ± ∆2 + 4t2 cos2 qa/2, (6.49)
where a is the lattice constant and q is the Bloch vector. The gap is equal to
2∆; at half filling the system is always insulating except for ∆ = 0. The squared
localization length (within OBCs) is the tight-binding version of Eq. (6.31), i.e.
N
2 a2 X 2
hx ic = Pjj ′ (j − j ′ )2 . (6.50)
4N ′
j,j =1
This is a monothonical function of t/∆; it is easily verified that it vanishes in

the extreme ionic case (t = 0). In the metallic case (∆ = 0) the ground-state
projector has a simple analytical form:
1
Pjj = ; Pjj ′ = 0 for even |j ′ − j| = 2s,
2
(−1)s
Pjj ′ = for odd |j ′ − j| = 2s + 1, (6.51)
π(2s + 1)
75
Figure 6.2: Diagonal element of
the KS localization tensor vs.
the inverse direct gap (theoret-
ical and experimental), for sev-
eral elemental and binary semi-
conductors (from Ref. [142])
The points corresponding to Si
and Ge with the theoretical
gaps are out of scale. From Ref.
[142].
which clearly implies divergence of Eq. (6.50). At any finite N within OBCs
Eq. (6.50) leads to a finite hx2 ic value; however Eq. (6.51) suggests that in the
metallic case hx2 ic diverges linearly with N , in qualitative agreement with the
case of the 1d electron gas. This has been verified by actual simulations, even
when ∆ 6= 0 but the Fermi level is not in the gap [149].
Other simulations [145] have addressed dimerized chains, i.e. ∆ = 0 but
alternant hoppings in Eq. (6.48). While nothing relevant occurs within PBCs,
partly filled end states within OBCs at some fillings are at the root of some
noticeable features.
The first ab-initio study (in 2001) addressed several elemental and binary
cubic semiconductors at the KS level [142]. The tensor is real and isotropic.
The computed hx2 ic (Fig. 6.2) is smaller than 3 bohr2 in all the materials
studied: the ground many-body wavefunction is therefore very localized in this
class of materials. The SWM inequality was also checked, and found to be well
verified using both the theoretical KS gap and the experimental one (the latter
is typically larger).
Other studies have addressed the ferroelectric perovskites in their different
(cubic and noncubic) structures [143], and some model Hamiltonians in 1d and
2d [144].
6.6.3 Correlated (Mott) insulators

Starting from the noninteracting Hamiltonian of Eq. (6.48) and augmenting it
with an on-site repulsive term we get the two-band Hubbard model
[(−1)j ∆ c†jσ cjσ − t(c†jσ cj+1σ + H.c.)] + U

X X
Ĥ = nj↑ nj↓ . (6.52)
jσ j
The explicitly correlated ground-state wavefunction has been found by exact

diagonalization [134], and the corresponding hx2 ic has been computed as a
function of U for fixed t/∆ = 1.75. The results are shown in Fig. 6.3 in
dimensionless units; it turns out that there is only one singular point U = 2.27t,
where hx2 ic diverges. Indeed, it has been verified that at such value the ground-
state becomes degenerate with the first excited singlet state, i.e. the system is
76
Figure 6.3: Squared localiza-
tion length for the Hamiltonian
in Eq. (6.52) at half filling for
t/∆ = 1.75. The system un-
dergoes a quantum phase tran-
sition from band-like insulator
to Mott-like insulator at U/t =
2.27. From Ref. [134].
metallic. The singular point is the fingerprint of a quantum phase transition:

on the left we have a band-like insulator, and on the right a Mott-like insulator.
The two insulating states are qualitatively different; by adopting the modern
jargon, nowadays we could say that they are topologically distinct. The static
ionic charges (on anion and cation) are continuous across the transition, while
the dynamical (Born) effective charge on a given site changes sign [160]. Other
studies of the localization tensor within the same Hubbard model can be found
in Ref. [161].
The transition from a band metal to a Mott insulator has been studied in
a model linear chain of Li atoms by Vetere et al. [147]. At a mean-field level
the infinite chain is obviously metallic at any lattice constant a, since there
is one valence electron per cell. However the mean-field description becomes
inadequate at large a, where the electrons localize and the system becomes a
Mott insulator. If electron correlation is properly accounted for at any a, the
system undergoes a sharp metal-insulator transition at a critical a.
The calculations addressed linear LiN systems (N up to 8) within OBCs,
where the finite size prevents a sharp transition; the tradeoff is that full config-
uration interaction was affordable with 6 atomic orbitals per site (yielding more
than 109 symmetry-adapted Slater determinants). The wavefunction of Vetere
et al. is therefore exempt from any bias insofar as the treatment of correlation
is concerned, although its quality is determined by the basis set. A study of the
longitudinal component hx2 ic of the localization tensor indicates rather clearly
the occurrence of the metal-insulator transition at a ≃ 7 bohr; other indicators
give concordant results [147]. For comparison, the nearest-neighbour distance
in 3d metallic lithium is 5.73 bohr.
More recently, Stella et al. [148] have performed variational quantum Monte
Carlo studies of hydrogen chains, up to 66 atoms, within PBCs. The crossover
between the weakly correlated (band) metallic regime—at short distances—
and the strongly correlated (Mott) insulating regime—at large distances—has
been thoroughly investigated by means of hx2 ic . The Mott transition occurs at
a ≃ 3.5 bohr.
6.6.4 Disordered (Anderson) insulators

We start from the same Hamiltonian as in Eq. (6.48), and we replace the ordered
string (−1)j by a random string of ±1, chosen with equal (and uncorrelated)
probability. This system models a random binary alloy at 50% concentration.
It is well known both from analytical arguments and actual simulations that its
77
Figure 6.4: Density of
states (arbitrary units) for
a model binary alloy in 1d.
The crystalline (band) case
corresponds to the Hamilto-
nian of Eq. (6.48) with ∆ =
0.25 and t = 1. The disor-
dered (Anderson) case cor-
responds to a random choice
of the anion/cation distri-
bution.
spectrum is gapless [130, 162]. The density of states for both the ordered and
disordered systems are shown in Fig. 6.4, and confirm the expected features.
The band structure of Eq. (6.49) yields obviously a gapped density of states; at
the band
√ edges it shows van Hove singularities, which in 1d have the character
of 1/ ǫ divergences. As discussed above, the system is insulating at half fill-
ing and conducting otherwise. The disordered system, instead, is gapless and
nonetheless insulating at any filling. In fact, this model Hamiltonian describes
a paradigmatic Anderson insulator in 1d.
The conventional theory of transport focusses on the nature of the one-
particle orbitals at the Fermi level; in Anderson insulators these are localized,
thus forbidding steady state currents [128]. More than fifty years of literature
have been devoted to investigate Anderson insulators under the most diverse
aspects. [130, 162, 163, 164].
At variance with such wisdom, Ref. [149] addressed this paradigmatic An-
derson insulator from the nonconventional viewpoint of the modern theory of
the insulating state. In the spirit of Kohn’s theory the individual Hamilto-
nian eigenstates become apparently irrelevant, while the focus is on the many-
electron ground state as a whole. The squared localization length hx2 ic has been
computed within OBCs from Eq. (6.50), and found to be finite, as expected.
Nonetheless its value is about 20 times larger than the one for the band insula-
tor, at the same value of the parameters (i.e. ∆ = 0.25, t = 1). This reflects the
fact that the scattering mechanisms are profoundly different: incoherent (An-
derson) versus coherent (band). In the latter case, the Hamiltonian eigenstates
are individually conducting but“locked” by the Pauli principle if the Fermi level
lies in the gap.
78
Chapter 7
Topological invariants and topological

insulators
7.1 Chern invariant

Within OBCs the localization tensor is always real: this is perspicuous in
Eqs. (6.31) and (6.34). Instead the imaginary part of the PBC hrα rβ ic ,
Eq. (6.37), is the BZ integral of the Bloch Berry curvature, Eq. (3.50):
Vcell 2Vcell
Z Z
Im hrα rβ ic = dk Im F αβ (k) = − dk Ω(k)
(2π)3 n BZ (2π)3 n BZ
n
Vcell
Z X
= dk Im h∂α ujk |∂β ujk i. (7.1)
(2π)3 n BZ j=1
A necessary condition for this to be nonzero is the absence of time-reversal

symmetry.
The integral in Eq. (7.1) has (in 3d) the dimensions of an inverse length,
and is quantized in units of π times a reciprocal vector. We call this reciprocal
vector “Chern invariant”; in fact it is a Chern number (Sect. 3.4) in k-space
and in appropriate units. In 2d the BZ integral is dimensionless and simply
related to the Chern number, defined as:
1 1
Z Z X
C1 = dk Ω(k) = − dk Im h∂1 ujk |∂2 ujk i. (7.2)
2π BZ π BZ n
Band insulators where the Chern invariant is nonzero are called “Chern insula-
tors” (normal insulators otherwise). The possible existence of Chern insulators
has been pointed out by Haldane in 1988 [165], by means of a remarkable model
Hamiltonian in 2d, illustrated below (Sect. 7.3.1). We emphasize that the non-
vanishing of the Chern number prevents it prevents the existence of a smooth
periodic gauge, Eq. (3.42), across the whole reciprocal cell (or BZ): if, for in-
stance, Eq. (3.42) is enforced on the reciprocal cell boundary, an “obstruction”
will show up at some point inside. Because of this same reason, exponentially
localized Wannier functions do not exist [166, 167]; this is at variance with nor-
mal insulators where localized Wannier functions exist, and can even be chosen
as exponentially localized [19, 108, 168].
79
Figure 7.1: Discretization
of the reciprocal cell. The
periodic gauge is enforced
on the boundary; otherwise
the gauge is unspecified and
possibly erratic. The Chern
number is the sum of the
Berry phases computed on
the small squares.
7.1.1 Computing the Chern number

The computation of a Chern number proceeds similarly to what presented in
Sec. 4.1.4 for a toy-model Hamiltonian. Suppose we discretize the reciprocal
cell with a regular mesh, as in Fig. 7.1. We start enforcing the periodic gauge
|ujk+G i = e−iG·r |ujk i on the boundary, i.e. for all the couples of boundary
points which are related by a reciprocal lattice vector. At all the interior points
the gauge is chosen by the diagonalization routine and is therefore erratic. The
mesh actually shown in Fig. 7.1 would actually require 8 × 8 = 64 Hamiltonian
diagonalizations.
We recall that the curvature is the Berry phase per unit reciprocal area, and
we compute it on each of the small squares using the four-point discrete formula
γ = −Im log huk1 |uk2 ihuk2 |uk3 ihuk3 |uk4 ihuk4 |uk1 i, (7.3)
which applies to the single-band case. In the many-band case this is replaced
by
γ = −Im log det S(k1 , k2 )S(k2 , k3 )S(k3 , k4 )S(k4 , k1 ), (7.4)
where S is the overlap matrix
Snn′ (ks , ks′ ) = hunks |unks′ i; (7.5)
see Eq. (3.46). One has to choose the “Im log” branch with γ in [−π, π].
The Chern number C1 is the sum of all the γs, covering the whole reciprocal
cell (64 in Fig. 7.1). In our discontinuous approach, the obstruction is not
located at any particular k point, although its effect becomes apparent after all
γs are summed.
7.1.2 Chern number as the curvature for unit area

The Berry curvature Ω(k) has the dimensions of a squared length; as said
above, it has the meaning of Berry phase per unit reciprocal area. Here instead
we consider a possibly disordered system, where the k vector is devoid of any
meaning. For a system which is macroscopically homogeneous, we are going to
define a “global curvature” Ω, which is a number—not a function of k—and is
extensive, scaling like the system size (or like the electron number). The Chern
number is then proportional to our global curvature divided by the system area
in r space.
80
We start rewriting Eq. (7.2) for a square lattice as
Z π/L Z π/L
1
C1 = dkx dky Ω(k), (7.6)
2π −π/L −π/L
where L is the lattice constant. As usual, a noncrystalline system can be dealt

with in a supercell framework, by addressing an artificial crystal with a large
L, actually larger than the relevant correlation length. In this limit the BZ
becomes very small; owing to the mean value theorem
2
1 2π
C1 ≃ Ω(0). (7.7)
2π L
We identify Ω(0) with the (extensive) global curvature Ω; if A = L2 is the
system area, then
Ω
C1 = 2π (7.8)
A
in the thermodynamic limit.
At this point, it looks like we still need to deal with a k vector since, within
the the usual definition of Ω(0), its essential ingredients are k derivatives of the
|unk i evaluated at k = 0. Instead, if we work within our discrete Berryology,
such derivatives are not needed. Incidentally, this shows once more that discrete
Berryology is more general and more powerful than differential Berryology. The
value of the global curvature Ω can be obtained from a single Hamiltonian
diagonalization for a large supercell—either crystalline or disordered—and no
derivative is actually involved.
This bears some relationship to the so-called single-point Berry phase, dis-
cussed in Sec. 5.3.5. Here we follow the approach of Ref. [169]: to give a flavor
of how it works, we start from Fig. 7.1, where one needs 64 independent diag-
onalizations, as said above. One may look at the same system by doubling the
elementary cell in each direction, in which case the number of occupied bands
increases fourfold, and the new reciprocal cell has one quarter of the original
area. Now the calculation of Ω requires 64/4 = 16 independent diagonaliza-
tions; the algorithm of Eq. (7.4) yields an Ω value identical to previous one (up
to computer accuracy), because we are folding the same orbitals in a smaller
reciprocal cell (or BZ). We can go on with the folding, diagonalizing in 64, 16,
4, 1 points. We still get a numerically identical result insofar as no disorder
is introduced. If the supercell size L is large enough, the Ω value computed
via a single Hamiltonian diagonalization and (super)lattice-periodical orbitals
is very close to the thermodynamic limit Ω = C1 A/(2π), Eq. (7.8), with C1
integer [169]. Once the large supercell is adopted, disorder can be introduced at
no extra cost; since C1 is a robust topological invariant, weak disorder cannot
change the Ω value.
What we have presented here is an r space interpretation of the Chern num-
ber, Eq. (7.8), which gets rid of k space and deals with crystalline and disordered
systems on the same footing. Nonetheless, the approach of the present Section
requires a macroscopically homogeneous system and the adoption of PBCs. Is
it possible to overcome these limitations, and address topological order even
for inhomogeneous systems (e.g. heterojuncions) and/or adopting OBCs? The
answer is “yes”; the presentation of this is deferred to Sec. 7.5, after we have in-
troduced the reference computational material (“Haldanium”) for dealing with
this kind of issues.
81
Figure 7.2: Upon dou-
bling the elementary cell,
the reciprocal-cell area is di-
vided by 4, and the number
of occupied bands is multi-
plied by 4.
7.1.3 Chern number and hermaphrodite orbitals

The hermaphrodite orbitals have been defined and discussed in Sec. 6.5.2. Here
we address the simple case of a 2d square lattice of constant a and a single
occupied band. We simplify notations by indicating with χky (x, y) the central-
cell (ky -dependent) orbital; the complete set in the occupied manifold is, after
Eq. (6.46), χm ky (x, y) = χky (x − ma, y), where m ∈ Z and ky ∈ [0, 2π/a). The
inverse transformation yields the Bloch orbitals as
X
ψk (r) = eimkx a χky (x − ma, y) (7.9)
m
X
uk (r) = e−ikx (x−ma) χ̃ky (x − ma, y), χ̃ky (x, y) = e−iky y χky (x, y).
m
The uk orbitals are clearly lattice-periodical and normalized over the unit cell;
their k derivative yields
X
∂1 uk (r) = −i e−ikx (x−ma) (x − ma) χ̃ky (x − ma, y),
m
X ′ ∂
∂2 uk (r) = e−ikx (x−m a) χ̃k (x − m′ a, y),
∂ky y
m′
X ∂
h∂1 uk |∂2 uk i = i e−ikx ma hχ̃m ky |(x − ma)| χ̃0 ky i, (7.10)
m
∂ky
where “h i′′ means r integral in (−∞, ∞) × (0, a). The Berry curvature and the
Chern number are therefore
Ω(k) = −2 Im h∂1 uk |∂2 uk i

X ∂
= −2 Re e−ikx ma hχ̃m ky |(x − ma)| χ̃0 ky i,
m
∂ky
Z 2π/a Z 2π/a X
1 ∂
C1 = − Re dkx dky e−ikx ma hχ̃m ky |(x − ma)| χ̃0 ky i
π 0 0 m
∂ky
2 2π/a ∂
Z
= − dky Re hχ̃ky |x| χ̃k i, (7.11)
a 0 ∂ky y
where |χ̃ky i ≡ |χ̃0 ky i is the central-cell orbital. If we define x0 as its (ky

dependent) center, then
x0 (ky ) = hχ̃ky |x|χ̃ky i,

dx0 ∂
= 2 Re hχ̃ky |x| χ̃k i (7.12)
dky ∂ky y
82
2π/a
1 dx0 1
Z
C1 = − dky = − [ x0 (2π/a) − x0 (0) ]. (7.13)
a 0 dky a
The hermaphrodite orbitals carry therefore a very clear topological signature.
In a topologically trivial 2d insulator their center is periodical in ky ; while
instead in the nontrivial case their center shifts by −C1 a when ky → ky + 2π/a.
More generally, not only the center but even the function χ̃m ky (x, y) enjoys an
analogous property.
As for the k space periodicity of the Bloch functions, Eqs. (7.9) and (7.13)
yield
ψkx+2π/a,ky (r) = ψkx ,ky (r)

ψkx ,ky+2π/a (r) = eiC1 kx a ψkx ,ky (r). (7.14)
this confirms that in the Chern case the gauge cannot be periodical over the
reciprocal cell. The gauge implicit in Eq. (7.9) can be called “cylindrical”:
periodic in kx but not in ky .
7.1.4 Lowest Landau level

We consider a system of noninteracting electrons in a flat substrate potential and
in a normal magnetic field. Although the system is translationally invariant in
the (xy) plane, both the Hamiltonian and the density matrix P are not invariant.
Despite this, the orbitals of e.g. the lowest Landau level (LLL) share many of
the properties of the hermaphrodite orbitals.
If the LLL is fully occupied (i.e. at filling ν = 1) the electron density is uni-
form and equal to n0 = 1/(2πℓ2 ), where ℓ = (~c/eB)1/2 is the magnetic length;
the modulus of the density matrix has the translationally invariant expression
(for single occupancy)
′ 2
/(4ℓ2 )
|P (r, r′ )| = n0 e−|r−r | , (7.15)
and the cumulant second moment is clearly finite. The trace hx2 ic + hy 2 ic of
the localization tensor is in fact equal to ℓ2 , the squared magnetic length, and
the minimum quadratic spread in the x direction is ℓ2 /2.
The LLL orbitals in the Landau gauge are labeled by a one-dimensional wave
vector q ∈ (−∞, ∞) and have the form
ψq (r) = e−iqy ϕ(x − qℓ2 ), (7.16)
where ϕ(x) is the normalized ground eigenfunction of a 1d harmonic oscillator

with frequency ωc = eB/me c
1/4
1 2
/(2ℓ2 )
ϕ(x) = e−x . (7.17)
πℓ2
The quadratic spread of this function is indeed ℓ2 /2; the ψq (r) are therefore
eigenstates of P xP , and are the magnetic analogue of the maximally localized
hermaphrodite orbitals. Notice that in Eq. (7.16) ψq (r) has a plane-wave-like
normalization over y.
83
To make contact
√ with our notations for hermaphrodite orbitals it is enough
to define a = 2π ℓ and
q = ky + 2πm/a, ky ∈ [0, 2π/a), m ∈ Z

1 −i2πmy/a
χ̃mky (x, y) = e ϕ(x − ky ℓ2 − 2πmℓ2 /a)
a
1 −i2πmy/a
= e ϕ(x − ky a2 /(2π) − ma); (7.18)
a
as in the previous Section, the χ̃ orbitals are normalized over (−∞, ∞) × (0, a).
Finally, we verify the behavior of χ̃ vis-a-vis the Chern number. The center
of the central cell orbital is x0 = ky a2 /(2π); if ky is increased by 2π/a, then
x0 increases by a, as it must be: we remind that C1 = −1 in the LLL (sign
conventions are not uniform across the literature).
7.2 Quantum Hall insulators: Correlated wave-

function
The integer quantum Hall (QH) effect has been discussed at length above. The
most perspicuous feature—see Fig. 2.7—is that whenever the 2d electron fluid
is in the QH regime: (i) the transverse resistivity is quantized; and (ii) the
longitudinal resistivity vanishes. The same two features are common to the
fractional QH effect as well. We remind that, while the integer QH effect can
be understood at the independent-particle level, a correlated wave function is
essential in the fractional case. We also remind that in the QH regime the static
longitudinal conductivity vanishes: this is why we speak of “QH insulators”,
both in the integer and fractional cases.
We are going to show that electron localization—as defined in the modern
theory of the insulating state—is the common cause for both the vanishing of
the dc longitudinal conductivity and the quantization of the transverse one.
Therefore to predict whether the dc transverse conductivity is quantized, it is
enough to inspect electron localization in the ground state. This outstanding
result was proved in 2005 [136].
Our system is a square of size L × L, where the presence of a macroscopic B
field forbids simple PBCs; this, however, is no serious problem. We choose the
Landau gauge for the magnetic vector potential in Eq. (6.1), and we impose the
magnetic boundary conditions which are customary in the QH literature [170];
this requires the total flux φ = BL2 to be an integer times φ0 , where φ0 = hc/e
is the flux quantum.
When switching from 3d to 2d, many of the above formulae require obvious
modifications. For an isotropic system, the 2d analogue of Eqs. (6.25) and (6.26)
can be written as
Z ∞
dω πe2 N
Re σ11 (ω) = Re hx2 ic , (7.19)
0 ω ~L2
2e2 N
Re σ12 (0) = Im hxyic . (7.20)
~L2
A glance at the sum-over-states expression for hrα rβ ic , Eq. (6.10), shows that
whenever Re hrα rβ ic converges, then Im hrα rβ ic converges as well. However,
84
we are going to show that the latter may only converge to quantized values in
the present 2d case.
By definition we have
1
Im hxyic = Im h∂1 Ψ0 (0)|∂2 Ψ0 (0)i. (7.21)
N
In the large-L limit, we may write
Im h∂1 Ψ0 (0)|∂2 Ψ0 (0)i (7.22)
2 Z 2π Z 2π
L L L
= Im dκ1 dκ2 h∂1 Ψ0 (κ)|∂2 Ψ0 (κ)i.
2π 0 0
The dimensionless integral is over a closed surface (a torus), given the magnetic
boundary conditions. It is therefore equal to −πC1 , where C1 is the Chern
number of the first class (Sect. 3.4); hence
L2
Im hxyic = − C1 , (7.23)
4πN
and the transverse conductivity, Eq. (7.20) becomes
e2 e2
Re σ12 (0) = − C1 = − C1 . (7.24)
2π~ h
We have thus arrived at the famous Niu-Thouless-Wu formula [170] which holds
for both the integer and fractional QH effect. The original derivation was based
on an analysis of the Green function, under the hypothesis that the system
has a Fermi gap; in the present derivation the presence of a Fermi gap be-
comes apparently irrelevant, since the localization tensor is a pure ground-state
property. Underlying the modern theory of the insulating state, however, is a
fluctuation-dissipation theorem, relating the ground state to the excitations of
the system.
The main message of Ref. [136] (and of the present Section) can be summa-
rized by saying that any 2d insulator displays a quantized transverse conduc-
tance (nonvanishing only in absence of time-reversal symmetry).
An electron fluid is kept in the QH regime by disorder, and analytical imple-
mentations of the present formulae are obviously not possible. Nonetheless, in
order to illustrate how the theory works, it is expedient to consider the academic
case of noninteracting electrons in a flat substrate potential. As shown above, if
the first Landau level is fully occupied the trace hx2 ic + hy 2 ic of the localization
tensor is finite and equal to the square of the magnetic length ℓ = (~c/eB)1/2 .
If the filling is fractional the density cannot be uniform (for noninteracting elec-
trons), and, more important, the cumulant second moment diverges (see Ref.
[136] for details): the system is not insulating, and its transverse conductivity
is not quantized.
7.3 Topological insulators

The QH state of matter, discovered in 1980 and addressed above, provides the
first example of a quantum state which is topologically distinct from all pre-
viously known states of matter. This is the reason why its macroscopic quan-
tization properties are very robust (“topologically protected”), and insensitive
85
Figure 7.3: Four unit cells of crystalline Halda-
nium [165]. Filled (open) circles denote sites with
E0 = −∆ (+∆). Solid lines connecting near-
est neighbors indicate a real hopping amplitude t1 ;
dashed arrows pointing to a second-neighbor site in-
dicates a complex hopping amplitude t2 eiφ . Arrows
indicate sign of the phase φ for second-neighbor hop-
ping.
to small changes in material parameters. The debut of topological concepts in

electronic structure theory dates from two milestone papers: TKNN in 1982
[50] and Niu-Thouless-Wu in 1985 [170]. These papers addressed a 2d electron
fluid (noninteracting and interacting, respectively) in the presence of a macro-
scopic magnetic field: the Hamiltonian, therefore, cannot be lattice periodical.
Topological insulators (TIs) are defined as materials whose ground state is topo-
logically nontrivial in absence of an applied magnetic field.
7.3.1 Haldanium
The archetype of all TIs is the Haldane model Hamiltonian [165] (“Haldanium”):
its hallmark is quantized Hall conductance in absence of a macroscopic magnetic
field. Haldanium is the paradigm for Chern insulators, and has been used as
a workhorse in many simulations, providing invaluable insight into topological
features of the electronic wavefunction [165, 169, 166, 171, 172, 173], as well as
into features of orbital magnetization [174, 175, 169, 176].
The model is comprised of a 2d honeycomb lattice with two tight-binding
sites per primitive cell with site energies ±∆, real first-neighbor hopping t1 , and
complex second-neighbor hopping t2 e±iφ , as shown in Fig. 7.3. Within this two-
band model, one deals with insulators by taking the lowest band as occupied.
The appeal of the model is that the vector potential and the Hamiltonian are
lattice periodical and the single-particle orbitals have the usual Bloch form.
Essentially, the microscopic magnetic field can be thought as staggered (i.e. up
and down in different regions of the cell), but its cell average vanishes.
For a 2d lattice-periodical Hamiltonian the Chern number C1 has been de-
fined before, Eq. (7.2). As a function of the flux parameter φ, this system
undergoes a transition from a normal insulator (C1 = 0) to Chern insulator
Figure 7.4: Chern number C of the

bottom band of the Haldane model
as a function of the parameters φ
and ∆/t2 ; the plot is drawn for
t1 = 1 and t2 = 1/3. At φ = 0
Haldanium is a model for graphene
and hexagonal boron nitride.
86
(|C1 | = 1). Its phase diagram is shown in Fig. 7.4. On the vertical φ = 0 axis
the Hamiltonian is time-reversal invariant and Haldanium becomes a model for
hexagonal boron nitride; the center of the plot (∆ = 0, φ = 0) corresponds
to graphene. Notice also that on the horizontal ∆ = 0 axis Haldanium is a
nonpolar and centrosymmetric insulator (except at φ = 0).
Like in the QH case discussed in Sect. 7.2: (i) the imaginary part of the
localization tensor is related to the Chern number by Eq. (7.23); and (ii) the
trace of the localization tensor is finite. This is confirmed by the simulations of
Ref. [166], where the actual value of ΩI , Eq. (6.47), is computed. We remind
that, despite ΩI being finite, localized Wannier functions (with finite quadratic
spread) do not exist in Chern insulators.
Haldanium and QH insulators display quantized transverse conductivity;
all are localized in the sense of the modern theory of the insulating state. It is
worth pointing out, though, that the decay of the density matrix is qualitatively
different: exponential in Haldanium [166] vs. Gaussian in the (noninteracting)
QH case, Eq. (7.15).
No microscopic realization of a Chern insulator (in absence of a macroscopic
magnetic field) is known to date, although some unpublished work in this direc-
tion was announced at the APS 2012 March meeting. Chern insulators remained
a curiosity of academic interest only for many years. In 2005 it was realized that
the Haldane model is the forebear of a completely new class of topological in-
sulators, the time-reversal symmetric ones [51, 10, 11].
7.3.2 Time-reversal symmetric topological insulators

Although Chern insulators do not exist in nature, a different kind of topological
insulators does exist in nature in both 2d and 3d [10, 11, 52, 53, 54, 55, 56]; their
macroscopic quantization properties are topologically protected, and insensitive
to small changes in material parameters, exactly like in the QH case.
Modern topological insulators are characterized by novel invariants called
Z2 (close relatives of the Chern number in 2d and of the Chern invariant in 3d).
The paradigmatic Z2 topological insulator is the Kane-Mele model Hamiltonian
in 2d [51]. Most important, at variance with Chern insulators, the modern
topological insulators are time-reversal symmetric. As stated in Sec. 4.2.3,
the group Z2 is the additive group of the integers modulo 2: it has only two
elements, which can be labeled as 0 and 1, or as “even” and “odd”. Topological
insulators are Z2 -odd, while a trivial insulator is Z2 -even.
Since spin-orbit interaction plays a major role, the definition of localization
tensor needs to be augmented to explicitly include the spin coordinates; this
seems to be rather straightforward, and is deferred to further publications. After
this is done, the scope of the modern theory of the insulating state would include
topological insulators.
No investigation of this kind exists yet; nonetheless, a closely related inves-
tigation concerning the Wannier functions in a Z2 insulator has appeared in
2011 [177] . The results confirm that the ground state of a topological insula-
tor is localized in the sense of the modern theory of the insulating state: ΩI ,
Eq. (6.47), is in fact finite. Furthermore, at variance with the Chern-insulator
case: (1) localized WFs (with finite quadratic spread) do exist [177], and (2) a
smooth gauge over the whole BZ (with no obstruction) exists [178]. Both local-
ized WFs and the smooth gauge requires breaking the Kramers-pair symmetry
87
in the Z2 -odd case.
7.4 Digression: Bulk-boundary correspondence

As discussed throughout these Notes, geometrical and topological properties of a
given system manifest themselves in reciprocal space. The concepts of reciprocal
space and of k vector are based on PBCs: the sample has no boundaries by
construction. However, for the same physical system, one could consider a
bounded sample within OBCs. In this case there is no k vector to speak of:
the geometrical properties manifest themselves as surface (in 3d) or edge (in
2d) properties. Bulk-boundary correspondence is the hallmark of geometry and
topology in electronic structure.
The archetype of bulk-boundary correspondence is the IQHE. When a
toroidal geometry is adopted there is no boundary, and the fingerprint of non-
trivial topology is the Chern number. If instead the 2d electron fluid in the
quantum Hall regime has the shape of a ribbon, then there are edge states which
carry a chiral current. However, these edge states are a bulk property and are
“topologically protected”. Similar considerations apply to Chern insulators in
2d.
For 2d time-reversal invariant topological insulators the edge states cannot
carry any net current; instead there are counterpropagating spin currents which,
once more, are topologically protected: one speaks then of “quantum spin Hall
effect”. In the 3d case, there are topologically protected surface states, whose
nature is nontrivial.
The bulk-boundary correspondence also occurs in the case of polarization,
where the topologically protected boundary quantity is not current but charge.
Here one has to pay attention to the fact that polarization is the sum of an
electronic and a ionic term: only the former term is expressed as a k integral.
However, the bulk polarization determines the surface charge modulo the “quan-
tum”. In the simple case where the bulk is centrosymmetric, topology requires
the charge per surface charge to be an integer or an half integer. We stress that,
quite often, a polar surface is metallic and topology does not apply.
7.5 Topological order in r space

The Chern number, as well as other topological invariants, labels the manner
in which the electronic ground state is topologically “twisted” or “knotted” in
k-space [10, 11]. But topological order must reflect a peculiar organization of
the electrons even when the concept of k-space does not apply, such as for
inhomogeneous systems (e.g. heterojunctions), as well as for finite systems
within OBCs. In Ref. [173] it has been shown that the Chern number in a
2d insulator can be mapped by means of a “topological marker”, defined in r
space, and which may vary in different regions of the same sample. Notably,
this applies equally well to periodic and open boundary conditions.
We start rewriting Eq. (7.1) in 2d for a single band:
Acell
Z
Im hrα rβ ic = dk Im h∂α ujk |∂β ujk i, (7.25)
(2π)2 BZ
88
whence the Chern number is related to the imaginary part of the localization
tensor as
4π 4π
C1 = − Im hrα rβ ic = − Im Trcell{P xQy}, (7.26)
Acell Acell
where the second equality owes to Eq. (6.33).
At this point it looks like that we can use Eq. (7.26) even for a finite 2d
flake within OBCs, replacing the trace over the cell with the trace over the
whole system, and replacing Acell with the flake area. Instead, we have already
observed in Sec. 6.4.2 that the localization tensor is real within OBCs: this is
confirmed by noticing that
i
Im Tr{P xQy} = −Im Tr{P xP y} = Tr{ [P xP, P yP ] }, (7.27)
2
and that the trace of the commutator is zero, whenever P projects on a finite
manifold.
The key to the apparent paradox is that the thermodynamic limit and the
trace do not commute. In Eq. (7.26) the limit to the infinite system is performed
before taking the trace, while in Eq. (7.27) we are attempting at taking the trace
for a finite system, and eventually performing the limit afterwards.
The solution proposed in Ref. [173] is to address the commutator in
Eq. (7.27) before taking its trace. If we write
Z
Tr{ [P xP, P yP ] } = dr hr| [P xP, P yP ] |ri, (7.28)
then the integrand carries the information about the local topological order. A
“topological marker” is defined as the dimensionless function
C(r) = −2πihr| [P xP, P yP ] |ri; (7.29)
for a crystalline system, Eq. (7.26) guarantees that the cell average of C(r)
coincides with the Chern number. The function C(r) fluctuates over microscopic
dimensions; in the nonperiodc case, the cell average has to be replaced with the
macroscopic average hCimacro , defined as in electrostatics (see e.g. Jackson [89]).
We stress that the microscopic function C(r) itself, before macroscopic averaging,
bears no physical meaning.
The topological marker is validated over different Haldanium samples, by
means of simulations performed on finite flakes within OBCs. The samples are
crystalline, disordered, and inhomogeneous (heterojunctions), in both normal
and Chern regimes. The results are illustrated and commented in Ref [173].
The topological marker described here only addresses the Chern (i.e. Z)
topological order. A marker having similar properties and addressing instead
the Z2 topological order has not be found so far (2013).
89
Chapter 8
Orbital magnetization
Like polarization P, even macroscopic magnetization M is a fundamental con-

cept that all undergraduates learn about in elementary courses [88, 89]. Macro-
scopic magnetization M is the sum of two terms, which are unambiguously
defined in nonrelativistic (and semirelativistic) quantum mechanics: spin mag-
netization M(spin) and orbital magnetization M(orb) . Experimentally, magneto-
mechanical measurements, based on the Einstein-de Haas effect, provide the
two terms separately. For instance, the values of M(spin) and M(orb) for the
three ferromagnetic metals (Fe, Co, and Ni) are accurately known since half
a century [179]. From the viewpoint of the present review, M(spin) is a dull
quantity. Electronic-structure codes routinely compute the spin density, which
is a simple lattice-periodical function. Its cell average (times a trivial factor)
coincides with M(spin) . In other words, a “dipolar density” is unambiguously
identified, while the same does not happen in the orbital case. Throughout
this Chapter we address orbital magnetization, using the symbol M to identify
M(orb) only.
Polarization P and orbital magnetization M share some analogy, but also
some significant differences. For the time being, we emphasize the main analogy:
textbooks define both P and M as the dipole moment of a sample, divided by
the volume V :
1 1
Z Z
d (micro) m
P= = dr rρ (r), M = = dr r × j(micro) (r), (8.1)
V V V 2cV
where ρ(micro) (r) and j(micro) (r) are the microscopic charge and (orbital) current
densities. Contrary to what most textbooks pretend, there is no such thing as a
“dipolar density”: the basic microscopic quantities are ρ(micro) (r) and j(micro) (r)
[180]. If the sample is uniformly polarized/magnetized, then the microscopic
charge/current averages to zero in the bulk of the sample. Owing to the presence
of the unbound operator r, both P and M Eq. (8.1) are dominated by surface
contributions, while instead phenomenologically they are bulk properties.
The modern theory of magnetization dates from 2005-06 [175, 181], and is
still (2013) partly work on progress [181, 182, 183, 169, 176]. A couple of reviews
have appeared [8, 18], while only a few first-principle implementations appeared
so far (2013) [184, 185].
90
8.1 Magnetization and magnetic field
The modern theory of magnetization only addresses the polarization M in a
null macroscopic B field; in this case M can be nonzero only if time-reversal
symmetry is broken in the spatial wavefunction. For instance, in a ferromag-
net the spin-orbit interaction transmits the symmetry breaking from the spin
degrees of freedom to the spatial (orbital) ones.
It must be realized that, insofar as we address an infinite system with no
boundaries, the B field is quite arbitrary, in full analogy with the case of E dis-
cussed in Sec. 5.1. The usual choice invariably performed within all electronic-
structure codes is to impose a lattice-periodical vector potential, i.e. B = 0.
When addressing a finite sample with boundaries, the B field is in principle
measurable inside the material, without reference to what happens at the sample
boundary: in fact, B obtains by averaging over a macroscopic length scale the
microscopic electric field B(micro) (r), which fluctuates at the atomic scale [89].
In a macroscopically homogeneous system the macroscopic field B is constant,
and in crystalline materials it coincides with the cell average of B(micro) (r).
As for the electrical case, discussed in Sec. 5.1, there is no need of addressing
finite samples and external vs. internal fields from a theoretician’s viewpoint.
Nonetheless a brief digression is in order, given that experiments are performed
over finite samples, often in external fields. Suppose a finite macroscopic sample
is inserted in a constant external field B(ext) : the microscopic field B(micro)(r)
coincides with B(ext) far away from the sample, while it is different inside be-
cause of screening effects. If we choose an homogeneous sample of ellipsoidal
shape, then the macroscopic average of B(micro) (r), i.e. the macroscopic screened
field B, is constant in the bulk of the sample. The shape effects are embedded in
the demagnetization coefficients [88]: the simplest case is the extremely oblate
ellipsoid, i.e. a slab of a macroscopically homogeneous material; more details
are given in Ref. [8]. For the slab geometry in a vanishing external field B(ext)
the internal field B vanishes when M is normal to the slab (longitudinal polar-
ization), while B = 4πM is the demagnetization field when M is parallel to the
slab (transverse polarization): see Fig. 8.1. Notice that this is the opposite of
what happens in the electrical case (Fig. 5.1).
Figure 8.1: Macroscopic magnetization M in a slab normal to z, for a vanishing

external field B(ext) . Left: When M is normal to the slab, no demagnetizing
field and no surface current is present. Right: When M is parallel to the slab,
a demagnetizing field B = 4πM is present inside the slab, and dissipationless
currents Ksurface = c M × n flow at the surfaces.
91
8.2 Magnetization “itself”
In Eq. (8.1) we have emphasized the main analogy between P and M; here,
instead, we anticipate the main difference. In the case of P—as stressed in Sec.
5.2—the modern theory addresses the difference in polarization ∆P between
two states of the material that can be connected by an adiabatic switching
process. As for P “itself”, the modern theory of polarization states that the
bulk electron distribution determines P only modulo a “quantum”, whose value
depends on the boundary [94, 118] (Sec. 5.3.2, and also Figs. 5.4 and 5.5). The
modern theory of magnetization, instead, addresses M “itself” directly, and is
not affected by any quantum indeterminacy.
Another key difference is that, while P makes sense only for insulators at zero
temperature, M is well defined even for metals and even for finite temperature.
In the present Notes we limit ourselves to discuss the zero-T insulating case,
both normal and Chern. We refer to the original papers [175, 181, 182], as well
as to the review of Ref. [8] for the metallic and finite-T cases.
The main concepts are more clearly formulated in the simple 2d case: elec-
trons in the xy plane and magnetization M along z; the 3d theory is not concep-
tually different, although it requires a more complex algebra. We only address
a system of spinless independent electrons. If the chemical potential µ is in the
bulk gap, the value of M is µ-independent in the normal case and µ-dependent
in the Chern case, owing to boundary currents. However, the boundary currents
are “topologically protected” (i.e. are a bulk effect): therefore in both cases the
M value is a unique function of the bulk electron distribution and of µ.
The macroscopic magnetization at fixed chemical potential is M =
−(1/A) ∂G/∂B, where A is the system area and G = U − µN is the grand
thermodynamic potential at T = 0:
1 ∂U µ ∂N
M =− + = M1 + M2 . (8.2)
A ∂B A ∂B
We address the second term first: defining the areal density n = N/A, we get
M2 = µ ∂n/∂B; in the thermodynamic limit we make contact with Streda’s
formula [84], Eq. (4.52):
∂n eC1 C1
= = , (8.3)
∂B 2π~c φ0
where φ0 = hc/e is the flux quantum.
In order to address the first term M1 in Eq. (8.2) we start considering a 2d
macroscopic flake of finite size: its energy is
X
U= εj . (8.4)
ǫj <µ
In the normal case (C1 = 0) the flake as a whole remains gapped in the large
system limit, while in the Chern case (C1 6= 0) the bulk is gapped but the
spectrum of the flake becomes gapless in the large system limit. Nonetheless,
for a given (finite) flake the number of terms in Eq. (8.4) is B-independent for
B → 0. We may therefore invoke the Hellmann-Feynman theorem to get
∂U X ∂H e X
− =− hϕj | |ϕj i = − hϕj | r × v |ϕj i; (8.5)
∂B ǫ <µ
∂B 2c ǫ <µ
j j
92
not surprisingly, M1 coincides then with the 2d analogue of Eq. (8.1). Using
v = i[H, r]/~ we have
ie X
M1 = − hϕn | r × [H, r] |ϕn i (8.6)
2~cA ǫ <µ
j
ie X
= − ( hϕn | xHy |ϕn i − hϕn | yHx |ϕn i ).
2~cA ǫ <µ
j
Eq. (8.6) is a trace; since M1 is real,

e
M1 = Im iM1 = Im Tr {P xHyP }, (8.7)
~cA
where P is the ground-state projector, Eq. (6.30) (for single occupancy).
The trace in Eq. (8.7) has the same drawback as Eqs. (8.1) and (8.5): it is
dominated by boundary contributions. The key transformation comes next. We
write H = P HP + QHQ, where Q is the complementary projector, Eq. (6.32).
Then it is rather straightforward to transform Eq. (8.7) into
e
M1 = Im Tr {P xQHQyP − QxP HP yQ}. (8.8)
~cA
A different derivation of the same expression is due to Souza and Vanderbilt
[183].
8.2.1 The operator P r Q

We have already observed in Sec. 6.4.1 that the nasty position operator r is
ill defined within PBCs, while instead it is well defined when “sandwiched”
between a P and a Q. Here we give more details.
First we show explicitly that P rQ commutes with a lattice translation R for
a crystalline system.
Z
hr + R| P rQ |r′ + Ri = dr” P (r + R, r”) r” Q(r”, r′ + R)
Z
= dr” P (r + R, r” + R) (r” + R) Q(r” + R, r′ + R)
Z
= hr| P rQ |r′ i + R dr” P (r, r”)Q(r”, r′ ), (8.9)
and the last term obviously vanishes. Clearly, a similar property does not hold
for P rP ; instead
hr + R| P rP |r′ + Ri = hr| P rP |r′ i + R P (r, r′ ). (8.10)
Another useful property is that in a crystalline insulator the operator
hr| P rQ |r′ i decays quasi-exponentially for large values of |r − r′ |:
Z
hr| P rQ |r′ i = dr” P (r, r”) r” Q(r”, r′ )
Z
= r′ P (r, r′ ) − dr” P (r, r”) r” P (r”, r′ )
Z
= dr” P (r, r”) (r′ − r”) P (r”, r′ ). (8.11)
93
The integral in Eq. (8.11) can be regarded as the overlap in the variable r′′ of
two functions: the former is P (r, r”), centered at r and exponentially vanishing
away from it; the latter is (r′ − r”) P (r”, r′ ), centered at r′ and vanishing expo-
nentially (times a linear function) away from it. Such an overlap decays quasi-
exponentially (i.e. like an exponential times a polynomial) for |r − r′ | → ∞.
With respect to the above two properties, P rQ behaves analogously to P .
Like for P (r, r′ ), in order to get the operator hr| P rQ |r′ i in the bulk of a sample
one can equivalently perform the thermodynamic limit within either OBCs or
PBCs. For a crystalline system P rQ—like P itself—can be equivalently ex-
pressed in terms of either localized orbitals or Bloch orbitals; we provide here
the explicit Bloch-orbital expression.
The P and Q projectors are, for a 2d crystal with Nc (singly) occupied bands
Nc Z
Acell X
P = dk |ψnk ihψnk |,
(2π)2 n=1 BZ
∞
Acell X
Z
Q = dk′ |ψn′ k′ ihψn′ k′ |. (8.12)
(2π)2 ′
n =Nc +1
Then, by some manipulations which are standard in linear-response theory [124],

we have
hψn′ k | r |ψnk i = ihun′ k |∇k unk i n 6= n′

(2π)2
hψn′ k′ | r |ψnk i = i δ(k − k′ ) hun′ k |∇k unk i. (8.13)
Acell
We stress once more that, while the position operator r is ill-defined within
PBCs [96], its off-diagonal elements over the Hamiltonian eigenstates are well
defined; more accurately, Eq. (8.13) should be interpreted as a definition of such
elements. Therefore
Nc ∞
iAcell X X Z
QrP = dk |ψn′ k ihun′ k |∇k unk ihψnk |
(2π)2 n=1 BZ
n′ =Nc +1
Nc ∞
iAcell X X Z
P rQ = − dk |ψnk ih∇k unk |un′ k ihψn′ k |. (8.14)
(2π)2 n=1 BZ
n′ =Nc +1
8.2.2 Magnetization in k space

Let us start from a finite flake, as addressed above, whose magnetization is
given in Eq. (8.8). If the flake is crystalline, cut from the bulk, and if its
linear dimensions are much larger than the quasi-exponential decay length of
hr| P rQ |r′ i, then the genuine bulk value of M1 is retrieved from
e
M1 = Im Trcell {P xQHQyP − QxP HP yQ}, (8.15)
~cAcell
where the trace is taken over the central cell in the crystalline flake. At this
point we only need the thermodynamic limit of P rQ, which is conveniently
94
expressed in terms of Bloch orbitals, as shown in Sec. 8.2.1:
Nc ∞
Acell X X Z
P xQHQyP = dk ×
(2π)2 n=1 BZ
n′ =Nc +1
× |ψnk ih∂1 unk |un′ k i εn′ k hun′ k |∂2 unk ihψnk |
Nc ∞
Acell X X Z
QxP HP yQ = dk ×
(2π)2 n=1 ′ BZ
n =Nc +1
× |ψn′ k ihun′ k |∂1 unk i εnk h∂2 unk |un′ k ihψn′ k | (8.16)
Nc
X ∞
X Z
Im Trcell {P xQHQyP − QxP HP yQ} = dk ×
n=1 n′ =Nc +1 BZ
× [ εn′ k h∂1 unk |un′ k ihun′ k |∂2 unk i − εnk h∂2 unk |un′ k ihun′ k |∂1 unk i ].
At this point we may replace the sum over the empty states with the sum over
all states, since the added contribution is real. Therefore
e
Z
M1 = Im dk h∂1 unk | (Hk + εnk ) |∂2 unk i, (8.17)
~c(2π)2 BZ
where Hk is the effective Hamiltonian acting on the u’s, i.e. Hk = e−ik·r Heik·r .
In order to get the total magnetization we still have to add the second term
M2 in Eq. (8.2). Using Streda’s formula, Eq. (8.3), and the k-space expression
for the Chern number, Eq. (7.2), we get the full k-space expression for orbital
magnetization
e
Z
M = M1 + M2 = Im dk h∂1 unk | (Hk + εnk − 2µ) |∂2 unk i. (8.18)
~c(2π)2 BZ
This is the ultimate formula for orbital magnetization at zero temperature. The
finite temperature formula is in Ref. [182], and is also reviewed in Refs. [8, 18].
The proof of Eq. (8.18) provided here is original and yet (2013) unpublished.
The formula applies to normal insulators, to Chern insulators, and—without
proof here—to metals. We conjecture that the present proof can be extended
to the metallic case as well, although the ground state projector P (r, r′ ) is
only polynomial in |r − r′ | and not exponential. Previous proofs were either
semiclassical [83] or limited to normal insulators [174, 175, 169]. In the Chern
insulating case Eq. (8.18) was heuristically guessed and numerically validated
in Ref. [175].
8.2.3 Magnetization in r space

The basic expression for the magnetization of a finite flake, Eq. (8.8), has been
used so far to derive the previously known k space expression, Eq. (8.18); it is
remarkable that the same expression can be exploited directly in r space [176].
We start discussing a normal (zero Chern number) insulator first: in this
case the term M1 , Eqs. (8.8) and (8.15), is the only contribution to M . In order
to illustrate how the local approach works, we choose Haldanium samples with
∆/t2 = 3.67, φ = 0.1π; with reference to the phase diagram of Fig. 7.4, the
95
Figure 8.2: Convergence of M with
the flake size, normal insulating
phase. Filled circles: total mag-
netic moment divided by the flake
area, Eq. (8.7). Open circles: trace
over the central cell, Eq. (8.15). All
flakes have the same aspect ratio;
the abscissae indicate the length L
of one rectangle side in lattice pa-
rameter units. After Ref. [176].
material is clearly in the nontopological phase. Its macroscopic magnetization

M = M1 , Eq. (8.8), has been computed for rectangular Haldanium flakes of
various sizes and fixed aspect ratio within OBCs [176]; the trace over the cell in
Eq. (8.15) is just the average over the two central sites.
The result, shown in Fig. 8.2 is remarkable: when comparing two alterna-
tive definitions of macroscopic magnetization, namely the elementary one (total
dipole over sample area) vs. our Eq. (8.15), the latter converges much faster to
the bulk limit. In fact, the “nearsightedness” (i.e. exponential decay [156]) of
the density matrix implies an exponential convergence of Eq. (8.15), while the
elementary definition only yields a 1/L convergence.
Next we notice that the local approach in r space is much more general
than the k space one, since it can be applied to cases where the concept of k
vector does not make any sense, such as noncrystalline and/or inhomogeneous
materials (e.g. heterojunctions). Similarly to what we did in Sec. 7.5, we define
the microscopic function
e
M1 (r) = Im ( hr| P xQHQyP |ri − hr| QxP HP yQ |ri ) . (8.19)
~c
In the crystalline case, our previous Eq. (8.15) is just the cell average of M(r);
this can be clearly generalized by replacing the cell average with the macroscopic
average hM1 imacro , defined as in electrostatics (see e.g. Jackson [89]).
Next, we address Chern insulators. Therein the spectrum of a finite sample
within OBCs becomes gapless in the large sample limit; when µ is in the bulk
gap, the bulk is insulating but M depends on µ, owing to boundary currents.
This is quite clear from Eqs. (8.2) and (8.3); since, as shown in Sec. 7.5, even
the Chern number admits a local description, we may write both contributions
M1 and M2 to magnetization as
µ
M = hM1 imacro + h C imacro , (8.20)
φ0
where h C imacro is the macroscopic average of our topological marker, Eq. (7.29).
An equivalent definition is M = hMimacro , where
e
M(r) = Im ( hr| P xQHQyP |ri − hr| QxP ( H − 2µ )P yQ |ri ) ; (8.21)
~c
this is clearly reminiscent of the k space expression, Eq. (8.18).
We stress that the macroscopic average may vary over a macroscopic scale;
at a given point r inside the sample it is uniquely defined by the behavior of
96
M(r) in a microscopic neighborhood of r, and is insensitive to the conditions of
the sample boundary.
At the beginning of this chapter we have investigated the concept of “dipolar
density”, showing that it is well defined for spin magnetization but ill defined
for orbital magnetization. Our local function M(r) plays indeed the role of a
magnetic dipolar density; we stress, however, that only its macroscopic average
bears a physical meaning.
97
98
Bibliography
[1] R. Resta, Rev. Mod. Phys. 66, 899 (1994).

[2] R. Resta, in: Quantum-Mechanical Ab-initio Calculation of the Properties
of Crystalline Materials, Lecture Notes in Chemistry, Vol. 67, edited by
C. Pisani (Springer, Berlin, 1996), p. 273.
[3] R. Resta, J. Phys.: Condens. Matter 12, R107 (2000).
[4] R. Resta, J. Phys.: Condens. Matter 14, R625 (2002).
[5] D. Vanderbilt and R. Resta, in: Conceptual foundations of materials: A
standard model for ground- and excited-state properties, S.G. Louie and
M.L. Cohen, eds. (Elsevier, 2006), p. 139.
[6] R. Resta and D. Vanderbilt, in: Physics of Ferroelectrics: a Modern Per-

spective, Topics in Applied Physics Vol. 105, Ch. H. Ahn, K. M. Rabe,
and J.-M. Triscone, eds. (Springer-Verlag, 2007), p. 31.
[7] R. Resta, J Phys.: Conference Series 117, 012024 (2008).
[8] R. Resta, J. Phys.: Condens. Matter 22 123201 (2010).
[9] R. Resta, Eur. Phys. J. B 79, 121 (2011).
[10] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010).
[11] J. E. Moore, Nature 464, 194 (2010).
[12] N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Rev.
Mod. Phys. 82, 1539 (2010).
[13] D. Xiao, M.-C. Chang, and Q. Niu, Rev. Mod. Phys. 82, 1959 (2010).
[14] E. Prodan, J. Phys. A 44, 113001 (2011).
[15] X.-L. Qi and S.-C. Zhang, Rev. Mod Phys. 83, 1057 (2011).
[16] M. Z. Hasan and J. E. Moore, Annu. Rev. Condens. Matter Phys. 2, 55
(2011).
[17] J. Maciejko, T. L. Hughes, and S.-C. Zhang, Annu. Rev. Condens. Matter
Phys. 2, 31 (2011).
[18] T. Thonhauser, Int. J. Mod. Phys. B 25, 1429 (2011).
99
[19] N. Marzari, A. A. Mostofi, J. R. Yates, I. Souza, and D. Vanderbilt, Rev.
Mod. Phys. 84, 1419 (2012).
[20] N. A. Spaldin, J. Solid State Chem. 195, 2 (2012).
[21] R. Resta, Berry’s Phase and Geometric Quantum Distance: Macroscopic
Polarization and Electron Localization, Lecture Notes for the “Troisième
Cycle de la Physique en Suisse Romande” (Lausanne, 2000).
http://www-dft.ts.infn.it/~resta/publ/notes2000.ps.
[22] M. V. Berry, Nature Phys. 6, 148 (2010).
[23] Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959); reprinted in
Geometric Phases in Physics, edited by A. Shapere and F. Wilczek (World
Scientific, Singapore, 1989), p.104.
[24] R. P. Feynman, R. B. Leighton, and M. Sands, The Feynman Lectures in
Physics, Vol. 2 (Addison Wesley, Reading, 1964), Sect. 15-4.
[25] R. G. Chambers, Phys. Rev. Lett. 5, 3 (1960).
[26] M. Peshkin and A. Tonomura, The Aharonov–Bohm Effect (Springer,
Berlin, 1989).
[27] http://en.wikipedia.org/wiki/SQUID.
[28] P. Bocchieri and A. Loinger, Nuovo Cimento 47, 475 (1978).
[29] http://www.phy.bris.ac.uk/people/berry mv/index.html.
[30] M. V. Berry, Proc. Roy. Soc. Lond. A 392, 45 (1984).
[31] H. C. Longuet-Higgins, U. Öpik, M. H. L. Pryce, and R. A. Sack, Proc.
Roy. Soc. A 244, 1 (1958).
[32] G. Herzberg and H. C. Longuet-Higgins, Discuss. Faraday Soc. 35, 77
(1963).
[33] C. A. Mead and D. G. Truhlar, J. Chem. Phys. 70, 2284 (1979).
[34] C. A. Mead, Chemical Physics 49, 23 (1980).
[35] V. Heine, Phys. Rev.145,593 (1966).
[36] J. A. Appelbaum and D. R. Hamann, Phys. Rev. B 10, 4973 (1974).
[37] L. Kleinman, Phys. Rev. B 11, 858 (1975).
[38] F. Claro, Phys. Rev. B 17, 699 (1977).
[39] J. A. Appelbaum, G. A. Baraff, and D. R. Hamann, Phys. Rev. B 14,
1623 (1976).
[40] K. von Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494
(1980).
[41] T. Ando, J. Phys. Soc. Jpn. 37, 622 (1974).
100
[42] R. B. Laughlin, Phys. Rev. B 23, 5632 (1981).
[43] R. E. Prange, S. M. Girvin, M. E. Cage, and K. von Klitzing, The Quan-
tum Hall Effect, Second Edition (Springer, New York, 1990).
[44] D. Yoshioka, The Quantum Hall Effect (Springer, Berlin, 2002).
[45] D. Bures, Trans. Am. Math. Soc. 135, 199 (1969).
[46] Geometric Phases in Physics, edited by A. Shapere and F. Wilczek (World
Scientific, Singapore, 1989).
[47] D. J. Thouless, Topological Quantum Numbers in Nonrelativistic Physics
(World Scientific, Singapore, 1998).
[48] A. Bohm, A. Mostafazadeh, H. Koizumi, Q. Niu, and J. Zwanzinger, The
Geometric Phase in Quantum Systems (Springer, Berlin, 2003).
[49] J. J. Sakurai, Modern Quantum Mechanics (Addison-Wesley, Reading,
1994), p.140.
[50] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, Phys.
Rev. Lett. 49, 405 (1982).
[51] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005).
[52] D. N. Sheng, Z. Y. Weng, L. Sheng, and F. D. M. Haldane, Phys. Rev.

Lett. 97, 036808 (2006).
[53] S.-C. Zhang, Physics 1, 6 (2008).
[54] Y. L. Chen et al., Science 325, 178 (2009).
[55] J. E. Moore, Physics 2, 82 (2009).
[56] X.-L. Qi and S.-C. Zhang, Phys. Today 63(1), 38 (2010).
[57] J. P. Provost and G. Vallee, Commun. Math Phys. 76, 289 (1980).
[58] F. Wilczek and A. Zee, Phys. Rev. Lett. 52, 2111 (1984); reprinted
in:Geometric Phases in Physics, edited by A. Shapere and F. Wilczek
(World Scientific, Singapore, 1989), p. 141.
[59] http://www.abinit.org/.
[60] http://www.crystal.unito.it/.
[61] http://cms.mpi.univie.ac.at/vasp/.
[62] http://www.quantum-espresso.org.
[63] M.-C. Chang and Q. Niu, Phys. Rev. B 53, 7010 (1996).
[64] G. Sundaram and Q. Niu, Phys. Rev. B 59, 14915 (1999).
[65] C. A. Mead, Rev. Mod. Phys. 64, 51 (1992).
[66] M. Kohmoto, Ann. Phys. 160, 343 (1985).
101
[67] D. N. Zubarev, Non-Equilibrium Statistical Mechanics (Consultants Bu-
reau, New York, 1974).
[68] P. Schmelcher, L. S. Cederbaum, and H.–D. Meyer, Phys. Rev. A 38, 6066
(1988).
[69] L. Yin and C. A. Mead, J. Chem. Phys. 100, 8125 (1994).
[70] D. Ceresoli, R. Marchetti and E. Tosatti, Phys. Rev. B 75, 161101(R)
(2007).
[71] R. Karplus and J. M. Luttinger, Phys. Rev. 95, 1154 (1954).
[72] T. Jungwirth, Q. Niu, and A. H MacDonald, Phys. Rev. Lett. 88, 207208
(2002).
[73] M. Onoda and N. Nagaosa, J. Phys. Soc. Jpn. 71, 19 (2002).
[74] Y. Yao, L. Kleinman, A. H. MacDonald, J. Sinova, T. Jungwirth, D.-S.
Wang, E. Wang, and Q. Niu, Phys. Rev. Lett. 92, 037204 (2004).
[75] X. Wang, J. R. Yates, I. Souza, and D. Vanderbilt, Phys. Rev. B 74,
195118 (2006).
[76] F. D. M. Haldane, Phys. Rev. Lett. 93, 206602 (2004).
[77] X. Wang, D. Vanderbilt, J. R. Yates, and I. Souza, Phys. Rev. B 76,
195109 (2007).
[78] N. W. Ashcroft and N. D. Mermin, Solid State Physics (Saunders,
Philadelphia, 1976).
[79] J. C. Slater, Phys. Rev. 76, 1592 (1959).
[80] J. M. Luttinger, Phys. Rev. 84, 814 (1951).
[81] J. Zak, Phys. Rev. 168, 686 (1968).
[82] G. Panati, H. Spohn, and S. Teufel, Commun. Math Phys. 242, 547
(2003).
[83] D. Xiao, J. Shi, and Q. Niu, Phys. Rev. Lett. 95, 137204 (2005).
[84] P. Streda, J. Phys. C 15, L717 (1982).
[85] D. J. Thouless, Phys. Rev. B 27, 6083 (1983).
[86] J. B. Pendry and C. H. Hodges, J. Phys. C 17, 1269 (1984).
[87] J. Meister and W. H. E. Schwarz, J. Phys. Chem. 98, 8245 (1994).
[88] L. D. Landau and E. M. Lifshitz, Electrodynamics of Continuous Media
(Pergamon Press, Oxford, 1984).
[89] J. D. Jackson, Classical Electrodynamics (Wiley, New York, 1975).
[90] C. Kittel, Introduction to Solid State Physics, 7th. edition (Wiley, New
York, 1996).
102
[91] R. M. Martin, Phys. Rev. B 9, 1998 (1974).
[92] M. Posternak, A. Baldereschi, A. Catellani and R. Resta, Phys. Rev. Lett.
64, 1777 (1990).
[93] R. Resta, Ferroelectrics 136, 51 (1992).
[94] R. D. King-Smith and D. Vanderbilt, Phys. Rev. B 47, 1651 (1993).
[95] G. Ortı́z and R. M. Martin, Phys. Rev. B 43, 14202 (1994).
[96] R. Resta, Phys. Rev. Lett. 80, 1800 (1998).
[97] http://www.uam.es/departamentos/ciencias/fismateriac/siesta/.
[98] http://www.cpmd.org/.
[99] R. Resta, Modelling Simul. Mater. Sci. Eng. 11, R69 (2003).
[100] W. H. Duan and Z. R. Liu, Curr. Opin. Solid State Mater. Sci. 10, 40
(2006).
[101] M. Rabe, and J.-M. Triscone, eds., Physics of Ferroelectrics: a Modern
Perspective, Topics in Applied Physics Vol. 105, Ch. H. Ahn, K. (Springer-
Verlag, 2007).
[102] M.P. Marder, Condensed Matter Physics (Wiley, New York, 2000).
[103] G. Grosso and G. Pastori-Parravicini, Solid State Physics (Academic, San
Diego, 2000).
[104] I. Souza, J. Ìñı̀guez, and D. Vanderbilt, Phys. Rev. Lett. 89, 117602
(2002).
[105] P. Umari and A. Pasquarello, Phys. Rev. Lett. 89, 157602 (2002).
[106] R. Resta, M. Posternak, and A. Baldereschi, Phys. Rev. Lett. 70, 1010
(1993).
[107] G. H. Wannier, Phys. Rev. 52, 191 (1937).
[108] N. Marzari and D. Vanderbilt, Phys. Rev. B 56, 12847 (1997).
[109] A. A. Mostofi, Y.-S. Lee, I. Souza, D. Vanderbilt, and N. Marzari, Comput.
Phys. Commun. 178, 685 (2008).
[110] J. R. Yates, C. J. Pickard, and F. Mauri, Phys. Rev. B 76, 024401 (2007).
[111] R. Resta, J. Chem. Phys. 124, 104104 (2006).
[112] E. I. Blount, in Solid State Physics, edited by H. Ehrenreich, F. Seitz and
D. Turnbull, vol 13 (Academic, New York, 1962), p. 305.
[113] O. F. Mossotti, Memorie di Matematica e di Fisica della Società Ital-
iana delle Scienze Residente in Modena, 24, 49 (1850); R. Clausius, Die
Mechanische Behandlung der Electrica (Vieweg, Berlin, 1879).
[114] A. Pasquarello and R. Resta, Phys. Rev. B 68, 174302 (2003).
103
[115] I. Dabo, B. Kozinsky, N. E. Singh-Miller, and N. Marzari, Phys. Rev. B
77 , 115139 (2008); ibid. 84, 159910(E) (2011).
[116] P. Umari, A. Dal Corso, and R. Resta, in: Fundamental Physics of Fer-
roelectrics: 2001 Williamsburg Workshop, H. Krakauer, ed. (AIP, Wood-
bury, New York, 2001), p. 107.
[117] Q. Niu, Phys. Rev. 33, 5368 (1986).
[118] D. Vanderbilt and R. D. King-Smith, Phys. Rev. B 48, 4442 (1993).
[119] K. N. Kudin, R. Car, and R. Resta, J. Chem. Phys. 127, 194902 (2007).
[120] M. Stengel, Phys. Rev. B 84, 205432 (2011).
[121] P. L. Silvestrelli, M. Bernasconi, and M. Parrinello, Chem. Phys. Lett.
277, 478 (1997).
[122] A. Pasquarello and R. Car, Phys. Rev. Lett. 79, 1766 (1997).
[123] X. Gonze, Ph. Ghosez, and R. W. Godby, Phys. Rev. Lett. 74, 4035
(1995).
[124] S. Baroni, S. de Gironcoli, A. Dal Corso, and P. Giannozzi, Rev. Mod.
Phys. 73, 515 (2001).
[125] F. Bloch, Z. Phys. 52, 555 (1928).

[126] A. H. Wilson, Proc. Roy. Soc. A 133, 458, and 134, 277 (1931).
[127] N. F. Mott, Proc. Phys. Soc. (London) 62, 416 (1949).
[128] P. W. Anderson, Phys. Rev. 109, 1492 (1958).
[129] N. Mott, Metal-Insulator Transitions, 2nd ed. (Taylor & Francis, London,
1990).
[130] E. Abrahams (Ed.), 50 Years of Anderson Localization, (World Scientific,
Singapore, 2010).
[131] W. Kohn, Phys. Rev. 133, A171 (1964).
[132] W. Kohn, in Many–Body Physics, edited by C. DeWitt and R. Balian
(Gordon and Breach, New York, 1968), p. 351.
[133] Only about 80 citations (ISI) in the 25 years from 1964 to 1991.
[134] R. Resta and S. Sorella, Phys. Rev. Lett. 82, 370 (1999).
[135] I. Souza, T. Wilkens, and R. M. Martin, Phys. Rev. B 62, 1666 (2000).
[137] D. N. Zubarev, Soviet Phys. Ushpekhi 3, 320 (1960).
[138] R. McWeeny, Methods of Molecular Quantum Mechanics, Second Edition
(Academic, London, 1992).
104
[139] R. Kubo, M. Toda, and N. Hashitsume, Statistical Physics II, Nonequi-
librium Statistical Mechanics, Springer Series in Solid-State Sciences, Vol.
31, (Springer, Berlin, 1985).
[140] D. Forster, Hydrodynamic Fluctuations, Broken Symmetry, and Correla-
tion Functions (Benjamin, Reading, 1975).
[142] C. Sgiarovello, M. Peressi, and R. Resta, Phys. Rev. 64, 115202 (2001).
[143] M. Veithen, X. Gonze, and Ph. Ghosez, Phys. Rev. B 66, 235113 (2002).
[144] N. D. M. Hine and W. M. C. Foulkes, J. Phys: Condens. Matter 19,
506212 (2007).
[145] A. Monari, G. L. Bendazzoli, and S. Evangelisti, J. Chem. Phys. 129,
134104 (2008).
[146] C. Aebischer, D. Baeriswyl, and R. M. Noack , Phys. Rev. Lett. 86, 468
(2001).
[147] V. Vetere, A. Monari, G.L. Bendazzoli, S. Evangelisti, and B. Paulus, J.
Chem. Phys. 128, 214701 (2008).
[148] L. Stella, C. Attaccalite, S. Sorella, and A. Rubio, Phys. Rev. B 84, 245117
(2011).
[149] G. L. Bendazzoli, S. Evangelisti, A. Monari, and R. Resta, J. Chem. Phys.
133, 064703 (2010).
[150] E. Akkermans, J. Math. Phys. 38, 1781 (1997).
[151] W. Kohn, Phys. Rev. Lett. 2, 393 (1959).
[152] J. des Cloizeaux, Phys. Rev. 135, A685 (1964); ibid. 135, A697 (1964).
[153] S. Ismail-Beigi and T.A. Arias, Phys. Rev. Lett. 82, 2127 (1999).
[154] L. He and D. Vanderbilt, Phys. Rev. Lett. 86, 5341 (2001).
[155] G. F. Giuliani and G. Vignale, Quantum Theory of the Electron Liquid
(Cambridge University Press, Cambridge, 2005).
[156] W. Kohn, Phys. Rev. Lett. 76, 3168 (1996).
[157] S.F. Boys, Rev. Mod. Phys. 32, 296 (1960); J.M. Foster and S.F. Boys,
ibid. 300.
[158] J. G. Àngyàn, Int. J. Quantum Chem. 109, 2340 (2009).
[159] J. G. Àngyàn, Curr. Org. Chem. 15, 3609 (2011).
[160] R. Resta and S. Sorella, Phys. Rev. Lett. 87, 4738 (1995).
[161] T. Wilkens and R. M. Martin, Phys. Rev. B 63, 235108 (2001).
[162] B. Kramer and A. MacKinnon, Rep. Prog. Phys. 56, 1469 (1993).
105
[163] D. J. Thouless, Phys. Rep. 13, 93 (1974).
[164] A. Lagendijk, B. van Tiggelen, and D. S. Wiersma, Phys. Today 62(8),
24 (2009).
[165] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988).
[166] T. Thonhauser and D. Vanderbilt, Phys. Rev. B 74, 235111 (2006).
[167] D. J. Thouless, J. Phys. C 17, L325 (1984).
[168] C. Brouder, G. Panati, M. Calandra, Ch. Mourougane, and N. Marzari,
Phys. Rev. Lett. 98, 046402 (2007).
[169] D. Ceresoli and R. Resta, Phys. Rev. B 76, 012405 (2007).
[170] Q. Niu, D. J. Thouless, and Y. S. Wu, Phys. Rev. B 31, 3372 (1985).
[171] N. Hao et al., Phys. Rev. B 78, 075438 (2008).
[172] S. Coh and D. Vanderbilt, Phys. Rev. Lett. 102, 107603 (2009).
[173] R. Bianco and R. Resta, Phys. Rev. B 84, 241106(R) (2011).
[174] T. Thonhauser, D. Ceresoli, D. Vanderbilt, and R. Resta, Phys. Rev. Lett.
95, 137205 (2005).
[175] D. Ceresoli, T. Thonhauser, D. Vanderbilt, and R. Resta, Phys. Rev. B
74, 024408 (2006).
[176] R. Bianco and R. Resta, Phys. Rev. Lett. 110, 087202 (2013).
[177] A. A. Soluyanov and D. Vanderbilt, Phys. Rev, B 83, 035108 (2011).
[178] A. A. Soluyanov and D. Vanderbilt, Phys. Rev, B 85, 115415 (2012).

[179] A. J. P. Meyer and G.Asch, J. Appl. Phys. 32, S330 (1961).
[180] L. L. Hirst, Rev. Mod. Phys. 69, 607 (1997).
[181] D. Xiao, Y. Yao, Z. Fang, and Q. Niu, Phys. Rev. Lett. 97, 026603 (2006).
[182] J. Shi, G. Vignale, D. Xiao, and Q. Niu, Phys. Rev. Lett. 99, 197202
(2008).
[183] I. Souza and D. Vanderbilt, Phys. Rev. B 77, 054438 (2008).
[184] T. Thonhauser, D. Ceresoli, A.A. Mostofi, N. Marzari, R. Resta, and D.
Vanderbilt, J. Chem. Phys. 131, 101101 (2009).
[185] D. Ceresoli, U. Gerstmann, A. P. Seitsonen, and F. Mauri, Phys. Rev. B
81, 060409(R) (2010).
[186] R. Resta, Phys. Rev. B 84, 214428 (2011).
106

Resta Lecture Notes

Uploaded by

Copyright:

Available Formats

Resta Lecture Notes

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Resta Lecture Notes

Uploaded by

Copyright:

Available Formats

Geometry and Topology

in Electronic Structure Theory

Notes subject to ongoing editing

6 Quantum metric and the theory of the insulating state 61

7 Topological invariants and topological insulators 79

1.1 About the present Notes

1.1.2 Why writing these Notes?

logical sequence many results scattered in the literature leads me to scrutinize

1.2 What topology is about

1.2.1 Gauss-Bonnet theorem

The Gaussian curvature Ω is by definition the determinant of the Hessian at the

1.2.2 Euler characteristic

1.3 Electronic wavefunctions

Figure 1.3: Some surfaces and their genus

1.6 Gauge and flux

1.6.1 Classical mechanics

mr̈ = −∇V (r). (1.12)

1.6.2 Quantum mechanics, open boundary conditions

H(κ)|ψn (κ)i = ǫn (κ)|ψn (κ)i. (1.14)

The eigenvectors and eigenvalues of the Schrödinger equation depend on the

1.6.3 Quantum mechanics, periodic boundary conditions

1.6.4 Example: Free particle in 1d

The eigenfunctions within PBCs and the spectrum are

1.6.5 Flux and flux quantum

Figure 1.5: The electron motion is con-

2.1 The Aharonov-Bohm effect: A paradox?

Figure 2.1: The Aharonov-Bohm interference experiment (From Ref. [24])

2.2 Conical intersections in molecules

Figure 2.2: A homonuclear trimer in its equilateral configu-

Figure 2.4: The Born-Oppenheimer sur-

E± (ξ) ∝ |ξ|2 ± const |ξ|. (2.2)

2.3 Quantization of the surface charge

Setting dv/dt = 0 we get the steady-state solution:

If n is the carrier density, the current is j = −ne v:

in zero B field we retrieve the standard Drude (diagonal) conductivity:

Inversion of the conductivity tensor

Figure 2.5: Hall effect in a 2d system.

2.4.2 Landau levels

2.4.3 The experiment

Figure 2.7: A modern realization of the integer quantum Hall effect

2.4.4 Early theoretical interpretation

Figure 2.8: Geometry for Laughlin’s

Berryology is used as synonymous for geometry in nonrelativistic quantum me-

H(ξ)|Ψ(ξ)i = E(ξ)|Ψ(ξ)i, (3.1)

where the d-dimensional real parameter ξ is defined in a suitable domain of Rd :

3.1 Phases and distances

3.2 Berry phase

Figure 3.1: State vectors in the two-

hΨ0 (ξ)|Ψ0 (ξ+∆ξ)i

where A(ξ) is called the Berry connection:

A(ξ) = i hΨ0 (ξ)|∇ξ Ψ0 (ξ)i. (3.11)

Figure 3.3: A smooth closed curve C in

A(ξ) = −Im hΨ0 (ξ)|∇ξ Ψ0 (ξ)i. (3.12)

A number of manifestations of the Berry phase occurring in molecular and

3.3 Connection and curvature

where Ω is the Berry curvature, defined as

Ω(ξ) = ∇ξ × A(ξ) = −Im h∇ξ Ψ0 (ξ)| × |∇ξ Ψ0 (ξ)i

3.4 Chern number

Figure 3.4: A sphere cut at the equator